On , 2021Jan28, at 10:35, Jed Brown <[email protected]<mailto:[email protected]>> wrote:
Why plots like this are not _absolutely standard_ on all HPC sites' webpages is a source of continuing mystery to me. I've been asking for it for years. Curious. Why? What would it tell you? I mean, other than “42”. I’ll ask around how this is done in TX. Considering all the tinkering Intel does to their MPI and what with all the UCX crap, we keep a close eye on MPI performance, especially collectives. latency is super variable (depending on the partition you get and what other jobs are running elsewhere on the machine; Blue Gene famously didn't have that problem). The harsh truth of fat-trees with static routing: https://arxiv.org/abs/1909.12195 Btw, I think we’ve turned off dynamic routing for now because of too many problems. Victor.
