[HTCondor-users] On Condor 8.0.x, the Parallel Universe and partitionable slots

[Mark Calleja <mc321@xxxxxxxxx>]

Hi All,

In order to get cross-host MPI jobs to run, I've been trying to get Parallel Universe jobs to work with partitionable slots for Condor 8.0.6 under Debian "wheezy", but without success. I should say that using static slots works just fine, as does using the Vanilla Universe with partitionable slots for MPI jobs sitting on one host (using shared memory). Has anyone got partitionable slots to work with the Parallel Universe? The problem occurs when the host resources are being allocated, i.e. before any file transfer takes place, and what nodes are allocated just go into Claimed state and hang.

Thanks for any help,
Mark