From: Gengbin Zheng (gzheng_at_ks.uiuc.edu)
Date: Mon Apr 12 2004 - 13:28:10 CDT
If you compiled MPI version, you can run NAMD just like a MPI program
using mpirun typically, which normally is PBS friendly.
If you are using net version, charmrun has to be used to launch the
program. by default it uses "rsh", but you can change it to "ssh".
Also you need to write a script in your job script to generate a list of
nodes. something like this in csh syntax:
#!/bin/tcsh
set node=nodelist.$PBS_JOBID
echo "group main ++shell ssh" >! $node
list=`echo $PBS_NODES | sed 's/ /\n/g' | sort | uniq`
foreach h ( $list )
echo "host $h" >> $node
end
and when you run NAMD, pass ++nodelist nodelist.$PBS_JOBID
to tell it to read the nodes from the nodelist file. for example:
./charmrun +p64 ++nodelist nodelist.$PBS_JOBID ./namd2 alanin
Gengbin
On Sun, 11 Apr 2004, Amarda Shehu wrote:
>
> Dear all,
>
> I have installed namd on the Rice Teracluster, which is a 1 TeraFLOP Linux
> cluster based on IntelR ItaniumR 2 Processors. The jobs in the teracluster
> are run through a PBS script. In the teracluster, parallel jobs are run by
> using mpiexec and specifying the number of processes needed.
>
> I am running into problems trying to run namd in parallel on the
> teracluster nodes.
>
> Does namd run in parallel by trying to do rsh to the
> nodes of the specified nodelist? As this is basically trying to log in to
> the nodes of the cluster, it fails because rsh to the nodes is not
> allowed. Is this the only way to run namd in parallel?
>
> I thought I would check with you before contacting my system's administrators.
>
> Thank you,
> Amarda Shehu.
>
>
> ***********************************
> Amarda Shehu
> Graduate Student,
> Computer Science, Rice University.
> ***********************************
>
This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:38:34 CST