Queue Manager Tips

From wiki
Jump to: navigation, search

Explanation of queue configuration

Following the new version of gridengine ("Son of Gridengine"), the queue configuration and why it is as such is described here. First let's use the script "qsee.pl" to hav look at what queues there are:

# qsee.pl
Marvin: Total of 4 Queues Available. Details:
Queue all.q: 78 cores(c) on 11 nodes: 0c @ marvin / 10c @ node1 / 10c @ node2 / 10c @ node3 / 10c @ node4 / 0c @ node5 / 10c @ node6 / 10c @ node7 / 10c @ node8 / 4c @ node9 / 4c @ node10.
Queue highmemory.q: 24 cores(c) on 2 nodes: 12c @ node9 / 12c @ node10.
Queue lowmemory.q: 42 cores(c) on 8 nodes: 6c @ node1 / 6c @ node2 / 6c @ node3 / 6c @ node4 / 0c @ node5 / 6c @ node6 / 6c @ node7 / 6c @ node8.
Queue marvin.q: 16 cores(c) on 1 nodes: 16c @ marvin.
Queue single.q: 16 cores(c) on 1 nodes: 16c @ node5.
Node view: how many cores (c) taken up, per node, by all queues? (red implies node is off-line)
marvin:16c node1:16c node10:16c node2:16c node3:16c node4:16c node5:16c node6:16c node7:16c node8:16c node9:16c

Note that the marvin node has its own queue and does in fact (though it's not mentioned) has the most CPUs (32). Only 16 are used int he queue because the other 16 are for programs outside the queue, Typically these are normal interactive sessions that consist of low-load operations such as listing directories, viewing files, etc.

  • marvin.q has its own queue because it is the only node with direct internet access.
  • single.q consists on one full node (node5) for program that prefer to occupy one full node/

Pipe a jobscript to qsub

echo -e 'date; sleep 2; date' | qsub -cwd -j y -V -S /bin/bash -q all.q -N pipetest

This will send a string with the command date, followed by sleep 2 seconds, and then date again. A trivial operation to see if the command works. If the -S option is left out, there will be a warning about lack of a tty, and lack of job control, although the script will still work.

Change job priority

qalter -p <priority> <jobid>