Roy,
I have your patch. I'll look at it and let you know.
Thanks,
Brian Christiansen
Roy Dragseth wrote:
On Sunday 22 February 2009 13:25:59 Roy Dragseth wrote:
Hi,
I think I've got the solution coded now. By introducing a new config
parameter, IDLEJOBDEPTH, one can get maui to continue to parse down the
idle job chain until it reaches the IDLEJOBDEPTH limit. The default value
is 1 to preserve the current behaviour of maui or else the patch will not
be accepted by CR. My plan is to use this in combination with a per user
based MAXIJOB to prevent single users from flooding the queue. If you set
IDLEJOBDEPTH to the total number of job slots in your cluster you should be
able to fill the available job slots within a reasonably few iterations.
It's not the optimal solution, one should perhaps introduce some logic that
understands why some jobs might be idle while others further down the chain
can run, but given the potential complexity introduced by combining
classes, standing reservations and node features I'm not clever enough to
figure out how.
I'll submit a patch against v3.2.6p21 to bugzilla for review by CR as soon
as its up and running again, I get this error right now
It seems to take a while before the bugzilla comes online so here's the patch.
Can anyone at CR tell me if this will be accepted or not? Please let me know
if something is missing or not acceptable for inclusion into the maui
mainline.
We've been running maui with this patch since Sunday 23rd and the utilization
has increased by almost 15%.
r.
------------------------------------------------------------------------
_______________________________________________
mauiusers mailing list
[email protected]
http://www.supercluster.org/mailman/listinfo/mauiusers
_______________________________________________
mauiusers mailing list
[email protected]
http://www.supercluster.org/mailman/listinfo/mauiusers
