I am wondering what effects it will have when I configure openmpi
with --enable-mpi-threads: we are using the TM interface to torque
and therefore each MPI program gets assigned a certain number of cores
on a set of nodes by the scheduler. If the MPI program now starts
additional threads how does that work?
- will this lead to an oversubscription of nodes
- will the creation of additional threads fail, if the number of
processes on a node has reached the process count on that node
assigned through torque?
How is this done properly?
Head, Research Computing
WestGrid Site Lead
IT Services phone: 778 782-4691
Simon Fraser University fax: 778 782-4242
Burnaby, British Columbia email: siegert_at_[hidden]
Canada V5A 1S6