[gmx-users] mdrun_mpi crashes for no apparent reason

[Vishal Vaidyanathan]

Hi,
  I am using mdrun_mpi from release 3.1.4 and I am facing the following
problem: the program stops executing and crashes without warning or error
to stderr or log. This does not seem to be a simulation issue since
resuming the run from the last frame works fine.  There are no LINCS
warnings or errors. I get several hours to 1-2 days of runtime before a
crash.

  Any ideas why this could be happening? Seems to be some technical issue
rather than a simulation problem.

  System:
          SMP dual athlons, Suse linux 8.0, kernel 2.4.18-64GB-SMP
          LAM MPI, lam-6.5.4-usysv.1, from gmx webpage
	  fftw packages from gmx webpage

	  Everything was installed from the rpm's provided

  I'm using MPI only for processors on the same board - ie. nothing should
really be depending on the network, so I don't think it's a networking
issue either. And there's ample disk space.
  Sometimes while using tpbconv to resume, I get a SIGSEGV when it reads
the last frame of the trajectory (no incomplete frame warning, just
SIGSEG). Have to use the -time option to resume in that case. But
mosttimes I get the "incomplete frame" warning and tpbconv exits normally.

  Has anybody else noticed such behaviour? It's hard to guess what's wrong
because there's absolutely no error message...

  Any suggestions are welcome.

  Thanks,
								Vishal