I’ve written about NUMA effects and process affinity on this blog lots of times in
In my last post, I described the Simple mode of Open MPI v1.7’s process affinity system.
The Simple mode
In my last post, I mentioned that we just finished a complete revamp of the Open MPI process
Process affinity is a hot topic. With commodity servers getting more and more complex internally (think: NUMA and NUNA),
In prior blog posts, I’ve talked about the implications of registered memory for both MPI applications and implementations.
Today we feature a deep-dive guest post from Ralph Castain, Senior Architecture in the Advanced R&D group