DE102005043750A1 - Distributing and finding method of meta information for management of data storage systems in computer systems, involves application of user data zone and associated security area in data storage system - Google Patents

Abstract

The method of meta information involves application of a user data zone and associated security area in data storage system (4). Safety area serves to the additional decentralized storage of meta information and its absolute condition and separation in memory sections specified on the storage system. The virtual address of the data, which is to be stored, of the logical data area is converted in the associated physical address on the data storage system consecutively with each allocation or shift of an extent of the user data zone during storing of data on the data memory system by a super ordinate instance. The additional meta information is stored in the associated safety area. All the user data stored on the data storage system and the allocation to the logical data areas can be reconstructed after a loss of the central instance for each data storage system with the help of the meta information stored in a safety area of and can be stored in a central instance. INDEPEDENT CLAIMS are also included for the following: (1) Computer program product; and (2) Computer readable storage medium.

Description

The The present invention describes methods for distribution and for Locating data in distributed storage systems such as storage networks and / or an intranet and / or the Internet according to the generic term of claim 1 and a corresponding computer program product and a corresponding computer readable storage medium.

The increasing amount of information to be backed up Computer applications has led that complexity of distributed used to store such information Data storage systems have increased steadily in recent years is. Even data storage systems such as hard drives with a capacity of several hundred GB are no longer able to all in a company or in public Institutions to secure data. Data storage systems with several hundred TByte or even PByte of information need the Coordination of hundreds or even thousands working together Data storage systems.

at The management of these data storage systems often fall large quantities so-called meta-information that provides information about the Structure of data storage systems, access rights and distribution the data over contain the data storage systems. In the following, all data, by the construction of a data storage system as a distributed Data storage system arise, referred to as meta-information. All other data to be backed up are called user data. Furthermore, as today predominantly used embodiment of data storage systems, also called storage system for short, also referred to as hard disks, where these terms are to be understood synonymously and each type of data storage. Data can be in this context Any kind of stored information, no matter how it is Represents information becomes.

On Today, meta-information is typically accessed through centralized information management accessed, but has the disadvantage that after the Failure of the central information units access to the Data storage systems can not or only conditionally done. Furthermore, the central storage of the information creates a bottleneck in accessing the meta-information, the e.g. in environments with a high number of servers too high Can cause latency in accessing the payload.

Storage Area Networks form a typical application of generic methods, in which through the use of appropriate techniques as possible high data security even after the failure of all coordinating Instances to a technically possible high level guaranteed shall be.

Storage Area Networks 1 connect like in 1 specified a lot of data storage systems 4 with a lot of computers or clients 3 , In a storage network 1 it is possible for all clients 3 Access to all storage systems 4 have, as well as that a dedicated rights allocation between the data storage systems 4 and the clients 3 is carried out.

For the purposes of the description, a number of units accessing the data blocks, the clients, will be described below 3 , and a lot of units delivering data blocks, the servers, distinguished. The terms server and data storage system 4 are used synonymously.

By the physical decoupling between the data storage systems 4 and the clients 3 For example, the data storage systems can be managed much more efficiently, previously fixed allocations between data storage systems 4 and computers are broken up.

However, this increase in degrees of freedom entails an increase in the complexity that can be met through the use of storage virtualization solutions. These storage virtualization solutions take an abstraction between the clients' view 3 to the memory and the physically available memory. So can a so-called logical drive, which is the client 3 as accessible by it, through the virtualization solution on almost all physically existing data storage systems 4 be distributed.

These virtualization solutions can be eg in the client 3 even as a driver component, in a combination between the clients connected to the network 3 and one of the several in the 2 presented metadata appliances 5 , in one or more of the in the 3 represented SAN appliances 5 between the clients 3 and the storage systems 4 or in the switches and routers 2 be implemented by the storage network.

