Filter
Top Products
A – Benchmark Programs
Benchmark 3: Messaging Rate Microbenchmarks
IB6054601-00 D A-3
Q
MPI_Isend function, while the receiving node consumes them as quickly as it can
using the non-blocking MPI_Irecv, and then returns a zero-length acknowledgement
when all of the set has been received.
You can run this program with:
$ mpirun -np 2 -ppn 1 -m mpihosts osu_bw
Typical output might look like:
# OSU MPI Bandwidth Test (Version 2.0)
# Size Bandwidth (MB/s)
1 2.250465
2 4.475789
4 8.979276
8 17.952547
16 27.615041
32 52.676363
64 104.704225
128 198.347505
256 335.396929
512 521.273433
1024 829.369420
2048 884.249845
4096 926.723948
8192 934.093084
16384 941.191459
32768 938.179872
65536 945.163478
131072 950.206048
262144 951.938802
524288 952.912385
1048576 953.716825
2097152 953.922714
4194304 954.119999
Note that the increase in measured bandwidth with messages size results from the
fact that latency’s contribution to the measured time interval becomes relatively
smaller.
A.3
Benchmark 3: Messaging Rate Microbenchmarks
mpi_multibw is the microbenchmark used to highlight QLogic’s messaging rate
results. This benchmark is a modified form of the OSU NOWlab’s
osu_bw