PC GAMESS-related discussion club


Re: pcgamess -t switch and parallel runs

Igor Polyakov
polyakoviv@gmail.com


hello!

Frankly speaking, just setting -t switch like this -t /scr/avishai in the end should work. I should also point out, that in your commandline i saw something like /home/scr/avishai, but if you use nfs on this cluster and you have /home/scr/ as a common directory you shouldn't use it. For scratch data you should only use local disk of your node. For me works just fine -t /tmp/pcgamess or -t /scr/pcgamess. Using -t switch you will be able to run 4 compute processes on each node to fully utilise your computational power. Of course u need to have appropriate procgrp file.

>almost forgot, i'm running sles 10sp1 on all nodes, core2 quad cpu's using kernel 2.6.16, mpich 1.2.7 and pcgamess 7.15 dynamically linked to mpich.

I thought that dynamically linked version of pcgamess works only when initiated by mpirun (or mpiexec) command in the first place, like mpirun pcgamess , but in your commandline there is simple pcgamess -p4pg which is usually used for static linked versions. But maybe i'm wrong with this one.

>regards,
>avishai
>

Good luck!


[ Previous ] [ Next ] [ Index ]           Sat May 10 '08
[ Reply ] [ Edit ] [ Delete ]           This message read 1312 times