[MPI学习笔记]mpi并行实现方法

我不是女神ヾ 2021-11-16 09:14 676阅读 0赞

首先介绍下rsh命令功能：

rsh
    
    
    NAME
         rsh - remote shell
    
    SYNOPSIS
         rsh [-Kdnx] [-l username] host [command]
    
    DESCRIPTION
         Rsh executes command on host.
         Rsh copies its standard input to the remote command, the   standard output of the remote command to its standard output, and the  standard error of the remote command to its standard error.  Interrupt, quit and  terminate signals are propagated to the remote command; rsh normally terminates when the remote command does.  The options are as follows:    
    
         -d    The -d option turns on socket debugging (using setsockopt (2)) on
               the TCP sockets used for communication with the remote host.
    
         -l    By default, the remote username is the same as the local username
               The -l option allows the remote name to be specified.
    
         -n    The -n option redirects input from the special device /dev/null
               (see the BUGS section of this manual page).
    
         If no command is specified, you will be logged in on the remote host
         using rlogin(1).
    
    FILES
         /etc/hosts

通过执行mpirun命令和do.sh脚本分析mpirun与rsh命令关系：  
Script：do.sh

1 #! /bin/bash
    2 cat hosts$1 | sed '/#/d' >idotemp
    3 for i in `cat idotemp`
    4 do
    5 rsh $i $2
    6 done
    7 rm -f idotemp

其中hoststemp文件内容为：

data
    node100
    node101
    node102
    node104

执行编译命令

1 mpif77 -O2 -o 02 02_mpi.f

（1）执行rsh命令

1 rsh temp ./02

得到输出结果：

Hello! My rank is   0.  My hostname is "data.gucas.ac.cn"
     Hello! My rank is   0.  My hostname is "node100.gucas.ac.cn"
     Hello! My rank is   0.  My hostname is "node101.gucas.ac.cn"
     Hello! My rank is   0.  My hostname is "node102.gucas.ac.cn"
     Hello! My rank is   0.  My hostname is "node104.gucas.ac.cn"

（2）或执行mpirun命令

1 mpirun -np 5 -machinefile hoststemp 02

得到输出结果：

Hello! My rank is   2.  My hostname is "node101.gucas.ac.cn"
     Hello! My rank is   3.  My hostname is "node102.gucas.ac.cn"
     Hello! My rank is   1.  My hostname is "node100.gucas.ac.cn"
     Hello! My rank is   4.  My hostname is "node104.gucas.ac.cn"
     Hello! My rank is   0.  My hostname is "data.gucas.ac.cn"

通过对以上命令输出结果分析发现其中my rank不同，而my hostname相同。  
（1）my hostname相同说明mpirun通过rsh在各个节点上并行执行命令，这点可以从mpirun脚本中得到验证

1 which mpirun
    2 vi /opt/mpichintel/bin/mpirun

mpirun脚本中有一行：

RSHCOMMAND="/usr/bin/rsh"

（2）而my rank不同，说明mpirun相对于rsh还有另外一个功能：对进程rank赋值。

总结：

（1）mpi程序编译执行一般流程是：

1 mpif77 -O2 -o 02 02_mpi.f

===> 编译mpi程序，将mpi.f头文件加载到程序中，形成可执行并行程序。

1 mpirun -np 5 -machinefile hoststemp 02

===> 通过mpirun脚本建立并行环境，分配节点/进程rank，通过rsh/ssh

在各个节点上加载程序并并行执行。根据指定不同进程rank号执行