[GE users] SGE-6.0: How to achieve increasing and decreasing seq_no?

Andy Schwierskott andy.schwierskott at sun.com
Wed Oct 27 14:45:45 BST 2004


looks like you used the implict behavior the for a parellel job the next
queue in the sort order is selected. I'm not sure if this behavior can
always be guaranteed (e.g. think about resource reserveration, fragmentation
of the sorted free queue slots due to finishing jobs). However I assume you
were satisfied with the actual behavior;-)

If it's important for you that every individual host has a different
sequence number (and you don't want to use hostgroups where thorugh
hostgroups a couple of hosts have the same sequence number) you nned to
accept the new syntax - through the "qconf -?attr" interface you have full
possibilites to script this configuration.


> Well, thanks for your answer, but I don't quite understand
> how my problem is related to host groups!?  Erik.
> On Wed, 27 Oct 2004, Kirk Patton wrote:
>> Did you consider using host groups?
>> man hostgroup
>> Kirk
>> On Wed, Oct 27, 2004 at 03:25:55PM +0200, Erik Soyez wrote:
>>> Good Day,
>>> with SGE-5.3 I used to configure clusters with e.g. increasing seq_no for
>>> a "normal" queue and decreasing seq_no for an "express" queue in order to
>>> keep parallel jobs stick together and keep express and normal jobs as far
>>> away from each other as long as possible.  That was pretty easy with a
>>> script that just added up seq_no when modifying queues.
>>> Now, how do I translate this configuration into SGE-6.0 where a
>>> "qconf -mq normal at node1" does not work - I don't like the idea of adding
>>> something like "seq_no [node1=1],[node2=2],[node3=3]" etc. to
>>> "qconf -mq normal" for 200 hosts - if there might be another way.
>>> Thanks for your help,
>>> Erik.

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