Product specifications

6 – MPI Sample Applications
OSU Latency
6-2 D000006-000 Rev A
Q
For all but the mandel and tachyon demos, when MPI applications are run with the
scripts provided, the results of the run will be logged to a file in
/opt/iba/src/mpi_apps/logs. The file name will include the date and time
of the run for uniqueness.
6.1
OSU Latency
This is a simple benchmark of end-to-end latency for various MPI message sizes.
The values reported are one-direction latency.
A script is provided to run this application that will execute an assortment of sizes:
1. cd /opt/iba/src/mpi_apps
2. ./run_lat
This will run assorted latencies from 0 to 256 bytes. To run a different set of message
sizes an optional argument specifying the maximum message size can be provided.
This benchmark will only use the first two nodes listed in mpi_hosts.
During this benchmark the /opt/iba/src/mpi_apps/mpi.param.pallas
config file is used.
6.2
OSU Latency2
This is a simple performance test of end-to-end latency for various MPI message
sizes. The values reported are one-direction latency.
A script is provided to run this application, which will execute an assortment of sizes:
1. cd /opt/iba/src/mpi_apps
2. ./run_lat2
This will run assorted latencies from 0-to-4Megabytes.
This benchmark will only use the first two nodes listed in mpi_hosts.
During this benchmark the /opt/iba/src/mpi_apps/mpi.param.pallas
config file is used.
6.3
OSU Bandwidth
This is a simple benchmark of maximum unidirectional bandwidth.
A script is provided to run this application which will execute an assortment of sizes:
1. cd /opt/iba/src/mpi_apps
2. ./run_bw