[GE users] Slot Allocation Strategy for Differing Memory Sizes ..

Graham Jenkins Graham.Jenkins at its.monash.edu.au
Mon Jun 23 06:02:49 BST 2008


    [ The following text is in the "ISO-8859-1" character set. ]
    [ Your display is set for the "ISO-8859-10" character set.  ]
    [ Some special characters may be displayed incorrectly. ]

We're using N1GE 6.1 on the Monash Sun Grid, with some 220  lx24-amd64
slots in about 90 (2-CPU and 4-CPU) execute nodes.  These have been grouped
in queues according to memory capacity and CPU-count .. so that, for
instance ..

   dqu4 .. priority 400 .. contains 4-CPU servers with h_vmem=3900.00M
   bqu4 .. priority 500 .. contains 4-CPU servers with h_vmem=16G

Our users have been instructed to submit jobs which have non-trivial
run-time and/or memory requirements with statements like:

    qsub -l h_rt=24:00:00 -l h_vmem=16G BigMemJob.sh

Thus, if a job has no particular large-memory requirement, it can
execute on slots in either of the above (or other) queues.

Unfortunately, the scheduler looks at Load and Slots-in-User per
node .. and will often allocate small-memory jobs to 'bqu4' .. thereby
consuming resources which might actually be needed later by
another job.

So .. is there an easy way of telling it "Only  allocate 'bqu4' slots
if large-memory has been requested, or if All 'bqu2' slots are in use"?

Regards ..

-- 
Graham Jenkins
Senior Software Specialist, eResearch
Monash University (Clayton Campus, Bldg 11, Rm S503)

Email: Graham.Jenkins at its.monash.edu.au
Tel:   +613 9905-5942 (office)   +614 4850-2491 (mobile)

---------------------------------------------------------------------
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