[Mauiusers] Roundrobin node allocation

Jon Wright jon at gate.sinica.edu.tw
Tue Sep 13 20:11:36 MDT 2005



Hi,

Sometimes (but always it seems at 7pm on a Fri) one of our nodes will have a 
problem where it will accept a job from maui but the job will crash 
immediately. If the user resubmitting the job maui will send it straight back 
to the problem node and we get a repeat, which leaves the user very very 
frustrated. Is there away of getting maui to use a round robin algorithm on 
the available applicable nodes for each job so that while this type of 
problem is not solved it is at least minimised for the users?

Our cluster has 4-5 types of nodes so we run 4-5 queues (different queues go 
to different nodes) so any round robin could not round robin on the whole set 
of available nodes but on the subset of nodes which are available and belong 
to the right class.

Current our maui.cfg has
NODEALLOCATIONPOLICY       MINRESOURCE

somelike 
NODEALLOCATIONPOLICY       ROUNDROBIN

would be excellent,

Many thanks
Jon


More information about the mauiusers mailing list