NONMEM and PsN on a Cluster with TORQUE

From: Ahmed Suleiman Date: September 15, 2012 technical Source: mail-archive.com
Dear NM users, I have just installed NONMEM successfully on a cluster together with perl and PsN. Our cluster uses TORQUE as the resource manager. I have been running some tests to check the installation and came with 2 questions: 1. When I submit a bootstrap using a batch script requesting 10 nodes and 10 ppn for example, with a command at the end of the script saying "bootstrap run1.mod -samples=100 -threads=100 -nodes=10", the run times of the bootstrapped samples are a couples of times longer than the run time of the original fit. Apparently the multi-node operation doesn't work as it should have been working and the processes run only on the first node. When I omit the "threads option", 5 threads are used (as specified in the psn.conf) and the run times drop to their usual values, which further proves the over commitment of the executing node. How can I solve this within the batch script ? 2. It would be easier if I were able to run PsN without having to write batch scripts, however I am not aware of the options which would allow me to request resources other than those set as default by the administrator. For example the default wall time in our cluster is 1 hour, and the only option which I found to change this was -max_runtime, but unfortunately this works only with SLURM. Is it possible to specify different resources from the command interface ? Kind Regards, Ahmed Abbas Suleiman, MSc PhD Candidate at Bonn University Institute of Pharmacology - Clinical Pharmacology Unit Hospital of the University of Cologne Gleueler Str. 24, 50931 Cologne Germany