WO2004027652A1 - Programmable streaming data processor for database appliance having multiple processing unit groups - Google Patents

Programmable streaming data processor for database appliance having multiple processing unit groups Download PDF

Info

Publication number
WO2004027652A1
WO2004027652A1 PCT/US2003/030001 US0330001W WO2004027652A1 WO 2004027652 A1 WO2004027652 A1 WO 2004027652A1 US 0330001 W US0330001 W US 0330001W WO 2004027652 A1 WO2004027652 A1 WO 2004027652A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
psdp
jpu
records
host
Prior art date
Application number
PCT/US2003/030001
Other languages
French (fr)
Inventor
Foster D. Hinshaw
David L. Meyers
Barry M. Zane
Original Assignee
Netezza Corporation
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Netezza Corporation filed Critical Netezza Corporation
Priority to EP03759450.4A priority Critical patent/EP1546930B1/en
Priority to AU2003275181A priority patent/AU2003275181A1/en
Publication of WO2004027652A1 publication Critical patent/WO2004027652A1/en

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24Querying
    • G06F16/245Query processing
    • G06F16/2455Query execution
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/27Replication, distribution or synchronisation of data between databases or within a distributed database system; Distributed database system architectures therefor
    • G06F16/273Asynchronous replication or reconciliation
    • YGENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y10TECHNICAL SUBJECTS COVERED BY FORMER USPC
    • Y10STECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y10S707/00Data processing: database and file management or data structures
    • Y10S707/99941Database schema or data structure
    • Y10S707/99944Object-oriented database structure

