Q1. I built MPI LAM on AIX 5.1 on a POWER-3 system with XL compilers. The
configure was run under OBJECT_MODE=64 and "--with-noromio --
with-shm-poolsize=100000000
--with-shm-short=8000
The lamboot -V
responds to
SSI rpi : crtcp lamd sysv tcp usysv
If I use this build on a POWER4 system, strictly within a node (for shared
memory) are
there any tunable parameters to improve performance ?
1. for message passing
2. for collective communication
The build seems to run OK on a POWER4 system - but major performance issues
relative to the
Parallel Environment on several communication intensive applications. MPI
LAM seems to be
choking with some buffer size limits. Is there a remedy ?
|
