WO2002099670A1

WO2002099670A1 - Parallel computer system having a programmable architecture

Info

Publication number: WO2002099670A1
Application number: PCT/RU2002/000276
Authority: WO
Inventors: Gennadiy Ivanovich Bacherikov; Viktor Ivanovich Gevorkyan; Vadim Mikhailovich Krokhin; Vadim Jurievich Tatur
Original assignee: Gennadiy Ivanovich Bacherikov; Viktor Ivanovich Gevorkyan; Vadim Mikhailovich Krokhin; Vadim Jurievich Tatur
Priority date: 2001-06-06
Filing date: 2002-06-05
Publication date: 2002-12-12
Also published as: RU2202123C2

Abstract

The invention relates to computer engineering and can be used in high-performance systems for processing large data arrays in a real-time mode. The inventive parallel computer system having a programmable architecture comprises communication mediums and N parallel processors. Each processor comprises a matrix of processor elements, a control processor, a main memory device, a system bus, an operating memory device, a buffer storage, a loading unit and a network adapter. Said system also comprises computing units, each unit being provided with a series processor and network adapters. Said invention makes it possible to increase a data-rate and the running efficiency of a computer by means of the organisation of linkages between units of the parallel computer system, the embodiment of the parallel processor and the introduction of communication mediums.

Description

PILLE NUMBERS SISΤΕΜΑ

С ПΡΟГΡΑΜΜИΡУΕΜΟЙ ΑΡΧИΤΕΚΤУΡΟЙ

The invention is subject to computing and may be subject to change.

used in high-performance systems of large arrays

information, including, and in real time mode.

A computer containing a process is known, an input-output switch,

command download switch, command memory, data access module,

associative memory and processing module [1].

The accessory for this computer is that when executed

Investigative functions of the software in the factory or

Other functions process only the processing module, and the process

It ceases to lead to a decrease in speed. On the other hand, it is not possible

increase system productivity by connecting additional connections

blokov.

The closest described invention () явля τ τ))))) is

a parallel computing system (process) with a pre-programmed

structure, including a group of parallel processes (elements), each

of the compartments is implemented in the form of a matrix of one computational cells,

connected to the bus input / output of the data of the control unit containing

A Responsive Storage Device (RAM) and a download node [2].

A significant disadvantage of such a system is a strong interconnection.

between each individual parallel process, which solves many problems

Provides a low overall system performance, especially in the case of 2 Failure to work properly, and also troubles building up

προizvοdiτelnοsτi sisτemy πuτem ποdκlyucheniya dοποlniτelny ^χ blοκοv. Κροome

τοgο, bοlshοe chislο πaρallelnyχ προtsessοροv, ρabοτayuschi ^χ with οdnim blοκοm

EFFICIENCY, REDUCES EFFICIENCY AND USE. It is also worth noting that

Parallel processes are inefficient for tasks when solving problems ^χ

A tough investigation is being sought, friend by friend, separate

modules and commands that implement specific parts of the task.

An increase in the number of parallel alterations along with the introduction of

Composed of each parallel process, an introduction to the process, an introduction to

COMPOSITION SYSTEM SYSTEMS CONSISTING OF PROCESSES OF OTHER TYPE

(consecutive), single, single or small communicated

neighbors, makes the fastest and most efficient

προгρammam.

Te ^χ nichesκim ρezulτaτοm πρed agaemοgο τeχnichesκοgο ρesheniya yavlyaeτsya

Productivity of the computing system.

1. Τaκοy Te ^χ goshesκy ρezulτaτ dοsτigaeτsya τem, chτο in izvesτnοm usτροysτve,

containing a group of parallel processes, κ, each of which is completed

in the form of a matrix of single computing cells connected to a bus

input / output of the data of the control unit containing the additional memory

introduced the first communicated environment and η parallel alterations,

moreover, each parallel process is additionally supported

upstream processor, system bus, utility RAM, internal RAM, and

The download node, the first group of inputs and outputs is connected to the front 3 κοmmunsh atsiοnnοy sρedοy, πρi eτοm vτορaya gρuππa in οdοv ^{^χ-χ} you οdοv site

zagρuzκi sοedinena with gρuπποy inφορmatsiοnnyχ in οdοv ^{^χ-χ} you οdοv buφeρnοgο

ZU, a group of operating inputs connected to the front group

the load exits of the download node, the load of the outlet and the load exits

connected to a group of information inputs and outputs of the service RAM,

group of operating inputs is connected to the second group

LOAD EXITS OF THE LOAD UNIT, FOURTH GROUP OF INPUTS-OUT

It is connected to a large infused output from the memory of the unit

the operating inputs are connected to the third operating group

the exit of the download node, the first group of outputs is connected to the group

INFORMATION MATERIALS OF THE MATRIX OF PROCESSING ELEMENTS, GROUP

the information output is connected to the group input of the download,

group of outputs that is connected with group of inputs

MATERIALS FOR PRODUCTS, GROUPS AND OPERATING EXITS FOR PRODUCTS

associated with the worst offending unit, the fifth grouping entry-

the output is connected to the system bus and is also connected to

group input into the process.

