On Fri, 8 Jun 2018 at 03:18, Jed Brown <[email protected]> wrote:
Jeff Hammond <[email protected]> writes:
The PETSc folks appear to be the biggest proponents of blocking poll (solely for purposes of running dozens of MPI processes on their laptops, it seems) and they seem to prefer ch3:sock. I defer to their experience as to whether a better implementation exists.
Indeed, the test suite has several tests that use 9-16 ranks and take less than 1 second even when oversubscribed using ch3:sock, but are disturbingly slow with modern MPICH. The conundrum of developers not being able to debug tests on their laptop and many CI solutions not being able to execute those tests is reminiscent of the testing fiasco that is multi-GPU development.
Intel MPI seems to have some backoff support with I_MPI_SPIN_COUNT, I_MPI_SHM_SPIN_COUNT, and some others vars. I never tested how well these features work, though. I guess PETSc developers would just like to have a way to somehow alter behavior at runtime by exporting some environ vars in a CI script. This would benefit testing and latency-insensitive use cases. -- Lisandro Dalcin ============ Research Scientist Computer, Electrical and Mathematical Sciences & Engineering (CEMSE) Extreme Computing Research Center (ECRC) King Abdullah University of Science and Technology (KAUST) http://ecrc.kaust.edu.sa/ 4700 King Abdullah University of Science and Technology al-Khawarizmi Bldg (Bldg 1), Office # 0109 Thuwal 23955-6900, Kingdom of Saudi Arabia http://www.kaust.edu.sa Office Phone: +966 12 808-0459