We have a problem with running Amber 9 with OpenMPI using multiple
threads, on our system with the following configuration:
CPU: 2 x Intel Xeon Quad Core 2.66 GHz
MEMORY: 8 GB
OS: Redhat 4 Update 4 (X86_64)
MPI: OpenMPI 1.2.4
COMPILERS: Intel C++, Fortran (version 10.1.008) and MKL 9.1.023
SCHEDULER: Sun Grid Engine 6u4
Sander and pmemd are compiled using intel compilers both for serial and
parallel configurations and for x86_64. The command "make test" returns no
error. However, when a parallel simulation is submitted through openmpi
(smp configuration - all threads running on the same machine), amount of
total virtual memory is limited to 1 GB no matter how many threads are
used (as seen from the "top" command). Hence, the speed of the simulation
is too slow. What would be the possible cause of this memory limitation?
We have tried different openmpi programs and saw that these programs could
allocate 8 GB memory, so no restriction is enforced by the system.
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
Received on Fri Apr 18 2008 - 21:15:38 PDT
