[molpro-user] Oversubscription of granted cores

[Reuti]

Hi,

while running Molpro in parallel in combination with MRCC in parallel too, we observe that on the one hand the distribution of load is exactly like expected (thanks to using the "-t" option to Molpro and not exporting OMP_NUM_THREADS).

On the other hand, the waiting Molpro processes seem to perform an active waiting by polling while waiting for the one process which started the `mrcc` external application. These waiting processes show a 100% load too and using `strace` I see "poll([{fd=6, events=POLLIN}], 1, 0)     = 0 (Timeout)" endlessly.

Is there any way to avoid this oversubscription which happens this way, or is the general advice to run Molpro in serial only, when using in combination with MRCC (being it serial or parallel)?

-- Reuti