[torqueusers] pbs and MatStudio Modeling interactions

Chris Samuel csamuel at vpac.org
Tue Jan 3 16:37:26 MST 2006


On Wednesday 04 January 2006 09:59, Steven A. DuChene wrote:

> I just spoke with Mr. Jian-Jie Liang from Accelerys and he indicated
> that Accelrys is in a short period of time doing a major upgrade release
> of 4.0 and said "many" of the issues highlighted by Chris will be fixed
> in that release.

That's basically what I'd heard, but I didn't know whether I was able to 
report that level of detail or not. :-)

> He did expressly indicate there still are known issues with running with
> Maui that prevent using Maui as a scheduler above PBS (due to faults in Maui
> according to him).

Erm, Maui has no bearing on this at all - I think they misunderstand what it 
does.

> I indicated we are using Moab here but he still expressed doubts that it
> would work.

We are running Materials Studio (with my patches) using Torque and Moab and it 
works - sort of.

The previous versions ran OK in MPI mode but since they moved to HP-MPI (why 
why why!) I can't get it to work for any more than 2 CPUs on a single node.  
My reading of their Perl code makes me think that it's not correctly (if at 
all) setting up the HP-MPI environment.   But as I've never touched HP-MPI 
before I've no idea what's really going on.

Save to say it's nothing to do with Torque or Moab/Maui.

> He said it had never  been tested or "certified" He also said they have
> stopped testing with OpenPBS and have transitioned to PBSPro.

I had to be quite forceful to get them to accept that our configuration 
(dedicated management node with a separate login node for our users) was a 
reasonable configuration and that their way of detecting PBS (looking for a 
pbs_sched process) was not sufficient.  They wanted us to run it all on the 
management node.

IMHO too often ISVs seem to assume that a cluster is built to run just their 
product and their "supported configuration" just doesn't work when we've got 
over 400 different users running all sorts of code, most of which they've 
written themselves...

Chris
-- 
 Christopher Samuel - (03)9925 4751 - VPAC Deputy Systems Manager
 Victorian Partnership for Advanced Computing http://www.vpac.org/
 Bldg 91, 110 Victoria Street, Carlton South, VIC 3053, Australia



More information about the torqueusers mailing list