MPIBlib: MPI Benchmark library

Collective benchmark

Performs regular universal or operation-specific collective benchmark. Basic arguments are described in Options for executables.

Example:

 $ mpirun -n 16 collective -O MPIB_Scatter_binomial -l libmpib_coll.so -m 1024 -M 2049 > collective.out

will benchmark the default binomial tree scatter operation with message sizes per process of 1024 and 2048

collective.plot draws the graph of the execution time (sec) against message size (kb) with error bars.

  • input: collective.out
  • output: collective.eps

Using the gnuplot script:

 $ gnuplot collective.plot