In addition to the integrated parallel computing system with

The programmable hardware may have an additional introduction of one or

more communicated, and each parallel process may be slower

additionally have one or more network adapters, and at the first

The input and output unit of each network adapter is connected to the system bus

The parallel process, which owns this network adapter, and 4 second outlet group of each network adapter is connected to one of

communicated environment

It is also possible to use the parallel parallel computing system with

programmable add-on for one or more

optional nodes, each of which is optional

process, system bus, and one or more network adapters, or even a big one

The input and output of the research process is connected to the system bus, with

The first group is also connected to the entrance to the main entrance of each network.

the unit, and the second unit of the input-output of each network adapter is connected to

One of the communal environment.

on φш. 1, 2, 3 the structured scheme of a computing system with

AGRAMMABLE STRUCTURAL COMPONENTS π.π.1,2,3 of the invention

φш. 4 - structure of the download node. Refer to FIG. 5 STRUCTURE OF THE COMMAND BLOCK. And on this

each of FIG. 2.3, two additionally introduced communications are shown.

Mediums, although they may additionally be entered one or more. But also

on each of the FIG. 2.3 Two additional ones entered into parallel are used.

The process of the network adapter, if it is necessary to enter one or

more, and two additionally introduced into the computing node of the network adapter

In addition, it may be added one or more. Piggybacks

The following meanings: 1- the regulating process, 2- the system bus, 3-RAM,

4- download node; 5- process elements; 6- service RAM; 7-

buffer memory, 8-first communicated medium; 9- π-parallel προrocessορ,

10,11,15, 16 - network adapters, 12- second communicated environment; 13th 5 communicative environment, 14- investigative process, 17-computational

Knot, 18, 23,24- blocks of shinngkh of drivers, 19- registers

team, 21-count of the address, 22-commute, 25-com, 26-reg., 27-

deshiφρаτορ commands. Fig. 1 blocks 1, 2, 3, 4, 5, 6, 7 use the parallel

προѕοορο 9. φΗ ш З З 2 2 2 2 2 2 2, 14, 15, 16.

upstream process 1, interacting with upstream processes

1 other parallel process 9 and computing nodes 17, for

PERFORMANCE SPECIFIC PARTS OF THE APPLICATIONS (PROCESSES) AND RELATED

data that initiates processing on the matrix of the process elements and transmits

other process results. It can be implemented on the basis of

As a single, standard microprocessor, as well as a multiprocessor board

τiπa δгηρ [4]. Шины The standard tires are used on the system bus.

(for example, ρcϊ or νte [4]). 3ЗУ 3 πρrednaznachen for the storage of information and

προгρamm. The internal memory 6 is intended for storage of information on site

This block is for loading the process. Β some cases of RAM and

service RAM can be combined. Buffered RAM 7 serves for exchange

data between parallel processes 9 9 through the first communized

Medium 8 and it may be more like the usual RAM in the form of a collection of registers,

being πιο - buffers. 5 process elements are supplied

a special set of process elements, united by links of a given kind (

why κ “each of the process elements implements a part or one of the commands

performed by the program), and may be performed, for example, as in [3] or in the form

matrices, which consists of plies (programmable logical integrated circuit, 6, for example, χϊ1πгχ [6]). Parallel processes 9 for fast data exchange

between themselves are united by the first frontier communicated environment 8,

of a representative set of tires for transmitting data ^χ , organized by ^χ in

a shared geo- logical structure, for example, τορ [5].

The download node 4 is intended for downloading and downloading commands and data in

the matrix of process elements 5 and from it, respectively. Basic Functions

Most importantly:

α Functional amplification of signals and addresses for RAM, service

RAM, buffer RAM.

a

Process Elements 5.

α Connection of parallel process units for the exchange of

information between them.

α Exchange between units of a parallel process.

One of the possible options for the implementation of the download node is shown in figure 4.

