Re: Running NAMD on cluster

From: Philip Peartree (P.Peartree_at_postgrad.manchester.ac.uk)
Date: Wed Apr 30 2008 - 07:21:28 CDT

What would be useful here is to run the script on a varying number of
nodes, with a varying number of processors, at some point the
communication between nodes will become a bottleneck and the program
will slow down, if you do this and post the figures you get (possibly
with the test job (ApoA1) which is available from the NAMD site) then
we'd be able to tell whether it's behaving normally or not.

Quoting sudipta sinha <sudipta.mml_at_gmail.com>:

> Dear NAMD user,
> I want to run NAMD job parallely on our hp-proliant cluster. It
> has 16 nodes and one master server with operating system RHEL4 and
> also the home directory is automounted. To do this job, I downloaded
> the NAMD_2.6_Linux-i686.tar.gz file from the UIUC site. This file is
> untar in a home directory of an user account. Now I want to simulate
> a water box by this NAMD program. To fire this job, the command which
> I have used is written below
>
> /home/sudipta/NAMD_2.6_Linux-i686/charmrun ++remote-shell ssh
> ++nodelist namd.nodelist /home/sudipta/NAMD_2.6_Linux-i686/namd2 +p3
> water51.conf ++verbose
>
> The namd.nodelist file contains,
> group main
> host iitkgp
> host node1
> host node2
>
> The program is running fine on each node parallely. But with
> increasing the number of nodes the CPU efficiency decreases. As a
> result the runtime increases with increasing the number of nodes. That
> means just opposite is happening with increasing the number of nodes.
> Does it mean each job running on each node independently with huge
> network crowding? But actually I want to run one job on multiple
> processors. Could anyone suggest me what is my wrong.
>
> Thanks and regards
> Sudipta
>
>

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:49:25 CST