[GE users] Task scheduling

Ron Chen ron_chen_123 at yahoo.com
Tue Feb 15 03:19:51 GMT 2005


--- Toby Sargeant <sargeant at wehi.edu.au> wrote:
> This has a few problems:
> 
> 1) If the queue isn't specified when a job is
> started, it executes tasks in all queues, which
> seems to cause 2 tasks to execute per CPU (although
> I'm not sure why this happens).

Are they Xeon CPUs with HT? How many CPUs does each
machine have?

Run $SGE_ROOT/gridengine/utilbin/<arch>/loadcheck --
it will show you the number of CPUs available.

> 2) Migrating the longer running job back to all.q
> after the other job finishes has to be done by hand,
> and until it's done we waste half of our CPUs.

Or script it?

> 3) While we've only had two jobs at once thus far as
> soon as we have 3 or more this solution doesn't
work.

If the jobs are from different users, you can setup a
share tree to allow each user to have a fairshare of
the cluster.

If they are from the same user, setting
"max_aj_instances" (see sge_conf(5)) will limit the
number of tasks allowed to run per array job.

So what you need to do is to have a script that checks
the number of jobs running or pending in the cluster,
and then divide the number of CPUs by the number of
jobs. Say if you have 64 CPUs and 4 jobs for example,
your script will set max_aj_instances to 16, and thus
only 16 tasks will be able to run concurrently.

 -Ron





		
__________________________________ 
Do you Yahoo!? 
Yahoo! Mail - Easier than ever with enhanced search. Learn more.
http://info.mail.yahoo.com/mail_250

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