In all cases, the virtualization solution needs information about the location of each item on the data storage systems 4 be secured. This information is currently being backed up by the interaction of the virtualization solution with one or more data storage systems 4 the meta-information is managed. In this case, it is minimally necessary for the position of a first data block to be known, after the localization of which further information can be loaded. After the failure of this data block and possibly existing copies of this master block a reconstruction of the existing data is not possible or only to a very limited extent.

In US 2004/0128587 A1 describes methods by means of which the reconstruction of data after the failure of one or more Data storage systems allows becomes. In the process, a schedule is created centrally, which is distributed. This solution requires only a limited number of data storage systems fails. The procedure does not allow without completely redundant storage of all meta-information on all data storage systems a reconstruction possible many user data even after the failure of a large part of the data storage systems perform.

The US Patent 5,537,652 deals with ways of dealing with files on a faulty data storage system, despite the To reconstruct errors on the file storage system. These errors can be e.g. Affect metainformation on the same data storage system. This is ensured by redundancy in the storage of directory entries.

In the US Pat. No. 6,718,436 B2 Extents are introduced as a storage unit to denote a population density and not to accomplish a distribution of data across the hard disk. The problem thus does not correspond to the storage virtualization, but to the simpler so-called volume managers, which often carry out an allocation (as described here) on the basis of logical partitions. The location of data on the respective data storage systems is calculated solely on the basis of corresponding formulas.

task The present invention is therefore, in addition to the continue existing, central backup of meta-information another Introduce structure of backup of meta-information by their help even after the failure of all, possibly even redundant designed central systems to ensure data security is or one as possible size Amount of information backed up on the data storage systems accessible again can be made.

The solution the task of the invention arises from the characterizing features of claim 1 in Interaction with the characteristics of the generic term. Further advantageous Embodiments of the invention will become apparent from the dependent claims.

The invention relates to a generic method for distributing and retrieving metadata for managing data storage systems in computer systems and / or storage networks and / or a data network, preferably an Internet and / or an intranet, in which the total amount of data storage systems assigned to a virtual data room and is divided into different logical data spaces, wherein the meta information about the distribution of the logical data spaces are stored in a central instance. Such a generic method is further developed according to the invention in that at least one user data area and at least one associated security area is applied to each data storage system, wherein the at least one security area for additional decentralized storage of meta information used and its absolute location and subdivision into memory sections on the Storage system are defined in advance uniquely, and subsequently at each allocation or shift of an extent of the payload area when storing data on the data storage system from a parent instance, the virtual address of the data to be stored of the logical data space into the corresponding physical address on the data storage system converted and the corresponding Meta information is stored in the associated security area, whereby, after a failure of the central instance for each data storage system using d he reconstructed in the at least one security area stored meta information all stored on the data storage system user data and the assignment to the logical data spaces and in a new can be stored in the central instance.

One Such method is particularly applicable for the reconstruction of data after the failure of parts of the infrastructure of distributed Storage systems through the distributed provision of additional information. Furthermore possible it decentralized the procedure presented here, administrative information secure and in a further embodiment, faster access to it. A peculiarity of the procedure is that it after a Failure of parts of the storage infrastructure all meta-information can reconstruct associated with still reconstructable payload become. This effect comes with a minimal amount of additional information achieved because each data storage system only meta information about the User data required on this data storage system is required. A redundant storage of meta information on multiple data storage systems for backup is not necessary.

Farther It is possible with other configurations that on a central Instance to store the meta information may be completely omitted can be and by the underlying distributed storage the metainformation a significant increase in speed at access to the metadata.

in the Below, different embodiments of the method are presented, by their use even after the failure possibly existing master blocks and / or the coordinating authorities have maximum information about the Structure of the storage network can be recovered. The Particularity of the method presented here and its preferred Embodiments is that even during the Reconstruction of the metainformation the servers by decentralized Information present in the system, both information about the Payload as well over can access the meta-information. In another embodiment the procedure is extended so that the use of a central instance even completely can be waived.

basic processing to back up the data by storing information the involved data storage systems

It is assumed that each physical partition of a data storage system is divided into several areas of different importance. Thus, the memory area of the data storage system according to 4 be divided into a signature area, a data area, a security information area and a block-level information area.

task the area with the security information is, even after the failure of all control bodies to have a chance on to reconstruct the information stored in the data storage systems and thus a rescue of the stored information in case of failure data storage systems, a so-called disaster recovery on the available data storage systems, e.g. to perform the connected hard disks.

