[GE users] Help with intelligently scheduling jobs.

Aaron Seligman aaron at ssc.ucla.edu
Wed Apr 13 20:49:50 BST 2005


    [ The following text is in the "utf-8" character set. ]
    [ Your display is set for the "ISO-8859-10" character set.  ]
    [ Some characters may be displayed incorrectly. ]

Hello,

I would like some help configuring my cluster.
I have everything setup and running, but I would like to setup the some restrictions / scheduling to better handle the types of jobs our users run.

We have 4 nodes. Each node is a Dual AMD64.
Our nodes are named sge1, sge2, sge3, sge4
sge1 has 8GB of Ram
sge2, sge3, sge4 have 2GB of Ram each.
We currently run only batch jobs on the cluster, no parallel.
We are running gridengine 6

sge4 is a interactive nodes we use qrsh (bound to ssh) for users to connect. When they connect they have to request the consumable h_data. 
We do not need to change this node.
Sge4 is its own cluster queue

sge1 - sge3 is the other cluster queue, there are no limitations for users apart from the consumable resource h_data. H_data (memory) is set to the total amount of ram for each node in each node's respective configuration (queue instance)

I use the sort by sequence number in the scheduler configuration and edit the sequence_nr value on the queue instances to have jobs go to sge2 and sge3 first, before going to sge1.

2. I would like to have the following scheduling policy for my jobs in addition to my current setup
-Jobs < 1 Gig of h_data can run on any node.
-Jobs > 1 Gig and requiring < 120 minutes (wallclock) can run anywhere, BUT the policy will be enforced by automatically terminating their job once the user-provided ?required? time is up. This will simultaneously capture ?run-away? jobs.
-Jobs > 1 Gig and requiring >120 minutes will ONLY be run on the sge1.
Is this possible? How would I go about setting this up.?

In large clusters people seem to restrict a slot to a certain type of job. But that is not want I want to do here, plus we don't have the money to buy more equipment.

I really appreciate any input or advice that I may receive.

Thanks,

Aaron Seligman
aaron at ssc.ucla.edu




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