Heya all,
Just to wrap up this thread, our problem with the was caused by the the
fact that our shared lsf cluster is divided into enclosures to aid in
other user's MPI jobs (so that jobs try to run across machines plugged
into the same switch). This confused the lrmsinfo-lsf script. Ulrich
provided us with a pre-prerelease glite-info-dynamic-scheduler-lsf rpm
and tool, bhostsinfo, that enabled the lrmsinfo-lsf to extract the
information we need and pass it on to our information system.
My thanks again to Ulrich, Massimo and Jeff.
Cheers,
Matt
|