each entry of the benchmark now corresponds to a single MPI_Send/Receive pair, instead of going back-and-forth (thus the numbers are roughly halved)
increased statistics
here are some results from a 4-node (i.e. 16 rank) job on tursa
made some changes since the start of this PR:
* each entry of the benchmark now corresponds to a single `MPI_Send/Receive` pair, instead of going back-and-forth (thus the numbers are roughly halved)
* increased statistics
~~here are some results from a 4-node (i.e. 16 rank) job on tursa~~
Now it is really ready to merge.
Also some further documentation can be found here: https://git.dev.dirac.ed.ac.uk/portelli/dev-project-management/src/branch/main/doc/benchmarks
portelli
approved these changes 2024-11-19 10:37:24 +00:00
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Latency benchmark in a point-to-point configuration
latency_benchmarkto Point-to-Point latencymade some changes since the start of this PR:
MPI_Send/Receivepair, instead of going back-and-forth (thus the numbers are roughly halved)here are some results from a 4-node (i.e. 16 rank) job on tursaAfter discussion, update timer resolution
Now it is really ready to merge.
Also some further documentation can be found here: https://git.dev.dirac.ed.ac.uk/portelli/dev-project-management/src/branch/main/doc/benchmarks