Definitions

  • This invention relates to distributed data processing systems that use multiple processing unit groups, and in particular to programmable data streaming processor that performs initial processing before tuples are handled by a job processor.
  • I/O Input/Output
  • CPU Central Processing Unit
  • the demand for access to information is growing at least twice as fast the ability of a single CPU to process and deliver it.
  • data is stored on one or more mass storage devices, such as hard disk drives.
  • One or more computers are then programmed to read data from the disks and analyze it - the programs may include special database software written for this pu ⁇ ose.
  • the speed at which the data analysis can be performed is typically limited to the speed at which the entire set of data can be transferred into a computer's memory and then examined by the CPU(s).
  • the aggregate data transfer rate of the disks does not govern the speed at which the analysis can be performed.
  • Disks are inexpensive, and as such, data can be spread across a large number of disks arranged to be accessed in parallel.
  • the effective data transfer rate of a set of disks, collectively, can therefore be almost arbitrarily fast.
  • the bandwidth of an interface or communications network between the disks and the CPUs is also typically less than the aggregate data transfer rate of the disks. The bottleneck is thus in the communications network or in the CPUs, but not in the disks themselves.
  • SMP Symmetric Multiprocessing
  • ASMP Asymmetric Multiprocessing
  • MPP Massively Parallel Processing
  • SMP systems consist of several CPUs, each with their own memory cache. Resources such as memory and the I/O system are shared by and are equally accessible to each of the processors.
  • the processors in an SMP system thus constitute a pool of computation resources on which the operating system can schedule "threads" of executing code for execution.
  • the SMP approach Two weaknesses of the SMP approach impair its performance and scalability when processing very large amounts of data.
  • the first problem results from a limited ability to actually provide information to the processors.
  • the I/O subsystem and the memory bus are shared among all processors, yet they have a limited bandwidth.
  • a second problem with the SMP approach is cache coherence.
  • Within each processor is typically a cache memory for storing records so that they may be accessed faster.
  • the more that processors are added to an SMP system the more that time must be spent synchronizing all of the individual caches when changes are made to the database. In practice, it is rare for SMP machines to scale linearly beyond about 64 processors.
  • Asymmetric Multiprocessing (ASMP) systems assign specific tasks to specific processors, with a master processor controlling the system. This specialization has a number of benefits. Resources can be dedicated to specific tasks, avoiding the overhead of coordinating shared access. Scheduling is also easier in an ASMP system, where there are fewer choices about which processor to assign to a task. ASMP systems thus tend to be more scalable than SMP systems. One basic problem with asymmetry is that it can result in one processor being overloaded while others sit idle.
  • Massively Parallel Processing (MPP) systems consist of very large numbers of processors that are loosely coupled. Each processor has its own memory and devices and runs its own operating system. Communication between the processors of an MPP system is accomplished by sending messages over network connections. With no shared resources, MPP systems require much less synchronization than
  • U.S. Patent No. 6,507,834 issued to Kabra et al. uses a multi-processor architecture to process Structured Query Language (SQL) instructions in a publish/subscribe model such that new entries in a database are automatically processed as added.
  • SQL Structured Query Language
  • a first processor is used as a dispatcher to execute optimized queries, setup communication links between operators, and ensure that results are sent back to the application that originated the query.
  • the dispatcher merges results of parallel execution by other processors to produce a single set of output tuples that is then returned to a calling procedure.
  • the present invention is a data processing system having two or more groups of processors that have attributes that are optimized for their assigned functions.
  • a first processor group consists of one or more host computers, which are responsible for interfacing with applications and/or end users to obtain queries, for planning query execution, and for, optionally, processing certain parts of queries.
  • the hosts in the first group may be SMP type machines.
  • a second processor group consists of many streaming record-oriented processors called Job Processing Units (JPUs), preferably arranged as an MPP structure.
  • the JPUs typically carry out the bulk of the data processing required to implement the logic of a query. Functions of the host computers in the first group can be divided into a
  • Front End and an “Execution Engine”.
  • the Front End is responsible for parsing queries, generating query execution plans, optimizing parallelizing execution plans, controlling transactions, sending requests for processing to the Execution Engine and receiving results of such requests from the Execution Engine.
  • the Execution Engine is responsible for scheduling the execution of jobs and other operations to run on the JPUs or locally within the Execution Engine itself, (such as sorting, grouping, and relational joining).
  • Each of the JPUs in the second group typically include a general pu ⁇ ose microcomputer, local memory, one or more mass storage devices, and one or more network connections.
  • the JPUs preferably use a multi-tasking operating system that permits multiple tasks to run at a given instant in time, in a priority-based demand scheduling environment.
  • the JPUs are responsible for:
  • each JPU also has a special pu ⁇ ose programmable processor, referred to herein as a Programmable Streaming Data
  • PSDP acts as a storage controller, to serve as an interface between the CPU of a JPU and the mass storage device.
  • PSDP is a processor that is distinct from the more general pu ⁇ ose CPU in each JPU. It is also distinct from the CPU of the "host" in the first group.
  • FPGA field-programmable gate array
  • ASIC Application-Specific Integrated Circuit
  • ASSP Application Specific Standard Product
  • the PSDP is specially programmable to also inte ⁇ ret data in a specific format as it is read from or written to the associated disk(s). This enables PSDP to perform portions of jobs on data directly, as it is read off the disk, prior such data ever being forwarded to the JPU.
  • data can be filtered by the PSDP as records and fields of a database, so that only certain records, or certain portions of records, are actually forwarded to be written into the associated JPU's main memory.
  • the JPUs in the second group use the special pu ⁇ ose PSDP hardware to interface to a disk and filter data after it reads from a disk, but still prior to a more general pu ⁇ ose execution unit.
  • the system designer may now use industry standard disk controllers and standard hard disk drives. This allows the designer to effectively leverage the ever increasingly higher density of standard IDE and SCSI compatible storage media, as soon as they become available.
  • the PSDP avoids inefficiencies of other approaches that: allocate memory for unused information • waste time copying unused information into memory waste time stepping around unused information
  • the PSDP avoids these problems since database filtering operations are performed "on the fly" in a streaming fashion, as data is read as records stream out of the mass storage devices.
  • the PSRP can also be programmed perform operations such as Boolean comparisons of record field values against either literal values or other record field values, or values held in registers of the processing element, and reject records that fail these Boolean comparisons before they are stored in memory.
  • the PSDP element can thus additionally filter out the subset of fields that are irrelevant to a particular query.
  • the PSDP also can perform other operations on records as they are read from mass storage.
  • the PSDP can be programmed to decompress records entering memory and to compress records being sent out of memory. It can be instructed to decrypt records entering memory or to encrypt records being sent out of memory. It can convert lowercase fields to mixed or uppercase. It can, in fact, be programmed to perform myriad other such operations. Because these operations occur as each record streams into memory, the PSDP offloads such tasks from the JPUs main CPU, freeing it for other useful work.
  • the PSDP is programmed to perform simple Boolean operations, such as to compare field values of the record stream against values held in its local registers.
  • This allows a limited class of join operations to be performed of records before they are stored in memory. For example, if the values of the fields being joined are limited in range (such as when a set of consecutive integers is used to represent each of the 50 United States), the presence or absence of a particular field value can be encoded as a bit within a sequence of bits, whose position within the sequence corresponds to the integer field value.
  • One advantage of this is that it allows field-level filtering and more complex processing to proceed in parallel within the JPU, for additional performance benefit.
  • a more important advantage is that this configuration of processors is most effective at reducing the amount of data that must flow through the system.
  • the JPU's CPU is thus free to perform as much record processing as possible before it must return records over the network (for aggregation with the results of other JPUs) into a final reply to the SMP host. Because moving vast amounts of data requires much overhead, it is advantageous to add a dedicated processing element before each step in the data movement pathway, from input to final result.
  • the JPU/PSDP architecture in effect, separates streaming record processing from other query processing functions. Because the PSDP can be programmed to recognize record formats, it is capable of producing tuple sets as an output.
  • any operation may be arranged to take as input(s) the output(s) from any other operation.
  • a common set of algorithms may be used for all operations whether on the host(s) or JPUs.
  • the two group architecture also allows an application to be insulated from the details of the programming model of the JPU.
  • the application interacts only with the first group, and the first group translates the application's requests into requests against the JPU.
  • This approach has several advantages: changes are easily made to functionality because of the inherent modularity of the system; any step of a query may be optimally executed on either of the groups or a combination thereof; bugs in application code cannot cause data corruption, crashes, or affect the requests of other applications; so that queries written in existing standard languages using existing
  • APIs Application Programming Interfaces
  • Fig. 1 is a system level block diagram of an asymmetric record processing system according to the present invention.
  • Fig. 2 is a more detailed view of a Job Processing Unit (JPU).
  • Fig. 3 is a more detailed view of software components in a host.
  • JPU Job Processing Unit
  • Fig. 4 is a more detailed view of Job Processing Unit (JPU) software components.
  • JPU Job Processing Unit
  • Fig. 5 is a block diagram of a Programmable Streaming Data Processor (PSDP) component.
  • Fig. 6 is a more detailed view of portions of a PSDP.
  • PSDP Programmable Streaming Data Processor
  • Fig. 7 is a detailed diagram of a Data Engine component.
  • the present invention is a data processing system having at least two "groups" of processing units, in which the individual components of each group are individual network “nodes" within the system.
  • the present invention has to do with how the a first group of one or more host processors accepts and responds to queries for data, and transforms such queries into one or more jobs, a second group of nodes comprising one or more Job Processing Units (JPUs), wherein each JPU has a sfreaming data interface, for receiving data from a streaming data source, one or more general pu ⁇ ose CPUs, for responding to requests from the host computers in the first group, and one or more Programmable Streaming Data Processors (PSDPs), which perform primitive functions directly on data received from the sfreaming data interface.
  • JPUs Job Processing Units
  • PSDPs Programmable Streaming Data Processors
  • the first group 10 consists of one or more SMP "host" computers 12, each with its own memory, network interface, and local storage (not shown in Fig. 1).
  • Each host 12 runs its own operating system, and typically, but not necessarily, each host 12 uses the same type of operating system as the other hosts 12.
  • the hosts 12 typically accept queries that are requests for data stored on mass storage devices, such as hard disk drives 23.
  • the requests may originate from any number of business intelligence applications that may be residing on local processors 28 or client computers 36 or separately ruiming application software 30, that may originate through a computer network 33 or locally.
  • Queries are typically provided in a format such as Structured Query Language (SQL), Open DataBase Connectivity (ODBC), Java DataBase Connectivity (JDBC), or the like.
  • SQL Structured Query Language
  • ODBC Open DataBase Connectivity
  • JDBC Java DataBase Connectivity
  • the hosts 12 accept queries that can retrieve, modify, create and/or delete data stored on disk 23 and the schema for such data.
  • the hosts 12 also accept requests to start, cormnit, and rollback transactions against the data.
  • the hosts 12 also perform typical admimstrative functions such as reporting on the status of the system 10, start and shutdown operation, backing up the current state of the data, restoring previous states of the data, replicating the data, and performing maintenance operations.
  • a catalog management component 15 contains descriptions of the fields and layout of data.
  • Catalog management 15 also contains information about which users and applications have which permissions to operate in which ways on which types of records, datasets, and relations.
  • the various hosts 12 interact with catalog management 15 in order to process the requests they receive.
  • catalog management 15 is embedded within one of the hosts 12, with parts replicated to the other hosts 12 and second group 20 components. As will be understood shortly, the catalog manager provides information to permit the components of the second group 20 to perform filtering functions.
  • the hosts 12 are generally able to respond to requests without having to communicate among themselves. In very rare instances, inter-host 12 communication may occur to resolve a transaction sequencing issue.
  • the second group 20 consists of a plurality of Job Processing Units (JPUs) 22.
  • each JPU 22 consists of a network interface 25 for receiving requests and delivering replies, a general pu ⁇ ose Central Processing Unit (CPU) 26 such as a microprocessor 26, memory 27, and a Programmable Sfreaming Data Processor (PSDP) 28.
  • CPU Central Processing Unit
  • PSDP Programmable Sfreaming Data Processor
  • Each JPU 22 runs a multi-tasking schedule-based operating system.
  • Each JPU 22 also has an attached disk 23 and disk controller from which the JPU 22 may read streaming data.
  • the JPU 22 can receive streaming record data from alternate or additional sources such as other on-board processors or via other network interfaces in place of the disk drives 23.
  • Such streaming data might include stock quotes, satellite data, patient vital signs, and other kinds of "live-feed" information available via a network connection.
  • the JPU 22 accepts and responds to requests from host computers 12 in the first group 10 to process the sfreaming record-oriented data under its control. These requests are typically "jobs" of a larger query, and are expressed as sequences of primitive operations on an input stream. The primitive operations could be inte ⁇ reted, but in the preferred embodiment, they are packaged as compiled code that is ready for execution. An exemplary job-based query is described in more detail below.
  • a JPU 22 also accepts and responds to requests from hosts for other operations such as:
  • Each JPU 22 also accepts and responds to requests from the hosts 12 to: Perform mirroring or other replication operations Redistribute data from one JPU to another • Send data local to one JPU to another JPU to help process a query job
  • JPU(s) 22 typically use multi-tasking Operating System (OS) to allow receiving, processing, and reporting the results from multiple jobs in a job queue.
  • OS Operating System
  • the OS should also support overlapping job execution. To coordinate this, the OS typically is responsible for scheduling and prioritizing requests according to a number of factors that are determined in real time. These may include a job priority as assigned by the user and/or host 12, as well as a job's expected impact on the
  • JPU's 22 local resources includes the amount of memory, disk, network, and/or I/O queues needed to complete the job.
  • the JPU 22 can also contain software for performing concurrency control, transaction management, recovery and replication of data for which the JPU is responsible.
  • JPUs 22 in the second group 20 are not directly visible or accessible to the users of, or the applications that run on, for example, the clients 36 or business intelligence applications 30 that present queries to the system 10.
  • the JPUs are, instead, an embedded component that maintain significant autonomy and control over their own data.
  • a given record (or other data primitive) in the system 10 is thus normally directly accessible to, and processed by only one JPU 22. While JPUs may replicate their records to increase reliability or performance, they do not share responsibility for processing a given record with other JPUs 22 when carrying at a job as part of a query.
  • the storage manager 320 within each JPU 22 provides support for other functions such as error checking, creation and deletion of tables, the use of indices, record insert and delete, mass loading of existing user data among various JPUs, and the like.
  • the components and sub-components are designed to optimize performance thru extensive use of streaming operations coupled with tuple set operations.
  • most operations are designed to take tuple sets (records or groups of records) as their input and output streams; these operations try not to materialize data, but instead they stream the output to the next operation.
  • tuple sets records or groups of records
  • a storage layer can be designed as a tuple set manager where (from the view of other JPU processes) it stores and retrieves tuple sets. From the storage layer onward, data is normally handled in tuple sets, providing a consistent, well organized, and easily accessible format for internal operations. This is in contrast to other systems where the storage layer stores and retrieves undifferentiated blocks of data which are later converted to tuple sets by some other downstream process.
  • Another example of the streaming/tuple set architecture is the network layer, which sends and receives tuple sets instead of blocks of data.
  • Yet another example is a merge aggregation mode, where a sorted data stream is aggregated as requested, and whenever a new key index value is received, the aggregation from the previous key index value may be streamed to the next node.
  • a streaming/tuple set operation can be illustrated by tracking a typical dataflow during a load operation.
  • data is read into a host 12 over TCP/IP network connection 32, that data is parsed, error-checked, and transformed, and the distribution value calculated, all while the specific byte/field is in processor cache, and saved to the internal network output frame buffers as one step.
  • the result is that the input data is read/transformed in a sfreaming fashion and converted to network-ready tuple set packets at streaming speed with n nimal overhead.
  • the received data is read, converted into an approved storage format, and placed in memory buffers on a record-by-record basis.
  • a storage layer in the JPU double-checks that the data corresponds to the indicated table, and that the table "owns” the physical space on the disk 23, and then writes that data to the disk 23. Note that during this process, a given byte of data was "touched” only a few times, and that the data was manipulated in tuple sets thereby optimizing performance and reliability.
  • a second illustration of a sfreaming tuple set operation is a join/aggregate operation where three joins and one co-located aggregation are performed on JPUs
  • ODBC client 36 e.g., Business Objects
  • the disk 23 is scanned and data read off the disk through the associated PSDP, which filters records of interest and fields of interest within those records, and places the resulting tuples into a tuple set buffer in JPU memory.
  • that tuple set is passed through each of three JPU join nodes and the aggregate node in turn.
  • the previous aggregate value and associated key value tuple are transformed as necessary per the ODBC request, and placed in the JPU network packet output buffer associated with the requesting host 12.
  • a network packet output buffer in the JPU When a network packet output buffer in the JPU is filled, its contents are sent to the host 12, where it is immediately placed in the user-side network buffer and is immediately sent to the ODBC client 36.
  • Fig. 2 is a software component diagram for a host 12. A summary description of the functional blocks and their interactions now follows. This list is intended here to be an introduction to a more detailed description of how a query is processedinto a set of jobs that can then be carried out as. synchronously and autonomously by JPUs 22.
  • Postmaster 201 accepts requests from user applications via API 200
  • Cost-based optimizer with the addition of locale costs which optimizes for most efficient operation/highest level performance
  • PSDP Programmable Data Streaming Processing
  • PSDP 28 has returned its data to the JPU 22, and what will be done in the Host 12 after the JPU 22 has returned its data '
  • Logger/Replication Server 218 • Logs transaction plans, messages, failures, etc. to Netezza log in conventional fashion Implemented as a standard transaction logger/replication server
  • System Manager 220 ⁇ Defines and maintains JPU Configuration information, striping information
  • Mirror Master maintains mirrors info - what JPUs are being mi ⁇ ored where, maintains SPA data, maintains info on system spares Initiates failover processing when informed by Comm layer of a non- communicative JPU - directs mirror of failed JPU to take over as primary and begin copying to designated spare, directs primary of JPU mirrored on failed JPU to copy its data to that same designated spare, to reduce load on mirror of original failed JPU also directs mirror of the primary on that failed
  • Commimicates to callhome component to initiate replacement process Manages system expansion and allows for redistribution of data as appropriate or as requested by user during expansion
  • MOX/OLAP 240 Provides OLAP/MDX, ROLAP Engine on Host
  • Host Disk Manager 250 • Manages Host Disk (used for Catalog, Temp Tables, Transaction Log,
  • Executes remainder of Execution Plan that has to be done at Host 12 Provides intermediate and final sort-merge of JPU 22 sorted data as required Handles joins of data returned from JPUs 22 as required Communicates to JPUs through Comm Layer 260 to request partial result sets from JPU buffers when idle (e.g., to get and sort/process partial records that the JPU currently has instead of waiting for JPU 22 to fill a buffer and then send to Host 12 Host Transaction Manager 264
  • TTD Manager 268 • Provides unique fransaction identifiers (TIDs)
  • Fig. 3 is a diagram of the software components of a JPU 22.
  • Communications Layer 300 Provides internal communication among nodes Includes Job Listener 301 to await requests
  • Network Poster 302 to send data when buffer filled, job completed, or at Host request
  • JPU Lock Manager 308 • Controls concurrent access to data
  • Interfaces with EventTask 36 before a query is executed and for each result . set returned from a scan Provides support for arithmetic locking
  • Mirror Sender receives copies of record updates from Storage Manager 320 and fransmits these to the mirror for this JPU when an updating transaction commits •
  • Mirror Receiver receives record updates, buffers these in memory, and flushes out to disk through the Storage Manager when the Mirror Receiver buffer is full
  • Schedules jobs to run on the PSDP 28 communicates with JPU/PSDP Scheduler 324 to queue up PSDP requests to retrieve required data • Optimizes the queue to keep the PSDP/disk as busy as possible, with requests from multiple queries intermixed in the queue based on disk characteristics and location of data on the disk Takes into account the needs of any data loading for new tables being created and transformed to internal data format (i.e., to optimize the loading process) Supports heuristic-based scheduling, ensuring that jobs are scheduled on a priority basis, but also ensuring that all jobs do get serviced (e.g., raising a job in priority if it has not been run in a certain interval of time)
  • Executes the portion of the Execution Plan that could not be handled by the PSDP but that does not have to be handled at the Host level Handles sorts, joins, transformations, and aggregations that could not be done as data stream through the PSDP 28 • Maintains a memory buffer of result set records and returns these to Host through the Comm Layer when buffer filled, job completed, or at Host request
  • JPU Diags 332 Runs diagnostics on JPU as required/requested
  • JPU Boot/Init 334 Executes image burned into flash memory at boot time to bootstrap the JPU, run diagnostics, register the JPU with the primary Host server, and download new image from Host to run
  • the PSDP allows data to be processed during Direct Memory Access (DMA) disk read operations.
  • DMA Direct Memory Access
  • PSDP 28 functions fall into two general categories: disk driver logic interface 281 and data "filter” 282. Each of these functions is described in some detail below. It is sufficient here to note that the disk driver logic interface 281 accepts standard disk drive interface signaling, such as IDE (Integrated Device
  • the interface 281 becomes a network interface that is suitable to receive and/or transmit data over a communications network.
  • the disk driver logic 281 is usually implemented in an Integrated Circuit (IC) in a computer or communications device, in or part of an IC that contains other logic, such as other interface logic or the CPU 26 itself.
  • IC Integrated Circuit
  • the disk driver 281 can even be inside the disk 23 itself, making the disk a special-pmpose unit attachable only to JPUs or communications devices for which the interface is specific.
  • the PSDP 28 is however an Integrated Circuit
  • DMA Direct Memory Access
  • the PSDP 28 also provides programmable hardware directly in the disk read path, to and from the controller. This function of the PSDP hardware, called the "filter” unit 282, is programmed to understand the structure of the data the analysis software running on the JPU 22 wishes to read and analyze.
  • the PSDP 28 can be this programmed to operate on data as it is received from the disk 23, before it is stored into the JPU's memory, and in the process discard data that the JPU 22 would otherwise have to analyze.
  • data can be filtered by the PSDP 28 as records and fields of a database, so that only certain fields from certain records are actually forwarded to be written into the associated JPU's main memory.
  • the PSDP 28 can be programmed to recognize that a certain set of records in a database have a specified format, for example, a preamble or "header" of determined length and format, perhaps a field including the length of the record, followed by data including some number of fields of a certain type and length (e.g., 4-byte integers), followed by some number of fields of a different type and length (e.g., 12-byte character strings), followed by some number of fields of variable length, whose first few bytes specify the length of the field in some agreed-upon manner, and so forth.
  • a preamble or "header" of determined length and format perhaps a field including the length of the record, followed by data including some number of fields of a certain type and length (e.g., 4-byte integers), followed by some number of fields of a different type and length (e.g., 12-byte character strings), followed by some number of fields of variable length, whose first few bytes specify the length of the field in some agreed-upon manner, and so forth.
  • the filter unit 281 can then execute this program as it reads data from the disk 23, locate record and field boundaries, and even employ further appropriate Boolean logic or arithmetic methods to compare fields with one another or with literal value. This allows the filter unit 282 to determine precisely which fields of which records are worth ttansferring to memory. The remaining records are discarded, or tagged in a manner that signals the JPU 22 that a record need not be analyzed. Again, there will be more discussion of how this is done in detail below. In the preferred embodiment, there are two basic reasons for which the filter unit 282 can discard a record (or mark it as unworthy of attention). The first is an analysis of the contents of the fields as described above.
  • the filter unit 282 can, for example, be programmed for a store sales database to check a purchase date field against a range of numbers that correspond to dates in the month of July in the year 2000, another field for a number or string uniquely associated with a particular store in North Carolina, another field for a set of SKU (Stock-Keeping Unit) values belonging to various styles or manufacturers of blue raincoats, and in this fashion mark only certain records for further processing.
  • the filter unit 282 can further be programmed to know which fields contain the name and address of the customer who made the purchase, and return only these fields from the interesting records.
  • the filter unit 282 can perform them at the same rate as the data is supplied by the disk 23. Far less data ends up in the JPU's memory as a result leaving the JPU 22 free for more complex tasks such as sorting the resulting list of names and addresses by last name or by postal code.
  • a second example of how the filter unit 282 can be used is to discard or mark a record, as in record creation and deletion in a multi-user environment.
  • Databases are not static, and it is common for some users to be analyzing a database while others are updating it.
  • records can be tagged with transaction numbers that indicate when or by whom a record was created or marked obsolete.
  • a user querying a database may not wish to see records created by another user whose activity began subsequently, or whose activity began previously but is not yet complete; if so, he probably will want to see records marked obsolete by such a user. Or the user may wish to see only the results of transactions entered by certain users, or only the results of transactions not entered by certain users.
  • record headers can contain creation and deletion identifiers that the filter unit 282 can be programmed to compare with the current user's identifier to determine whether records should be "visible" to the current user. Once again, the filter unit can avoid transferring useless data to memory or relieve the JPU 22 of a time-consuming analysis task.
  • the filter 282 unit can use to filter out data that is unnecessary for a given query, thereby reducing traffic on the communications network and reducing the workload on the JPU 22.
  • the filter unit 282 can simply discard the data. This is not always practical, however. Imagine a very long record with many fields, or large fields, many of which are to be returned to the JPU 22. Further consider a situation where a record meets the criteria is arranged in such a way that the contents of the last field are relevant to the decision to transfer or discard the selected fields of the record. Practical implementations of the filter unit 282 may not be able to store ("buffer") the largest possible set of returnable fields in a very long record, since there will be a limit on local buffer size. In such a case, the filter unit must begin sending the selected fields to the JPU 22 before it can tell whether they actually should be sent.
  • the filter can tag the transferred data with a states bit that says "never mind", thus saving the JPU 22and the communications network a great deal of work.
  • the filter unit can append a length indication to every record fragment it does transfer to the JPU 22, so that the JPU 22 can find the boundaries between the record fragments the filter unit deposits in memory. This is a natural place for this status bit (or bits, if the JPU 22 must distinguish among multiple reasons) indicating the transfer of a useless record.
  • the filter unit 282 can create and return additional fields not present in the database, by performing calculations on the contents of the fields that are present. This can further relieve the JPU 22 of work.
  • An example of this is the calculation of a "hash" function on the values of specified fields from a record, some of whose fields are to be transferred to the JPU 22.
  • a hash function is a numerical key assigned to a collection of numeric or non-numeric field values that speeds up the process of searching through a list of records.
  • Other examples of useful information that can be computed by the filter unit 282 include running sums or averages of field values from one record to the next. All of these benefits accrue from the filter unit's
  • Another example is a transformation, such as an ASCII substitution.
  • ASCII substitution is to change the collation sequence of a given field. For example, if the LAST_NAME starts with the French 'c' (ASCII 135) then the French 'c' (ASCII 135) then the French 'c' (ASCII 135)
  • the PSDP has 2 groups of registers, each 256 bytes. If transformation of a given field is selected then the PSDP setup loads transformation fields into these registers before the data is streamed in. Each register in the transformation fields corresponds to an extended ASCII value and the register contains the value that each extended ASCII character is to be converted into. In the example above, register numberl35 contains the value 76.
  • each byte is individually transformed to its converted value.
  • Two registers are provided so that two types of transforms may be applied to different fields in a given stream, such as UPPER() and LOWER().
  • the transforms may be applied either (a) in the "filter” path, before evaluation and comparisons or (b) in the "project” path so that a given field is converted before being output from the PSDP. This is especially useful for correcting collation sequences in preparation for the CPU performing a sort. More details of such a substitution table are contained in the above referenced co-peanding U.S. patent application entitled “Field Oriented Pipeline Architecture for a Programmable Data Sfreaming Processor,” (Attorney Docket No. 3336.1008-002).
  • PSDP 28 consists of a finite state machine called the Data Engine 400, which implements filter logic and other control operations; a JPU interface 404; a disk interface (here the ATA interface 408); First- In-First-Out (FIFO) memories 406 and 407; and a DMA driver 402.
  • the Data Engine 400 which implements filter logic and other control operations
  • JPU interface 404 a finite state machine
  • disk interface here the ATA interface 408
  • FIFO First- In-First-Out
  • the PSDP 28 is in one sense an On-Line Analytic Processing (OLAP)- oriented disk drive interface. It contains logic that is capable of identifying records, filtering out the unwanted records, and selecting fields for return as the tuple sets.
  • OLAP On-Line Analytic Processing
  • the PSDP 28 supports both a Programmed I/O (PIO) Mode-2 for register access by - the JPU 22 and a UDMA (Ultra-Direct Memory Access) mode-4 for data transfers.
  • PIO Programmed I/O
  • UDMA Ultra-Direct Memory Access
  • Data that is filtered has been processed, perhaps by culling records via a comparison and/or transaction ID processing (as described below), but typically by reformatting the records into tuple format, during which uninteresting fields can be dropped and PSDP-generated fields can be added.
  • This process of culling records is called a "restrict" operation.
  • the process of formatting fields into tuples is called a
  • Disk FIFO 407 In filtering mode, disk blocks are pulled from a Disk FIFO 407, feeding them through the Block Header, Record Header, NULL Vector, Transaction ID, Field Parse, and Filter circuits in the Data Engine 400. Fields to be returned are pushed into the Memory FIFO 406. There is of course also a DMA write mode, in which data from the JPU 22 flows through the DMA driver 402 directly to the ATA interface 408.
  • the PSDP 28 shadows the read/write disk command in order to control its own DMA state machines. It does not shadow the disk address or sector count, nor does it have access to the memory addresses. For writes and raw reads, the PSDP 28 blindly moves data from one interface to the other until the JPU 22 disables the mode. The JPU 22 knows the quantity of data to be moved for these modes and uses the disk and DMA controller 402 interrupts to identify the end of transfer. For filtered reads, the quantity of data to be transferred to memory is generally unknown, and the JPU 22 identifies the end of transfer from the disk and filter interrupts. All of the record info — header and data — can be projected during a filtered read, but the block header info can only be returned by a raw read. DMA data integrity is protected across the disk interface by an IDE CRC check.
  • the Data Engine 400 includes filter logic 500, a data parser 502, header storage 504, transaction ID processing 510, error checking 506, and output tuple generator 508.
  • the data parser 502 is responsible for taking information from the disk 23 and formatting it into headers and fields so that the filter logic 500, header storage 504 and error checking 506 blocks can perform their respective tasks.
  • the tuple generator 508 takes the output of the filter and TID processing 510 blocks and formats the results in a "tuple", suitable for processing by the JPU 22 or host 12.
  • Raw user table data as read from the disk 23 is understood and inte ⁇ reted by the data parser 502.
  • user table data is stored on disk in 128 KB segments called "blocks". Each block begins with an 8- word header, followed by 0 or more records.
  • the format of the block header may be as follows:
  • the CRC-32 field is meant to be computed by software and written to the disk along with the rest of the block header. Its calculation was to include all data from the block number through the end of the last sector of the block, including any trailing O's. Its primary pu ⁇ ose was to detect data corruption resulting from hardware or software bugs, but it could have detected disk data-retention problems as well. It is unrelated to the UDMA-mode CRC-16 calculation required by the
  • the sector count is the number of sectors in the block, which must be from 1 to 256. Thus a 0 in this 1-byte field means 256.
  • the sector count occupies the most-significant byte of the last word of the block header.
  • the record count is the number of records in the block, which may be 0. Although the record count occupies the least-significant three bytes of the last word of the block header, only 13 bits are used.
  • a record as read from disk 23 into the Data Engine 400 is typically composed of a record header and one or more data fields, where the record header consists of three special fields, a length, and a null vector.
  • the special fields are the row number, created transaction JD, and deleted transaction ID. All of the record header entries are optional on a per-table (not per-record) basis. However, if the record has a null vector, it must also have a record length, but not vice versa. The allowed data types are described elsewhere below.
  • the row number (sometimes called row_num) is the unique number of the row or record in the user's table. It is distinct from the row address (sometimes called row_addr), which is the complete physical address of a row in node-table- block-record format. The row number is also distinct from the record number, which is the 0-based ordinal number of a record within a block. The record number is the final component of the row address.
  • the row address is computed by the PSDP.
  • the created XID contains the number, or ID, of the fransaction that created the record.
  • the deleted XID In the preferred embodiment, records are not actually deleted. Rather, they are marked as deleted so they can be restored if the transaction that performed the deleting is rolled back. (There are system management tools to reclaim the space.) A value of 0 indicates the record has not been deleted. A value of 1 indicates that the record was created by a transaction that was rolled back.
  • the record length is the length of the record in bytes, excluding the row number and the fransaction IDs, but including the record length, the record null vector, the data fields, and any pad bytes at the end of the record needed for proper alignment of the first item of the following record. Thus, it is the distance in bytes from the beginning of the record length field to the beginning of the next record. Note that although all records in a table must have the same makeup, record lengths may vary because of variable-length character fields.
  • the RecordLengthSize register defines record length sizes of 0, 1, 2, and 4 bytes, but only 0 and 2 are used.
  • the record null vector specifies which fields in the record are null, thereby indicating validity, not existence. For instance, a null varchar is not the same as an empty one.
  • the record null vector consists of an even number of bytes. If it exists, the record null vector has the same number of bits as the record has data fields, and computes the number of half-words in the null vector as (FieldCount + 15) » 4. This vector is an array of bytes. Bit 0 of the byte immediately following the record length corresponds to the 0 th data field; bit 7 of that byte corresponds to the 7 th data field; bit 0 of the last byte of the word that contains the record length corresponds to the 8 data field; and so on.
  • the row number, created XID, and deleted XID are all 8 byte fields and do not require pad bytes to align them. If there is a record length but no record null vector, two pad bytes are required following the record length. If the record null vector exists, it immediately follows the record length and naturally starts on a two-byte boundary, but two pad bytes may be required following the record null vector to properly align the first data field.
  • the physical order of data fields which often is not the same as the logical order, takes care of aligning non-character data fields; the physical order is N16, T12, N8, 18, F8, N4, 14, F4, D4, 12, D2, II, Cl, C2, ... C16, V2.
  • the fixed-length character fields are packed in as tightly as possible and are not aligned.
  • Variable- length character fields start with a 2-byte length; they are .i- word-aligned and may require a preceding pad byte. Up to three pad bytes may follow the record's last data field in order to align the next record. If so, they are counted in the length of the earlier record.
  • TLO processing as performed by the TID processing block 510, includes rollback are contained in our co-pending U.S. Patent Application (Attorney Docket No. 3336.1017-001) entitled “Controlling Visibility in Multi- Version Database Systems", by Foster D. Hinshaw et al. filed on August 22, 2003.
  • a detailed circuit diagram of the filter / comparision logic 500 is shown in
  • the filter logic 500 supports up to 32 comparison (and hash) instructions. Each operates on a single field, which can be a header field, the row address, or a data field. Multiple instructions can operate on fields. Each instruction can perform two comparisons, using either a Data-String Register (DSR) 701 or temp registers 702 for the second operand(s); the two temp registers 702-0, 702-1 are used to store an early record field for comparison to a later record field. There are two programmable, byte-wide substitution tables 703 that simplify character field comparisons by switching to all upper or lower case, for example. The instruction results are combined in the use/lose circuit 705.
  • DSR Data-String Register
  • every comparison can be programmed for every supported data type (which may include integer, floating point, fixed- and variable-length character types etc.), not all combinations of data types and comparison operations are useful. The equality and null comparisons are appropriate for all types.
  • the string length and ASCII value of the characters determine inequality, such that "ABC” is less than “ABCD”, and "A" is less than both "a” and "B”. If a field is null, every comparison except null will fail.
  • the bit- vector join and its inverse are for the integer data type. Begins, contains, and their inverses are for character types, both fixed- and variable- length.
  • a bit vector is created with ' 1 's in the positions corresponding to those positions where the join condition is true.
  • the PSDP is loaded with a 4096-bit vector and the specific field position of the streaming data which is to be matched to that bit vector.
  • the lower 12 bits of that field are mapped into the bit vector to determine if the corresponding bit is set to "1". If it is, then the join condition is "TRUE", otherwise it is "FALSE".
  • This particular join type is implemented by modifying the nested loop hash, and merge join and bit join types.
  • the smaller table is placed into memory with either a hash index or sorted with access methods of hash or binary. If the smaller table has a range smaller than 4096, then it is converted to a bit index and put into the PSDP as with the bit join above.
  • the smaller table is scanned to see if it contains (or does not contain) the target field from the streaming tuple.
  • a "use/lose” logic circuit consists of up to eight sum or product terms. Each term can accept an input from each of the 32 instructions. The term outputs can be individually inverted before they're combined by either a sum-of-products (SOP) or product-of-sums (POS) calculation. Normally the filter indicates which records are to be kept, but the output of the SOP/POS calculation can be inverted to indicate which to reject. Taken altogether, the use/lose options provide deMorgan's Law term minimization.
  • SOP sum-of-products
  • POS product-of-sums
  • the list contains up to 32 entries. The first is always the ID of the current transaction, that is the fransaction that created the current scan. The remainder define the transactions that are concurrent to the "current" transaction. There are five modes: off, normal, dirty, silent, and rollback. Normal and dirty set a status bit in the returned tuple, as described in the section below on tuple formats; silent and rollback affect the tuple's return, in conjunction with the filter results.
  • a project function encompasses the selection of record fields, the generation of new fields, and the tuple formation and return.
  • Tuples typically consist of a row number, some data fields, and a 2-byte length/status, but they can also include the created and/or deleted transaction IDs, the row address, up to 255 pad words, the 32 instructions results formed into a boolean word, the hash result, and a null vector.
  • the hash is used to organize similar tuples into groups for processing joins or grouping selects, and with the exception of the record null vector and length/status, all record-header and data fields can be used in its calculation.
  • Hash operations are defined on a per-field basis by the comparison instructions.
  • a "tuple” is used to describe the projected data as output by the tuple 500.
  • the tuple generator 508 uses principally the filter 500 output but can also use TID processing 510 and error checking 506 outputs (Fig. 5).
  • the term “tuple” is used here for the pu ⁇ ose of differentiating "raw" disk 23 and PSDP 28 output record formats.
  • a tuple contains fields projected from the source record and up to six "virtual" fields: row address, pad words (tuple scratch pad), the boolean results from each of the filter operations, a hash result, the tuple null vector, and the tuple length. All are optional on a per-table basis. The order of these fields is given in the following table.
  • the row address is a compressed version of the node, table, block, and record information.
  • RowAddress[63:32] is defined by the NodelDTablelD register, a 32-bit register that is programmed with a 32-bit, merged version of the node ID and the table ID as part of the filter setup.
  • RowAddress[31 : 13] is the 19-bit block number defined by the block header.
  • RowAddress[12:0] is the 13-bit record number calculated by the PSDP 28; it is 0-based within the current block.
  • the Boolean filter result contains the pass/fail result for each of the 32 filter instructions.
  • the hash result is the output of the hash circuit.
  • the tuple null vector contains the record null vector bits for each data field software requested. Note that record and tuple null vectors do not generally match up. The tuple null vector must consist of an even number of bytes and begin on a two-byte boundary. Software ignores any undefined bits as they may have been set by a previous scan. Once again, the existence of the null vector requires the existence of the length. Like the record null vector, the least-significant bit of byte 0 of the null vector refers to the 0 th field; the most-significant bit of byte 0 refers to the
  • the least-significant bit of byte 1 refers to the 8 th field, and so on, but the alignment and therefore the location of each of the bytes is different.
  • the tuple length is the total length of the tuple in bytes, including leading specials at the beginning of the tuple and any pad bytes at the end needed for proper alignment of the first item in the following tuple. Although all tuples returned by a scan must have the same makeup, tuples sizes may vary due to variable-length character fields.
  • the TupleLengthSize register defines tuple length sizes of 0, 1, 2, and 4 bytes. Because tuple fields are 4-byte aligned, tuple lengths are always multiples of four, and the least-significant two bits of the tuple length are available to indicate tuple status. Bit 0 is the overrun bit.
  • Bit 1 is the invalid bit.
  • set it means the record from which this tuple was constructed has transaction IDs that make it invalid (i.e., invisibly created or visibly deleted).
  • tuple field alignments are the same as record field alignments.
  • the length and nulls precede the data fields, and the record null vector is left-aligned against the preceding record length.
  • the tuple null vector is right-aligned against the tuple length, which ends the tuple.
  • the size of the tuple null vector and the requirement that it end in byte lane 1 together determine the location of its first byte: byte lane 0 or 2 (see the examples below). Aligning the tuple length in this manner makes it possible for software to locate the length while striding through the tuples backwards in memory.
  • CPU software leaves enough space in memory at the head of the first tuple for a tuple length and null vector. This space allows relocating the length and null vectors ahead of the corresponding data fields as it walks the tuples backwards, then reverse direction to process the tuples forward. Alignment can require as many as five pad bytes between the last byte of data and the tuple null vector or length and, if neither null vector nor length exists, as many as three pad bytes following the last data byte.
  • the database defines a SalesDetail data table, a Customer data table, and a Store data table as follows:
  • a sample query might be to "show me the total units and dollar amount of rain gear sold to females in North Carolina in 2000, by customer ID.” This can be translated into the SQL statement:
  • a basic execution plan can be created by the SQL Expression Evaluator 207, plan generator 204 and plan optimizer 205 of the host(s) 12.
  • the plan might specify for example, to perform joins and aggregations on the JPUs 22, with restriction functions being performed on the Programmable Streaming Data Processor (PSDP) 28.
  • PSDP Programmable Streaming Data Processor
  • the query is passed from the application (which may be running on, Business Intelligence Application 30 local application server 29 or client 36), the Plan Generator 204 then creates tentative execution plans. Plans not only specify the above job descriptions, but also may specify whether specific jobs can run on currently or must run in sequence on the JPUs.
  • the Plan Optimizer 205 selects one of the plans and optimizes that plan and passes it to the Plan Link 206.
  • the Plan Link 206 expands the plan as necessary, based on where parts of the plan will be executed, and then passes the expanded plan to the Host Dispatch 208.
  • the Host Dispatch 208 then sends individual Jobs within the plan to the respective locales (i.e., the JPUs 22) for execution.
  • jobs 1-6 are sent to the JPUs 22 for execution, with job 7 reserved for the host 12.
  • Job 1 scans the Customer table with the required restriction and projection, and materializes it.
  • Job 2 scans the Store table with the required restriction and projection, and since it is a small table, broadcasts the resulting tuple set to all JPUs 22, where the tuples from all JPUs 22 are then accumulated and saved in memory as TEMPStore. Jobs 1 and 2 are specified or determined to run concurrently if possible.
  • the Host Dispatch 208 may thus combine Jobs 3-6 into one streaming job because they can all be implemented in a sfreaming manner without materialization of intermediate sets.
  • This combined job scans the SalesDetail table, with its restrictions and projections.
  • each tuple is joined with TEMPStore and TEMPCustomer and aggregated.
  • the aggregation node as each new customer LD is received, the previous one and its sums are sent to the host, where Job 7 is then invoked in a sfreaming fashion, to return the aggregated tuples through the ODBC connection 38 back to the user.
  • Fig. 7 is a diagram illustrating how an exemplary job is processed by a set of JPUs 22, in the second group.
  • the example Job 3 replicated here again included instructions
  • the broadcast message is typically sent as a User Datagram Protocol (UDP) type message, but can also be sent in other ways, such as a Transmission Control Protocol (TCP) message or a unicast message.
  • UDP User Datagram Protocol
  • TCP Transmission Control Protocol
  • the JPU dispatch unit 304 Upon receipt of a job message at the job listener 301, the JPU dispatch unit 304 informs the fransaction manager 306 and storage manager 320 to then schedule Job 3 for execution. More details of job execution can be found in the related co- pending U.S. Patent application entitled "Programmable Data Sfreaming Architecture Having Autonomous and Asynchronous Job Processing Unit," (Attorney Docket No. 3336.1016-003) mentioned above.

Abstract

A data processing system having two or more groups of data processors that have attributes that are optimized for their assigned functions. The first group consists of one or more host computers (12) responsible for interfacing with applications and/or end users to obtain queries and for planning query execution. A second processor group consists of many streaming record-oriented processors called Job Processing Units (JPUs) (22), preferably arranged as an MPP structure. The JPUs typically carry out the bulk of the data processing required to implement the logic of a query. Each of the JPUs (22) typically include a general purpose microcomputer (26), local memory (27), one or more mass storage devices (23), and one or more network connections. Each JPU (22) also has a special purpose programmable processor, referred to herein as a Programmable Streaming Data Processor (PSDP) (28). The PSDP (28) serves as an interface between the CPU and a JPU and the mass storage device, to offload functions from the CPU of the JPU (22).

