[mpich-discuss] mpiexec : multi nodes and multi-cores questions
zhouh at anl.gov
Fri May 6 11:25:23 CDT 2022
Assuming ParaView is using the same behavior as upstream hydra
* -np 8 should launch 8 processes, 4 on each node
* -np 4 should launch 4 processes, 2 on each node
* To launch 4 processes on one node, try -hosts compute0:4,compute1:4 -np 4
* The number 4 behind each host instructs hydra to assign the first 4 ranks to the host before moving on to another assignment, the default is 1. Hydra uses round-robin assignment policy
Hope it helps,
From: Nicholas Yue via discuss <discuss at mpich.org>
Sent: Friday, May 6, 2022 11:11 AM
To: discuss at mpich.org <discuss at mpich.org>
Cc: Nicholas Yue <yue.nicholas at gmail.com>
Subject: [mpich-discuss] mpiexec : multi nodes and multi-cores questions
I am using mpiexec from Paraview which from the `help` print out, indicates that it is derived from MPICH
Please see the intructions provided at
for further details
so I thought I might ask some questions on this mailing list.
I have set up a two node cluster, each compute node has 4 cores
To fully utilise all the cores in both nodes, is this the correct way to launch pvserver ?
/shared/ParaView-5.10.1-osmesa-MPI-Linux-Python3.9-x86_64/bin/mpiexec -hosts compute0,compute1 -np 8 /shared/ParaView-5.10.1-osmesa-MPI-Linux-Python3.9-x86_64/bin/pvserver
If I change to `-np 4` would mpiexec use only compute0 or would it still run on both compute0 and compute1 but use only 2 cores of each node ?
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the discuss