Jed Brown wrote:
Are you saying the output of mpicc/mpif90 -show has the same optimization flags? MPICH2 usually puts it's own optimization flags into the wrappers.
Jed, thank you for your reply. Yes, mpif90 shows (other than differing libraries) identical flags.
Ralph Castain wrote:
Did you set -mca mpi_paffinity_alone 1? This will bind the processes to cores and (usually) significantly improve performance.
Ralph, thank you for the suggestion. I had focussed on RDMA, and this made a significant difference. I have only had time to re-run an ensemble of one configuration (rather than the suite I had been running) and it improved the OpenMPI performance by 19.5%. So, it would appear this was the primary cause.
I will read through the documentation to find how to make this the default.
Thank you for your assistance! I look forward to the 1.3.4 improvements. Cheers, Brian
