On Aug 26, 2004, at 1:47 PM, jnt7 wrote:
> I ran into some trouble when porting LAM-7.0.6 to the G5; on single
> processes
> (using OpenMP to utilize both processors, i.e. MPI/OpenMP) the code
> works
> fine, but as soon as I have more than one process (say, one process
> per CPU) -
> not using OpenMP (i.e. MPI/MPI), my program fails or produces incorrect
> results.
>
> It appears that the MPI processes are unable to read files that it had
> created
> itself.
You'll need to provide a little more information -- that claim doesn't
make a lot of sense without the context of whatever you're trying to
do...
> What would be my best bet at a remedy; mount my G5 disks using NFS,
> configuring LAM/MPI --without-romio, or perhaps something else?
So are you writing files with MPI I/O in one run and then trying to
read them in again in a later run? Are you using the same number of
processes to write as to read? Are the files on NFS or on local
filesystems?
Rajeev can probably best talk about how those things are supposed to
work.
--
{+} Jeff Squyres
{+} jsquyres_at_[hidden]
{+} http://www.lam-mpi.org/
|
