From: Axel Kohlmeyer (akohlmey_at_gmail.com)
Date: Tue Jul 10 2012 - 05:18:28 CDT
if this is on an SGI Altix with a single system image,
don't you have to use something like the dplace or
omplace command to have your job migrated/locked
to the suitable partition. have you talked to the sysadmins?
On Tue, Jul 10, 2012 at 7:38 AM, Norman Geist
> Hi flavio,
> if you are sure this is a multicore version without network support, there
> must be something else what the nodes share. Maybe check if your output
> frequencies like restart and dcd are too slow so that you mess up your
> shared filesystem. That would also worsen with every node that start
> If you have a NAMD version with network support, check how the jobs are
> distributed. Use the for instance the top command for that to see what
> processes are running on the nodes.
> If that also doesn't help, think about what else the nodes could share and
> if they are real independent nodes and not kind of virtual machine etc. As
> we don't know your setup, it's hard to make more precise advice.
> Norman Geist.
>> -----Ursprüngliche Nachricht-----
>> Von: owner-namd-l_at_ks.uiuc.edu [mailto:owner-namd-l_at_ks.uiuc.edu] Im
>> Auftrag von flavio seixas
>> Gesendet: Sonntag, 8. Juli 2012 21:48
>> An: undefined undefined
>> Betreff: namd-l: Multi core problem
>> I am running namd on SUSE LINUX Enterprise Server 11 SP1 (x86_64)
>> kernel 18.104.22.168 64bits, at 12 cores parallel in a single node of a SGI
>> Altix ICE 8400 LX machine.
>> The programs works fine, but when I submit another namd job to a
>> different node, the velocity of processing decrease by half. If I
>> submit a third job also in a different node, the velocity decrease by
>> 1/3 at all nodes.
>> Theoretically, the jobs running in different nodes do not suppose to
>> interfere in eat other.
>> Does someone experience a similar problem to give me some clue about
>> what happen?
-- Dr. Axel Kohlmeyer akohlmey_at_gmail.com http://goo.gl/1wk0 International Centre for Theoretical Physics, Trieste. Italy.
This archive was generated by hypermail 2.1.6 : Tue Dec 31 2013 - 23:22:13 CST