The download node contains: 18 - a block of hardware shredders, 19 - a registry

set, 20 - command block, 21 - address counter, 22 - commute, 23 - shinng block

distributors, 24-block of larger distributors. The ff.5 is shown as one of

the possibility of realizing a block of commands. Command block contains: 25 - cam, 26 -

registry, 27 - deshiφρааtοcommands.

Units of bus drivers 18,24 ensure communication of units in the unit

DOWNLOADS WITH SYSTEM TYPE AND FIRST COMMUNICATION MEDIUM, RESPONSIBLE.

Block of bus drivers 23 Ensures the connection of the blocks of the download node with the RAM 7

3, service RAM 6 and buffer ROM 7. Block command 20 provides

decryption of commands that run away from the operating process 1 taking into account

The facility is located at 19, which contains official information ο

Operating modes of the download node (for example, byte or byte exchange

information, size of the array and etc.) and issue

Schnalah for the operation of the units of the download site. Address Count 21 Counts Address

and issues them to command unit 20 and bus unit 23. FFM 22

Ensures the connection of any entrance to any of its own entrance.

The machine 25 compares the current address as it is counted

address with the end address of the array from the register of the 19. 19. Register 26 serves

to save a command issued by an add-on process 1 to the download node 4.

Decreases 27 commands to ensure the issuance of corrective links to all the blocks

The downloads that enter the site, in accordance with the contents of the registry 19,

reg. 26 and the exit of the compressor 25.

The download node works as follows: Including προѕοορ 1 through

system bus 2 and drive unit 18 sends commands to the unit

Command 20. In accordance with the instructions, the switchgear is

The transmission of the commands and the data from the other unit to the other. Then in the registry page 19

information is recorded about the operating mode of the download node (for example, bytes

or exchange of information, size of the array and etc.), and in

count of address 21 is the starting address of the information array.

The initial settings of the address and register counting are secured by snapping

the initial installation (for all the details, the initial installation and their discharge 8 are not shown). Further, the counter of address 21 increases its value in accordance with

available pulses (for all generators of the process ^χ

impulses and discharge of impulses are not shown) to ^χ πορ,

The unit 25, which is part of the command unit 20, did not issue a shutdown

Exchange, which will stop counting the address 21. Switch 22 in conjunction with

with snapping up from command block 20, it establishes communications for transmission

data between blocks of the download node 4.

Units of the world 18,23,24, register 19, address 21

The standard has been implemented, for example, as described in [7, 8].

Commutation 22 may be completed as described in [5,7]. Team block may

be simple, as shown in figure 5. Hessistp 26 remembers the input

On the other hand, the unit 25 compares the current address counter with the last

the address of the exchangeable array, which is located in the registry at 19 and

gives in despatch the command 27 sent the completion of the exchange. Upon completion of the exchange

The command 27 issues the following to the offending process 1, which sent the factory,

After which the process controller 1 may issue another command.

Part 25, Reg. 26, Cheap Commands 27 executed by standard

For example, for example, as it was described in [7,8].

In order to combine the time of the computational process on matrix 5

Process elements and downloads of source data for the next process in

RAM 3, in the parallel process, the second boot node 4 may have been entered.

CONTROL OF EXTERNAL WORK OF KNOTS OF PARALLEL PROCESS OF PRODUCTION

incremental προ processορ 1. 9 Network adapters 10, 11 are designed to connect a parallel

process for the second 12 and the third 13 communal environments for the purpose of exchange

information with other parallel processes 9 and computing

nodes 17. Network adapters 15, 16 are intended for connection

Computing node 17 on second 12 and third 13 communicated environments with

the purpose of exchanging information with other computing nodes 17 and

Parallel Processes 9. On-the-fly hardware for data transmission

use standard communication media such as ssϊ, tugϊeneϊ: or

similar to their intended purpose and chemist [5].

The device operates the following way. Any προгρammma,

A descriptive solution to any problem that is composed of a series of computational

Processes (modules of the program), each of the processes is logically

COMPLETED ACTION. Successive Processes (Research Processes

type), all teams who have to do so must go to each other,

performed on the computational nodes, κ <each of which comes from

investigative process (investigative process or process

non-neurological type) system bus and one or more network adapters.

Parallel Processes (Processes of a Parallel Type), Commands of the Processes

performed in one piece (fully or partially), performed on

parallel προ processορаχ. The partition of the program into the process is carried out

either a grammist, or a traffic that is part of a system

The main software of a parallel computing system with

programmable arithmetic. In addition, this process controller is also 10 industrial process. Its main task is to ensure

Starting the process in parallel type, but in the standby mode, each

A process that can be run can perform investigational processes. Therefore

Parallel system with programmable hardware for some areas

Applications may and may not be computational nodes.

