On Apr 17, 2006, at 5:18 PM, Timothy G Thompson wrote:
> Thanks for your reply...the problem is fixed (as you suggested) by
> adding
> an argument to the mpirun:
> mpirun -ssi rpi tcb -v pathSchema.
>
> Interestingly (I suppose), when I tried: mpirun -ssi rpi usysv -v
> pathSchema (and ...sysv) things
> failed in MPI_Init.
That's very strange - I'm not sure why usysv would cause a failure.
If you are interested in tracking the issue down, it would be good to
know what error message you are seeing.
As to why the tcp performance is better than the usysv/sysv
performance... When TCP is used alone, LAM/MPI will block on
blocking operations (like PWSC). However, it must poll if two
devices are needed. In some cases (and it differs on each platform
when this occurs), you can get degenerate cases where your
performance gets much worse due to this polling. It sounds like you
were hitting one of those cases on your Solaris machines.
Brian
--
Brian Barrett
LAM/MPI developer and all around nice guy
Have a LAM/MPI day: http://www.lam-mpi.org/
|
