Computer Center for Parallel Technologies Computer Systems Laboratory Siberian State University of Institute of Semiconductor Physics Telecommunications and Informatics Kirov str., 86 Lavrentiev ave., 13 630102, Novosibirsk, Russia 630090, Novosibirsk, Russia Tel. & Fax: +7 (3832) 66 38 37 Tel. & Fax: +7 (3832) 33 21 71 E-mail: khor@neic.nsk.su. E-mail: khor@isp.nsc.ru. ARCHITECTURE AND FUNCTIONING OF HIGH-PERFORMANCE DISTRIBUTED RECONFIGURABLE COMPUTER SYSTEMS AND DATA PARALLEL PROCESSING Prof. Dr. Victor KHOROSHEVSKY Corresponding Member of Russian Academy of Sciences Novosibirsk – RGW – 2005 – 27.06 – 06.07.2005 7/6/2005 1 7/6/2005 1
LECTURE OUTLINE Russian – German Workshop – 2005 1. Conception and Architecture of Reconfigurable Distributed Computer Systems (DCS) 2. Soviet and Russian Reconfigurable DCS 2.1. Architecture Facilities of DCS 2.2. Space-distributed Multicluster Computer System 3. Parallel Multiprogramming 3.1. DCS Multiprogram Modes 3.2. Game-theoretical Optimization of Functioning DCS 3.3. Stochastic Programming Optimization of Functioning DCS 4. Functioning Effectiveness Analysis of DCS 4.1. DCS Continuous Model 4.2. Robust DCS 4.3. Robustness of DCS 4.4. Parallel Solving Realizability of Tasks 5. Conclusion Novosibirsk, RGW – 2005, 27.06. – 06.07.2005 7/6/2005 2 7/6/2005 2
COMPUTER ARCHITECTURE DEVELOPMENT SISD; Computer MISD; Pipeline System Processor Control Elementary Processing Units EPU EPU … EPU … 1 i n ALU Data Flow Result Flow . . . . . . Memory Memory SIMD; Array System MIMD; Distributed Programmable Structure Computer System Common Control Unit … … … EP EP EM EM EP EM 12 1n 11 12 1n 11 … … … EP EM EM EM EP EP 21 22 21 22 2n 2n … … … … … … … … … … … … EP EP … EP EM EM EM nn n1 n2 n1 n2 nn EP – Elementary Processor EM – Elementary Machine 7/6/2005 3 7/6/2005 3
DISTRIBUTED RECONFIGURABLE COMPUTER SYSTEMS Распределенные вычислительные системы с программируемой структурой - сочетание архитектурных свойств универсальных и специализированных средств обработки информации . Академик Н . Н . Яненко : “ Чем шире класс задач , охватываемой специализированной машиной , тем сложнее ее структура и как наиболее совершенную форму ЭВМ следует рассматривать ЭВМ с перестраиваемой архитектурой ”. ( Доклад на Конференции по проблеме “ ЭВМ . Перспективы и гипотезы ”, посвященной дню Советской науки , Новосибирск , СО АН СССР , 17 апреля 1981 г .). 7/6/2005 4 7/6/2005 4
DISTRIBUTED RECONFIGURABLE COMPUTER SYSTEMS Distributed Computer System (DCS) - a composition of sets of elementary processors, local memory, control means and communication network Mainframe Mainframe Mainframe DCS Architecture COMMUNICATION NETWORK • MIMD-architecture Mainframe Mainframe • programmability of intermachine network structure • program transform possibility of MIMD Mainframe Mainframe architecture into SIMD and MISD ones Mainframe • massively parallelism CONTROL MEANS Elementary Machine • homogeneity, modularity and scalability ELEMENTARY PROCESSOR • decentralization, distributeness (of control and data). LOCAL MEMORY 7/6/2005 5 7/6/2005 5
SOVIET AND RUSSIAN PRORAMMABLE STRUCTURE DISTRIBUTED COMPUTER SYSTEMS DCS History “Minsk-222” 1965 MICROS 1986 Control DCS 1967 MICROS-2 1992 ASTRA Family 1973 MICROS-T 1994 MINIMAX 1975 MBC-100 1992 – 1995 SUMMA 1977 MBC-1000 1997 - current time MICROS Family Architecture Realizability of MIMD-architecture Program-transformation possibility of MIMD-architecture into SIMD and MISD ones Massively parallelism of data processing and control process Programmability of intermachine network structure Homogeneity, modularity and scalability Decentralization (absence of a unique unit for system) Distributeness (of data and control means), short-range interaction Asynchonism and locality (“point-to-point”) of intermachine and interprocess interactions MBC-15000BM System Number of elementary machines (PowerPC970 2.2 GHz) – 924 Structure – Myrinet Performance (Rpeak, LINPACK) – 5,355 TeraFLOPS 7/6/2005 6 7/6/2005 6
