I believe the mpich test cases are correct with myrinet, so it's not clear
where
the problem is. However, Amber is the only code I currently have running
with mpi.
David Pearlman wrote:
> Ha! I am trying to compile on basically the same hardware
> (Linux cluster, Myrinet, PGI compilers) and am seeing
> the identical problem you are: Code works OK with
> np 1 (1 processor) but not with np > 1 (temperature rises
> quickly, then code hangs)...
>
> I have to imagine this has something to do with the Myrinet
> aspect...
>
> Have not yet worked it out...If anyone has experience with the
> same hardware, perhaps you can share some insights as to what
> to look at?
>
> dap
>
> John Nicholas wrote:
> >
> > All,
> >
> > Can anyone provide help with the following problems?
> >
> > I'm trying to compile and run Amber 6 on an IBM Netfinity PIII cluster,
> > using Redhat, the PGI compilers, MPICH 1.2.1..7, and a Myrinet network.
> >
> > 1. The code appears to compile correctly using gcc and pgf77 without
> > linking
> > in MPICH (a serial version). This code hangs during the nmr and nmode
> > test
> > cases, producing some output and then continuing to use cpu time (hours)
> >
> > without further output. Compiling the whole code without optimization
> > seems to cure the problem, but compiling just the nmr specific modules
> > of sander without optimization does not. I haven't tried to isolate the
> > nmode
> > problem. Has anyone experienced and hopefully cured this problem?
> >
> > 2. The code also appears to compile correctly using the same protocol
> > as above with mpich linked in. Running the code serially produces
> > the same problems as above. However, running on more than one proc
> > the sander md results (i.e bnpi) are much different than the standard,
> > with temperature
> > rising quickly. The tests do not run long enough for the system to get
> > totally
> > out of control, but the results are clearly wrong.
> >
> > Any problems with the PGI compilers?
> >
> > Any problms with specific revs of mpich?
> >
> > Thanks, John
> >
> > --
> > ___________________________________________________
> >
> > John B. Nicholas Ph.D.
> > Bioorganic Chemistry Department
> > Genentech, Inc. Phone: (650) 225 2581
> > 1 DNA Way Fax: (650) 225 2061
> > South San Francisco, CA 94080 e-mail: jbn_at_gene.com
> > ___________________________________________________
--
___________________________________________________
John B. Nicholas Ph.D.
Bioorganic Chemistry Department
Genentech, Inc. Phone: (650) 225 2581
1 DNA Way Fax: (650) 225 2061
South San Francisco, CA 94080 e-mail: jbn_at_gene.com
___________________________________________________
Received on Fri Oct 26 2001 - 09:19:11 PDT