Set up of the security area

Of the Security area is so before use as a security area initializes that not described parts of the security area not be mistakenly interpreted as information. This can be done, for example, by padding with zeros.

Around perform this reconstruction to be able to the memory area of the partitions in memory areas, so-called. Extents divided, with an extent the smallest connected unit is on the data storage system that can be assigned. Farther can Memory areas allocated only in whole multiples of these extents become. For every possible one Extent that can be stored in the data area becomes Metablock reserved in the security area. That is, the number of metablocks is proportional to the size of the disk divided by the minimum possible Size of one Extents. The minimum extent size is while e.g. to 4 MB (4096 x 1024 Byte). Furthermore, the size of an extent should be a power of two be.

An extent may only be placed on the data storage system in this preferred embodiment to start on a 4 Mbyte boundary of the subdivision of the data storage system. This minimizes the complexity of allocating the extents, even with different extents on a physical partition can give a blend. The start address of the extents on the partition is referred to below as startExtent. The size of a metablock is called metaBlockSize.

Around even after the failure of all coordinating units on the Data storage systems to be able to reconstruct saved data, the beginnings clearly distinguishable from each other. In a preferred variant, this is done by the minimum Extent size in advance is known and used in the calculation of the boundaries of different Areas are included.

The start address of the data area startDaten is here:

The start address startMetaBlock for a metablock entry is calculated as follows: startMetaBlock = Offset + (startExtent · metaBlockSize) / (4096 · 1024) bytes

The following applies to the offset address:

At This location does not attempt to minimize the size of the security area because it depends on a quick search and entry, if a new extent is requested or if an extent is moved becomes. Through the structure presented here, access to a Table access to be reduced, which are performed in constant time can.

• – Id des virtuellen Laufwerkes: 32 Bit
• – Art der Datenkodierung: 8 Bit
• – Nummer dieses Blocks im Striping-Set: 8 Bit
• – logische Größe des Extents: 8 Bit
• – Virtuelle Adresse dieses Extents: 64 Bit

The size of a metablock metaBlockSize results from the amount of information needed to reconstruct the block. These include, for example:

• - Id of the virtual drive: 32 bits
• - Type of data encoding: 8 bits
• - Number of this block in the striping set: 8 bits
• - logical size of the extent: 8 bits
• - Virtual address of this extent: 64 bits

From this This results in a total size of 128 bits for one Metablock. At a minimum size of 4 MByte extent this gives an extra Effort of far less than 0.1 ‰ of the partition size.

The Data is constructed here as a table in which the block number is the data storage system, e.g. a disk directly into the table index can be converted. This can quickly create a relationship of the block number will be pulled to the extent number.

Write of information in the security area

dates are written to the security area only if a new extent is allocated or when an extent is moved. Further accesses are usually not.

This usually takes place at the first access to one in the area this Extents falling address. At this time, the control instance a transformation of the virtual address to a logical drive into a physical address on a physical partition carry out. This transformation creates a tuple from the accessed data storage system and a location on the data storage system. Based on this Tupels can turn the information to be stored in the security area and the starting address will be calculated, starting from the meta-information be secured.

One further access occurs when an extent through an instance in is moved to the data storage system. This can for example to increase performance. In the preferred method is a new Entry on the partition to which the data was moved were. Subsequently the metainformation on the data storage system, on the the data was previously saved, deleted.

