[交流] 【求助】如何使用openmpi在多节点并行计算？ 已有3人参与

5. 运行MPICH程序
多机环境中运行MPICH程序与单机环境类似，可以用mpirun来进行。运行程序前先创建一个machinefile文件，其中列出要使用的结点机名，然后用命令“mpirun -machinefile 文件名 ...”来在指定的结点上运行程序。例如，假设用户登录在结点node2上，文件mfile中包含下述内容：
node3
node4
则命令：
mpirun -machinefile mfile -np 3 cpi
将用node2，node3 和node4来运行程序cpi，每个结点一个进程，这是因为默认情况下mpirun总是将当前结点添加到程序的结点机列表中。如果不希望使用当前结点(node2)，可以加上-nolocal选项：
mpirun -nolocal -machinefile mfile -np 3 cpi
选项-np给出的进程数与-machinefile给出的文件中的结点机数不一定要相等。如果进程数少于结点机数，则程序只使用其中的一部分结点。如果进程数多于结点机数，则一些结点上会运行多于一个进程。

[zhuqx_hp@node3 Se]$ nohup /home/software/openmpi-1.2.2-intel9/bin/mpirun -machiefile INCAR KPOINTS POTCAR POSCAR -np 8 /home/bin/vasp.openmpi >out

--------------------------------------------------------------------------
Open RTE detected a parse error in the hostfile:
    INCAR
It occured on line number 1 on token 3.
--------------------------------------------------------------------------
[node3:26677] [0,0,0] ORTE_ERROR_LOG: Error in file rmgr_urm.c at line 358
[node3:26677] mpirun: spawn failed with errno=-1

node2
node4
node5
node6
node7

Originally posted by beyondstar at 2010-07-17 08:43:13:
你的节点名不对，字面上看，是不是node02

Originally posted by goldenfisher at 2010-07-07 22:41:35:
在-machinefile后面还要加上你命名的各个节点的文件。记住，-machinefile是一个命令，而后面跟上的才是其对应的节点。
比如命名文件为mf,其中内容为上面所说的内容，则运行
mpirun -n 8 -machinefile mf vasp即可

SYNOPSIS
       Single Process Multiple Data (SPMD) Model:

       mpirun [ options ] [ ]

       Multiple Instruction Multiple Data (MIMD) Model:

       mpirun [ global_options ]
              [ local_options1 ] [ ] :
              [ local_options2 ] [ ] :
              ... :
              [ local_optionsN ] [ ]

       Note  that in both models, invoking mpirun via an absolute path name is
       equivalent to specifying the --prefix option with a value equiva-
       lent  to  the  directory where mpirun resides, minus its last subdirec-
       tory.  For example:

           % /usr/local/bin/mpirun ...

       is equivalent to

           % mpirun --prefix /usr/local

QUICK SUMMARY
       If you are simply looking for how to run an MPI application, you proba-
       bly want to use a command line of the following form:

           % mpirun [ -np X ] [ --hostfile ]  

       This  will  run X copies of in your current run-time environ-
       ment (if running under a supported resource manager, Open MPI's  mpirun
       will  usually  automatically  use  the  corresponding  resource manager
       process starter, as opposed to, for example, rsh or ssh, which  require
       the  use  of a hostfile, or will default to running all X copies on the
       localhost), scheduling (by default) in a  round-robin  fashion  by  CPU
       slot.  See the rest of this page for more details.

[zhuqx_hp@node2 Se]$ nohup /home/software/openmpi-1.2.2-intel9/bin/mpirun  -np 8 -hostfile mfile /home/bin/vasp.openmpi >out &

--------------------------------------------------------------------------
Failed to find the following executable:

Host:       node2
Executable: mfile

Cannot continue.
--------------------------------------------------------------------------