Hi, I have a question regarding the mpool_hugepage_page_size MCA parameter.
Would varying this parameter affect performance over the infiniband network for large messages (messages greater than the eager limit)? I have set the parameter to 4K and 4MB (default is 2MB on the machine) and ran a simple latency benchmark (osu_latency) between two nodes over an Infiniband interconnect. The results show a marginal slowdown with larger page sizes but is insufficient as the variance between the multiple runs is sufficiently large on the 4K runs. I have had a quick glance through the ompi source code and cannot definitively say whether this parameter affects performance of large messages. Many Thanks, Dean
