并行程序设计基础

Size: px

Start display at page:

Download "并行程序设计基础"

腰黄
7 years ago
Views:

1 1800

2 1800 学计发软调术计术应软务 2

3 1. 论学 2. 实验学 3. 计学计学计学计动学计学计学计学计学计学计电学 3

4 4

5 ( 级计 ) CRAY, 银 I SMP( ) IBM p690 SUN MPP( 规 ) T3E 1000 HP ccnuma( 储 ) SGI Qrigin 统联腾 1800, IBM SP2 5

6 6

7 7 CPU MHz MIPS/R16000 SGI MHz UltraSparc Sun MHz Athlon2800+ AMD MHz Alpha HP GHz P4 Xeon Intel 1219 no GHz Opteron AMD 1285 no GHz P4 Xeon Intel GHz Power4 IBM GHz Itanium2 Intel Specfp 2000Peak TPP n=1000 Peak (Mflops) Clock CPU Manufacturer

8GHz P4 Xeon Intel 1219 no 3600 1.8GHz Opteron AMD 1285 no 6400 3.

8 Rank Manufacturer Computer/Procs R max R peak Installation Site Country/Year 51 Legend Group DeepComp P4 Xeon 2 GHz - Myrinet/ Academy of Mathematics and System Science China/ Legend Group DeepComp P4 Xeon 2.4 GHz - Myrinet/ LASG/Institute of Atmospheric Physics, CAS China/ Hewlett-Packard SuperDome 875 MHz/HyperPlex/ Youngwoo China/ IBM xseries Cluster Xeon 2.4 GHz - Gig-E/ ShengriLi China/ IBM xseries Cluster Xeon 2.4 GHz - Gig-E/ Sinopec China/ Legend Group DeepComp P4 Xeon 2.4 GHz - Giganet/ Exploration & Development Res. Inst. of Daqing Oil Field China/ Hewlett-Packard SuperDome 750 MHz/HyperPlex/ Beijing Social Insurance China/2001

00 LASG/Institute of Atmospheric Physics, CAS China/2002 278 Hewlett-Packard SuperDome 875 MHz/HyperPlex/ 192 408.20 672.00 Youngwoo China/2003 292 IBM xseries Cluster Xeon 2.4 GHz - Gig-E/ 256 402.

9 9

10 1800 学计发软调术计术应软务 10

11 HotSpot BottleNeck 11

12 IO 12

13 13

14 System System Level Level Application Level Level Micro-Architecture Level Level 14

15 Perfmon* NetMon* ApiMon* System Level Less Intrusive VTune Performance Analyzer Quantify* Application Level More Intrusive Micro-Architecture Level Less Intrusive 15

16 16

17 for(i=0;i<num;i++) { for(j=0;j<num;j++) { for(k=0;k<num;k++) { c[i][j] =c[i][j] + a[i][k] * b[k][j]; } } } for(i=0;i<num;i++) { for(k=0;k<num;k++) { } } for(j=0;j<num;j++) { c[i][j] =c[i][j] + a[i][k] * b[k][j]; } SELF IMSL RELEASE OPT 4 CXML ATLAS MKL50 MKL s s 5.445s 5.457s 3.328s 0.762s 0.848s 0.738s 17

18 1800 学计发软调术计术应软务 18

19 PRAM BSP logp C 3 SIMD MIMD SPMD MPMD 19

20 20

21 CPU Memory CPU Memory C P U C P U C P U C P U Network Bus CPU Memory CPU Memory Memory OpenMP 21

22 组 22

23 The Problem 组 Communication Initial tasks Combined Tasks Final Program 23

24 Atmosphere Model Hydrology Model Ocean Model Land Surface Model 24

25 25

26 26

27 - 27

28 N N N N N N 28

29 N1 N2 N2 N4 N4 N4 N4 N2 N2 N1 29

30 P1 T1 T2 T3 T4 TN-1 TN P2 T1 T2 T3 TN-2 TN-1 TN P3 T1 T2 TN-3 TN-2 TN-1 TN 30

31 - Master Slave Slave Slave 31

32 N1 N2 N3 32

33 33

34 数 HPF 储 OpenMP Pthreads 传递 MPI PVM 34

35 Navier-Stokes

36 (i,j+1) (i-1,j) (i,j) (i+1,j) (i,j-1) N Y 36

37 (i,j+1) (i-1,j) (i,j) (i+1,j) (i,j-1) 37

38 38 P2 P2 N Y P1 P1 N Y P3 P3 N Y P4 P4 N Y

39 处数拟时间 / (8 节点 )

40 1800 学计发软调术计术应软务 40

42 DL_POLY( ) 80,000 Fortran AIX / Power4 ~> Linux / Intel Xeon DOCK( ) 100,000 C IRIX / SGI ~> Linux / Intel Xeon Ccsm( ) 170,000 Fortran AIX / Power4 ~> Linux / Intel Xeon 42

43 MOLDY 调 3 DL_POLY 调 2.25 软调调 6 庆 Geodepth 调单节点时络达 126MB/s 43

44 Intel / PGI / LF95 Compiler MKL svml Goto(Math Library) IPP(Performance Library) TotalView(Parallel Debug) VTune(Performance Analysis) VGV(Cluster Performance Analysis) Perfmon ; Apimon ; Netmon 44

45 45

46 46

并行计算

并行计算并行计算 Parallel Computing 主讲人孙广中 Spring, 2016 并行计算结构算法编程第一篇并行计算的基础第一章并行计算与并行计算机结构模型第二章并行计算机系统互连与基本通信操作第三章典型并行计算机系统介绍第四章并行