This problem went away yesterday. There was no intervening reboot of
my cluster or a recompile of the code. So all I can surmise is
something got cleaned up in a cron script. Wierd.
Anyways, now I've benchmarked the HPL using OpenMPI vs LAM-MPI. The
OpenMPI runs about 13% to sometimes 50% slower than the LAM-MPI. I'm
running over TCP and using SSH.
Does anyone expect OpenMPI to be slower than LAM-MPI under these
conditions?
Lee Peterson
On Apr 10, 2006, at 9:57 PM, Lee D. Peterson wrote:
Dear OpenMPI,
I'm transitioning from LAM-MPI to OpenMPI and have just compiled
OMPI 1.0.2 on OS X server 10.4.6. I'm using gcc 3.3 and XLF (both
f77 and f90), and I'm using ssh to run the jobs. My cluster is all
G5 dual 2GHz+ xserves, and I am using both ethernet ports for
communication. One is used for NFS and the other is for MPI.
I've had few problems the past year running this config with LAM-
MPI (latest release). But what worked before doesn't with OpenMPI
1.0.2.
When I run any parallel job that spans multiple machines, the
process runs indefinitely. I've checked this using the BLACS and
PBLAS test routines, the HPL benchmark, and even a simple mpi-pong
program. All of them execute but produce no output past some
initial output, consuming 100% of the CPU on every node that's
launched. In contrast, all of these programs run in a few seconds
on a single node, with two processors, and up to -np 8. When I
cntrl-C to stop the program, openmpi safely stops all the
processes, no matter how many machines have been used.
I noticed a couple postings from the past few months that seem to
be related but didn't seem to be quite the same symptoms. Any ideas
what could be going on?
OpenMPI is a really great project, and it is obvious the quality of
software development that has gone into it. I appreciate all your
help. My config.log and omni-info.out files are attached.
Lee Peterson
Professor
Aerospace Engineering Sciences
University of Colorado
Boulder, CO
<ompi-output.tgz>
_______________________________________________
users mailing list
us...@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/users