Description

PROGRAMMABLE STREAMING DATA PROCESSOR FOR DATABASE APPLIANCE HAVING MULTIPLE PROCESSING UNIT GROUPS
RELATED APPLICATIONS
This application claims the benefit of U.S. Provisional Application No.
60/412,057 entitled "Asymmetric Streaming Record Processing Computer System," filed on September 19, 2002, and U.S. Provisional Application No. 60/411,686 entitled "Intelligent Storage Device Controller," filed on September 18, 2002. The entire teachings of these provisional applications is hereby incoφorated by reference.
This application is also related to U.S. Patent Application entitled "Intelligent Storage Device Controller," (Attorney Docket No. 3336.1008-001); U.S. Patent Application entitled "Field Oriented Pipeline Architecture for a Programmable Data Streaming Processor," (Attorney Docket No. 3336.1008-002); U.S. Patent Application entitled "Asymmetric Streaming Record Data Processor
Method and Apparatus," (Attorney Docket No. 3336.1016-001); and U.S. Patent Application entitled "Programmable Data Streaming Architecture Having . Autonomous and Asynchronous Job Processing Unit," (Attorney Docket No. 3336.1016-003), all of which are being filed together on the same date as this application. The entire teachings of each of these co-pending patent applications is also hereby incoφorated by reference. This application and the above applications are also all assigned to Netezza Coφoration. BACKGROUND OF THE INVENTION
This invention relates to distributed data processing systems that use multiple processing unit groups, and in particular to programmable data streaming processor that performs initial processing before tuples are handled by a job processor. With continued development of low cost computing systems and proliferation of computer networks, the world continues to see an exponential growth in the amount and availability of information. Indeed, the Massachusetts- based Enteφrise Storage Group has observed a doubling of information every few months. Demand for easy and efficient access to this ever-growing amount of digital information is another certainty. For example, World Wide Web traffic increased 300% in 2001 according to Forrester Research. Included among the applications that continue to make the greatest demands are systems for processing:
financial transactions;
"click stream" data that encapsulates the behavior of visitors to web sites; data relating to the operational status of public utilities such as electric power networks, communications networks, transportation systems and the like;
scientific data supporting drug discovery and space exploration.
Greg Papadopolous, the Chief Technical Officer of Sun Microsystems, Inc., has observed that the demand for access to decision support databases, referred to as the Input/Output (I/O) demand growth, doubles every nine months. To put this in context, Moore's Law predicts that Central Processing Unit (CPU) power doubles only about every 18 months. In other words, the demand for access to information is growing at least twice as fast the ability of a single CPU to process and deliver it. In a typical general puφose data processing system, data is stored on one or more mass storage devices, such as hard disk drives. One or more computers are then programmed to read data from the disks and analyze it - the programs may include special database software written for this puφose. The problem with a general puφose system architecture, however, is that all the data must be retrieved from the disk and placed in a computer's memory, prior to actually being able to perform any operations on it. If any portion of the data retrieved is not actually needed, the time spent fetcMng it is wasted. Valuable time is thus lost in the process of retrieval and storage of unnecessary data.
The speed at which the data analysis can be performed is typically limited to the speed at which the entire set of data can be transferred into a computer's memory and then examined by the CPU(s). Usually, the aggregate data transfer rate of the disks does not govern the speed at which the analysis can be performed. Disks are inexpensive, and as such, data can be spread across a large number of disks arranged to be accessed in parallel. The effective data transfer rate of a set of disks, collectively, can therefore be almost arbitrarily fast. The bandwidth of an interface or communications network between the disks and the CPUs is also typically less than the aggregate data transfer rate of the disks. The bottleneck is thus in the communications network or in the CPUs, but not in the disks themselves.
It has been recognized for some time that achieving adequate performance and scalability in the face of vast and rapidly growing data thus requires some kind of system architecture that employs multiple CPUs. The three most prevalent classes of so-called multiprocessing systems today include: Symmetric Multiprocessing (SMP) Asymmetric Multiprocessing (ASMP) Massively Parallel Processing (MPP)
But even these approaches have weaknesses that limit their ability to efficiently process vast amounts of data. .
SMP systems consist of several CPUs, each with their own memory cache. Resources such as memory and the I/O system are shared by and are equally accessible to each of the processors. The processors in an SMP system thus constitute a pool of computation resources on which the operating system can schedule "threads" of executing code for execution.
Two weaknesses of the SMP approach impair its performance and scalability when processing very large amounts of data. The first problem results from a limited ability to actually provide information to the processors. With this architecture, the I/O subsystem and the memory bus are shared among all processors, yet they have a limited bandwidth. Thus, when the volume of data is too high, the speed of the processors is wasted waiting for data to arrive. A second problem with the SMP approach is cache coherence. Within each processor is typically a cache memory for storing records so that they may be accessed faster. However, the more that processors are added to an SMP system, the more that time must be spent synchronizing all of the individual caches when changes are made to the database. In practice, it is rare for SMP machines to scale linearly beyond about 64 processors.
Asymmetric Multiprocessing (ASMP) systems assign specific tasks to specific processors, with a master processor controlling the system. This specialization has a number of benefits. Resources can be dedicated to specific tasks, avoiding the overhead of coordinating shared access. Scheduling is also easier in an ASMP system, where there are fewer choices about which processor to assign to a task. ASMP systems thus tend to be more scalable than SMP systems. One basic problem with asymmetry is that it can result in one processor being overloaded while others sit idle.
Massively Parallel Processing (MPP) systems consist of very large numbers of processors that are loosely coupled. Each processor has its own memory and devices and runs its own operating system. Communication between the processors of an MPP system is accomplished by sending messages over network connections. With no shared resources, MPP systems require much less synchronization than
SMP and ASMP systems.
One weakness of the MPP model is that communication among processors occurs by passing messages over a network connection, which is a much slower technique than communication through shared memory. If frequent inter-processor communication is required, then the advantages of parallelism are negated by communication latency. Another problem with the MPP approach is that traditional programming models do not map cleanly onto message passing architectures. Using approaches such as Common Object Request Broker Architecture (CORBA), which are designed to handle message passing, are considered awkward by some designers. There have also been attempts over the years to use distributed processing approaches of various types. These began with proposals for "Database Machines" in the 1970s, for "Parallel Query Processing" in the 1980s, and for "Active Disks" and "Intelligent Disks" in the last five to ten years. These techniques typically place a programmable processor directly in a disk sub-assembly, or otherwise in a location that is tightly coupled to a specific disk drive. This approach pushes processing power towards the disks, and thus can be used to reduce the load on a host computer's CPU.
More recently, system architectures have been adopted for parallel execution of operations that originate as standard database language queries. For example, U.S. Patent No. 6,507,834 issued to Kabra et al. uses a multi-processor architecture to process Structured Query Language (SQL) instructions in a publish/subscribe model such that new entries in a database are automatically processed as added. As explained in the Abstract of that patent, a first processor is used as a dispatcher to execute optimized queries, setup communication links between operators, and ensure that results are sent back to the application that originated the query. The dispatcher merges results of parallel execution by other processors to produce a single set of output tuples that is then returned to a calling procedure.
SUMMARY OF THE INVENTION
Brief Description of a Preferred Embodiment
In a prefeπed embodiment, the present invention is a data processing system having two or more groups of processors that have attributes that are optimized for their assigned functions. A first processor group consists of one or more host computers, which are responsible for interfacing with applications and/or end users to obtain queries, for planning query execution, and for, optionally, processing certain parts of queries. The hosts in the first group may be SMP type machines. A second processor group consists of many streaming record-oriented processors called Job Processing Units (JPUs), preferably arranged as an MPP structure. The JPUs typically carry out the bulk of the data processing required to implement the logic of a query. Functions of the host computers in the first group can be divided into a
"Front End" and an "Execution Engine". The Front End is responsible for parsing queries, generating query execution plans, optimizing parallelizing execution plans, controlling transactions, sending requests for processing to the Execution Engine and receiving results of such requests from the Execution Engine. The Execution Engine is responsible for scheduling the execution of jobs and other operations to run on the JPUs or locally within the Execution Engine itself, (such as sorting, grouping, and relational joining).
Each of the JPUs in the second group typically include a general puφose microcomputer, local memory, one or more mass storage devices, and one or more network connections. The JPUs preferably use a multi-tasking operating system that permits multiple tasks to run at a given instant in time, in a priority-based demand scheduling environment. The JPUs are responsible for:
- receiving data processing requests from the hosts in the form of jobs, retrieving data items from disk or other data sources, and other tasks such as local transaction processing, concurrency control and replication; - communicating results back to the first Execution Engine; and
- occasionally communicating with other second group components. In a preferred embodiment, each JPU also has a special puφose programmable processor, referred to herein as a Programmable Streaming Data
Processor (PSDP). The PSDP acts as a storage controller, to serve as an interface between the CPU of a JPU and the mass storage device. The PSDP is a processor that is distinct from the more general puφose CPU in each JPU. It is also distinct from the CPU of the "host" in the first group.
The PSDP can be implemented as a Field Programmable Gate Array
(FPGA), as in the preferred embodiment, or as an Application-Specific Integrated Circuit (ASIC), a fully-custom Application Specific Standard Product (ASSP), or even as discrete logic on a printed-circuit board. It can also be included in an integrated processor (i.e., a CPU that includes peripheral interface logic) on a single chip or in a single package, or it can be included with the circuitry of the mass storage device. In addition to assisting the JPU in accessing data, the PSDP is specially programmable to also inteφret data in a specific format as it is read from or written to the associated disk(s). This enables PSDP to perform portions of jobs on data directly, as it is read off the disk, prior such data ever being forwarded to the JPU. In an embodiment specifically adapted for processing of record-oriented data, data can be filtered by the PSDP as records and fields of a database, so that only certain records, or certain portions of records, are actually forwarded to be written into the associated JPU's main memory.
However, many other operations beyond simple filtering are possible to implement in the PSDP. For example, records with certain characteristics can be tagged as they are written in the JPU's main memory, to indicate that such records are to be ignored in further processing, or to indicate certain attributes of such records, such as if they are to be handled differently in a transactions from other records.
While of use in processing field-oriented database records, it should be understood that the particular invention can also be used to advantage in processing many different types of data, including other field delimited data such as tables, indices, and views. The system is also advantageously used to process less structured data such as character strings, Binary Large Objects (BLOBS), XML, graphics files, and the like.
Discussion of Advantages
A number of advantages result from this architecture. First, unlike prior art database machines that integrate special processing hardware into the disk assembly itself (e.g. on the heads, on the arms, or electronically nearby), the JPUs in the second group use the special puφose PSDP hardware to interface to a disk and filter data after it reads from a disk, but still prior to a more general puφose execution unit. As a result, the system designer may now use industry standard disk controllers and standard hard disk drives. This allows the designer to effectively leverage the ever increasingly higher density of standard IDE and SCSI compatible storage media, as soon as they become available. Second, like the custom controller approach, any need to first read records into memory locations prior to performing any operation on them is still avoided. But when only a fraction of the available data is relevant to a query,, the PSDP avoids inefficiencies of other approaches that: allocate memory for unused information waste time copying unused information into memory waste time stepping around unused information The PSDP avoids these problems since database filtering operations are performed "on the fly" in a streaming fashion, as data is read as records stream out of the mass storage devices. In a preferred embodiment, the PSRP can also be programmed perform operations such as Boolean comparisons of record field values against either literal values or other record field values, or values held in registers of the processing element, and reject records that fail these Boolean comparisons before they are stored in memory. Of the records that pass the filtering conditions, the PSDP element can thus additionally filter out the subset of fields that are irrelevant to a particular query. In addition to field-level record filtering, the PSDP also can perform other operations on records as they are read from mass storage. For example, the PSDP can be programmed to decompress records entering memory and to compress records being sent out of memory. It can be instructed to decrypt records entering memory or to encrypt records being sent out of memory. It can convert lowercase fields to mixed or uppercase. It can, in fact, be programmed to perform myriad other such operations. Because these operations occur as each record streams into memory, the PSDP offloads such tasks from the JPUs main CPU, freeing it for other useful work.
Other advantages result if the PSDP is programmed to perform simple Boolean operations, such as to compare field values of the record stream against values held in its local registers. This allows a limited class of join operations to be performed of records before they are stored in memory. For example, if the values of the fields being joined are limited in range (such as when a set of consecutive integers is used to represent each of the 50 United States), the presence or absence of a particular field value can be encoded as a bit within a sequence of bits, whose position within the sequence corresponds to the integer field value.
One advantage of this is that it allows field-level filtering and more complex processing to proceed in parallel within the JPU, for additional performance benefit. A more important advantage is that this configuration of processors is most effective at reducing the amount of data that must flow through the system.
In essence, by using a PSDP that is dedicated to performing as much field- level filtering as possible before records are stored into the JPU's memory, the JPU's CPU is thus free to perform as much record processing as possible before it must return records over the network (for aggregation with the results of other JPUs) into a final reply to the SMP host. Because moving vast amounts of data requires much overhead, it is advantageous to add a dedicated processing element before each step in the data movement pathway, from input to final result. The JPU/PSDP architecture, in effect, separates streaming record processing from other query processing functions. Because the PSDP can be programmed to recognize record formats, it is capable of producing tuple sets as an output. As a result, after data leaves the PSDP, it can always be handled in tuple set form. This permits very fast handling data procedures to be implemented, because a consuming operation (be it in the JPU or the host) never has to process a block of undifferentiated binary data.
Additionally, since there can now be one common data handling paradigm throughout the system, i.e., the streaming tuple set, all functions such as storage, network, data operations, and transaction operations can efficiently and consistently use the tuple set model. Therefore, any operation may be arranged to take as input(s) the output(s) from any other operation. Also, a common set of algorithms may be used for all operations whether on the host(s) or JPUs.
This is in contrast to most database systems, which may materialize data as blocks of binary information that needs to be parsed by differing operations; which use different paradigms for network, storage, and internal operations; and which are unable to stream efficiently because of those different paradigms.
The two group architecture also allows an application to be insulated from the details of the programming model of the JPU. The application interacts only with the first group, and the first group translates the application's requests into requests against the JPU. This approach has several advantages: changes are easily made to functionality because of the inherent modularity of the system; any step of a query may be optimally executed on either of the groups or a combination thereof; bugs in application code cannot cause data corruption, crashes, or affect the requests of other applications; so that queries written in existing standard languages using existing
Application Programming Interfaces (APIs) will run correctly while gaining the performance advantages of the invention; and requests to the JPU can be made large enough to amortize the cost of the network communication, so that the performance benefits of parallelism are not lost to network latency.
BRIEF DESCRIPTION OF THE DRAWINGS
The foregoing and other objects, features and advantages of the invention will be apparent from the following more particular description of preferred embodiments of the invention, as illustrated in the accompanying drawings in which like reference characters refer to the same parts throughout the different views. The drawings are not necessarily to scale, emphasis instead being placed upon illustrating the principles of the invention.
Fig. 1 is a system level block diagram of an asymmetric record processing system according to the present invention.
Fig. 2 is a more detailed view of a Job Processing Unit (JPU). Fig. 3 is a more detailed view of software components in a host.
Fig. 4 is a more detailed view of Job Processing Unit (JPU) software components.
Fig. 5 is a block diagram of a Programmable Streaming Data Processor (PSDP) component. Fig. 6 is a more detailed view of portions of a PSDP.
Fig. 7 is a detailed diagram of a Data Engine component.
DETAILED DESCRIPTION OF A PREFERRED EMBODIMENT
A. SYSTEM LEVEL ARCHITECTURE
1. First Group Components
The present invention is a data processing system having at least two "groups" of processing units, in which the individual components of each group are individual network "nodes" within the system. As will be explained in detail below, the present invention has to do with how the a first group of one or more host processors accepts and responds to queries for data, and transforms such queries into one or more jobs, a second group of nodes comprising one or more Job Processing Units (JPUs), wherein each JPU has a sfreaming data interface, for receiving data from a streaming data source, one or more general puφose CPUs, for responding to requests from the host computers in the first group, and one or more Programmable Streaming Data Processors (PSDPs), which perform primitive functions directly on data received from the sfreaming data interface.
As more particularly shown in Fig. 1, the first group 10 consists of one or more SMP "host" computers 12, each with its own memory, network interface, and local storage (not shown in Fig. 1). Each host 12 runs its own operating system, and typically, but not necessarily, each host 12 uses the same type of operating system as the other hosts 12.
The hosts 12 typically accept queries that are requests for data stored on mass storage devices, such as hard disk drives 23. The requests may originate from any number of business intelligence applications that may be residing on local processors 28 or client computers 36 or separately ruiming application software 30, that may originate through a computer network 33 or locally. Queries are typically provided in a format such as Structured Query Language (SQL), Open DataBase Connectivity (ODBC), Java DataBase Connectivity (JDBC), or the like.
The hosts 12 accept queries that can retrieve, modify, create and/or delete data stored on disk 23 and the schema for such data. The hosts 12 also accept requests to start, cormnit, and rollback transactions against the data. The hosts 12 also perform typical admimstrative functions such as reporting on the status of the system 10, start and shutdown operation, backing up the current state of the data, restoring previous states of the data, replicating the data, and performing maintenance operations. Optionally, there is a load balancing function 11 in front of the host 12 processors, which directs individual transactions to specific host or hosts 12 so as to evenly distribute workload.
A catalog management component 15 contains descriptions of the fields and layout of data. Catalog management 15 also contains information about which users and applications have which permissions to operate in which ways on which types of records, datasets, and relations. The various hosts 12 interact with catalog management 15 in order to process the requests they receive. In one embodiment, catalog management 15 is embedded within one of the hosts 12, with parts replicated to the other hosts 12 and second group 20 components. As will be understood shortly, the catalog manager provides information to permit the components of the second group 20 to perform filtering functions.
With the exception of their need to consult catalog management 15, the hosts 12 are generally able to respond to requests without having to communicate among themselves. In very rare instances, inter-host 12 communication may occur to resolve a transaction sequencing issue.
. 2. Second Group Components
The second group 20 consists of a plurality of Job Processing Units (JPUs) 22. As shown in Fig. 2, each JPU 22 consists of a network interface 25 for receiving requests and delivering replies, a general puφose Central Processing Unit (CPU) 26 such as a microprocessor 26, memory 27, and a Programmable Sfreaming Data Processor (PSDP) 28. Each JPU 22 runs a multi-tasking schedule-based operating system. Each JPU 22 also has an attached disk 23 and disk controller from which the JPU 22 may read streaming data. In other embodiments, the JPU 22 can receive streaming record data from alternate or additional sources such as other on-board processors or via other network interfaces in place of the disk drives 23. Such streaming data might include stock quotes, satellite data, patient vital signs, and other kinds of "live-feed" information available via a network connection.
The JPU 22 accepts and responds to requests from host computers 12 in the first group 10 to process the sfreaming record-oriented data under its control. These requests are typically "jobs" of a larger query, and are expressed as sequences of primitive operations on an input stream. The primitive operations could be inteφreted, but in the preferred embodiment, they are packaged as compiled code that is ready for execution. An exemplary job-based query is described in more detail below. In addition to processing jobs, a JPU 22 also accepts and responds to requests from hosts for other operations such as:
Start, pre-commit, commit, abort, and recover transactions Perform mirroring or other replication operations Start, initialize, reinitialize, stop, and retrieve status information Create, modify, or delete descriptions of records, indices, views and other metadata
Each JPU 22 also accepts and responds to requests from the hosts 12 to: Perform mirroring or other replication operations Redistribute data from one JPU to another Send data local to one JPU to another JPU to help process a query job
Send data to a logging device Send data to a replication device
Acknowledge the successful completion of an operation requested by another node. JPU(s) 22 typically use multi-tasking Operating System (OS) to allow receiving, processing, and reporting the results from multiple jobs in a job queue. The OS should also support overlapping job execution. To coordinate this, the OS typically is responsible for scheduling and prioritizing requests according to a number of factors that are determined in real time. These may include a job priority as assigned by the user and/or host 12, as well as a job's expected impact on the
JPU's 22 local resources includes the amount of memory, disk, network, and/or I/O queues needed to complete the job. The JPU 22 can also contain software for performing concurrency control, transaction management, recovery and replication of data for which the JPU is responsible.
JPUs 22 in the second group 20 are not directly visible or accessible to the users of, or the applications that run on, for example, the clients 36 or business intelligence applications 30 that present queries to the system 10. The JPUs are, instead, an embedded component that maintain significant autonomy and control over their own data. A given record (or other data primitive) in the system 10 is thus normally directly accessible to, and processed by only one JPU 22. While JPUs may replicate their records to increase reliability or performance, they do not share responsibility for processing a given record with other JPUs 22 when carrying at a job as part of a query.
More details of this autonomous, asynchronous nature of the JPU's can be found in the above referenced co-pending U.S. Patent Application (Attorney Docket No. 3336.1016-003) entitled "Programmable Data Streaming Architecture Having Autonomous and Asynchronous Job Processing Unit."
The storage manager 320 within each JPU 22 provides support for other functions such as error checking, creation and deletion of tables, the use of indices, record insert and delete, mass loading of existing user data among various JPUs, and the like. Throughout the system, the components and sub-components are designed to optimize performance thru extensive use of streaming operations coupled with tuple set operations. As will be understood shortly, most operations are designed to take tuple sets (records or groups of records) as their input and output streams; these operations try not to materialize data, but instead they stream the output to the next operation. As a consequence many operations can be handled as one continuous data flow, whereas in a conventional system, it would be necessary to handle them in various layers.
For instance, a storage layer can be designed as a tuple set manager where (from the view of other JPU processes) it stores and retrieves tuple sets. From the storage layer onward, data is normally handled in tuple sets, providing a consistent, well organized, and easily accessible format for internal operations. This is in contrast to other systems where the storage layer stores and retrieves undifferentiated blocks of data which are later converted to tuple sets by some other downstream process. Another example of the streaming/tuple set architecture is the network layer, which sends and receives tuple sets instead of blocks of data.
Yet another example is a merge aggregation mode, where a sorted data stream is aggregated as requested, and whenever a new key index value is received, the aggregation from the previous key index value may be streamed to the next node.
A streaming/tuple set operation can be illustrated by tracking a typical dataflow during a load operation. In this example load case, as data is read into a host 12 over TCP/IP network connection 32, that data is parsed, error-checked, and transformed, and the distribution value calculated, all while the specific byte/field is in processor cache, and saved to the internal network output frame buffers as one step. The result is that the input data is read/transformed in a sfreaming fashion and converted to network-ready tuple set packets at streaming speed with n nimal overhead. As each packet is received, it is sent over the internal network 34 to an appropriate JPU 22 (as determined by the a distribution value in a Query Plan). At the JPU 22, the received data is read, converted into an approved storage format, and placed in memory buffers on a record-by-record basis. As memory buffers are filled, a storage layer in the JPU double-checks that the data corresponds to the indicated table, and that the table "owns" the physical space on the disk 23, and then writes that data to the disk 23. Note that during this process, a given byte of data was "touched" only a few times, and that the data was manipulated in tuple sets thereby optimizing performance and reliability.
A second illustration of a sfreaming tuple set operation is a join/aggregate operation where three joins and one co-located aggregation are performed on JPUs
22, and the results are returned through the host 12 via ODBC to the ODBC client 36 (e.g., Business Objects).
In this example, on each of three JPUs, the disk 23 is scanned and data read off the disk through the associated PSDP, which filters records of interest and fields of interest within those records, and places the resulting tuples into a tuple set buffer in JPU memory. As each tuple set buffer is filled, that tuple set is passed through each of three JPU join nodes and the aggregate node in turn. Each time a new key value is received by the aggregate node, the previous aggregate value and associated key value tuple are transformed as necessary per the ODBC request, and placed in the JPU network packet output buffer associated with the requesting host 12. When a network packet output buffer in the JPU is filled, its contents are sent to the host 12, where it is immediately placed in the user-side network buffer and is immediately sent to the ODBC client 36.
Note that, as in the previous example, the data was "touched" only a few times. Because the data was handled in tuple sets, it could be operated on as integral units with very minimal overhead. Because the operations are extremely integrated, mixed operations such as joins, aggregates, output transformation, and network packet creation are all performed while the data is in processor cache memory. More information regarding the sfreaming nature of data transfer can be found in the above referenced co-pending U.S. Patent Application (Attorney Docket No. 3336.1016-001) entitled "Asymmetric Streaming Record Data Processor Method and Apparatus," (Attorney Docket No. 3336.1016-001).
B. HOST SOFTWARE FUNCTIONS
Fig. 2 is a software component diagram for a host 12. A summary description of the functional blocks and their interactions now follows. This list is intended here to be an introduction to a more detailed description of how a query is processedinto a set of jobs that can then be carried out as. synchronously and autonomously by JPUs 22.
Postmaster 201
Serves as Front-end for query processing
Postmaster 201 accepts requests from user applications via API 200
Creates an Execution Plan
May use authentication
Plan Generator 204
Parse/query rewrite/planner - plans how query will be processed. Supports SQL-92 DDL/DML
Supports SQL Functions
Provides compatibility with Oracle, SQL Server
Integrated with SQL triggers, stored procedures
Plan Optimizer 205
Cost-based optimizer, with the addition of locale costs which optimizes for most efficient operation/highest level performance
Indicates which operations will be done within host and which will be done within JPU
Communicates with Plan Link, providing tips on what filtering should be done within the Programmable Data Streaming Processing ("PSDP") if there are multiple filters that can be done there (more than the PSDP can handle)
Maintains usage/reference statistics for later index creation, refreshing cluster indices
Plan Link 206
Takes an Execution Plan as input
Analyzes Execution Plan and splits plan further, identifying what will be done within the PSDP 28, what will be done within the JPU 22 after the
PSDP 28 has returned its data to the JPU 22, and what will be done in the Host 12 after the JPU 22 has returned its data '
SQL Expression Evaluator/SQL Converter 207 Expression Evaluator
Creates object code for evaluating given expression to be executed on the Host, JPU, and PSDP based on the expressions, their type, and the capabilities of the installed hardware
Host Dispatch 208
Similar to standard UNIX scheduler/dispatcher Queues execution plan and prioritizes based on (a) the plan's priority, history, and expected resource requirements, and (b) available resources and other plans' requirements
Controls number of jobs being sent to any one JPU 22 to avoid JPU Scheduler or JPU memory overload
Sends Host jobs to host execution engine
Communications Layer 210
Provides communications among the nodes Includes Job Listener to await data from nodes
Uses striping data from a Topology Manager to direct multicast and unicast messages
Detects non-responsiveness of nodes and communicates with Topology
Manager to trigger failover processing
Call Home 212
Initiates message to a Technical Assistance Center (not shown) to identify failed part and trigger service call or delivery of replacement component (as appropriate given user support level) Optionally communicates via SNMP to a defined app to receive a failure indicator and callhome trigger
Logs error(s)
Logger/Replication Server 218 Logs transaction plans, messages, failures, etc. to Netezza log in conventional fashion Implemented as a standard transaction logger/replication server
System Manager 220 Defines and maintains JPU Configuration information, striping information
Mirror Master - maintains mirrors info - what JPUs are being miπored where, maintains SPA data, maintains info on system spares Initiates failover processing when informed by Comm layer of a non- communicative JPU - directs mirror of failed JPU to take over as primary and begin copying to designated spare, directs primary of JPU mirrored on failed JPU to copy its data to that same designated spare, to reduce load on mirror of original failed JPU also directs mirror of the primary on that failed
JPU's mirror to do double duty and act as new primary until failover copying has been completed
Commimicates to callhome component to initiate replacement process Manages system expansion and allows for redistribution of data as appropriate or as requested by user during expansion
Initiates JPU diagnostics when appropriate
Provides an API to allow client management interface to get configuration data for user display/control
Host Diags 226
Runs diagnostics on Host as required/requested
Loader 230
Provides fast loader capability for loading user data onto disks Conimiinicates directly to Host Dispatch to load database/insert records
Communicates with System Manager to get configuration and mirroring data Controls index. creation on primary (and sets up job to run later to create indices on mirror)
Supports input via a number of methods (e.g., tab-separated data, backup/recovery)
Does ETL, converts data from Oracle, SQL Server, DB/2, etc. to the internal data format
MOX/OLAP 240 • Provides OLAP/MDX, ROLAP Engine on Host
Creates and maintains MOLAP cubes Supports multi-user MDX Creates Execution Plans for OLAP requests and communicates these directly to Host Dispatch
Supports metadata writeback
Provides administrative support for user creation, security
Access System Catalog through API
Cube Builder User Interface (UI) 242
Provides interface for defining and managing cubes to be used in OLAP Processing
JPU Downloader 250
Downloads Firmware to System JPUs 22 at system initiation/boot Downloads PSDP 28 and JPU 22 images
Communicates with System Manager to understand number of JPUs and JPU configurations
Initializes spares for failover Initializes replacements
Host Disk Manager 250 Manages Host Disk (used for Catalog, Temp Tables, Transaction Log,
Netezza Log, Swap space)
Host Event Handler 252
Receives partial record sets from JPUs 22 through the Comm Layer Job Listener
Executes remainder of Execution Plan that has to be done at Host 12 Provides intermediate and final sort-merge of JPU 22 sorted data as required Handles joins of data returned from JPUs 22 as required Communicates to JPUs through Comm Layer 260 to request partial result sets from JPU buffers when idle (e.g., to get and sort/process partial records that the JPU currently has instead of waiting for JPU 22 to fill a buffer and then send to Host 12 Host Transaction Manager 264
Manages transactions on the host 12 Controls requests sent to JPUs 22 that will be involved in the transaction
Provides lock management and deadlock detection Initiates abort processing Sends state data to Recovery Manager 266 Sends ID requests to the Transaction LD.(TID) Manager 268 Provides transaction IDs and deleted transaction IDs to ensure that disk records are preceded Manages catalog requests as fransaction requests as required
TTD Manager 268 Provides unique fransaction identifiers (TIDs)
Coordinates with other hosts to avoid generating duplicate TIDs
Host Recovery Manager 266
Ensures transaction atomicity after component (e.g., JPU) failure Maintains journal of transaction state
Initiates rollback as required
Backup/Recovery 270
Supports Host side of Backup/Recovery process Interfaces with Transaction Manager and JPU Storage Manager
C. JPU SOFTWARE COMPONENTS
Fig. 3 is a diagram of the software components of a JPU 22.
Communications Layer 300 Provides internal communication among nodes Includes Job Listener 301 to await requests
Includes Network Poster 302 to send data when buffer filled, job completed, or at Host request
JPU Dispatch/Scheduler 304
Receives plan through Communications Layer 300 Queues Plan
Schedules/dispatches jobs according to their priority, "fairness" to date, expected resource requirements, and available resources
JPU Transaction Manager 306
Processes changes in transaction state to begin a transaction, pre-commit a transaction, commit a transaction, or abort a transaction Handles processing of dependencies among transactions as flagged by the lock manager; broadcasts information about these dependencies to relevant host(s); initiates deadlock checks
JPU Lock Manager 308 Controls concurrent access to data
Interfaces with EventTask 36 before a query is executed and for each result . set returned from a scan Provides support for arithmetic locking
JPU Recovery Manager 312