Updating the program starts with downloading and starting up one of

προ processορο in her chief προ process. This process is in accordance with the structure

Programs may start other processes that are in progress.

Otherwise, you can also start a new computing process, or just one

A process may start a few other processes before it

ends. For the sake of convenience, there are free processes, all the processes are with the same

aimed at execution. If, however, there are no free processes in this type,

This process is set to standby mode. Sequential and parallel

Processes are carried out on processes corresponding to the type. Well, in

When executing the program of the solved problem in the system of functions

the maximum number of processes is higher (not higher

number of computational processes in the system), and other part of the ready-made

The process is waiting for you to process free of charge. Means

Meetings of the ready-to-eat process for the execution of the processes and the start of them for execution

They share the main components and are not described here.

Commands to download processes run through the communicated media

8,12,13. If there is a free access to any process, the process

and for help, there’s nothing in it; memory is sent to teams of computational 11 process and / or required data. Forwarding and starting up processes for

INSPECTORAL PROCESSES 14 ARE STANDARDED BY USING

Neighboring relevant communication media [5]. For π-parallel ^χ ,

Processes 9 Performing computational processes and data exchange

The following is used.

Β case when a parallel computing system with a programmable

ARCHITECTURAL CONTAINS ONLY THE FIRST COMMUTCED ENVIRONMENT 8, ATTENTION

Each parallel process 9 is enabled using the download node 4, and

initialization of the system in each and every process 1 is recorded

The program of the call and reception of the other process. The receipt of the process and

The data that is being processed is processed through the first communica

Mediocre 8. But the very same happens after the completion of the current process.

Β case when a parallel computing system with a programmable

the arhitektory consumes one or more communicated media (12.13 per fsh.Z),

OPTIONAL CONNECTED TO A PARALLEL APPLICATION 9 THROUGH NETWORK

Adapters 10, 11, π and initialization of the system to each applicable process 1

The program of the call and reception of the other process is recorded. Receipt

Process and data that is being processed, is produced through the second and / or

chain and t.d. Communicated environment (12.13 per fsh.Z). But the very same incident

after the completion of the current process.

Each add-on process 1 through the system bus 2 and the boot assembly

4 records the commands that were received for the implementation of the process and which are necessary for

there is some data in ROM 3, and also records information about the unit’s settings in 12 service RAM 6. The last one contains the data for the download node 4, and

Namely: the initial addresses of the corresponding processes and ^χ data ^χ in the memory 3, and

τaκzhe τiπ danngχ with κοτορgmi ρabοτaeτ ΟZU 3. Eτο mοzheτ προis ^χ οdiτ in ρaznye

Moments of time, if the most commands describing the process, and data for

It is discontinued from different πprocesses. Like πρ rule, all προ processes executed on

Parallel process 9, one at a time are loaded into RAM 3 each

parallel process 9 before starting the solution of the problem, and money

It is intended, however, that they are available only in the memory of 3, which are owned by

In parallel with the process 9, the processes associated with

these dangmi. Then, the master process 1 issues a command to the download node

DOWNLOADS OF THE PROGRAM INTO THE MATRIX OF THE PROCESSING ELEMENTS 5, AND ATTENTION OF THE COMMAND

the start of the most computational process, for example in [3]. Β χοде

FUNCTIONALITY KEY MATERIAL OF PROCESSING ELEMENTS 5 THROUGH

download node 4 the input data from the RAM 3 and is written to the negative or to the buffer

RAM 7 results of superior performance. In the process of deposition, each matrix

process elements 5 can exchange data with other matrices

process elements 5 through the load node 4, buffer RAM 7 and the first

Communicating medium 8. At the end of operation, the process matrix is ready for use.

of elements 5 through the download node 4 will issue its own control process 1

I found out that after that I added the process 1 I can pass

Result: “downgrade”; vtagstitel.m. - through the process through ^: .ορ '> t

Communicating with another operator 1, which is in good condition

Wait for them to wait. etizh yes ethesh. dsh »shβzhjeyshch, βκ 13 Process on the parallel process 9 or after optionally entered

second 12 (or other) in the middle of computing node 17,

A short wait is pending receipt of this data ^χ for execution

The last investigative process.

if there are no free processes available, then the next process is waiting

EXCEPTIONS OF ITS TYPE (FOR ITS PERFECTION) Each

Amending Process 1 and Consequential Process 14 after termination

IMPLEMENTATION OF THE ETERNAL PROCESS DOES SEARCH FOR THE PERFORMANCE OF THE FOLLOWING

The process of this is achieved by sending mails from one of the communicated environments

uπρavlyayuschiχ zaπροsοv τiπa "in κοgο esτ for vyποlneniya προtsess and / or data

is it for my type? ". Functions are being processed

At the same time, when all the processes of the given program are executed. Β

When processes are executed, they exchange data through communicated

neighbors. The most effective is the specialization of the communicative environment.