Reconstruct of data

• • Bestimme den Startpunkt des Sicherheitsbereiches
• • Setze i = 0
• • Wiederhole solange i < maxl – Lese Metainformationen mit Index i – Falls Metainformationen gültig – Rekonstruiere Informationen

The reconstruction of the data saved on a data storage system after the loss of control instances can be done as follows:

• • Determine the starting point of the safety area
• • Set i = 0
• • Repeat as long as i <maxl - read meta information with index i - if meta information is valid - reconstruct information

Advanced Basic procedure for securing the data by storing information on the participating data storage systems

at the preferred method presented in the context of this invention It may be due to the failure of a subarea of a data storage system done that the entire security area is no longer legible and thus the data after the additional failure of the control bodies are no longer reconstructable.

Around To counter this is an extended method of securing the data by storing information on the involved Data storage systems proposed in which the additional security information not stored in a locked security area, but spread over scattered the entire data storage system. Important for the extended Procedure is to make sure that the location of the additionally stored Security information can be clearly reconstructed. in the The following presents a procedure that is only dependent on from the chosen, minimum extent size these Can ensure uniqueness. This minimum extent size can either be unique as a constant of the security solution or in be encoded the signature area of the data storage system.

One way to distributed security information security is in 5 shown. Here, the security information is stored at the beginning of each extent followed by the actual payload to be stored in the extent.

The Storage of security information and reconstruction The data after the failure of the control instance can be like the preferred method for securing the data by storage information on the involved data storage systems.

accelerated Basic procedure for securing the data by storing information on the participating data storage systems

In the preferred and the extended method for securing the Data by storing metainformation on the involved Data storage systems use the security information that the data backed up on the data storage system also after can be reconstructed the failure of a control instance.

Furthermore However, there is also the possibility that the security information is active for speeding up access be used on the data storage systems. It should be noted that a client prior to accessing a virtual address on a logical Drive a tuple of physical drive and address on this physical drive must determine to which the virtual address is shown.

This can be done with a so-called out-of-band virtualization solution according to 2 by request to the SAN appliance 5 respectively. If a large number of clients make simultaneous requests to the SAN appliance, it can lead to a bottleneck in the SAN appliance.

in the Framework of the accelerated process variant described below to back up the data by storing information The data storage systems involved will be an extension to that extent suggested that the number of requests to the SAN appliance can significantly reduce.

The prerequisite for this accelerated procedure is that the client knows the physical drive to which a virtual address is mapped. This can be done, for example, via the ones in [BSS02] and in the DE 102 36 796 A1 be achieved described method.

In In this case, the client may have the position he needs on the data storage system by the security information stored on the data storage system determine. In the methods proposed above, the Client for this in the worst case all entries edit in the security area before getting the information he needs from the security area reads.

The The method described below couples the security gain the previously described basic method and the subsequent one based method variants with a speed gain at access to the data.

For this purpose, the Security area no longer tabular over the block number on the Data storage system such as a hard drive addressed, but there is a mapping of the extent information to the security area, the as picture key at least the virtual address on the logical data storage system contains.

When Procedure for the assignment to the security area will be setup procedures used by dictionaries, such as. Hashing or tree-based search structure (see other chapter 16 in [Sed03]), which can ensure that finding the security information based on the key in can be done almost constant or logarithmic time.

to Managing these non-table-based data structures on the Physical partitions now require security information be extended to so-called pointer, with the help of a link of the individual safety information can be made. This increases the amount of information to be stored.

The Security information can in this method, either distributed over the physical partition the data storage system or stored in a security area become.

The Calculation of the address for an item of security information to be stored is provided by a coupling of the presented methods with methods of construction of dictionaries.