Maintains a Journal to track fransaction status on the JPU 22, using the
Storage Manager API
Performs transaction recovery when requested by JPU Transaction Manager JPU Mirror Manager 314
Mirror Sender receives copies of record updates from Storage Manager 320 and fransmits these to the mirror for this JPU when an updating transaction commits Mirror Receiver receives record updates, buffers these in memory, and flushes out to disk through the Storage Manager when the Mirror Receiver buffer is full
Transmits all data to a spare system during failover processing
Storage Manager 320
Stores and manages information on disk in optimal fashion
Has an API that supports storage and retrieval of tuple sets
Supports error checking to insure that the data conforms to the indicated table and the indicated table "owns" the physical space to which the data is being written
Supports creation and deletion of tables, views, and indices Handles record inserts and deletes Supports ETL and mass loading of existing user data Provides storage support for commit/rollback Provides support for Precise Indexes
Provides mirroring support for failover
Optimizes sort operations and utilizes smart hash algorithm for data distribution/striping
Provides support for compression and smart storage optimization Controls disk I/O
JPU Resource Scheduler 322
Schedules jobs to run on the PSDP 28; communicates with JPU/PSDP Scheduler 324 to queue up PSDP requests to retrieve required data Optimizes the queue to keep the PSDP/disk as busy as possible, with requests from multiple queries intermixed in the queue based on disk characteristics and location of data on the disk Takes into account the needs of any data loading for new tables being created and transformed to internal data format (i.e., to optimize the loading process) Supports heuristic-based scheduling, ensuring that jobs are scheduled on a priority basis, but also ensuring that all jobs do get serviced (e.g., raising a job in priority if it has not been run in a certain interval of time)
Supports synchronous/piggy-backed scans, combining similar requests to optimize PSDP processing
Manages memory buffers/memory allocation on JPU; allocates memory to Execution Plans based on expected needs and hints received from Plan Optimizer
JPU Paging (if required)
PSDP Prep 330
Defines the instructions that will be given to the PSDP 28 in order to process a request (instructions tell the PSDP 28 what to do with each field being read
Figure imgf000027_0001
Identifies what filtering, transformation, projection, and aggregation operations are to by run by the PSDP 28
EventTask 310
Executes the portion of the Execution Plan that could not be handled by the PSDP but that does not have to be handled at the Host level Handles sorts, joins, transformations, and aggregations that could not be done as data stream through the PSDP 28 Maintains a memory buffer of result set records and returns these to Host through the Comm Layer when buffer filled, job completed, or at Host request
JPU Diags 332 Runs diagnostics on JPU as required/requested
JPU Boot/Init 334 Executes image burned into flash memory at boot time to bootstrap the JPU, run diagnostics, register the JPU with the primary Host server, and download new image from Host to run
Loads and transfers control to the image downloaded from the primary Host server to load the JPU application code, the operating system, the network stack, and disk driver code
Backup/Recovery 336
Supports JPU side of Backup/Recovery process Interfaces with Transaction Manager and JPU Storage Manager
DBA Lite 338
Provides automatic and dynamic disk and Storage Manager support Supports dynamic index creation, defragging, index garbage collection, timers, agents
JPU PSDP Scheduler 324
Schedules jobs to run on the PSDP; queues up PSDP requests to retrieve required data
D. DETAILED DESCRIPTION OF PSDP ARCHITECTURE
As discussed above, the PSDP allows data to be processed during Direct Memory Access (DMA) disk read operations. There are many different possible operations that can be performed by the PSDP 28, including transforming and comparing data with other data or with constants.
PSDP 28 functions fall into two general categories: disk driver logic interface 281 and data "filter" 282. Each of these functions is described in some detail below. It is sufficient here to note that the disk driver logic interface 281 accepts standard disk drive interface signaling, such as IDE (Integrated Device
Electronics) or SCSI (Small Computer Systems Interface), adapting it to a particular CPU native "bus" such as a Advanced Technology Attachment (ATA) bus or the like. Alternatively, if there is a communications network, such as Ethernet or Fibrechannel, instead of array of disks 23 to provide access to input data stream(s), the interface 281 becomes a network interface that is suitable to receive and/or transmit data over a communications network. The disk driver logic 281 is usually implemented in an Integrated Circuit (IC) in a computer or communications device, in or part of an IC that contains other logic, such as other interface logic or the CPU 26 itself. The disk driver 281 can even be inside the disk 23 itself, making the disk a special-pmpose unit attachable only to JPUs or communications devices for which the interface is specific. In the preferred embodiment, the PSDP 28 is however an Integrated Circuit
(IC) that interfaces a standard disk 23 to a peripheral bus of the JPU 22. All such controllers have the basic function of allowing the CPU 26 in the JPU 22 to read and write the disk 23, typically by setting up long data transfers between contiguous regions on the disk and contiguous regions in the CPU's 26 memory, a process usually referred to as Direct Memory Access (DMA).
The PSDP 28 also provides programmable hardware directly in the disk read path, to and from the controller. This function of the PSDP hardware, called the "filter" unit 282, is programmed to understand the structure of the data the analysis software running on the JPU 22 wishes to read and analyze. The PSDP 28 can be this programmed to operate on data as it is received from the disk 23, before it is stored into the JPU's memory, and in the process discard data that the JPU 22 would otherwise have to analyze. In an embodiment specifically adapted for processing of record-oriented data, data can be filtered by the PSDP 28 as records and fields of a database, so that only certain fields from certain records are actually forwarded to be written into the associated JPU's main memory.
Many other operations beyond simple filtering are possible however. For example, records with certain characteristics can be tagged as they are processed, to indicate that such records are to be ignored in further processing, or to indicate certain attributes of such records, such as if they are to be handled differently in a i transactions from other records. Other, non-filter like processes can be implemented such as compression/decompression; encryption/decryption; simple join operations, and the like. Thus, while the PSDP 28 of particular use in processing field-oriented database records, it should be understood it may process many different types of data, including other field delimited data such as tables, indices, and views; or less structured data such as character strings, Binary Large Objects (BLOBS), XML, graphics files, and the like. So although referred to herein as a "filter" unit that processes "records", it should thus be understood that filter 282 can also perform many other functions on various types of data, not just records.
As one example of filtering record-oriented data, the PSDP 28 can be programmed to recognize that a certain set of records in a database have a specified format, for example, a preamble or "header" of determined length and format, perhaps a field including the length of the record, followed by data including some number of fields of a certain type and length (e.g., 4-byte integers), followed by some number of fields of a different type and length (e.g., 12-byte character strings), followed by some number of fields of variable length, whose first few bytes specify the length of the field in some agreed-upon manner, and so forth.
The filter unit 281 can then execute this program as it reads data from the disk 23, locate record and field boundaries, and even employ further appropriate Boolean logic or arithmetic methods to compare fields with one another or with literal value. This allows the filter unit 282 to determine precisely which fields of which records are worth ttansferring to memory. The remaining records are discarded, or tagged in a manner that signals the JPU 22 that a record need not be analyzed. Again, there will be more discussion of how this is done in detail below. In the preferred embodiment, there are two basic reasons for which the filter unit 282 can discard a record (or mark it as unworthy of attention). The first is an analysis of the contents of the fields as described above. Using a previous example, the filter unit 282 can, for example, be programmed for a store sales database to check a purchase date field against a range of numbers that correspond to dates in the month of July in the year 2000, another field for a number or string uniquely associated with a particular store in North Carolina, another field for a set of SKU (Stock-Keeping Unit) values belonging to various styles or manufacturers of blue raincoats, and in this fashion mark only certain records for further processing. The filter unit 282 can further be programmed to know which fields contain the name and address of the customer who made the purchase, and return only these fields from the interesting records. Although other database software could perform these operations, the filter unit 282 can perform them at the same rate as the data is supplied by the disk 23. Far less data ends up in the JPU's memory as a result leaving the JPU 22 free for more complex tasks such as sorting the resulting list of names and addresses by last name or by postal code.
A second example of how the filter unit 282 can be used is to discard or mark a record, as in record creation and deletion in a multi-user environment. Databases are not static, and it is common for some users to be analyzing a database while others are updating it. To allow such users concurrent access to the database, records can be tagged with transaction numbers that indicate when or by whom a record was created or marked obsolete. A user querying a database may not wish to see records created by another user whose activity began subsequently, or whose activity began previously but is not yet complete; if so, he probably will want to see records marked obsolete by such a user. Or the user may wish to see only the results of transactions entered by certain users, or only the results of transactions not entered by certain users. To facilitate this kind of record filtering, record headers can contain creation and deletion identifiers that the filter unit 282 can be programmed to compare with the current user's identifier to determine whether records should be "visible" to the current user. Once again, the filter unit can avoid transferring useless data to memory or relieve the JPU 22 of a time-consuming analysis task.
In the preferred embodiment there are two basic methods the filter 282 unit can use to filter out data that is unnecessary for a given query, thereby reducing traffic on the communications network and reducing the workload on the JPU 22.
As described above, the filter unit 282 can simply discard the data. This is not always practical, however. Imagine a very long record with many fields, or large fields, many of which are to be returned to the JPU 22. Further consider a situation where a record meets the criteria is arranged in such a way that the contents of the last field are relevant to the decision to transfer or discard the selected fields of the record. Practical implementations of the filter unit 282 may not be able to store ("buffer") the largest possible set of returnable fields in a very long record, since there will be a limit on local buffer size. In such a case, the filter unit must begin sending the selected fields to the JPU 22 before it can tell whether they actually should be sent. After the record has been completely processed by the filter unit, and all the selected fields transferred to the JPU 22, the filter can tag the transferred data with a states bit that says "never mind", thus saving the JPU 22and the communications network a great deal of work. In practice, the filter unit can append a length indication to every record fragment it does transfer to the JPU 22, so that the JPU 22 can find the boundaries between the record fragments the filter unit deposits in memory. This is a natural place for this status bit (or bits, if the JPU 22 must distinguish among multiple reasons) indicating the transfer of a useless record.
In addition to selecting certain fields from certain records for transfer to the JPU 22, the filter unit 282 can create and return additional fields not present in the database, by performing calculations on the contents of the fields that are present. This can further relieve the JPU 22 of work. An example of this is the calculation of a "hash" function on the values of specified fields from a record, some of whose fields are to be transferred to the JPU 22. A hash function is a numerical key assigned to a collection of numeric or non-numeric field values that speeds up the process of searching through a list of records. Other examples of useful information that can be computed by the filter unit 282 include running sums or averages of field values from one record to the next. All of these benefits accrue from the filter unit's
282 ability to parse the data into records and fields as it transfers the data from the disk 23 to the JPU 22.
Another example is a transformation, such as an ASCII substitution. One usage for an ASCII substitution is to change the collation sequence of a given field. For example, if the LAST_NAME starts with the French 'c' (ASCII 135) then the
SQL clause "WHERE LAST_NAME IS >'H'" will erroneously fail unless ' ' has been mapped to 'C (ASCII 76). Similar issues involve the use of the UPPER() and LOWER() functions. In the preferred embodiment, the PSDP has 2 groups of registers, each 256 bytes. If transformation of a given field is selected then the PSDP setup loads transformation fields into these registers before the data is streamed in. Each register in the transformation fields corresponds to an extended ASCII value and the register contains the value that each extended ASCII character is to be converted into. In the example above, register numberl35 contains the value 76. During the streaming phase, as each tuple streams through the PSDP, for those fields where a transformation is indicated, each byte is individually transformed to its converted value. Two registers are provided so that two types of transforms may be applied to different fields in a given stream, such as UPPER() and LOWER().
The transforms may be applied either (a) in the "filter" path, before evaluation and comparisons or (b) in the "project" path so that a given field is converted before being output from the PSDP. This is especially useful for correcting collation sequences in preparation for the CPU performing a sort. More details of such a substitution table are contained in the above referenced co-peanding U.S. patent application entitled "Field Oriented Pipeline Architecture for a Programmable Data Sfreaming Processor," (Attorney Docket No. 3336.1008-002).
One preferred embodiment of the PSDP 28 is now described in further detail in connection with Fig. 5. As shown in Fig. 5, a PSDP 28 consists of a finite state machine called the Data Engine 400, which implements filter logic and other control operations; a JPU interface 404; a disk interface (here the ATA interface 408); First- In-First-Out (FIFO) memories 406 and 407; and a DMA driver 402.
The PSDP 28 is in one sense an On-Line Analytic Processing (OLAP)- oriented disk drive interface. It contains logic that is capable of identifying records, filtering out the unwanted records, and selecting fields for return as the tuple sets.
The PSDP 28 supports both a Programmed I/O (PIO) Mode-2 for register access by - the JPU 22 and a UDMA (Ultra-Direct Memory Access) mode-4 for data transfers. The terms "flow through" and "filtered" are used to differentiate DMA read modes. In flow-through mode, also referred to as "raw read" mode, data moves directly from the input to the output of the Data Engine 400 without being filtered.
Data that is filtered has been processed, perhaps by culling records via a comparison and/or transaction ID processing (as described below), but typically by reformatting the records into tuple format, during which uninteresting fields can be dropped and PSDP-generated fields can be added. This process of culling records is called a "restrict" operation. The process of formatting fields into tuples is called a
"project". In filtering mode, disk blocks are pulled from a Disk FIFO 407, feeding them through the Block Header, Record Header, NULL Vector, Transaction ID, Field Parse, and Filter circuits in the Data Engine 400. Fields to be returned are pushed into the Memory FIFO 406. There is of course also a DMA write mode, in which data from the JPU 22 flows through the DMA driver 402 directly to the ATA interface 408.
For all three DMA modes (write, raw read, and filtered read), the PSDP 28 shadows the read/write disk command in order to control its own DMA state machines. It does not shadow the disk address or sector count, nor does it have access to the memory addresses. For writes and raw reads, the PSDP 28 blindly moves data from one interface to the other until the JPU 22 disables the mode. The JPU 22 knows the quantity of data to be moved for these modes and uses the disk and DMA controller 402 interrupts to identify the end of transfer. For filtered reads, the quantity of data to be transferred to memory is generally unknown, and the JPU 22 identifies the end of transfer from the disk and filter interrupts. All of the record info — header and data — can be projected during a filtered read, but the block header info can only be returned by a raw read. DMA data integrity is protected across the disk interface by an IDE CRC check.
As shown in Fig. 6, the Data Engine 400 includes filter logic 500, a data parser 502, header storage 504, transaction ID processing 510, error checking 506, and output tuple generator 508. In general, the data parser 502 is responsible for taking information from the disk 23 and formatting it into headers and fields so that the filter logic 500, header storage 504 and error checking 506 blocks can perform their respective tasks. The tuple generator 508 takes the output of the filter and TID processing 510 blocks and formats the results in a "tuple", suitable for processing by the JPU 22 or host 12.
Raw user table data as read from the disk 23 is understood and inteφreted by the data parser 502. In one preferred embodiment at the present time, user table data is stored on disk in 128 KB segments called "blocks". Each block begins with an 8- word header, followed by 0 or more records. The format of the block header may be as follows:
Figure imgf000035_0001
The CRC-32 field is meant to be computed by software and written to the disk along with the rest of the block header. Its calculation was to include all data from the block number through the end of the last sector of the block, including any trailing O's. Its primary puφose was to detect data corruption resulting from hardware or software bugs, but it could have detected disk data-retention problems as well. It is unrelated to the UDMA-mode CRC-16 calculation required by the
ATA-5 specification, which only guards the physical interface between the PSDP 28 and disk-drive IO buffers.
The sector count is the number of sectors in the block, which must be from 1 to 256. Thus a 0 in this 1-byte field means 256. The sector count occupies the most-significant byte of the last word of the block header.
The record count is the number of records in the block, which may be 0. Although the record count occupies the least-significant three bytes of the last word of the block header, only 13 bits are used.
A record as read from disk 23 into the Data Engine 400 is typically composed of a record header and one or more data fields, where the record header consists of three special fields, a length, and a null vector. The special fields are the row number, created transaction JD, and deleted transaction ID. All of the record header entries are optional on a per-table (not per-record) basis. However, if the record has a null vector, it must also have a record length, but not vice versa. The allowed data types are described elsewhere below.
Figure imgf000036_0001
The row number (sometimes called row_num) is the unique number of the row or record in the user's table. It is distinct from the row address (sometimes called row_addr), which is the complete physical address of a row in node-table- block-record format. The row number is also distinct from the record number, which is the 0-based ordinal number of a record within a block. The record number is the final component of the row address. The row address is computed by the PSDP.
The created XID contains the number, or ID, of the fransaction that created the record.
The deleted XID. In the preferred embodiment, records are not actually deleted. Rather, they are marked as deleted so they can be restored if the transaction that performed the deleting is rolled back. (There are system management tools to reclaim the space.) A value of 0 indicates the record has not been deleted. A value of 1 indicates that the record was created by a transaction that was rolled back.
These XIDs support visibility in a multi-version database system, as is described in a related application (visibility application)
The record length is the length of the record in bytes, excluding the row number and the fransaction IDs, but including the record length, the record null vector, the data fields, and any pad bytes at the end of the record needed for proper alignment of the first item of the following record. Thus, it is the distance in bytes from the beginning of the record length field to the beginning of the next record. Note that although all records in a table must have the same makeup, record lengths may vary because of variable-length character fields. The RecordLengthSize register defines record length sizes of 0, 1, 2, and 4 bytes, but only 0 and 2 are used.
The record null vector specifies which fields in the record are null, thereby indicating validity, not existence. For instance, a null varchar is not the same as an empty one. The record null vector consists of an even number of bytes. If it exists, the record null vector has the same number of bits as the record has data fields, and computes the number of half-words in the null vector as (FieldCount + 15) » 4. This vector is an array of bytes. Bit 0 of the byte immediately following the record length corresponds to the 0th data field; bit 7 of that byte corresponds to the 7th data field; bit 0 of the last byte of the word that contains the record length corresponds to the 8 data field; and so on.
There are strict rules governing field order and alignment. Both the record and its first data field must start on a word boundary (addr[l :0]-0). All record fields are self-aligned up to word boundaries. This means that 16, 12, 8, and 4 byte fields are word-aligned, 2-byte fields are H-word-aligned (addr[0]=0), and 1-byte fields can start anywhere. The row number, created XID, and deleted XID are all 8 byte fields and do not require pad bytes to align them. If there is a record length but no record null vector, two pad bytes are required following the record length. If the record null vector exists, it immediately follows the record length and naturally starts on a two-byte boundary, but two pad bytes may be required following the record null vector to properly align the first data field.
The physical order of data fields, which often is not the same as the logical order, takes care of aligning non-character data fields; the physical order is N16, T12, N8, 18, F8, N4, 14, F4, D4, 12, D2, II, Cl, C2, ... C16, V2. The fixed-length character fields are packed in as tightly as possible and are not aligned. Variable- length character fields start with a 2-byte length; they are .i- word-aligned and may require a preceding pad byte. Up to three pad bytes may follow the record's last data field in order to align the next record. If so, they are counted in the length of the earlier record.
More details of TLO processing as performed by the TID processing block 510, includes rollback are contained in our co-pending U.S. Patent Application (Attorney Docket No. 3336.1017-001) entitled "Controlling Visibility in Multi- Version Database Systems", by Foster D. Hinshaw et al. filed on August 22, 2003. A detailed circuit diagram of the filter / comparision logic 500 is shown in
Fig. 7. The filter logic 500 supports up to 32 comparison (and hash) instructions. Each operates on a single field, which can be a header field, the row address, or a data field. Multiple instructions can operate on fields. Each instruction can perform two comparisons, using either a Data-String Register (DSR) 701 or temp registers 702 for the second operand(s); the two temp registers 702-0, 702-1 are used to store an early record field for comparison to a later record field. There are two programmable, byte-wide substitution tables 703 that simplify character field comparisons by switching to all upper or lower case, for example. The instruction results are combined in the use/lose circuit 705. In the preferred embodiment, sixteen (16) different comparison function types are implemented by the logic units 720: e.g, true and false NOPs, the equality operators (=, !=, <, >=, >, <=), bit-vector join and its inverse, field is/is not null, field begins/does not begin with the operand, field contains/does not contain the operand. Although every comparison can be programmed for every supported data type (which may include integer, floating point, fixed- and variable-length character types etc.), not all combinations of data types and comparison operations are useful. The equality and null comparisons are appropriate for all types. For character comparisons, the string length and ASCII value of the characters determine inequality, such that "ABC" is less than "ABCD", and "A" is less than both "a" and "B". If a field is null, every comparison except null will fail. The bit- vector join and its inverse are for the integer data type. Begins, contains, and their inverses are for character types, both fixed- and variable- length.
As in traditional systems, for bit joins a bit vector is created with ' 1 's in the positions corresponding to those positions where the join condition is true. In the preferred embodiment, during the setup phase the PSDP is loaded with a 4096-bit vector and the specific field position of the streaming data which is to be matched to that bit vector. During the streaming phase, as each tuple streams through the PSDP, the lower 12 bits of that field are mapped into the bit vector to determine if the corresponding bit is set to "1". If it is, then the join condition is "TRUE", otherwise it is "FALSE".
The SQL predicate "EXISTS" and other related predicates such as "ANY", "ALL", "NOT IN", "NOT EXISTS", "IN" are operated as an "EXISTS JOIN" type.
This particular join type is implemented by modifying the nested loop hash, and merge join and bit join types. Typically, the smaller table is placed into memory with either a hash index or sorted with access methods of hash or binary. If the smaller table has a range smaller than 4096, then it is converted to a bit index and put into the PSDP as with the bit join above. During the streaming phase, as each tuple streams either through the PSDP 28 (in the case of exist bit joins) or through the CPU 26; the smaller table is scanned to see if it contains (or does not contain) the target field from the streaming tuple.
A "use/lose" logic circuit consists of up to eight sum or product terms. Each term can accept an input from each of the 32 instructions. The term outputs can be individually inverted before they're combined by either a sum-of-products (SOP) or product-of-sums (POS) calculation. Normally the filter indicates which records are to be kept, but the output of the SOP/POS calculation can be inverted to indicate which to reject. Taken altogether, the use/lose options provide deMorgan's Law term minimization.
While the record's data fields are parsed and optionally filtered, another circuit determines whether the record is valid by comparing the created and deleted transaction identifiers (LDs) to a data structure called the Invisibility List. The list contains up to 32 entries. The first is always the ID of the current transaction, that is the fransaction that created the current scan. The remainder define the transactions that are concurrent to the "current" transaction. There are five modes: off, normal, dirty, silent, and rollback. Normal and dirty set a status bit in the returned tuple, as described in the section below on tuple formats; silent and rollback affect the tuple's return, in conjunction with the filter results.
A project function encompasses the selection of record fields, the generation of new fields, and the tuple formation and return. Tuples typically consist of a row number, some data fields, and a 2-byte length/status, but they can also include the created and/or deleted transaction IDs, the row address, up to 255 pad words, the 32 instructions results formed into a boolean word, the hash result, and a null vector. The hash is used to organize similar tuples into groups for processing joins or grouping selects, and with the exception of the record null vector and length/status, all record-header and data fields can be used in its calculation. There are 7 defined hash modes, such as full CRC, which calculate a 32-bit CRC hash starting with a seed of zero and using all of the bytes of all of the fields selected. Blank spaces in character fields are skipped, as are leading O's in unsigned and positive numerics and leading l's in negative numbers. Hash operations are defined on a per-field basis by the comparison instructions.
Within the PSDP 28, a "tuple" is used to describe the projected data as output by the tuple 500. The tuple generator 508 uses principally the filter 500 output but can also use TID processing 510 and error checking 506 outputs (Fig. 5). The term "tuple" is used here for the puφose of differentiating "raw" disk 23 and PSDP 28 output record formats. A tuple contains fields projected from the source record and up to six "virtual" fields: row address, pad words (tuple scratch pad), the boolean results from each of the filter operations, a hash result, the tuple null vector, and the tuple length. All are optional on a per-table basis. The order of these fields is given in the following table.
Figure imgf000040_0001
The row number, created XID, deleted XID, and data fields are the same as described above.
The row address is a compressed version of the node, table, block, and record information. RowAddress[63:32] is defined by the NodelDTablelD register, a 32-bit register that is programmed with a 32-bit, merged version of the node ID and the table ID as part of the filter setup. RowAddress[31 : 13] is the 19-bit block number defined by the block header. RowAddress[12:0] is the 13-bit record number calculated by the PSDP 28; it is 0-based within the current block.
Software may define up to 255 pad words in the tuple immediately following the special fields. The Boolean filter result contains the pass/fail result for each of the 32 filter instructions.
The hash result is the output of the hash circuit.
The tuple null vector contains the record null vector bits for each data field software requested. Note that record and tuple null vectors do not generally match up. The tuple null vector must consist of an even number of bytes and begin on a two-byte boundary. Software ignores any undefined bits as they may have been set by a previous scan. Once again, the existence of the null vector requires the existence of the length. Like the record null vector, the least-significant bit of byte 0 of the null vector refers to the 0th field; the most-significant bit of byte 0 refers to the
7th field; the least-significant bit of byte 1 refers to the 8th field, and so on, but the alignment and therefore the location of each of the bytes is different.
The tuple length is the total length of the tuple in bytes, including leading specials at the beginning of the tuple and any pad bytes at the end needed for proper alignment of the first item in the following tuple. Although all tuples returned by a scan must have the same makeup, tuples sizes may vary due to variable-length character fields. The TupleLengthSize register defines tuple length sizes of 0, 1, 2, and 4 bytes. Because tuple fields are 4-byte aligned, tuple lengths are always multiples of four, and the least-significant two bits of the tuple length are available to indicate tuple status. Bit 0 is the overrun bit. When set, it means the tuple was returned despite failing to meet the filter conditions. This can happen if the tuple is so large that the PSDP must begin transferring it to JPU memory before the use/lose ' decision can be made, as described above. Bit 1 is the invalid bit. When set, it means the record from which this tuple was constructed has transaction IDs that make it invalid (i.e., invisibly created or visibly deleted).
With the exception of the length and nulls, tuple field alignments are the same as record field alignments. In the record the length and nulls precede the data fields, and the record null vector is left-aligned against the preceding record length. In the tuple the length and nulls follow the data fields, and the tuple null vector is right-aligned against the tuple length, which ends the tuple. The size of the tuple null vector and the requirement that it end in byte lane 1 together determine the location of its first byte: byte lane 0 or 2 (see the examples below). Aligning the tuple length in this manner makes it possible for software to locate the length while striding through the tuples backwards in memory. CPU software leaves enough space in memory at the head of the first tuple for a tuple length and null vector. This space allows relocating the length and null vectors ahead of the corresponding data fields as it walks the tuples backwards, then reverse direction to process the tuples forward. Alignment can require as many as five pad bytes between the last byte of data and the tuple null vector or length and, if neither null vector nor length exists, as many as three pad bytes following the last data byte.
Alignment examples for valid end of field data, tuple null vector, and tuple length:
No Length or Nulls No Length or Nulls
Figure imgf000042_0001
Figure imgf000042_0002
Length and 2-Byte Null
Figure imgf000042_0003
Figure imgf000042_0004
D. QUERY PROCESSING EXAMPLE
As an aid in the illusfrating how the system 10 processes data, an example database will be described that contains store sales data. The database defines a SalesDetail data table, a Customer data table, and a Store data table as follows:
SalesDetail
StoreJD CustomerlD SaleDate ProductCategory Units Amount
Customer
CustomerlD Gender Store
StorelD StoreLocation
A sample query might be to "show me the total units and dollar amount of rain gear sold to females in North Carolina in 2000, by customer ID." This can be translated into the SQL statement:
SELECT SalesDetaiLCustomerlD AS "CustID", Sum(SalesDetaiLUnits) AS "Sales Units",
Sum(SalesDetail. Amount) AS "Sales Amount"
FROM SalesDetail, Customer, Store
WHERE SalesDetail.StorelD = Store.StorelD
AND SalesDetaiLCustomerlD = Customer. CustomerlD AND Store.StoreLocation = "NC"
AND Customer. Gender = "Female" AND Year(SalesDetail.SaleDate)="2000" AND SalesDetail.ProductCategory = "Raingear"
GROUP BY SalesDetaiLCustomerlD;
An output from this sample query with the total units and dollar amount of rain gear sold to females in North Carolina in 2000 by customer ID might be shown in tabular format:
CustID Sales Units Sales Amount
021442 1,300 $45,000
021443 1,200 $41,000 021449 1,800 $60,000 021503 3,500 $98,000 021540 4,200 $112,000
021599 5,000 $150,000
021602 4,700 $143,000 021611 4,100 $104,000
021688 3,600 $101,000
021710 2,000 $65,000
021744 1,200 $41,000
021773 1,500 $43,000
Using the above example, a basic execution plan can be created by the SQL Expression Evaluator 207, plan generator 204 and plan optimizer 205 of the host(s) 12. The plan might specify for example, to perform joins and aggregations on the JPUs 22, with restriction functions being performed on the Programmable Streaming Data Processor (PSDP) 28. Job Locale Operation
1 JPU SCAN Customer
PSDP RESTRICT Gender = "Female"
JPU PROJECT CustomerlD
JPU SAVE AS TEMPCustomer
2 JPU SCAN Store
PSDP RESTRICT StoreLocation = "NC"
JPU PROJECT StorelD
JPU BROADCAST AS TEMPStore
3 JPU SCAN SalesDetail
PSDP RESTRICT ProductCategory = "R
Year(SaleDate)="2000" JPU PROJECT CustomerlD, StorelD, Units, Amount
JPU JOIN WITH TEMPStore, StoreID=TEMPStore.StoreID
JPU PROJECT CustomerlD, Units, Amount
JPU JOIN WITH TEMPCustomer, CustomerID=
TEMPCustomer.CustomerlD JPU PROJECT CustomerlD, Units AS "Units" , Amount AS
"Amt"
JPU GROUP By CustomerlD
JPU AGGREGATE Sum(Units) AS "Units", Sum(Amt) AS
"AmtTotal" JPU PROJECT CustomerlD, "Units", "AmtTotal"
JPU RETURN HOST
7 HOST RETURN USER
Referring back to Figs. 1 and 3, the query is passed from the application (which may be running on, Business Intelligence Application 30 local application server 29 or client 36), the Plan Generator 204 then creates tentative execution plans. Plans not only specify the above job descriptions, but also may specify whether specific jobs can run on currently or must run in sequence on the JPUs. The Plan Optimizer 205 selects one of the plans and optimizes that plan and passes it to the Plan Link 206. The Plan Link 206 expands the plan as necessary, based on where parts of the plan will be executed, and then passes the expanded plan to the Host Dispatch 208. The Host Dispatch 208 then sends individual Jobs within the plan to the respective locales (i.e., the JPUs 22) for execution. In this example, jobs 1-6 are sent to the JPUs 22 for execution, with job 7 reserved for the host 12.
For example, Job 1 scans the Customer table with the required restriction and projection, and materializes it. Job 2 scans the Store table with the required restriction and projection, and since it is a small table, broadcasts the resulting tuple set to all JPUs 22, where the tuples from all JPUs 22 are then accumulated and saved in memory as TEMPStore. Jobs 1 and 2 are specified or determined to run concurrently if possible.
The Host Dispatch 208 may thus combine Jobs 3-6 into one streaming job because they can all be implemented in a sfreaming manner without materialization of intermediate sets. This combined job scans the SalesDetail table, with its restrictions and projections. As the tuples are received from scan run by the PSDP 28 , each tuple is joined with TEMPStore and TEMPCustomer and aggregated. On the aggregation node, as each new customer LD is received, the previous one and its sums are sent to the host, where Job 7 is then invoked in a sfreaming fashion, to return the aggregated tuples through the ODBC connection 38 back to the user.
Fig. 7 is a diagram illustrating how an exemplary job is processed by a set of JPUs 22, in the second group. The example Job 3 replicated here again included instructions
3 JPU SCAN SalesDetail
PSDP RESTRICT ProductCategory = "Raingear" AND
Year(SaleDate)="2000" JPU PROJECT CustomerlD, StorelD, Units, Amount
Individual jobs are forwarded from the host 12 to typically many JPUs 22 in parallel as a broadcast message. The broadcast message is typically sent as a User Datagram Protocol (UDP) type message, but can also be sent in other ways, such as a Transmission Control Protocol (TCP) message or a unicast message. Upon receipt of a job message at the job listener 301, the JPU dispatch unit 304 informs the fransaction manager 306 and storage manager 320 to then schedule Job 3 for execution. More details of job execution can be found in the related co- pending U.S. Patent application entitled "Programmable Data Sfreaming Architecture Having Autonomous and Asynchronous Job Processing Unit," (Attorney Docket No. 3336.1016-003) mentioned above.
While this invention has been particularly shown and described with references to preferred embodiments thereof, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the scope of the invention encompassed by the appended claims.