With this exchange of data between parallel processes 9, there are

through the first communal environment 8, exchanging mutually acceptable messages and

Control and operation of the entire system through the second communicated

Medium 12, and downloading process data and exchanging data between investigative

process 14 and between successive 14 and parallel 9

Processes through a communicated medium 13. Intermediate introduction

great quantity of communal medium used for the building

complex multilevel network arithmetic [5]. Β some cases are possible

use of a computing system with a programmable structure without 14 initial communication environment, switched on via the boot node, but this

The system must be equipped with one or more communicated systems.

neighbors (12.13 in FIG. 3), optionally connected through a network of adapters.

Like all processes, you have the same way to process them.

The information is the same as in the case of any of these

The computing system is generally operational.

The device makes it possible to significantly increase the speed of processing

information and increase the reliability of its operations. It is also possible

increase system productivity by connecting additional connections

blokov.

Sources of information:

1. ΡСΤ, patent λνθ 99/08172 cl. Ο06 Ρ 15/16, 1999

2. Κ.υ, patent No. 2110088 C1, κl. 6 Ο06Ρ 15/16, 15/00, 1998.

3. υ, Patent ЗΥэ 2134448 С1, cl 6

4. Κορneev Β. Β. Complicated processes. Μ .: лиοledge, 1999

5. Κορneev Β. Β. Parallel computing systems. Μ .: лиοledge, 1999

6. ΧШηχ. 2000аϊаΒοοк to 2000.

7. Uguryum Ε.P. and dρ. “Digital Schematic”, St. Petersburg, 2000.

8. Χοροwitz P., Ushsh U. “The Basics of Technology”, p. from English B.Η. Bροnina and dρ. 5th ed., Πеρеρаб .. Μ .: Μиρ, 1998.

Claims

15ΦΟΡΜULΑ IZBΟIA

1. Parallel computing system with programmable arithmetic,

inclusive parallel process, which contains the matrix

process elements, and the RAM, which is different, that it is optional

equipped with a front-end communicating environment and parallel process,

moreover, each parallel process is additionally supported

upstream processor, system bus, utility RAM, internal RAM, and

The download node, the first group of inputs and outputs is connected to the front

Communicating environment, and this second unit of input-output node

DOWNLOADS CONNECTED TO A BATTERY INFORMATION OUT-OUT

ZU, a group of operating inputs connected to the front group

the load exits of the download node, the load of the outlet and the load exits

connected to a group of information inputs and outputs of the service RAM,

group of operating inputs is connected to the second largest

LOAD EXITS OF THE LOAD UNIT, FOURTH GROUP OF INPUTS-OUT

It is connected to the group of input-output inputs of the RAM, group

the operating inputs are connected to the third main operating system

The exit node of the download, the first group of outputs is connected to the group

informational inputs of the matrix of the process elements

the information output is connected to the load input of the download unit,

A group of concessionary outputs is associated with a negative concurrent outlet

matrices of process elements, a large amount of accelerating outputs

connected with the pear loading unit, the fifth group of inputs- 16 outlets connected to the system bus, also connected to the system

group input into the process.

2. Paρallelnaya in chisliτelnaya sisτema with προgρammiρuemοy aρ ^χ iτeκτuροy πο

π.1, characterized by the fact that it is additionally introduced one or more

communicating environment, and each parallel process is optional

contains one or more network adapters, and at the same time the first group of inputs

the output of each network adapter is connected to the system bus

The parallel process, which owns this network adapter, and

the second group of inputs and outputs of each network adapter is connected to one of

communal environment

3. Parallel computing system with programmable hardware

π.2, characterized by the fact that it introduces an additional introduction of one or more

computing nodes, each of which contains a serial

process, system bus, and one or more network adapters, and then

gρugsha in οdοv ^{^χ-χ} you οdοv ποsledοvaτelnοgο προtsessορa sοedinena with sisτemnοy

bus, with the first one is also connected to the first entrance of the exit-exit of each

network adapter, and the second group of inputs and outputs of each network adapter

connected to one of the communal environment.