| 6:35:12 AM PDT - Thu, Aug 18th 2016 |
|
Hi,
I have checked the disk space of the computer nodes, and I have here 64GB available on a free node, without any calculations running:
Filesystem Size Used Avail Use% Mounted on
tmpfs 64G 112K 64G 1% /dev/shm
I recompiled the NWChem changing the ARMCI_NETWORK for "MPI-MT", and seems to be working now. I am running a big MP2 calculation to test the memory usage by GA and the everything seems to be ok so far. However, I think the MPI-MT option is not so efficient for my computer systems. So it would be great to have the "MPI-PR" version working to compare.
I don't think that multiple jobs would be the problem because I have always used the node-exclusive option in the submission script. Do you think 64GB of disk space is enough at least for small calculations? I quite not understand why the version with "MPI-MT" works and the other one doesn't. Do you have any guess to explain that?
Thanks & Have a nice day
|