Claims

CLATMSWhat is claimed is:
1. An asymmetric data processor comprising: a first group of nodes comprising one or more host processors, each host comprising a memory, a network interface, and one or more Central Processing Units (CPUs), wherein each host accepts and responds to queries for data, and transforms such queries into one or more jobs; a second group of nodes comprising one or more Job Processing Units
(JPUs), wherein each JPU comprises: a memory, for storing data a network interface, for receiving data and instructions a sfreaming data interface, for receiving data from a streaming data source; one or more general puφose CPUs, for responding to requests from at least one host computer in the first group, and to requests from other JPUs in the second group, and one or more Programmable Sfreaming Data Processors (PSDPs), which perform primitive functions directly on data received from the sfreaming data interface, each PSDP thus performing initial processing on a set of data; arid a network connecting the nodes within each group and between the two groups, and wherein a JPU receives jobs from one or most nodes in the first group, performs work requested by the job, and forms a reply.
2. The apparatus of claim 1 wherein the data comprises structured records, and the structured records further comprise fields of various lengths and data types.
3. An apparatus as in claim 1 wherein the primitive functions performed by the PSDPs comprise field-level filtering.
4. An apparatus as in claim 1 wherein the streaming data interface is an industry-standard mass storage interface.
5. An apparatus as in claim 2 in which at least one selected PSDP performs
Boolean comparisons of record field values against other values.
6. An apparatus as in claim 5 wherein the Boolean comparision is against other record field values, and/or values held internally to that PSDP.
7. An apparatus as in claim 5 in which the selected PSDP restricts records that fail Boolean comparisons of field values, as such records stream into the PSDP and before such records are placed into the memory of the associated JPU.
8. An apparatus as in claim 2 in which the selected PSDP filters out fields of records that are not needed for particular queries, as such fields stream into the PSDP and before such fields are placed into the memory of the associated JPU, projecting forward into JPU memory those fields that are needed.
9. An apparatus as in claim 2 in which the PSDP output data may contain projected fields not contained in the source data, such as row address, transforms, results of expression evaluation, results of bit joins, and results of visibility tests..
10. An apparatus as in claim 2 in which a selected PSDP decompresses fields and/or records.
11. An apparatus as in claim 1 wherein the sfreaming data interface is connected to receive data from a peripheral device selected from the group consisting of disk drive, network interface, and other streaming data source.
12. An apparatus as in claim 2 in which a selected PSDP performs a join operation, where the field values being joined have a small range of values, so that the presence or absence of a particular value can then be encoded as a bit within a sequence of bits, whose position within the sequence corresponds to the field value.
13. An apparatus as in claim 2 in which a selected PSDP performs an "exist join" operation, where the field values being joined have a small range of values, so that the presence or absence of a particular value can then be encoded as a bit within a sequence of bits, whose position within the sequence corresponds to the field value.
14. An apparatus as in claim 1 in which space is reserved in JPU memory at the head of the first tuple produced by the PSDP for recording tuple length and null vector, so that the length and null vectors from the end of the tuple may be relocated to this space.
15. An apparatus as in claim 1 in which at least one PSDP is implemented as a
Field Programmable Gate Array (FPGA).
16. An apparatus as in claim 1 in which the host computers in the first group contain software comprising a plan optimizer component that determines which filtering primitives should be executed within a PSDP.
17. An apparatus as in claim 1 in which the JPUs in the second group contain software comprising a plan optimizer component that determines which filtering primitives should be executed within a PSDP.
18. An apparatus as in claim 1 in which the host computers in the first group contain software comprising a plan link component, which determines a query execution plan, the query execution plan further having portions that will be processed by a PSDP, portions that will be processed by a JPU after a PSDP has returned data to the JPU, and portions that will be processed by a host, after the JPU has returned data to the host group.
19. An apparatus as in claim 1 in which the JPUs in the second group contain software comprising a plan link component, which determines a query execution plan, the query execution plan further having portions that will be processed by a PSDP, portions that will be processed by a JPU after a PSDP has returned data to the JPU, and portions that will be processed by a host, after the JPU has returned data to the host group.
20. An apparatus as in claim 1 in which the hosts in the first group contain software comprising a PSDPPrep component, which, for a given query execution plan, defines primitive instructions.
21. An apparatus as in claim 1 in which the JPUs in the second group contain software comprising a PSDPPrep component, which, for a given query execution plan, defines primitive instructions.
22. An apparatus as in claim 21 wherein the instructions defined by the PSDPPrep component include instructions to process fields of records.
23. An apparatus as in claim 21 in which a PSDPPrep component further identifies filtering, transformation, projection and/or aggregation operations to be performed by a PSDP.
24. An apparatus as in claim 21 in which a PSDPPrep component further modifies the query execution plan to specify restrict operations that are to be performed by a PSDP instead of a JPU.
25. An apparatus as in claim 1 in which the JPUs contain software comprising a PSDP Filter component, which loads an executable code image into a PSDP.
26. An apparatus as in claim 1 in which the JPUs contain software comprising a
PSDP Scheduler component, which schedules jobs to run on a PSDP and queues PSDP requests to retrieve required data.
27. An apparatus as in claim 1 in which the JPUs in the second group contain software comprising a JPU Resource Scheduler component, which is responsible for scheduling jobs to be run on the JPU.
28. An apparatus as in claim 27 in which the JPU Resource Scheduler component further schedules jobs to run on a PSDP, communicating with a PSDP Scheduler component to queue up PSDP requests to retrieve required data.
29. An apparatus as in claim 27 in which the JPU Resource Scheduler component further schedules jobs, in which similar PSDP instructions in different query execution plans are combined to avoid duplicate PSDP processing requests.
30. An apparatus as in claim 2 in which an initial query is provided by a structured query language (SQL) statement, and the records specified thereby exist in various processing states within at least two components of the system including at least within a PSDP within a JPU, and/or within a host.
31. An apparatus as in claim 30 in which a PSDP processes fields within records are received from the sfreaming data source, without waiting to process any records until all records are received.
PCT/US2003/030001 2002-09-18 2003-09-18 Programmable streaming data processor for database appliance having multiple processing unit groups WO2004027652A1 (en)

Priority Applications (2)

Application Number Priority Date Filing Date Title
EP03759450.4A EP1546930B1 (en) 2002-09-18 2003-09-18 Programmable streaming data processor for database appliance having multiple processing unit groups
AU2003275181A AU2003275181A1 (en) 2002-09-18 2003-09-18 Programmable streaming data processor for database appliance having multiple processing unit groups

Applications Claiming Priority (4)

Application Number Priority Date Filing Date Title
US41168602P 2002-09-18 2002-09-18
US60/411,686 2002-09-18
US41205702P 2002-09-19 2002-09-19
US60/412,057 2002-09-19