• • Schaue nach, ob Informationen zu diesem Extent bereits in dem Client vorhanden sind
• • Wenn Ja – Greife auf die Daten zu
• • Wenn Nein – Bestimme aus der virtuellen Adresse das physikalische Laufwerk, auf das die Adresse abgebildet wird – Ermittle mit Hilfe der Sicherheitsinformationen, auf welche Adresse auf der physikalischen Partition die virtuelle Adresse abgebildet wird – Wenn die Adresse bestimmt werden kann – Greife auf die Daten zu – Wenn die Adresse nicht bestimmt werden kann – Frage bei dem Kontrollrechner für dieses Extent nach – Greife auf die Daten zu

Read access to an extent by a client can now be done as follows:

• • See if information about this extent already exists in the client
• • If yes - access the data
• • If No - From the virtual address, determine the physical drive to which the address is mapped - Using the security information, determine which address on the physical partition will map the virtual address - If the address can be determined - Access the data to - If the address can not be determined - Check with the control calculator for this extent - Access the data

Advanced accelerated basic procedure for saving the data by storage of information on the participating data storage systems

at The accelerated method is only possible for the client to read to access the security information. At the accelerated accelerated Procedure is now an additional, coordinating authority inserted, the above also allows write access to the data storage systems.

A Further relief of the control instance can be made when the client upon initial access to an extent the security information independently can write. Can multiple clients writing to a physical partition, however, this can only be consistent according to the laws of computer science happen when a coordinating entity is interposed.

In the extended accelerated procedure presented here, the overall system is increased by a sol coordinating authority that allows short-term exclusive access to the security area of the data storage system. This instance can be implemented in the data storage systems themselves or in a SAN appliance, in a switch, or in one or more dedicated clients.

references

• [BSS02] Brinkmann, A., K. Salzwedel and C. Scheideler: Compact, adaptive placement schemes for non-uniform distribution requirements. In Proc. of the 14 ^th ACM Symposium on Parallel Algorithms and Architectures (SPAR), pages 53-62, of 2002.
• [Sed03] R. Sedgewick: Algorithms in Java, Third Edition, Addison-Wesley, ISBN: 0-201-36120-5, 2003

list of figures

1 - Setup storage network

2 - Building storage network with out-of-band appliance

3 - Setup with in-band appliances

4 - Partitioning a logical partition

5 - Extended procedure for distributed storage of security information

Part number list

1

Storage networking

2

switches or router

3

client

4

Data storage system

5

SAN Appliance

Claims (31)

