Charm++ parallel programming system

[Phil Miller <mille121 AT illinois.edu>]

On Tue, Jul 20, 2010 at 06:08, Olivier Demengeon
<olivier.demengeon AT inria.fr>
 wrote:
> Hello,
>
> I hope i post this mail at the right place, i look for help on charm++
> freezing.
>
> I have a problem with charmrun freezing when gathering cpu tolopology.
>
> I use namd (2.7b3) on IB with charm++. I have a cluster with 32 nodes. Each
> nodes have 8 cores.

Is your charm++ build 'net-linux-x86_64-ibverbs' then?

> When I use 11 nodes or less, asking 8 process per node everything work fine.
> When i try to use more than 11 nodes, still asking 8 process per compter,
> charmrun start to sometime hang at gathering cpu topology. More I have node,
> more the chance of freeze increase.
>
> Here the commande line I use for 11 nodes :
>    /charmrun ++verbose +p88 ++nodelist nodes ++scalable-start ++batch 2
> ./namd2 <file>

Could you post the output of this command to the list? Could you also
try it without the "++batch 2" argument? For 32 nodes, no batching
should be needed.

Also, does your cluster have MPI installed, and if so what
implementation? A recent feature in Charm++, included with NAMD 2.7b3,
allows use of some MPI system's startup process to launch Charm++
programs.

Phil