Publications (1)

Publication Number Publication Date
WO2004027652A1 true WO2004027652A1 (en) 2004-04-01

Family

ID=32033566

Family Applications (5)

Application Number Title Priority Date Filing Date
PCT/US2003/029377 WO2004027649A1 (en) 2002-09-18 2003-09-18 Asymmetric streaming record data processor method and apparatus
PCT/US2003/029699 WO2004027576A2 (en) 2002-09-18 2003-09-18 Asymmetric data streaming architecture having autonomous and asynchronous job processing unit
PCT/US2003/029187 WO2004027648A1 (en) 2002-09-18 2003-09-18 Intelligent storage device controller
PCT/US2003/030001 WO2004027652A1 (en) 2002-09-18 2003-09-18 Programmable streaming data processor for database appliance having multiple processing unit groups
PCT/US2003/029186 WO2004027647A1 (en) 2002-09-18 2003-09-18 Field oriented pipeline architecture for a programmable data streaming processor

Family Applications Before (3)

Application Number Title Priority Date Filing Date
PCT/US2003/029377 WO2004027649A1 (en) 2002-09-18 2003-09-18 Asymmetric streaming record data processor method and apparatus
PCT/US2003/029699 WO2004027576A2 (en) 2002-09-18 2003-09-18 Asymmetric data streaming architecture having autonomous and asynchronous job processing unit
PCT/US2003/029187 WO2004027648A1 (en) 2002-09-18 2003-09-18 Intelligent storage device controller

Family Applications After (1)

Application Number Title Priority Date Filing Date
PCT/US2003/029186 WO2004027647A1 (en) 2002-09-18 2003-09-18 Field oriented pipeline architecture for a programmable data streaming processor

Country Status (4)

Country Link
US (6) US7577667B2 (en)
EP (6) EP1546822A4 (en)
AU (5) AU2003270828A1 (en)
WO (5) WO2004027649A1 (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9423983B2 (en) 2012-01-19 2016-08-23 Syncsort Incorporated Intelligent storage controller