SPACE-DISTRIBUTED MULTICLUSTER COMPUTER SYSTEM Cluster network SibSUTI’s local network Elementary machine Clusters of SibSUTI Clusters of Siberian Branch of Russian Academy of Sciences Internet, telephone lines 7/6/2005 7 7/6/2005 7
MULTICLUSTER SOFTWARE ДИСПЕТЧЕР РАСПРЕДЕЛЕННЫХ РЕСУРСОВ ( РАСПРЕДЕЛЕННАЯ ОЧЕРЕДЬ , БАЛАНСИРОВКА НАГРУЗКИ ) ДИАГНОСТИЧЕСКИЕ СРЕДСТВА ПРИКЛАДНЫЕ ПРОГРАММЫ СРЕДСТВА ПОДСИСТЕМА ( ДИАГНОСТИЧЕСКИЙ СЕРВЕР , КОНТРОЛЬНЫЕ ТЕСТЫ ) ПАРАЛЛЕЛЬНОГО АНАЛИЗА ПРОГРАММИРОВАНИЯ ЭФФЕКТИВНОСТИ РЕКОНФИГУРАТОР (LAM v.6.5.6, MPICH v.1.2.0, PVM v.1.1.1) СРЕДСТВА ПАРАЛЛЕЛЬНОГО ПРОГРАММИРОВАНИЯ СИСТЕМЫ (LAM v.6.5.6, MPICH v.1.2.0, PVM v.1.1.1) СРЕДСТВА МЕЖМАШИННОГО ВЗАИМОДЕЙСТВИЯ (RLOGIN, RSH, TELNET, SSH) Сетевые протоколы (PPP, TCP/IP) ОПЕРАЦИОННАЯ СИСТЕМА (Linux kernel v.2.4.20) СРЕДСТВА МУЛЬТИПРОГРАММИРОВАНИЯ Обработка наборов задач Обслуживание потоков задач 7/6/2005 8 7/6/2005 8
DISTRIBUTED CS MULTIPROGRAM MODES Software Hardware Software Hardware Mainframe Mainframe Mainframe COMMUNICATION NETWORK Mainframe Mainframe Mainframe Mainframe Mainframe Tasks are coming to be solved by DCS Operating System - Single rank Multiprogram modes Monoprogram mode Servicing stochastic flow Processing set of tasks of tasks Stochastic programming Game-theoretical approach approach Partition of multiple Optimal mixed machines strategies 7/6/2005 9 7/6/2005 9
STOCHASTIC-OPTIMAL FUNCTIONING ORGANIZATION OF DCS Game-theoretical approach 1. Statement of problem. Task stream with a queue. DCS of N Elementary Machines (EM) Task queue of all ranks Solution of a task of rank j requires j EM Operating System (OS) – for distribution tasks between the system machines. 2. Simplest Game Model. Game of two objects: DCS & Operating system i – pure strategy of DCS, it allocates i machines for the task solution j – pure strategy of OS, it assigns the task of rank j for DCS { } ∈ - payment matrix, . C = , 0 , 1 , , K i j N c ij c - payment of OS to DCS, if DCS and OS choose strategy numbers i and j ij respectively { } π = π π π , , , K - mixed strategy of DCS 0 1 N { } = , , , K p p p p - mixed strategy of OS 0 1 N 7/6/2005 10 7/6/2005 10
STOCHASTIC-OPTIMAL FUNCTIONING ORGANIZATION OF DCS Game-theoretical approach (continuation) 3. Optimal mixed strategies Average payment to DCS – N N ∑∑ π = π T c p p C ij i j = = 0 0 i j π if DCS and OS employ mixed strategies and respectively. p * π * There are optimal mixed strategies and for DCS and OS so that: p π ≤ for all p, π ≥ π T * ( ) for all T p C v p C v = C π * * Game cost ( ) T v p + − ≥ ⎧ ( ) , jc i j c for i j 1 2 = It is proposed to take the following matrix: ⎨ c ij + − < ( ) , ⎩ jc j i c for i j 2 3 c - payment for the usage of one EM per unit time 1 c and - penalties per unit time for the idle time of one EM and for j – i = 1 c 2 3 Theorem. The matrix C has no saddle point if and only if { } < min , . c c c 1 2 3 4. Parallel method for game solution is based on a composition of the modified Braun-Robinson and simplex methods 7/6/2005 11 7/6/2005 11
STOCHASTIC-OPTIMAL FUNCTIONING ORGANIZATION OF DCS Stochastic programming approach 1. Sample of Problem Statement - number of Elementary Machines (EM) N - number of terminals (for a task stream) L j a - number of subsystems of rank , which can be loaded from terminal l jl ( ) p a - probability distribution density of the variable , a jl jl ∞ ∫ = ( ) 1, ∈ ∈ {1,2, , }, {1,2, , }, p a da K K j N l L jl 0 j - service price of the subsystem of rank for the terminal d l jl c j - cost of formation and maintenance of the subsystem of rank for the terminal l jl - number of subsystems of rank allocated necessarily for the terminal j y l jl - number of subsystems of rank allocated additionally for the terminal j x l jl 7/6/2005 12 7/6/2005 12
Recommend
More recommend
