Re: LDB questions/problems

From: Jim Phillips (jim_at_ks.uiuc.edu)
Date: Tue Dec 16 2003 - 22:44:32 CST

I don't understand the psfgen reference, but system size shouldn't be a
problem for psfgen. I'm guessing the load balancer is either running out
of memory or choking in the face of too many options (2M atoms is a lot).
I can definitely say it's getting stuck in Rebalancer::multirefine().

Try setting "maxSelfPart 1" and "maxPairPart 1", which eliminates the
partitional of intra- and inter-patch interactions into multiple objects.
This should reduce the number of choices the load balancer has to make.

Here's what these look like in SimParameters.C:

   // Maximum Partition options
   opts.optional("main", "maxSelfPart",
     "maximum number of self partitions in one patch", &maxSelfPart, 50);
   opts.range("maxSelfPart",POSITIVE);
   opts.optional("main", "maxPairPart",
     "maximum number of pair partitions in one patch", &maxPairPart, 20);
   opts.range("maxPairPart",POSITIVE);
   opts.optional("main", "numAtomsSelf",
               "maximum number of atoms in one self compute distribution",
               &numAtomsSelf, 125);
   opts.range("numAtomsSelf",NOT_NEGATIVE);
   opts.optional("main", "numAtomsPair",
               "maximum number of atoms in one pair compute distribution",
               &numAtomsPair, 200);
   opts.range("numAtomsPair",NOT_NEGATIVE);
   opts.optional("main", "numAtomsPair2",
               "maximum number of atoms in one pair compute distribution",
               &numAtomsPair2, 400);
   opts.range("numAtomsPair2",NOT_NEGATIVE);

-Jim

On Tue, 16 Dec 2003, Brian Bennion wrote:

> I am running on a linux cluster, using an in house rh 7.3 kernel with elan
> quadrics switches. So I don't think the solaris info applies. It may be
> just to big for psfgen to do correctly.
> I have also had several elan buffering errors which I worked around by
> increasing the mpi buffers from 200MB to 400MB
> Brian
>
>
> On Tue, 16 Dec 2003, Gengbin Zheng
> wrote:
>
> >
> > Hi,
> >
> > I think I'v heard this before - there is some weird memory usage
> > behavior with the Solaris-Sparc-SMP-CC version of Charm++
> > built upon net-sol-smp version of Charm++.
> > You may try the latest version of NAMD, but you have to build every thing
> > from source out of the CVS - Charm++ and NAMD (they have separate CVS
> > servers).
> >
> > Gengbin
> >
> > On Mon, 15 Dec 2003, Brian Bennion wrote:
> >
> > >
> > > Interesting...so I assume you meant reduce the number of processors I was
> > > using. I went from 64 to 32 cpus and it made it through the load
> > > balancing step before the queue timedout. However, one stray pme error did
> > > persist.
> > > Now i am more perplexed. my system has >2 million atoms so I don't know
> > > how to get this system minimized effeciently with any peace of mind...
> > >
> > > Brian
> > >
> > >
> > >
> > > On Tue, 16 Dec 2003, himanshu chandola
> > > wrote:
> > >
> > > > I was getting the same kinda errors when i used to run it on a solaris box
> > > > .Though in my case the namd would give ldb messages - stay there for a
> > > > few minutes and then crash showing memory problems. Reducing the number of
> > > > processes helped and removed this problem from reappearing.
> > > >
> > > > You can just try this perhaps!
> > > >
> > > >
> > > > cheers
> > > >
> > > > himanshu
> > > >
> > > > ----------------------------------------
> > > > Morpheus: Do you believe in fate, Neo?
> > > > Neo: No.
> > > > Morpheus: Why Not?
> > > > Neo: Because I don't like the idea that I'm not in control of my life.
> > > >
> > > >
> > > > On Mon, 15 Dec 2003, Brian Bennion wrote:
> > > >
> > > > >
> > > > > Hello,
> > > > >
> > > > > I am attempting to minimize a large system. I call for 100 steps of min,
> > > > > however, at the 10th step the load distribution system kicks in, gives the
> > > > > following message in the log and stays there for at least 20 minutes at
> > > > > which point I get kicked out of the queue. I turn the balancer off, with
> > > > > ldbstrategy none and things progress, but not without many stray pme grid
> > > > > errors.
> > > > >
> > > > > Any insight?
> > > > > Brian
> > > > >
> > > > > log snippet....
> > > > >
> > > > > ENERGY: 9 4446565.9422 815057.1233 0.0000
> > > > > 0.0000 -8238197.7208 17559560.4307 0.0000 0.0000
> > > > > 0.0000 14582985.7753 0.0000 14582985.7753 14582985.7753
> > > > > 0.0000 228178.0317 226466.3996 35898114.1381 228178.0317
> > > > > 226466.3996
> > > > >
> > > > > LDB: LOAD: AVG 17.6773 MAX 22.956 MSGS: TOTAL 17196 MAXC 336 MAXP 5
> > > > > None
> > > > > LDB: LOAD: AVG 17.6773 MAX 21.2128 MSGS: TOTAL 20029 MAXC 633 MAXP 6
> > > > > Alg7
> > > > >
> > > > > --
> > > > > *****************************************************************
> > > > > **Brian Bennion, Ph.D. **
> > > > > **Computational and Systems Biology Division **
> > > > > **Biology and Biotechnology Research Program **
> > > > > **Lawrence Livermore National Laboratory **
> > > > > **P.O. Box 808, L-448 bennion1_at_llnl.gov **
> > > > > **7000 East Avenue phone: (925) 422-5722 **
> > > > > **Livermore, CA 94550 fax: (925) 424-6605 **
> > > > > *****************************************************************
> > > > >
> > > > >
> > > > >
> > > > >
> > > >
> > >
> > > --
> > > *****************************************************************
> > > **Brian Bennion, Ph.D. **
> > > **Computational and Systems Biology Division **
> > > **Biology and Biotechnology Research Program **
> > > **Lawrence Livermore National Laboratory **
> > > **P.O. Box 808, L-448 bennion1_at_llnl.gov **
> > > **7000 East Avenue phone: (925) 422-5722 **
> > > **Livermore, CA 94550 fax: (925) 424-6605 **
> > > *****************************************************************
> > >
> > >
> >
> >
>
> --
> *****************************************************************
> **Brian Bennion, Ph.D. **
> **Computational and Systems Biology Division **
> **Biology and Biotechnology Research Program **
> **Lawrence Livermore National Laboratory **
> **P.O. Box 808, L-448 bennion1_at_llnl.gov **
> **7000 East Avenue phone: (925) 422-5722 **
> **Livermore, CA 94550 fax: (925) 424-6605 **
> *****************************************************************
>
>

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:37:14 CST