Families Citing this family (329)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7082552B2 (en) * 2000-02-08 2006-07-25 Mips Tech Inc Functional validation of a packet management unit
US7155516B2 (en) * 2000-02-08 2006-12-26 Mips Technologies, Inc. Method and apparatus for overflowing data packets to a software-controlled memory when they do not fit into a hardware-controlled memory
US7032226B1 (en) * 2000-06-30 2006-04-18 Mips Technologies, Inc. Methods and apparatus for managing a buffer of events in the background
US7649901B2 (en) * 2000-02-08 2010-01-19 Mips Technologies, Inc. Method and apparatus for optimizing selection of available contexts for packet processing in multi-stream packet processing
US7165257B2 (en) * 2000-02-08 2007-01-16 Mips Technologies, Inc. Context selection and activation mechanism for activating one of a group of inactive contexts in a processor core for servicing interrupts
US7139901B2 (en) * 2000-02-08 2006-11-21 Mips Technologies, Inc. Extended instruction set for packet processing applications
US7058064B2 (en) * 2000-02-08 2006-06-06 Mips Technologies, Inc. Queueing system for processors in packet routing operations
US7069375B2 (en) * 2001-05-17 2006-06-27 Decru, Inc. Stream-oriented interconnect for networked computer storage
US7904454B2 (en) * 2001-07-16 2011-03-08 International Business Machines Corporation Database access security
US7146524B2 (en) * 2001-08-03 2006-12-05 Isilon Systems, Inc. Systems and methods for providing a distributed file system incorporating a virtual hot spare
US7685126B2 (en) 2001-08-03 2010-03-23 Isilon Systems, Inc. System and methods for providing a distributed file system utilizing metadata to track information about data stored throughout the system
EP1546822A4 (en) * 2002-09-18 2008-07-02 Netezza Corp Asymmetric data streaming architecture having autonomous and asynchronous job processing unit
WO2004046971A1 (en) 2002-11-14 2004-06-03 Isilon Systems, Inc. Systems and methods for restriping files in a distributed file system
US7010538B1 (en) * 2003-03-15 2006-03-07 Damian Black Method for distributed RDSMS
CA2523548C (en) * 2003-05-23 2014-02-04 Washington University Intelligent data processing system and method using fpga devices
WO2005006153A2 (en) * 2003-07-07 2005-01-20 Netezza Corporation Sql code generation for heterogeneous environment
WO2005029280A2 (en) 2003-09-19 2005-03-31 Netezza Corporation Performing sequence analysis as a multipart plan storing intermediate results as a relation
US7149736B2 (en) * 2003-09-26 2006-12-12 Microsoft Corporation Maintaining time-sorted aggregation records representing aggregations of values from multiple database records using multiple partitions
US7181647B2 (en) * 2003-10-15 2007-02-20 International Business Machines Corporation Error tracking method and system
US7457819B2 (en) 2003-10-17 2008-11-25 International Business Machines Corporation Configurable flat file data mapping to a database
US7634500B1 (en) * 2003-11-03 2009-12-15 Netlogic Microsystems, Inc. Multiple string searching using content addressable memory
US7454418B1 (en) * 2003-11-07 2008-11-18 Qiang Wang Fast signature scan
US7870161B2 (en) 2003-11-07 2011-01-11 Qiang Wang Fast signature scan
US9098340B2 (en) * 2003-12-08 2015-08-04 Teradata Us, Inc. Virtual regulator for a database system
US8954420B1 (en) 2003-12-31 2015-02-10 Google Inc. Methods and systems for improving a search ranking using article information
DE102004001212A1 (en) * 2004-01-06 2005-07-28 Deutsche Thomson-Brandt Gmbh Process and facility employs two search steps in order to shorten the search time when searching a database
US20050154696A1 (en) * 2004-01-12 2005-07-14 Hitachi Global Storage Technologies Pipeline architecture for data summarization
US7529764B2 (en) * 2004-01-12 2009-05-05 Hitachi Global Storage Technologies Netherlands B.V. GUI for data pipeline
US7941439B1 (en) 2004-03-31 2011-05-10 Google Inc. Methods and systems for information capture
US8346777B1 (en) 2004-03-31 2013-01-01 Google Inc. Systems and methods for selectively storing event data
US8161053B1 (en) 2004-03-31 2012-04-17 Google Inc. Methods and systems for eliminating duplicate events
US7333976B1 (en) 2004-03-31 2008-02-19 Google Inc. Methods and systems for processing contact information
US8275839B2 (en) 2004-03-31 2012-09-25 Google Inc. Methods and systems for processing email messages
US7725508B2 (en) * 2004-03-31 2010-05-25 Google Inc. Methods and systems for information capture and retrieval
US8386728B1 (en) 2004-03-31 2013-02-26 Google Inc. Methods and systems for prioritizing a crawl
US8099407B2 (en) 2004-03-31 2012-01-17 Google Inc. Methods and systems for processing media files
US8631076B1 (en) 2004-03-31 2014-01-14 Google Inc. Methods and systems for associating instant messenger events
US20050234929A1 (en) * 2004-03-31 2005-10-20 Ionescu Mihai F Methods and systems for interfacing applications with a search engine
JP2005309847A (en) * 2004-04-22 2005-11-04 Sharp Corp Data processor
US8015154B1 (en) * 2004-06-07 2011-09-06 Teradata Us, Inc. Starting database software in response to a broadcast message
US7958292B2 (en) * 2004-06-23 2011-06-07 Marvell World Trade Ltd. Disk drive system on chip with integrated buffer memory and support for host memory access
US7552356B1 (en) 2004-06-30 2009-06-23 Sun Microsystems, Inc. Distributed data storage system for fixed content
US7536693B1 (en) 2004-06-30 2009-05-19 Sun Microsystems, Inc. Method for load spreading of requests in a distributed data storage system
US7328303B1 (en) * 2004-06-30 2008-02-05 Sun Microsystems, Inc. Method and system for remote execution of code on a distributed data storage system
US7296180B1 (en) 2004-06-30 2007-11-13 Sun Microsystems, Inc. Method for recovery of data
US7734643B1 (en) 2004-06-30 2010-06-08 Oracle America, Inc. Method for distributed storage of data
US9600216B2 (en) * 2004-07-22 2017-03-21 Ca, Inc. System and method for managing jobs in heterogeneous environments
US7886296B2 (en) * 2004-07-22 2011-02-08 Computer Associates Think, Inc. System and method for providing alerts for heterogeneous jobs
US8028285B2 (en) * 2004-07-22 2011-09-27 Computer Associates Think, Inc. Heterogeneous job dashboard
US8427667B2 (en) * 2004-07-22 2013-04-23 Ca, Inc. System and method for filtering jobs
US7984443B2 (en) 2004-07-22 2011-07-19 Computer Associates Think, Inc. System and method for normalizing job properties
US9020929B2 (en) * 2004-08-05 2015-04-28 International Business Machines Corporation Method and system for tracking performance by breaking down a query
DE102004052576A1 (en) * 2004-10-29 2006-05-04 Advanced Micro Devices, Inc., Sunnyvale Parallel processing mechanism for multiprocessor systems
US8055711B2 (en) 2004-10-29 2011-11-08 Emc Corporation Non-blocking commit protocol systems and methods
US8238350B2 (en) 2004-10-29 2012-08-07 Emc Corporation Message batching with checkpoints systems and methods
US8051425B2 (en) * 2004-10-29 2011-11-01 Emc Corporation Distributed system with asynchronous execution systems and methods
US20060100845A1 (en) * 2004-11-08 2006-05-11 Mazzagatti Jane C Multiple stream real time data simulation adapted for a KStore data structure
JP2006163596A (en) * 2004-12-03 2006-06-22 Internatl Business Mach Corp <Ibm> Information processing system, control method and program
WO2006071817A2 (en) 2004-12-29 2006-07-06 Netcell Corporation Intelligent storage engine for disk drive operations with reduced local bus traffic
US20060146868A1 (en) * 2005-01-05 2006-07-06 Intel Corporation Device, system and method for selective aggregation of transmission streams
US7610397B2 (en) * 2005-02-28 2009-10-27 International Business Machines Corporation Method and apparatus for adaptive load shedding
US7389300B1 (en) * 2005-05-27 2008-06-17 Symantec Operating Corporation System and method for multi-staged in-memory checkpoint replication with relaxed consistency
US7877750B2 (en) * 2005-07-27 2011-01-25 Sap Ag Scheduled job execution management
US7970788B2 (en) 2005-08-02 2011-06-28 International Business Machines Corporation Selective local database access restriction
US9270532B2 (en) * 2005-10-06 2016-02-23 Rateze Remote Mgmt. L.L.C. Resource command messages and methods
US7353332B2 (en) * 2005-10-11 2008-04-01 Integrated Device Technology, Inc. Switching circuit implementing variable string matching
US7788303B2 (en) 2005-10-21 2010-08-31 Isilon Systems, Inc. Systems and methods for distributed system scanning
US7551572B2 (en) 2005-10-21 2009-06-23 Isilon Systems, Inc. Systems and methods for providing variable protection
US7797283B2 (en) 2005-10-21 2010-09-14 Isilon Systems, Inc. Systems and methods for maintaining distributed data
US7917474B2 (en) 2005-10-21 2011-03-29 Isilon Systems, Inc. Systems and methods for accessing and updating distributed data
US20070094343A1 (en) * 2005-10-26 2007-04-26 International Business Machines Corporation System and method of implementing selective session replication utilizing request-based service level agreements
US7933923B2 (en) * 2005-11-04 2011-04-26 International Business Machines Corporation Tracking and reconciling database commands
US7958513B2 (en) * 2005-11-17 2011-06-07 International Business Machines Corporation Method, system and program product for communicating among processes in a symmetric multi-processing cluster environment
US7606844B2 (en) 2005-12-19 2009-10-20 Commvault Systems, Inc. System and method for performing replication copy storage operations
US7651593B2 (en) 2005-12-19 2010-01-26 Commvault Systems, Inc. Systems and methods for performing data replication
US8849745B2 (en) * 2005-12-23 2014-09-30 International Business Machines Corporation Decision support methods and apparatus
WO2007073470A2 (en) 2005-12-23 2007-06-28 Perdiem, Llc System and method for defining an event based on a relationship between an object location and a user-defined zone
US7525425B2 (en) 2006-01-20 2009-04-28 Perdiem Llc System and method for defining an event based on relationship between an object location and a user-defined zone
US20070150585A1 (en) * 2005-12-28 2007-06-28 Microsoft Corporation Multi-dimensional aggregation on event streams
US20070180275A1 (en) * 2006-01-27 2007-08-02 Brian Metzger Transparent encryption using secure JDBC/ODBC wrappers
US7848261B2 (en) 2006-02-17 2010-12-07 Isilon Systems, Inc. Systems and methods for providing a quiescing protocol
JP4140919B2 (en) * 2006-03-14 2008-08-27 インターナショナル・ビジネス・マシーンズ・コーポレーション System and method for performing processing such as spreadsheet
JP4532423B2 (en) * 2006-03-16 2010-08-25 富士通株式会社 Server system
US8607350B2 (en) * 2006-03-30 2013-12-10 International Business Machines Corporation Sovereign information sharing service
US7756898B2 (en) 2006-03-31 2010-07-13 Isilon Systems, Inc. Systems and methods for notifying listeners of events
US7933981B1 (en) * 2006-06-21 2011-04-26 Vmware, Inc. Method and apparatus for graphical representation of elements in a network
US7590652B2 (en) * 2006-08-18 2009-09-15 Isilon Systems, Inc. Systems and methods of reverse lookup
US7899800B2 (en) 2006-08-18 2011-03-01 Isilon Systems, Inc. Systems and methods for providing nonlinear journaling
US7822932B2 (en) 2006-08-18 2010-10-26 Isilon Systems, Inc. Systems and methods for providing nonlinear journaling
US7680836B2 (en) 2006-08-18 2010-03-16 Isilon Systems, Inc. Systems and methods for a snapshot of data
US7953704B2 (en) 2006-08-18 2011-05-31 Emc Corporation Systems and methods for a snapshot of data
US7882071B2 (en) 2006-08-18 2011-02-01 Isilon Systems, Inc. Systems and methods for a snapshot of data
US7680842B2 (en) 2006-08-18 2010-03-16 Isilon Systems, Inc. Systems and methods for a snapshot of data
US8244718B2 (en) * 2006-08-25 2012-08-14 Teradata Us, Inc. Methods and systems for hardware acceleration of database operations and queries
US7783654B1 (en) 2006-09-19 2010-08-24 Netlogic Microsystems, Inc. Multiple string searching using content addressable memory
US7660793B2 (en) 2006-11-13 2010-02-09 Exegy Incorporated Method and system for high performance integration, processing and searching of structured and unstructured data using coprocessors
US8289966B1 (en) 2006-12-01 2012-10-16 Synopsys, Inc. Packet ingress/egress block and system and method for receiving, transmitting, and managing packetized data
US8706987B1 (en) 2006-12-01 2014-04-22 Synopsys, Inc. Structured block transfer module, system architecture, and method for transferring
US8127113B1 (en) 2006-12-01 2012-02-28 Synopsys, Inc. Generating hardware accelerators and processor offloads
US8849746B2 (en) * 2006-12-19 2014-09-30 Teradata Us, Inc. High-throughput extract-transform-load (ETL) of program events for subsequent analysis
US8141100B2 (en) * 2006-12-20 2012-03-20 International Business Machines Corporation Identifying attribute propagation for multi-tier processing
US8286029B2 (en) 2006-12-21 2012-10-09 Emc Corporation Systems and methods for managing unavailable storage devices
US7593938B2 (en) 2006-12-22 2009-09-22 Isilon Systems, Inc. Systems and methods of directory entry encodings
US20080163224A1 (en) * 2006-12-29 2008-07-03 Frank Joachim H Modeling interrupts in a business process
US7509448B2 (en) 2007-01-05 2009-03-24 Isilon Systems, Inc. Systems and methods for managing semantic locks
US8019781B2 (en) * 2007-02-15 2011-09-13 Microsoft Corporation Host context framework
US8495367B2 (en) * 2007-02-22 2013-07-23 International Business Machines Corporation Nondestructive interception of secure data in transit
US8484220B2 (en) * 2007-03-06 2013-07-09 Mcafee, Inc. Clustered index with differentiated subfields
US7962638B2 (en) * 2007-03-26 2011-06-14 International Business Machines Corporation Data stream filters and plug-ins for storage managers
US7779048B2 (en) 2007-04-13 2010-08-17 Isilon Systems, Inc. Systems and methods of providing possible value ranges
US8966080B2 (en) 2007-04-13 2015-02-24 Emc Corporation Systems and methods of managing resource utilization on a threaded computer system
US7900015B2 (en) 2007-04-13 2011-03-01 Isilon Systems, Inc. Systems and methods of quota accounting
JP4073033B1 (en) * 2007-04-27 2008-04-09 透 降矢 A database query processing system using multi-operation processing using composite relational operations that considers improvement of processing functions of join operations
US7716192B2 (en) * 2007-05-08 2010-05-11 Microsoft Corporation Concurrent, lock-free object copying
US20080288498A1 (en) * 2007-05-14 2008-11-20 Hinshaw Foster D Network-attached storage devices
US7933894B2 (en) * 2007-06-15 2011-04-26 Microsoft Corporation Parameter-sensitive plans for structural scenarios
US7676461B2 (en) * 2007-07-18 2010-03-09 Microsoft Corporation Implementation of stream algebra over class instances
US7840556B1 (en) * 2007-07-31 2010-11-23 Hewlett-Packard Development Company, L.P. Managing performance of a database query
US8055653B2 (en) * 2007-08-09 2011-11-08 International Business Machines Corporation Processing overlapping continuous queries
US7882068B2 (en) 2007-08-21 2011-02-01 Isilon Systems, Inc. Systems and methods for adaptive copy on write
US7966289B2 (en) 2007-08-21 2011-06-21 Emc Corporation Systems and methods for reading objects in a file system
US7949692B2 (en) 2007-08-21 2011-05-24 Emc Corporation Systems and methods for portals into snapshot data
US7966343B2 (en) 2008-04-07 2011-06-21 Teradata Us, Inc. Accessing data in a column store database based on hardware compatible data structures
US9424315B2 (en) 2007-08-27 2016-08-23 Teradata Us, Inc. Methods and systems for run-time scheduling database operations that are executed in hardware
US8458129B2 (en) 2008-06-23 2013-06-04 Teradata Us, Inc. Methods and systems for real-time continuous updates
US8862625B2 (en) 2008-04-07 2014-10-14 Teradata Us, Inc. Accessing data in a column store database based on hardware compatible indexing and replicated reordered columns
US8108411B2 (en) * 2007-09-14 2012-01-31 Yahoo! Inc. Methods and systems for merging data sets
US8700608B2 (en) * 2007-10-17 2014-04-15 Oracle International Corporation SQL execution plan verification
US9003054B2 (en) * 2007-10-25 2015-04-07 Microsoft Technology Licensing, Llc Compressing null columns in rows of the tabular data stream protocol
US7917462B1 (en) * 2007-11-09 2011-03-29 Teradata Us, Inc. Materializing subsets of a multi-dimensional table
US8392461B2 (en) * 2007-11-19 2013-03-05 Teradota Us, Inc. Virtual data maintenance
US8156084B2 (en) * 2008-01-17 2012-04-10 International Business Machines Corporation Transfer of data from positional data sources to partitioned databases in restartable environments
US8521682B2 (en) * 2008-01-17 2013-08-27 International Business Machines Corporation Transfer of data from transactional data sources to partitioned databases in restartable environments
US9251012B2 (en) * 2008-01-18 2016-02-02 Tivo Inc. Distributed backup and retrieval system
US20100198830A1 (en) * 2008-03-06 2010-08-05 Nitrosecurity, Inc. Dynamic data distribution aggregation
US8392397B2 (en) 2008-03-12 2013-03-05 International Business Machines Corporation System and method for provenance function window optimization
US7870345B2 (en) 2008-03-27 2011-01-11 Isilon Systems, Inc. Systems and methods for managing stalled storage devices
US7984324B2 (en) * 2008-03-27 2011-07-19 Emc Corporation Systems and methods for managing stalled storage devices
US7949636B2 (en) 2008-03-27 2011-05-24 Emc Corporation Systems and methods for a read only mode for a portion of a storage system
US7953709B2 (en) * 2008-03-27 2011-05-31 Emc Corporation Systems and methods for a read only mode for a portion of a storage system
US7904460B2 (en) * 2008-04-23 2011-03-08 Microsoft Corporation Determining computer information from processor properties
US8261326B2 (en) 2008-04-25 2012-09-04 International Business Machines Corporation Network intrusion blocking security overlay
WO2009139024A1 (en) * 2008-05-15 2009-11-19 富士通株式会社 Network for mutually connecting computers
US8301626B2 (en) 2008-05-22 2012-10-30 International Business Machines Corporation Method and apparatus for maintaining and processing provenance data in data stream processing system
FR2931613B1 (en) * 2008-05-22 2010-08-20 Inst Nat Rech Inf Automat DEVICE AND METHOD FOR INTEGRITY VERIFICATION OF PHYSICAL OBJECTS
US8775344B2 (en) * 2008-05-22 2014-07-08 International Business Machines Corporation Determining and validating provenance data in data stream processing system
US8443363B1 (en) * 2008-05-30 2013-05-14 Symantec Corporation Coordinated virtualization activities
US9720729B2 (en) * 2008-06-02 2017-08-01 Microsoft Technology Licensing, Llc Scheduler finalization
US8037024B1 (en) * 2008-06-04 2011-10-11 Google Inc. Data propagation in a multi-shard database system
US8812614B2 (en) * 2008-06-05 2014-08-19 Qualcomm Incorporated Data backup for a mobile computing device
WO2009153687A1 (en) * 2008-06-18 2009-12-23 Petascan Ltd Distributed hardware-based data querying
US9418107B2 (en) * 2008-07-30 2016-08-16 At&T Intellectual Property I, L.P. Method and apparatus for performing query aware partitioning
US8676749B2 (en) * 2008-07-31 2014-03-18 Sybase, Inc. Statement logging in databases
US9477727B2 (en) 2008-08-01 2016-10-25 Sybase, Inc. Abstracting data for use by a mobile device having occasional connectivity
US8434093B2 (en) 2008-08-07 2013-04-30 Code Systems Corporation Method and system for virtualization of software applications
US8776038B2 (en) 2008-08-07 2014-07-08 Code Systems Corporation Method and system for configuration of virtualized software applications
US20100088325A1 (en) 2008-10-07 2010-04-08 Microsoft Corporation Streaming Queries
US9047124B2 (en) * 2008-10-14 2015-06-02 Hewlett-Packard Development Company, L.P. Query scheduler
US9043353B2 (en) * 2008-10-14 2015-05-26 Hewlett-Packard Development Company, L.P. Query stream execution using priority gradient multiprogramming
US8365174B2 (en) * 2008-10-14 2013-01-29 Chetan Kumar Gupta System and method for modifying scheduling of queries in response to the balancing average stretch and maximum stretch of scheduled queries
US8204859B2 (en) 2008-12-10 2012-06-19 Commvault Systems, Inc. Systems and methods for managing replicated database data
US9495382B2 (en) 2008-12-10 2016-11-15 Commvault Systems, Inc. Systems and methods for performing discrete data replication
US9514434B2 (en) * 2009-01-06 2016-12-06 The Boeing Company Apparatus and method for automatic work instruction generation
US8412880B2 (en) * 2009-01-08 2013-04-02 Micron Technology, Inc. Memory system controller to manage wear leveling across a plurality of storage nodes
US8352710B2 (en) * 2009-01-19 2013-01-08 International Business Machines Corporation Off-loading of processing from a processor blade to storage blades
US8832173B2 (en) * 2009-01-20 2014-09-09 Sap Ag System and method of multithreaded processing across multiple servers
US8214687B2 (en) * 2009-02-13 2012-07-03 International Business Machines Corporation Disaster recovery based on journaling events prioritization in information technology environments
US8566286B1 (en) 2009-05-15 2013-10-22 Idera, Inc. System and method for high speed database backup using rapidly adjusted dynamic compression ratios controlled by a feedback loop
US8239373B2 (en) * 2009-05-18 2012-08-07 Oracle International Corporation Efficient way to evaluate uncorrelated path-based row sources with XML storage
US8301620B2 (en) * 2009-05-28 2012-10-30 Oracle International Corporation Efficient way to evaluate aggregations on XML data using path-based row sources
US8745031B2 (en) * 2009-05-28 2014-06-03 Oracle International Corporation Cache-based predicate handling for queries on XML data using uncorrelated path-based row sources
US20100306005A1 (en) * 2009-05-29 2010-12-02 Perceptive Software, Inc. Workflow Management System and Method
US9081616B2 (en) * 2009-05-29 2015-07-14 Lexmark International Technology, SA System and method for adjusting a number of processing modules based on processing load
US8161035B2 (en) * 2009-06-04 2012-04-17 Oracle International Corporation Query optimization by specifying path-based predicate evaluation in a path-based query operator
JP5395565B2 (en) * 2009-08-12 2014-01-22 株式会社日立製作所 Stream data processing method and apparatus
US9158816B2 (en) 2009-10-21 2015-10-13 Microsoft Technology Licensing, Llc Event processing with XML query based on reusable XML query template
US9940670B2 (en) * 2009-12-10 2018-04-10 Royal Bank Of Canada Synchronized processing of data by networked computing resources
KR101608495B1 (en) * 2009-12-11 2016-04-01 삼성전자주식회사 Apparatus and Method for processing data stream
US8032673B2 (en) * 2009-12-14 2011-10-04 Apple Inc. Transaction ID filtering for buffered programmed input/output (PIO) write acknowledgements
US8683498B2 (en) * 2009-12-16 2014-03-25 Ebay Inc. Systems and methods for facilitating call request aggregation over a network
US9336291B2 (en) * 2009-12-30 2016-05-10 Sybase, Inc. Message based synchronization for mobile business objects
US8788458B2 (en) 2009-12-30 2014-07-22 Sybase, Inc. Data caching for mobile applications
US8909662B2 (en) * 2009-12-30 2014-12-09 Sybase, Inc. Message based mobile object with native PIM integration
US8954958B2 (en) 2010-01-11 2015-02-10 Code Systems Corporation Method of configuring a virtual application
US8645377B2 (en) * 2010-01-15 2014-02-04 Microsoft Corporation Aggregating data from a work queue
US9104517B2 (en) 2010-01-27 2015-08-11 Code Systems Corporation System for downloading and executing a virtual application
US8959183B2 (en) 2010-01-27 2015-02-17 Code Systems Corporation System for downloading and executing a virtual application
US9229748B2 (en) 2010-01-29 2016-01-05 Code Systems Corporation Method and system for improving startup performance and interoperability of a virtual application
US8504515B2 (en) 2010-03-30 2013-08-06 Commvault Systems, Inc. Stubbing systems and methods in a data replication environment
US9081888B2 (en) 2010-03-31 2015-07-14 Cloudera, Inc. Collecting and aggregating log data with fault tolerance
US8874526B2 (en) 2010-03-31 2014-10-28 Cloudera, Inc. Dynamically processing an event using an extensible data model
US9317572B2 (en) 2010-03-31 2016-04-19 Cloudera, Inc. Configuring a system to collect and aggregate datasets
US9116946B2 (en) * 2010-04-02 2015-08-25 Scalebase Inc. System and method for interacting with a plurality of data sources
US8763009B2 (en) 2010-04-17 2014-06-24 Code Systems Corporation Method of hosting a first application in a second application
US8725592B2 (en) 2010-11-18 2014-05-13 Wal-Mart Stores, Inc. Method, system, and medium for recommending gift products based on textual information of a selected user
US8539102B2 (en) * 2010-06-21 2013-09-17 Comcast Cable Communications, Llc Downloading a code image to remote devices
US8468151B2 (en) * 2010-06-29 2013-06-18 Teradata Us, Inc. Methods and systems for hardware acceleration of database operations and queries based on multiple hardware accelerators
US9218359B2 (en) 2010-07-02 2015-12-22 Code Systems Corporation Method and system for profiling virtual application resource utilization patterns by executing virtualized application
KR101139610B1 (en) * 2010-07-29 2012-04-27 한국수력원자력 주식회사 System and method for synchronized interfaces between engineering analysis programs using databased shared memory
US8566376B2 (en) 2010-07-30 2013-10-22 Chevron U.S.A. Inc. System and method for data compression using a field programmable gate array
US9021015B2 (en) 2010-10-18 2015-04-28 Code Systems Corporation Method and system for publishing virtual applications to a web server
US20120036146A1 (en) * 2010-10-26 2012-02-09 ParElastic Corporation Apparatus for elastic database processing with heterogeneous data
US20120109981A1 (en) * 2010-10-28 2012-05-03 Goetz Graefe Generating progressive query results
US9209976B2 (en) 2010-10-29 2015-12-08 Code Systems Corporation Method and system for restricting execution of virtual applications to a managed process environment
WO2012068561A2 (en) * 2010-11-18 2012-05-24 Wal-Mart Stores, Inc. Processing data feeds
JP6045505B2 (en) 2010-12-09 2016-12-14 アイピー レザボア, エルエルシー.IP Reservoir, LLC. Method and apparatus for managing orders in a financial market
US10102242B2 (en) 2010-12-21 2018-10-16 Sybase, Inc. Bulk initial download of mobile databases
US10019503B2 (en) * 2010-12-22 2018-07-10 Microsoft Technology Licensing, Llc Database transfers using constraint free data
US8892569B2 (en) 2010-12-23 2014-11-18 Ianywhere Solutions, Inc. Indexing spatial data with a quadtree index having cost-based query decomposition
US8892503B1 (en) 2011-01-19 2014-11-18 Accenture-Global Services Limited Journaling tool
US8572274B2 (en) * 2011-02-17 2013-10-29 International Business Machines Corporation Estimating load shed data in streaming database applications
US8589403B2 (en) * 2011-02-28 2013-11-19 Splunk Inc. Compressed journaling in event tracking files for metadata recovery and replication
US20120323971A1 (en) * 2011-06-14 2012-12-20 Sybase, Inc. Optimizing data storage and access of an in-memory database
US9767214B2 (en) * 2011-06-29 2017-09-19 Oracle International Corporation Technique and framework to provide diagnosability for XML query/DML rewrite and XML index selection
US9020969B2 (en) * 2011-07-13 2015-04-28 Sap Se Tracking queries and retrieved results
US8959313B2 (en) 2011-07-26 2015-02-17 International Business Machines Corporation Using predictive determinism within a streaming environment
US8560526B2 (en) 2011-07-26 2013-10-15 International Business Machines Corporation Management system for processing streaming data
US9148495B2 (en) 2011-07-26 2015-09-29 International Business Machines Corporation Dynamic runtime choosing of processing communication methods
US8990452B2 (en) 2011-07-26 2015-03-24 International Business Machines Corporation Dynamic reduction of stream backpressure
US20130067346A1 (en) * 2011-09-09 2013-03-14 Microsoft Corporation Content User Experience
US9747334B2 (en) * 2011-09-30 2017-08-29 Teradata Us, Inc. Managing excess capacity of database systems in a capacity controlled computing environment
US8650204B2 (en) 2011-12-19 2014-02-11 Oracle International Corporation Techniques for efficiently supporting XQuery update facility in SQL/XML
US8949429B1 (en) * 2011-12-23 2015-02-03 Amazon Technologies, Inc. Client-managed hierarchical resource allocation
US8577840B2 (en) * 2012-01-03 2013-11-05 International Business Machines Corporation Replication of data sets
US9213735B1 (en) * 2012-01-25 2015-12-15 Google Inc. Flow control in very large query result sets using a release message to confirm that a client computer is ready to receive the data associated with a data collection operation
US9405553B2 (en) 2012-01-30 2016-08-02 International Business Machines Corporation Processing element management in a streaming data system
US9172608B2 (en) 2012-02-07 2015-10-27 Cloudera, Inc. Centralized configuration and monitoring of a distributed computing cluster
US8825811B2 (en) * 2012-03-15 2014-09-02 International Business Machines Corporation Connection management and optimization for services delivered over networks
US8904014B2 (en) 2012-03-15 2014-12-02 International Business Machines Corporation Content delivery mechanisms for multicast communication
US8862564B2 (en) * 2012-03-16 2014-10-14 Google Inc. Sponsoring resource downloads
US9055069B2 (en) * 2012-03-19 2015-06-09 Xcelemor, Inc. Hardware computing system with software mediation and method of operation thereof
US11436672B2 (en) 2012-03-27 2022-09-06 Exegy Incorporated Intelligent switch for processing financial market data
US20130262433A1 (en) * 2012-04-02 2013-10-03 James Qiu Filtering a Distributed Join
US8904451B2 (en) * 2012-04-13 2014-12-02 Theplatform, Llc Systems for prioritizing video processing events based on availability of media file and agent to process the event type
US9146775B2 (en) 2012-04-26 2015-09-29 International Business Machines Corporation Operator graph changes in response to dynamic connections in stream computing applications
US20130318276A1 (en) * 2012-05-22 2013-11-28 Xockets IP, LLC Offloading of computation for rack level servers and corresponding methods and systems
US20130318280A1 (en) 2012-05-22 2013-11-28 Xockets IP, LLC Offloading of computation for rack level servers and corresponding methods and systems
US8874682B2 (en) 2012-05-23 2014-10-28 Sybase, Inc. Composite graph cache management
US9110807B2 (en) 2012-05-23 2015-08-18 Sybase, Inc. Cache conflict detection
US8805850B2 (en) 2012-05-23 2014-08-12 International Business Machines Corporation Hardware-accelerated relational joins
US9087095B2 (en) 2012-06-21 2015-07-21 International Business Machines Corporation Processing columns in a database accelerator while preserving row-based architecture
US8977637B2 (en) * 2012-08-30 2015-03-10 International Business Machines Corporation Facilitating field programmable gate array accelerations of database functions
US10922331B2 (en) 2012-09-28 2021-02-16 Oracle International Corporation Cloning a pluggable database in read-write mode
US10635674B2 (en) 2012-09-28 2020-04-28 Oracle International Corporation Migrating a pluggable database between database server instances with minimal impact to performance
US9930081B2 (en) 2012-11-13 2018-03-27 International Business Machines Corporation Streams optional execution paths depending upon data rates
GB2508603A (en) 2012-12-04 2014-06-11 Ibm Optimizing the order of execution of multiple join operations
US9250954B2 (en) 2013-01-17 2016-02-02 Xockets, Inc. Offload processor modules for connection to system memory, and corresponding methods and systems
US9378161B1 (en) 2013-01-17 2016-06-28 Xockets, Inc. Full bandwidth packet handling with server systems including offload processors
GB2511072A (en) * 2013-02-22 2014-08-27 Ibm Non-deterministic finite state machine module for use in a regular expression matching system
US9165078B2 (en) * 2013-03-14 2015-10-20 International Business Machines Corporation Row-based data filtering at a database level
US11061965B1 (en) * 2013-03-15 2021-07-13 Teradata Us, Inc. Stored procedure execution in a distributed database system
US10268724B2 (en) * 2013-03-15 2019-04-23 Teradata Us, Inc. Techniques for improving the performance of complex queries
US9147010B2 (en) * 2013-04-17 2015-09-29 International Business Machines Corporation Reconfiguring an operator graph based on attribute usage
US9135203B2 (en) 2013-05-01 2015-09-15 Jonathan Glickman Computer system and a computer device
US9317472B2 (en) 2013-06-07 2016-04-19 International Business Machines Corporation Processing element data sharing
US9282051B2 (en) * 2013-06-26 2016-03-08 Netronome Systems, Inc. Credit-based resource allocator circuit
US9235564B2 (en) 2013-07-19 2016-01-12 International Business Machines Corporation Offloading projection of fixed and variable length database columns
US9275168B2 (en) * 2013-07-19 2016-03-01 International Business Machines Corporation Hardware projection of fixed and variable length columns of database tables
DE102013109978A1 (en) * 2013-09-11 2015-03-12 Technische Universität Dresden Method and apparatus for pre-selecting, filtering and distributing data in database management systems
US9703925B1 (en) * 2013-10-22 2017-07-11 Pivotal Software, Inc. Rapid processing of biological sequence data
US20150120697A1 (en) 2013-10-28 2015-04-30 Scalebase Inc. System and method for analysis of a database proxy
US9608664B2 (en) 2013-12-30 2017-03-28 International Business Machines Corporation Compression of integer data using a common divisor
CN104793998B (en) * 2014-01-20 2019-04-16 中兴通讯股份有限公司 Terminal system resource management method and device
US10303702B2 (en) 2014-02-07 2019-05-28 Ignite Scalarc Solutions, Inc. System and method for analysis and management of data distribution in a distributed database environment
KR102214511B1 (en) 2014-02-17 2021-02-09 삼성전자 주식회사 Data storage device for filtering page using 2-steps, system having the same, and operation method thereof
US9576039B2 (en) 2014-02-19 2017-02-21 Snowflake Computing Inc. Resource provisioning systems and methods
CN113268314A (en) 2014-03-07 2021-08-17 卡皮塔罗技斯Ip所有者有限责任公司 Safe intelligent network system
US10846257B2 (en) * 2014-04-01 2020-11-24 Endance Technology Limited Intelligent load balancing and high speed intelligent network recorders
US9628107B2 (en) 2014-04-07 2017-04-18 International Business Machines Corporation Compression of floating-point data by identifying a previous loss of precision
US9747335B2 (en) * 2014-06-24 2017-08-29 Sap Se Generic operator framework
US9658782B2 (en) 2014-07-30 2017-05-23 Excelero Storage Ltd. Scalable data using RDMA and MMIO
US10936200B2 (en) 2014-07-30 2021-03-02 Excelero Storage Ltd. System and method for improved RDMA techniques for multi-host network interface controllers
US10237347B2 (en) 2015-06-08 2019-03-19 Excelero Storage Ltd. System and method for providing a client device seamless access to a plurality of remote storage devices presented as a virtual device
US10979503B2 (en) 2014-07-30 2021-04-13 Excelero Storage Ltd. System and method for improved storage access in multi core system
US9350384B2 (en) 2014-09-30 2016-05-24 International Business Machines Corporation Hierarchical data compression and computation
US10679140B2 (en) * 2014-10-06 2020-06-09 Seagate Technology Llc Dynamically modifying a boundary of a deep learning network
WO2016060422A1 (en) * 2014-10-12 2016-04-21 엘지전자 주식회사 Broadcast signal transmission device, broadcast signal reception device, broadcast signal transmission method, and broadcast signal reception method
US9959299B2 (en) 2014-12-02 2018-05-01 International Business Machines Corporation Compression-aware partial sort of streaming columnar data
US10909078B2 (en) 2015-02-25 2021-02-02 International Business Machines Corporation Query predicate evaluation and computation for hierarchically compressed data
US10585887B2 (en) 2015-03-30 2020-03-10 Oracle International Corporation Multi-system query execution plan
US10255206B2 (en) 2015-05-02 2019-04-09 Hewlett Packard Enterprise Development Lp Storage memory direct access
US20170010885A1 (en) * 2015-07-07 2017-01-12 Yi-Che Chen Wireless Data Update System and Method
US9836528B1 (en) 2015-07-20 2017-12-05 Google Inc. Data constrained resource access
US10740311B2 (en) * 2015-09-14 2020-08-11 Sap Se Asynchronous index loading for database computing system startup latency managment
US10733209B2 (en) 2015-09-30 2020-08-04 International Business Machines Corporation Smart tuple dynamic grouping of tuples
US10657135B2 (en) 2015-09-30 2020-05-19 International Business Machines Corporation Smart tuple resource estimation
US10558670B2 (en) 2015-09-30 2020-02-11 International Business Machines Corporation Smart tuple condition-based operation performance
US10296620B2 (en) 2015-09-30 2019-05-21 International Business Machines Corporation Smart tuple stream alteration
US10614063B2 (en) * 2015-10-01 2020-04-07 Microsoft Technology Licensing, Llc. Streaming records from parallel batched database access
US10789131B2 (en) 2015-10-23 2020-09-29 Oracle International Corporation Transportable backups for pluggable database relocation
US10380070B2 (en) * 2015-11-12 2019-08-13 International Business Machines Corporation Reading and writing a header and record on tape
US9948620B2 (en) 2015-12-15 2018-04-17 International Business Machines Corporation Management of encryption within processing elements
US11449365B2 (en) * 2016-01-04 2022-09-20 Trilio Data Inc. Ubiquitous and elastic workload orchestration architecture of hybrid applications/services on hybrid cloud
WO2017148503A1 (en) * 2016-02-29 2017-09-08 Nec Europe Ltd. Method and stream processing system for managing data stream processing tasks of a predefined application topology
US10523724B2 (en) 2016-05-10 2019-12-31 International Business Machines Corporation Dynamic stream operator fission and fusion with platform management hints
CN106874320A (en) 2016-06-20 2017-06-20 阿里巴巴集团控股有限公司 The method and apparatus of distributive type data processing
US10423800B2 (en) 2016-07-01 2019-09-24 Capitalogix Ip Owner, Llc Secure intelligent networked architecture, processing and execution
US10776846B2 (en) * 2016-07-27 2020-09-15 Nike, Inc. Assortment optimization
US10649950B2 (en) 2016-08-29 2020-05-12 Excelero Storage Ltd. Disk access operation recovery techniques
US20180060315A1 (en) * 2016-08-31 2018-03-01 International Business Machines Corporation Performing file system maintenance
US10613882B2 (en) * 2016-10-14 2020-04-07 Seagate Technology Llc Active drive API
US10585890B2 (en) * 2016-10-27 2020-03-10 International Business Machines Corporation Generation of query execution plans
EP3560135A4 (en) 2016-12-22 2020-08-05 IP Reservoir, LLC Pipelines for hardware-accelerated machine learning
US10387679B2 (en) * 2017-01-06 2019-08-20 Capitalogix Ip Owner, Llc Secure intelligent networked architecture with dynamic feedback
CN108319623B (en) * 2017-01-18 2021-10-22 华为技术有限公司 Data redistribution method and device and database cluster
US20200133732A1 (en) * 2017-04-03 2020-04-30 Ocient Inc. Coordinating main memory access of a plurality of sets of threads
CN107025362B (en) * 2017-04-28 2020-12-22 无锡市同步电子科技有限公司 Method for checking consistency of schematic diagram and PCB production data
US10528400B2 (en) * 2017-06-05 2020-01-07 International Business Machines Corporation Detecting deadlock in a cluster environment using big data analytics
US10313495B1 (en) 2017-07-09 2019-06-04 Barefoot Networks, Inc. Compiler and hardware interactions to remove action dependencies in the data plane of a network forwarding element
US10681120B2 (en) * 2017-07-25 2020-06-09 Uber Technologies, Inc. Load balancing sticky session routing
US11386058B2 (en) * 2017-09-29 2022-07-12 Oracle International Corporation Rule-based autonomous database cloud service framework
US11327932B2 (en) 2017-09-30 2022-05-10 Oracle International Corporation Autonomous multitenant database cloud service framework
EP3483673B1 (en) * 2017-11-14 2023-06-14 TTTech Auto AG Method and computer system to consistently control a set of actuators
US11048475B2 (en) * 2017-11-30 2021-06-29 International Business Machines Corporation Multi-cycle key compares for keys and records of variable length
US10936283B2 (en) 2017-11-30 2021-03-02 International Business Machines Corporation Buffer size optimization in a hierarchical structure
JP6668309B2 (en) * 2017-11-30 2020-03-18 株式会社日立製作所 Storage system and control method thereof
US11354094B2 (en) 2017-11-30 2022-06-07 International Business Machines Corporation Hierarchical sort/merge structure using a request pipe
US10896022B2 (en) 2017-11-30 2021-01-19 International Business Machines Corporation Sorting using pipelined compare units
US11032392B1 (en) * 2019-03-21 2021-06-08 Amazon Technologies, Inc. Including prior request performance information in requests to schedule subsequent request performance
KR102224771B1 (en) 2019-04-24 2021-03-08 주식회사 티맥스티베로 Method to manage database failure
US11188494B2 (en) 2019-05-21 2021-11-30 Google Llc Asymmetric data communication for host-device interface
US11455303B2 (en) * 2019-07-19 2022-09-27 Oracle International Corporation Driving massive scale out through rewrites of analytical functions
US11042318B2 (en) 2019-07-29 2021-06-22 Commvault Systems, Inc. Block-level data replication
US11314555B2 (en) * 2019-08-26 2022-04-26 International Business Machines Corporation Synchronous re-execution of a data transformation operation to obtain further details regarding an exception
CN112711602B (en) * 2019-10-25 2023-04-28 金篆信科有限责任公司 Method and device for running stored procedure, database system and storage medium
US11271992B2 (en) * 2020-01-22 2022-03-08 EMC IP Holding Company LLC Lazy lock queue reduction for cluster group changes
US20220374424A1 (en) * 2021-05-24 2022-11-24 Molecula Corp. Join queries in data virtualization-based architecture
US11615092B1 (en) * 2021-10-29 2023-03-28 Snowflake Inc. Lightweight database pipeline scheduler
US11809285B2 (en) 2022-02-09 2023-11-07 Commvault Systems, Inc. Protecting a management database of a data storage management system to meet a recovery point objective (RPO)

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5822527A (en) * 1990-05-04 1998-10-13 Digital Equipment Corporation Method and apparatus for information stream filtration using tagged information access and action registration
US5835755A (en) * 1994-04-04 1998-11-10 At&T Global Information Solutions Company Multi-processor computer system for operating parallel client/server database processes
US20030009531A1 (en) * 2001-06-07 2003-01-09 Richmann James N. Data Caching
US6507834B1 (en) * 1999-12-22 2003-01-14 Ncr Corporation Method and apparatus for parallel execution of SQL from stored procedures
US6594651B2 (en) * 1999-12-22 2003-07-15 Ncr Corporation Method and apparatus for parallel execution of SQL-from within user defined functions
US6601064B1 (en) * 1999-11-24 2003-07-29 Ncr Corporation Method and apparatus for fetching array based objects by direct delivery and batching
US20030212668A1 (en) * 2002-05-13 2003-11-13 Hinshaw Foster D. Optimized database appliance

