[GE users] Slots as host/queue consumable

William Hay ccaawih at ucl.ac.uk
Thu Mar 23 10:19:38 GMT 2006


Hi,
we have a SGE 5.3 cluster which was installed and configured for us by an
external contractor (name omitted to protect the guilty).

Each of the nodes has two queues on it <nodename>.cre which is configured
for loosely integrates parallel jobs and <nodename>.q which is configured
for serial jobs.

If I type qstat -F each of the .cre queues lists an attribute
hc:slots while each of the .q queues lists an attribute
qc:slots with one exception (host sf009) which lists qc:slots 
for both queues.

When using qconf -se all but the problems node list slots=<numprocessors>
under complex_values and NONE under complex_list.

The .cre queues do have complex_list and complex_values but they make no
reference to slots.

qconf -sc queue |grep slots 
produces no output
qconf -sc host|grep slots
produces
slots            s          INT    0               <=    YES         YES        1
None of the other complexes define slots either

In an attempt to bring sf009 into alignment with the other hosts I tried:
qconf -rattr exechost slots 8 sf009
but the only response is:
qconf: "slots" is not a exechost attribute

The presence of slots as a qc on both queues means that this host ends up
running twice as many jobs as it should which leads to jobs in the serial
queue being suspended which is a bad thing.  Raising the suspend threshold
would presumably just slow the jobs down which is also a bad thing.

Any suggestions for how to fix this short of a complete re-install?


William Hay


---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
For additional commands, e-mail: users-help at gridengine.sunsource.net




More information about the gridengine-users mailing list