[gmx-users] problem with cluster

[Yang Ye]

For (a), consult your job scheduling documentation on allocation. I use SGE,
so I use fill-up for the parallel environment that I put my jobs to. I also
use Compaq alpha's batch system, so I assign both cpu no. and node no. to
the job.

For (b), it is because of the bandwidth is not enough to exchange data
between cpus if they are working at 100%. So in this case, only use one.

Also, since gromacs is parallel by process not by threads. Using
hyperthreading actually hurts the performance. Wait for new version for it
(as Eric said).

Regards,
Yang Ye
Computational Biology Lab
School of Biological Sciences
Nanyang Technological University
Singapore
Tel: 65 6316 2884

-----Original Message-----
From: gmx-users-bounces at gromacs.org [mailto:gmx-users-bounces at gromacs.org]
On Behalf Of Nguyen Hoang Phuong
Sent: Wednesday, April 13, 2005 9:10 PM
To: gmx-users at gromacs.org
Subject: [gmx-users] problem with cluster


Dear All,

we have a new cluster consists of 11 Intel cpus with hyperthreading. I
have some problems with running gromacs on this cluster:

a) When I submit a job, it is distributed at random on 2 cpus (I asked for
2 cpus). How can I run the job on specific cpus?

b) The cpus which run the mdrun program alway work with about 40% of
power. How can I increase the power of the cpus?

Thank you very much in advance for any hint.

Phuong

_______________________________________________
gmx-users mailing list
gmx-users at gromacs.org
http://www.gromacs.org/mailman/listinfo/gmx-users
Please don't post (un)subscribe requests to the list. Use the 
www interface or send it to gmx-users-request at gromacs.org.