azurehpc/apps/imb-mpi
Xavier Pillons 9c1dfdcccd removed merged scripts 2020-02-18 16:33:57 +01:00
..
allreduce.sh fixes 2020-02-18 16:28:28 +01:00
parser_allreduce.sh chmod+x 2020-02-18 15:27:43 +01:00
parser_ringpingpong.sh chmod +x 2020-02-18 14:17:12 +01:00
readme.md initial import 2019-07-20 15:15:29 +01:00
ringpingpong.lsf added ping pong with LSF 2019-09-05 19:47:24 +02:00
ringpingpong.sh fixes 2020-02-18 16:28:28 +01:00

readme.md

IMB-MPI benchmarks

The following PBS scripts are available to run the IMB-MPI1 benchmarks:

  • ringpingpong.pbs
  • allreduce.pbs

The ringpingpong will report the ping pong times between adjacent nodes in the hostlist. A separate file is output for each run but the 1024 byte results will be displayed in a table sorted from best to worst. Here is an example of how to run:

qsub -l select=8:ncpus=1:mpiprocs=1,place=scatter:excl ringpingpong.pbs

The allreduce will perform the 8 and 16 byte all reduce with all the cores for the job. Run as follows:

qsub -l select=8:ncpus=60:mpiprocs=60,place=scatter:excl allreduce.pbs