Family Cites Families (132)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPS5144850A (en) * 1974-10-15 1976-04-16 Ricoh Kk
US4594655A (en) * 1983-03-14 1986-06-10 International Business Machines Corporation (k)-Instructions-at-a-time pipelined processor for parallel execution of inherently sequential instructions
FR2561428B1 (en) 1984-03-16 1986-09-12 Bull Sa DISC MEMORY RECORDING METHOD AND DISC MEMORY SYSTEM
US4829427A (en) 1984-05-25 1989-05-09 Data General Corporation Database query code generation and optimization based on the cost of alternate access methods
US4811214A (en) * 1986-11-14 1989-03-07 Princeton University Multinode reconfigurable pipeline computer
JP2760794B2 (en) 1988-01-29 1998-06-04 株式会社日立製作所 Database processing method and apparatus
US5544347A (en) 1990-09-24 1996-08-06 Emc Corporation Data storage system controlled remote data mirroring with respectively maintained data indices
US5321813A (en) 1991-05-01 1994-06-14 Teradata Corporation Reconfigurable, fault tolerant, multistage interconnect network and protocol
US5701480A (en) * 1991-10-17 1997-12-23 Digital Equipment Corporation Distributed multi-version commitment ordering protocols for guaranteeing serializability during transaction processing
WO1993020511A1 (en) 1992-03-31 1993-10-14 Aggregate Computing, Inc. An integrated remote execution system for a heterogenous computer network environment
US5301317A (en) 1992-04-27 1994-04-05 International Business Machines Corporation System for adapting query optimization effort to expected execution time
US5412806A (en) 1992-08-20 1995-05-02 Hewlett-Packard Company Calibration of logical cost formulae for queries in a heterogeneous DBMS using synthetic database
US5845113A (en) 1992-10-27 1998-12-01 International Business Machines Corporation Method for external sorting in shared-nothing parallel architectures
US5495601A (en) 1992-12-11 1996-02-27 International Business Machines Corporation Method to off-load host-based DBMS predicate evaluation to a disk controller
JP3266351B2 (en) 1993-01-20 2002-03-18 株式会社日立製作所 Database management system and query processing method
US5485607A (en) 1993-02-05 1996-01-16 Digital Equipment Corporation Concurrency-control method and apparatus in a database management system utilizing key-valued locking
US5581461A (en) 1993-02-08 1996-12-03 Itt Sheraton Corporation Computerized system and method for storage, processing and transfer of inventory and other data among a central processor/database and a number of remote locations
US5364845C1 (en) * 1993-03-31 2002-09-10 Nutramax Lab Inc Glusosamine chondroitin and manganese composition for the protection and repair of connective tissue
WO1995004968A1 (en) 1993-08-03 1995-02-16 Forte Software, Inc. Flexible multi-platform partitioning for computer applications
GB2297180B (en) 1993-09-27 1998-05-20 Oracle Corp Method and apparatus for parallel processing in a database system
US5742806A (en) 1994-01-31 1998-04-21 Sun Microsystems, Inc. Apparatus and method for decomposing database queries for database management system including multiprocessor digital data processing system
US5687363A (en) 1994-03-30 1997-11-11 Siemens Stromberg-Carlson Distributed database architecture and distributed database management system for open network evolution
US5721909A (en) * 1994-03-30 1998-02-24 Siemens Stromberg-Carlson Distributed database architecture and distributed database management system for open network evolution
JP3085085B2 (en) 1994-05-09 2000-09-04 三菱電機株式会社 Data access device and distributed database system
US5537533A (en) 1994-08-11 1996-07-16 Miralink Corporation System and method for remote mirroring of digital data from a primary network server to a remote network server
JPH08221445A (en) 1995-02-20 1996-08-30 Hitachi Ltd Data retrieval method
US5634053A (en) 1995-08-29 1997-05-27 Hughes Aircraft Company Federated information management (FIM) system and method for providing data site filtering and translation for heterogeneous databases
US5875334A (en) 1995-10-27 1999-02-23 International Business Machines Corporation System, method, and program for extending a SQL compiler for handling control statements packaged with SQL query statements
US5937415A (en) * 1995-12-13 1999-08-10 Sybase, Inc. Data base development system with methods facilitating copying of data from one data source to another
US6493761B1 (en) * 1995-12-20 2002-12-10 Nb Networks Systems and methods for data processing using a protocol parsing engine
US5781897A (en) * 1996-04-18 1998-07-14 International Business Machines Corporation Method and system for performing record searches in a database within a computer peripheral storage device
US5701460A (en) * 1996-05-23 1997-12-23 Microsoft Corporation Intelligent joining system for a relational database
US6026379A (en) 1996-06-17 2000-02-15 Verifone, Inc. System, method and article of manufacture for managing transactions in a high availability system
US5884288A (en) * 1996-07-01 1999-03-16 Sun Microsystems, Inc. Method and system for electronic bill payment
EP0910828B1 (en) 1996-07-09 2001-09-26 BRITISH TELECOMMUNICATIONS public limited company Storage system and storage method for distributed storage
US7680879B2 (en) * 1996-07-18 2010-03-16 Computer Associates Think, Inc. Method and apparatus for maintaining data integrity across distributed computer systems
US6175835B1 (en) 1996-07-26 2001-01-16 Ori Software Development, Ltd. Layered index with a basic unbalanced partitioned index that allows a balanced structure of blocks
IL118959A (en) 1996-07-26 1999-07-14 Ori Software Dev Ltd Database apparatus
US6886035B2 (en) * 1996-08-02 2005-04-26 Hewlett-Packard Development Company, L.P. Dynamic load balancing of a network of client and server computer
US5774660A (en) * 1996-08-05 1998-06-30 Resonate, Inc. World-wide-web server with delayed resource-binding for resource-based load balancing on a distributed resource multi-node network
US6298386B1 (en) 1996-08-14 2001-10-02 Emc Corporation Network file server having a message collector queue for connection and connectionless oriented protocols
JPH1074159A (en) * 1996-08-30 1998-03-17 Hitachi Ltd Method for controlling computer system
US5870739A (en) 1996-09-20 1999-02-09 Novell, Inc. Hybrid query apparatus and method
WO1998020812A1 (en) * 1996-11-15 1998-05-22 Cook Incorporated Splittable sleeve, stent deployment device
US5835908A (en) 1996-11-19 1998-11-10 Microsoft Corporation Processing multiple database transactions in the same process to reduce process overhead and redundant retrieval from database servers
US5937401A (en) * 1996-11-27 1999-08-10 Sybase, Inc. Database system with improved methods for filtering duplicates from a tuple stream
ATE200155T1 (en) * 1996-11-28 2001-04-15 Siemens Ag METHOD FOR PARAMETERIZING A FUZZY AUTOMATIC USED TO COMPARE A MEASURING SIGNAL WITH A SAMPLE SIGNAL
US5930795A (en) 1997-01-21 1999-07-27 International Business Machines Corporation Supporting dynamic tables in SQL query compilers
US5884299A (en) * 1997-02-06 1999-03-16 Ncr Corporation Optimization of SQL queries involving aggregate expressions using a plurality of local and global aggregation operations
US6295533B2 (en) * 1997-02-25 2001-09-25 At&T Corp. System and method for accessing heterogeneous databases
WO1998038762A2 (en) 1997-02-26 1998-09-03 Siebel Systems, Inc. Determining visibility to a remote database client
US5983270A (en) * 1997-03-11 1999-11-09 Sequel Technology Corporation Method and apparatus for managing internetwork and intranetwork activity
US6073209A (en) 1997-03-31 2000-06-06 Ark Research Corporation Data storage controller providing multiple hosts with access to multiple storage subsystems
US6237001B1 (en) 1997-04-23 2001-05-22 Oracle Corporation Managing access to data in a distributed database environment
US5999937A (en) 1997-06-06 1999-12-07 Madison Information Technologies, Inc. System and method for converting data between data sets
SE510001C2 (en) 1997-07-21 1999-03-29 Ericsson Telefon Ab L M Method for storing elements in a database
US6128642A (en) 1997-07-22 2000-10-03 At&T Corporation Load balancing based on queue length, in a network of processor stations
US6006220A (en) 1997-09-30 1999-12-21 International Business Machines Corporation Determining the optimal access path for a query at execution time using an actual value for each variable in a query for estimating a filter factor
US6112281A (en) 1997-10-07 2000-08-29 Oracle Corporation I/O forwarding in a cache coherent shared disk computer system
US6427173B1 (en) 1997-10-14 2002-07-30 Alacritech, Inc. Intelligent network interfaced device and system for accelerated communication
US6253240B1 (en) 1997-10-31 2001-06-26 International Business Machines Corporation Method for producing a coherent view of storage network by a storage network manager using data storage device configuration obtained from data storage devices
US6240428B1 (en) 1997-10-31 2001-05-29 Oracle Corporation Import/export and repartitioning of partitioned objects
US6151602A (en) 1997-11-07 2000-11-21 Inprise Corporation Database system with methods providing a platform-independent self-describing data packet for transmitting information
US6374336B1 (en) 1997-12-24 2002-04-16 Avid Technology, Inc. Computer system and process for transferring multiple high bandwidth streams of data between multiple storage units and multiple applications in a scalable and reliable manner
US6415373B1 (en) 1997-12-24 2002-07-02 Avid Technology, Inc. Computer system and process for transferring multiple high bandwidth streams of data between multiple storage units and multiple applications in a scalable and reliable manner
US6260044B1 (en) 1998-02-04 2001-07-10 Nugenesis Technologies Corporation Information storage and retrieval system for storing and retrieving the visual form of information from an application in a database
US6411974B1 (en) * 1998-02-04 2002-06-25 Novell, Inc. Method to collate and extract desired contents from heterogeneous text-data streams
US6247077B1 (en) 1998-02-06 2001-06-12 Ncr Corporation Highly-scalable parallel processing computer system architecture
US6175832B1 (en) 1998-05-11 2001-01-16 International Business Machines Corporation Method, system and program product for establishing a data reporting and display communication over a network
US6389460B1 (en) 1998-05-13 2002-05-14 Compaq Computer Corporation Method and apparatus for efficient storage and retrieval of objects in and from an object storage device
US6192377B1 (en) 1998-05-13 2001-02-20 Oracle Corporation Method and apparatus for determing whether a transaction can use a version of a data item
JP2978882B1 (en) 1998-06-03 1999-11-15 三菱電機株式会社 Device-to-device data transfer device and device-to-device data transfer method
US6279011B1 (en) 1998-06-19 2001-08-21 Network Appliance, Inc. Backup and restore for heterogeneous file server environment
US6973455B1 (en) 1999-03-03 2005-12-06 Emc Corporation File server system providing direct data sharing between clients with a server acting as an arbiter and coordinator
US6209031B1 (en) 1998-07-17 2001-03-27 International Business Machines Corporation Configuring computer network operations based upon a sequence of interactive user entries into a network server computer with a one time entry of data commonly required by multiple clients
US6138118A (en) * 1998-07-30 2000-10-24 Telcordia Technologies, Inc. Method and system for reconciling concurrent streams of transactions in a database
US6272662B1 (en) 1998-08-04 2001-08-07 International Business Machines Corporation Distributed storage system using front-end and back-end locking
US20010051956A1 (en) 1998-09-29 2001-12-13 Paul Bird Global caching and sharing of sql statements in a heterogeneous application environment
US6434649B1 (en) 1998-10-14 2002-08-13 Hitachi, Ltd. Data streamer
US6542508B1 (en) * 1998-12-17 2003-04-01 Watchguard Technologies, Inc. Policy engine using stream classifier and policy binding database to associate data packet with appropriate action processor for processing without involvement of a host processor
US6311187B1 (en) 1998-12-29 2001-10-30 Sun Microsystems, Inc. Propogating updates efficiently in hierarchically structured data under a push model
US6542886B1 (en) * 1999-03-15 2003-04-01 Microsoft Corporation Sampling over joins for database systems
US6789116B1 (en) * 1999-06-30 2004-09-07 Hi/Fn, Inc. State processor for pattern matching in a network monitor device
US6339772B1 (en) * 1999-07-06 2002-01-15 Compaq Computer Corporation System and method for performing database operations on a continuous stream of tuples
US6575361B1 (en) 1999-08-19 2003-06-10 E-2 Interactive, Inc. System and method for managing stored-value card data
US7139999B2 (en) 1999-08-31 2006-11-21 Accenture Llp Development architecture framework
US6332163B1 (en) * 1999-09-01 2001-12-18 Accenture, Llp Method for providing communication services over a computer network system
US6353819B1 (en) 1999-09-29 2002-03-05 Bull Hn Information Systems Inc. Method and system for using dynamically generated code to perform record management layer functions in a relational database manager
US6351749B1 (en) 1999-09-30 2002-02-26 Ncr Corporation Multi-threading, multi-tasking architecture for a relational database management system
US6161602A (en) * 1999-10-07 2000-12-19 Wisconsin Knife Works, Inc. Unitary finger joint cutting bit and finger joint cutting head incorporating the same
US6516337B1 (en) * 1999-10-14 2003-02-04 Arcessa, Inc. Sending to a central indexing site meta data or signatures from objects on a computer network
US6571232B1 (en) 1999-11-01 2003-05-27 Sun Microsystems, Inc. System and method for browsing database schema information
US6714952B2 (en) 1999-11-10 2004-03-30 Emc Corporation Method for backup and restore of a multi-lingual network file server
US6615203B1 (en) 1999-12-17 2003-09-02 International Business Machines Corporation Method, computer program product, and system for pushdown analysis during query plan generation
US6732084B1 (en) * 1999-12-22 2004-05-04 Ncr Corporation Method and apparatus for parallel execution of trigger actions
US6985964B1 (en) * 1999-12-22 2006-01-10 Cisco Technology, Inc. Network processor system including a central processor and at least one peripheral processor
US6477540B1 (en) * 1999-12-22 2002-11-05 Ncr Corporation Method and apparatus for using Java as a stored procedure language and as an embedded language on a client
US6763390B1 (en) * 2000-01-24 2004-07-13 Ati Technologies, Inc. Method and system for receiving and framing packetized data
US6529896B1 (en) * 2000-02-17 2003-03-04 International Business Machines Corporation Method of optimizing a query having an existi subquery and a not-exists subquery
US20020095400A1 (en) * 2000-03-03 2002-07-18 Johnson Scott C Systems and methods for managing differentiated service in information management environments
WO2001069919A1 (en) * 2000-03-10 2001-09-20 Datacube, Inc. Image processing system using an array processor
US6907546B1 (en) 2000-03-27 2005-06-14 Accenture Llp Language-driven interface for an automated testing framework
US20010039497A1 (en) 2000-03-30 2001-11-08 Hubbard Edward A. System and method for monitizing network connected user bases utilizing distributed processing systems
US20020002625A1 (en) * 2000-04-17 2002-01-03 Mark Vange System and method for reformatting data traffic
US6505285B1 (en) 2000-06-26 2003-01-07 Ncr Corporation Scratch segment subsystem for a parallel processing database system
US6721749B1 (en) 2000-07-06 2004-04-13 Microsoft Corporation Populating a data warehouse using a pipeline approach
TW561360B (en) * 2000-08-22 2003-11-11 Ibm Method and system for case conversion
US7689560B2 (en) * 2000-10-13 2010-03-30 Miosoft Corporation Persistent data storage techniques
US6493701B2 (en) 2000-11-22 2002-12-10 Sybase, Inc. Database system with methodogy providing faster N-ary nested loop joins
US7711818B2 (en) 2000-12-22 2010-05-04 Oracle International Corporation Support for multiple data stores
US6829611B2 (en) 2000-12-29 2004-12-07 Bellsouth Intellectual Property Corporation Data loader application
US6731229B2 (en) * 2001-02-09 2004-05-04 International Business Machines Corporation Method to reduce storage requirements when storing semi-redundant information in a database
US7162467B2 (en) 2001-02-22 2007-01-09 Greenplum, Inc. Systems and methods for managing distributed database resources
US7418441B2 (en) * 2001-02-28 2008-08-26 International Business Machines Corporation Data loading from a remote data source record by record
US6892205B1 (en) 2001-02-28 2005-05-10 Oracle International Corporation System and method for pre-compiling a source cursor into a target library cache
US7526427B2 (en) * 2001-03-06 2009-04-28 Ati Technologies, Inc. System and method for reception, processing and transmission of digital audio stream
US7203678B1 (en) 2001-03-27 2007-04-10 Bea Systems, Inc. Reconfigurable query generation system for web browsers
GB2395041A (en) * 2001-08-14 2004-05-12 Bloomberg Lp Distribution and mapping of financial records from data stream
WO2003025709A2 (en) * 2001-09-19 2003-03-27 Bay Microsystems, Inc. Vertical instruction and data processing in a network processor architecture
US7120704B2 (en) * 2002-01-31 2006-10-10 International Business Machines Corporation Method and system for workload balancing in a network of computer systems
US20030158842A1 (en) 2002-02-21 2003-08-21 Eliezer Levy Adaptive acceleration of retrieval queries
US6957236B1 (en) 2002-05-10 2005-10-18 Oracle International Corporation Providing a useable version of a data item
US7272605B1 (en) 2002-05-13 2007-09-18 Netezza Corporation Network interface for distributed intelligence database system
US20030233632A1 (en) 2002-06-12 2003-12-18 Lockheed Martin Corporation Automatically generated client application source code using database table definitions
US6917935B2 (en) * 2002-06-26 2005-07-12 Microsoft Corporation Manipulating schematized data in a database
US7305386B2 (en) 2002-09-13 2007-12-04 Netezza Corporation Controlling visibility in multi-version database systems
EP1546822A4 (en) * 2002-09-18 2008-07-02 Netezza Corp Asymmetric data streaming architecture having autonomous and asynchronous job processing unit
WO2004027650A1 (en) 2002-09-18 2004-04-01 Netezza Corporation Disk mirror architecture for database appliance
US8005854B2 (en) 2003-03-14 2011-08-23 Sybase, Inc. System with methodology for executing relational operations over relational data and data retrieved from SOAP operations
CA2425046C (en) 2003-04-08 2009-10-06 Ibm Canada Limited - Ibm Canada Limitee Method and system for caching database query statements
WO2005006153A2 (en) 2003-07-07 2005-01-20 Netezza Corporation Sql code generation for heterogeneous environment
US8615731B2 (en) 2004-08-25 2013-12-24 Mohit Doshi System and method for automating the development of web services that incorporate business rules

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5822527A (en) * 1990-05-04 1998-10-13 Digital Equipment Corporation Method and apparatus for information stream filtration using tagged information access and action registration
US5835755A (en) * 1994-04-04 1998-11-10 At&T Global Information Solutions Company Multi-processor computer system for operating parallel client/server database processes
US6601064B1 (en) * 1999-11-24 2003-07-29 Ncr Corporation Method and apparatus for fetching array based objects by direct delivery and batching
US6507834B1 (en) * 1999-12-22 2003-01-14 Ncr Corporation Method and apparatus for parallel execution of SQL from stored procedures
US6594651B2 (en) * 1999-12-22 2003-07-15 Ncr Corporation Method and apparatus for parallel execution of SQL-from within user defined functions
US20030009531A1 (en) * 2001-06-07 2003-01-09 Richmann James N. Data Caching
US20030212668A1 (en) * 2002-05-13 2003-11-13 Hinshaw Foster D. Optimized database appliance

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP1546930A4 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9423983B2 (en) 2012-01-19 2016-08-23 Syncsort Incorporated Intelligent storage controller

Also Published As

Publication number Publication date
US7730077B2 (en) 2010-06-01
WO2004027648A1 (en) 2004-04-01
US20040139214A1 (en) 2004-07-15
EP1546928A1 (en) 2005-06-29
EP2264628A1 (en) 2010-12-22
AU2003270828A8 (en) 2004-04-08
US7698338B2 (en) 2010-04-13
AU2003273333A1 (en) 2004-04-08
US20040148420A1 (en) 2004-07-29
WO2004027648A8 (en) 2004-07-08
EP1546930A4 (en) 2008-07-02
US20040133565A1 (en) 2004-07-08
EP1546929A4 (en) 2008-07-02
EP1546822A2 (en) 2005-06-29
EP1546926A4 (en) 2008-07-02
US20040205110A1 (en) 2004-10-14
US7634477B2 (en) 2009-12-15
AU2003275181A1 (en) 2004-04-08
WO2004027649A1 (en) 2004-04-01
EP1546926A1 (en) 2005-06-29
WO2004027647A1 (en) 2004-04-01
US8880551B2 (en) 2014-11-04
EP1546929A1 (en) 2005-06-29
WO2004027576A3 (en) 2004-09-02
AU2003270828A1 (en) 2004-04-08
US20040117037A1 (en) 2004-06-17
EP1546930A1 (en) 2005-06-29
AU2003270757A1 (en) 2004-04-08
AU2003275005A1 (en) 2004-04-08
EP1546930B1 (en) 2015-10-21
US7577667B2 (en) 2009-08-18
US7529752B2 (en) 2009-05-05
US20100257537A1 (en) 2010-10-07
EP1546822A4 (en) 2008-07-02
EP1546928A4 (en) 2008-07-09
WO2004027576A2 (en) 2004-04-01
WO2004027576A9 (en) 2004-06-17

Similar Documents

Publication Publication Date Title
EP1546930B1 (en) Programmable streaming data processor for database appliance having multiple processing unit groups
To et al. A survey of state management in big data processing systems
US7702610B2 (en) Performing sequence analysis as a multipart plan storing intermediate results as a relation
US10803066B2 (en) Methods and systems for hardware acceleration of database operations and queries for a versioned database based on multiple hardware accelerators
WO2020139655A1 (en) Technique of comprehensively support autonomous json document object (ajd) cloud service
US20130297565A1 (en) Database Management System
Armenatzoglou et al. Amazon Redshift re-invented
CN102089741A (en) Executing computer-intensive database user-defined programs on an attached high-performance parallel computer
Samwel et al. F1 query: Declarative querying at scale
Armenatzoglou et al. Amazon Redshi Re-invented
Mozafari et al. SnappyData: Streaming, Transactions, and Interactive Analytics in a Unified Engine
John Active storage for database applications
Watson et al. ICL, Wenlock Way, West Gorton
Hartstein Oracle Database New Features Guide, 10g Release 1 (10.1) Part No. B10750-01 Copyright© 2001, 2003 Oracle Corporation. All rights reserved. Contributors: Mateo Burtch, Paul Lane, Diana Lorentz, Mark Townsend, Sandra Cheevers, Michael

Legal Events

Date Code Title Description
AK Designated states

Kind code of ref document: A1

Designated state(s): AE AG AL AM AT AU AZ BA BB BG BR BY BZ CA CH CN CO CR CU CZ DE DK DM DZ EC EE EG ES FI GB GD GE GH GM HR HU ID IL IN IS JP KE KG KP KR KZ LC LK LR LS LT LU LV MA MD MG MK MN MW MX MZ NI NO NZ OM PG PH PL PT RO RU SC SD SE SG SK SL SY TJ TM TN TR TT TZ UA UG US UZ VC VN YU ZA ZM ZW

AL Designated countries for regional patents

Kind code of ref document: A1

Designated state(s): GH GM KE LS MW MZ SD SL SZ TZ UG ZM ZW AM AZ BY KG KZ MD RU TJ TM AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HU IE IT LU MC NL PT RO SE SI SK TR BF BJ CF CG CI CM GA GN GQ GW ML MR NE SN TD TG

121 Ep: the epo has been informed by wipo that ep was designated in this application
WWE Wipo information: entry into national phase

Ref document number: 2003759450

Country of ref document: EP

WWP Wipo information: published in national office

Ref document number: 2003759450

Country of ref document: EP

NENP Non-entry into the national phase

Ref country code: JP

WWW Wipo information: withdrawn in national office

Country of ref document: JP