From: Axel Kohlmeyer (akohlmey_at_gmail.com)
Date: Sat Mar 17 2012 - 15:27:40 CDT
On Fri, Mar 16, 2012 at 1:39 PM, Thomas Lemmin <thomas.lemmin_at_gmail.com> wrote:
> Hi,
>
> I am performing some benchmarks of NAMD 2.7b1 on a Bluegene/P.
> I tested two systems; One with 500K atoms and one with 1M atoms. The
> scaling is good, but I might have some memory problems.
it is a known "feature" of NAMD that its rank 0 task requires
much more memory than the remaining tasks, which in turn
limits it usability for large problems on machines with comparatively
little memory per node.
one way to alleviate this issue is to build an smp version
(of charm++ and then NAMD with it. check out:
http://www.ks.uiuc.edu/Research/namd/2.8/notes.html#compiling
not sure, if this is tested on blue gene machines.
cheers,
axel.
> Did anyone benchmark such systems on Bluegene/P with 2GB of memory per node?
> Can I trust the memory usage displayed in the following line:
>
> Info: Initial time: 4096 CPUs 0.03012 s/step 0.174305 days/ns 145.43 MB memory
>
> Is this the memory used per process.
>
> Regards
>
> Thomas Lemmin
>
-- Dr. Axel Kohlmeyer akohlmey_at_gmail.com http://goo.gl/1wk0 College of Science and Technology Temple University, Philadelphia PA, USA.
This archive was generated by hypermail 2.1.6 : Mon Dec 31 2012 - 23:21:19 CST