中粒度並列処理用ハードウェア同期処理機構の提案  [in Japanese] Proposal of hardware Synchronization Support System for middle grain parallelism  [in Japanese]

Access this Article

Search this Article

Author(s)

Abstract

近年、マルチプロセッサ・システムが普及し、プログラムを分割して並列に実行する並列処理手法が広く用いられるようになった。中でもプログラムをスレッド等の細かい粒度に分割する中・細粒度並列処理は、実行するマルチプロセッサ・システムの形態に強く依存しない手法として注目されている。しかしながらこの手法では、プログラムを細かく分割することによるタスク数の増加により、スケジューリングやコンテキスト・スイッチ、同期処理等に起因するオーバーヘッドが増大し、大幅な性能低下を招くことが問題となっている。この問題を低減する手法として、著者らはハードウェアでスケジューリングを高速化するSSH(Scheduling Support Hardware)、およびハードウェアでコンテキスト・スイッチ処理を行うCSS(Context switch Support System)を提案している。しかしながら、同期処理によるオーバーヘッドが依然問題となっており、これを解決する必要がある。そこで本研究では、同期処理を支援するS3(Synchronization Support System)を提案し、マルチプロセッサ・システムに実装することで、同期処理に起因するオーバーヘッドの増加を抑え更なる高速化を目指す。Today, multiprocessor systems have spread and Parallelism on the system are widely used in every usage. Particularly, Middle-fine grain parallelism, diving program into a large number of threads, is paid attention to. Because it hardly relies on multiprocessor system environment. However, the overhead of task scheduling, context switching and synchronization process becomes large compared to the executiontime of threads and it often makes performance down greatly. In order to reduce these overheads, we have proposed the Scheduling Support Hardware architecture (SSH) and Context switch Support System (CSS). SSH accelerates the performance of the OS with hardware by scheduling threads concurrently with the thread execution on the CPU. CSS saves the registers the former thread used and fetches the registers used in next thread in parallel with the thread execution on the CPU. By using them, we have obtained good results. However, the overhead of the synchronization process still causes performance down. This paper proposes the Synchronization Support System (S3) which supports synchronization process by snooping shared memory bus in parallel with the thread execution on CPU, and adds S3 to the multiprocessor system with SSH and CSS and evaluates them.

Today, multiprocessor systems have spread and Parallelism on the system are widely used in every usage. Particularly, Middle-fine grain parallelism, dividing program into a large number of threads, is paid attention to. Because it hardly relies on multiprocessor system environment. However, the overhead of task scheduling, context switching and synchronization process becomes large compared to the execution time of threads and it often makes performance down greatly. In order to reduce these overheads, we have proposed the Scheduling Support Hardware architecture (SSH) and Context switch Support System (CSS). SSH accelerates the performance of the OS with hardware by scheduling threads concurrently with the thread execution on the CPU. CSS saves the registers the former thread used and fetches the registers used in next thread in parallel with the thread execution on the CPU. By using them, we have obtained good results. However, the overhead of the synchronization process still causes performance down. This paper proposes the Synchronization Support System (S3) which supports synchronization process by snooping shared memory bus in parallel with the thread execution on CPU, and adds S3 to the multiprocessor system with SSH and CSS and evaluates them.

Journal

  • IPSJ SIG Notes

    IPSJ SIG Notes 2006(127(2006-ARC-170)), 79-84, 2006-11-30

    Information Processing Society of Japan (IPSJ)

References:  5

Codes

  • NII Article ID (NAID)
    110005717708
  • NII NACSIS-CAT ID (NCID)
    AN10096105
  • Text Lang
    JPN
  • Article Type
    Technical Report
  • ISSN
    09196072
  • NDL Article ID
    8571303
  • NDL Source Classification
    ZM13(科学技術--科学技術一般--データ処理・計算機)
  • NDL Call No.
    Z14-1121
  • Data Source
    CJP  NDL  NII-ELS  IPSJ 
Page Top