Personal Communication
MPI-F Performance

MPI MPI-F Performance

This page compares basic performance of MPI-F, MPICH, and MPL for the IBM-SP2. MPL is the native message passing library for the SP2. MPICH is a public domain library by ANL which sits on top of MPL. As the figures demonstrate MPI-F performs equally well in latency and bandwidth as the MPL. MPICH has somewhat understandably worse latency but similar bandwidth. MPI-F has a break at 8K in bandwidth because it tries to eliminate resource problems that can arise when various out of order messages are sent or if multiple communicators engage in communication.

[ Return to MPI-F home page ]


Latency for MPI-F, MPICH, MPL on the IBM SP2

Latency


Bandwidth for MPI-F, MPICH, MPL on the IBM SP2

Bandwidth


Hubertus Franke, email: frankeh@watson.ibm.com, December 15th, 1995
[ IBM homepage | Watson | Order | Search | Contact IBM | Help | (C)| (TM) ]