[GE users] Batch jobs claiming unused cycles in a mainly interactive job cluster?

Ben Singer bdsinger at Princeton.EDU
Wed Mar 16 21:53:35 GMT 2005


Context:
Our head node's /etc/csh.login runs qrsh when users log in to our 
64-node Apple Xserve Dual-G5 cluster, routing users to available nodes 
using the default 5.3 policy (2 slots/exec host). Users mostly run in 
interactive mode, for days at a time, though we are gently educating 
and encouraging batch submissions.

Question lead-up:
I implement the relocation howto-- and have batch jobs run only when 
interactive jobs are idle, but that seems inefficient. Better would be 
to suspend the batch job only if the interactive job is sufficiently 
busy. Using a suspend threshold based on total load will cause 
resource-intensive batch jobs to suspend themselves.

Actual Question:
Is there a way to suspend a queue based on the load of another queue?


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