Method for distributing and retrieving meta-information for managing data storage systems in computer systems and / or storage networks and / or a data network, preferably an Internet and / or an intranet, in which a total of the data storage systems is allocated a virtual data space and divided into different logical data spaces , wherein the meta information about the distribution of the logical data spaces are stored in a central instance, characterized in that on each data storage system at least one user data area and at least one associated security area is created in advance, wherein the at least one security area for additional decentralized storage of meta information serves and its absolute location and subdivision into memory sections on the storage system are pre-determined uniquely, and subsequently upon each allocation or shift of an extent of the payload data area in memory From a data source on the data storage system, the virtual address of the logical data space data to be stored is converted into the associated physical address on the data storage system and the corresponding meta information is stored in the associated security area, whereby after a failure of the central instance for each data storage system With the aid of the meta information stored in the at least one security area, all useful data stored on the data storage system and the assignment to the logical data spaces can be reconstructed and stored in a newly created central instance. Method according to claim 1, characterized in that the at least one security area before the use as a security area is initialized such that non-described parts of the safety area are not accidentally as information can be misinterpreted. Method according to one of claims 1 or 2, characterized in that the payload data area into extents, with an extent being the smallest contiguous entity on the data storage system that can be allocated. Method according to claim 3, characterized in that the memory areas only in whole Multiples of these extents will be awarded Method according to one the claims 3 or 4, characterized in that for each possible extent in the payload area can be stored, an area for meta information in the at least one security area is reserved. Method according to one the claims 3 to 5, characterized in that an extent only on the Data storage system can be arranged on a logical and / or physical boundary of the subdivision of the data storage system starts. Method according to one the claims 3 to 6, characterized in that the minimum extent size of the user data area be clearly defined in advance. Method according to one the claims 3 to 7, characterized in that the beginnings of the regions of the subdivision of the data storage system are clearly separated from each other by the minimum extent size is unique in advance is determined and in the calculation of the boundaries of the different Areas are included. Method according to one the claims 3 to 8, characterized in that the minimum extent of the extent in one Signature area of the storage system is stored. Method according to one the claims 3 to 9, characterized in that the data on the subdivision of the data storage system be constructed as a table such that the area number of the different areas on the storage system directly into the table index can be converted. Method according to one the preceding claims, characterized in that at the first access to the data an extent a tuple from the accessed data storage system and a position of the extent on the data storage system and from that to be stored in the at least one security area Information and the starting address are calculated, starting from which Meta information saved in the at least one security area become. Method according to one the preceding claims, characterized in that when moving an extent in the Data storage system a new entry on the data storage system is generated, to which the Extent is moved and then the Delete metadata on the data storage system, on which the extent was previously stored. Method according to one the preceding claims, characterized in that for reconstruction on a data storage system secured user data after the failure of control entities the starting point of at least one security area and determined in their absolute location and subdivision into memory sections on the storage system known meta-information and, if valid for reconstruction used the meta information stored in the central instance become. Method according to one the preceding claims, characterized in that in the at least one security area saved additional Meta information distributed via the entire data storage system will be stored Method according to claim 14, characterized in that the additionally decentrally stored meta-information stored at the beginning of each extent of the user data area are followed by the actual ones to be stored in the extent Payload. Method according to one the preceding claims, characterized in that the meta information is stored on a data storage system to protect against the failure of individual areas on the data storage systems even stored redundantly. Method according to one the preceding claims, characterized in that the meta-information is active for acceleration access to the data storage systems. Method according to claim 17, characterized in that a request for user data Client needed by him Position of the desired User data on the data storage system by the on the data storage system secured metadata itself determined. Method according to one the claims 17 or 18, characterized in that the mapping of the extent information on the at least one security area is such that as a picture key at least contain the virtual address on the logical data storage system is. Method according to one the claims 17 to 19, characterized in that in the at least one Security area so-called metadata stored Pointers can be extended, with the help of a link individual Meta information for the purpose of accelerated access by a client can be done. Method according to one the claims 17 to 20, characterized in that by storing the meta-information on the data storage systems in the form of a search structure and through the use of functional or table-based associations between data space areas and the associated data storage systems clients accessing the data storage systems for the meta-information directly from the data storage systems. Method according to claim 21, characterized in that the calculation of the address for a to be stored Element of metarmations through a coupling of the procedures to Storage of meta-information with procedures for creating dictionaries takes place. Method according to one the claims 17 to 22, characterized in that an additional coordinating entity is provided, which is a short-term exclusive writing Access to the data storage systems allows. Method according to claim 23, characterized in that an additional coordinating entity in the data storage systems themselves or in a SAN appliance, in a switch or in one or more dedicated clients is implemented. Method according to one the preceding claims, characterized in that distributed on the data storage systems secured metainformation by a central instance on the Data storage systems are stored. Method according to one the preceding claims, characterized in that the central entity also the central Backup of the meta information. Method according to one the preceding claims, characterized in that by the use of a distributed Locking mechanism more than one instance writing on the distributed available meta information. Method according to claim 27, characterized in that the writing instance is the accessing Servers themselves are. Method according to one the preceding claims, characterized in that the meta information about the Distribution of logical data spaces through the central instance to accessing the data storage systems Control instances are distributed. Computer program product that is a computer readable Storage medium comprises, where a program is stored that allows a computer to after being loaded into the memory of the computer Method for distributing and finding meta-information for managing data storage systems in computer systems and / or To perform storage networks and / or a data network, wherein the method for distributing and finding meta-information the method steps according to a the claims 1 to 29. A computer-readable storage medium having stored thereon a program that allows a computer after being loaded into the memory of the computer, a method for distributing and retrieving metadata for managing data storage systems in computer systems and / or storage networks and / or a computer Data network, taking the management of data storage systems comprising the method steps according to one of claims 1 to 29.