[GE users] SGE assigns a loaded machine for a new job.

Reuti reuti at staff.uni-marburg.de
Wed Oct 27 21:01:35 BST 2004

    [ The following text is in the "ISO-8859-1" character set. ]
    [ Your display is set for the "ISO-8859-10" character set.  ]
    [ Some special characters may be displayed incorrectly. ]


> We have a linux cluster with most of machines with identical configurations.
> All of our cluster nodes are dual cpu so I have assigned 2 slots per machine.
> We have set load threshold of np_load_avg to 0.8. 
> In some cases we have noticed that GE does not assign a least loaded machine
> to submit  a job. 
> Here is an example of the problem we are facing. 
> Suppose there are 10 machines out of 20 machines available for GE to submit
> jobs, so we have 20 slots at that moment. Out of 10 availabe machines some
> users are running a job by bypassing GE (They are running a job, by
> rlogin-ing to machine), so the load on all 10 available machine is less than
> load threshold but all have different loads.(For GE all these 10 machines are
> available as all the 20 slots are available and load is less than load
> threshold). Some times we have noticed that SGE assigns a machine to a newly
> submitted job to a machine which  has a np_load_avg value of 0.5 (At this
> moment there are some machine which has np_load_avg value to 0.0). Is there a
> way that I can specify that GE chooses a machine for  a new job to the least
> loaded machine. 

selecting the least loaded machine should be the default behavior of SGE. Maybe 
the scheduler will speed up the things while assuming, that a machine without 
any used slot at all, should have no load. But anyway: can't you force your 
users to request an interactive slot for their tasks?

Cheers - Reuti

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