[GE users] job_load_threshold under 6.0u1

Stephan Grell - Sun Germany - SSG - Software Engineer stephan.grell at sun.com
Fri Nov 5 07:48:29 GMT 2004

Sean Dilda wrote:

>On Thu, 2004-11-04 at 03:35, Stephan Grell - Sun Germany - SSG -
>Software Engineer wrote:
>>sorry, for replying sooooo late.
>>There is a misunderstanding in how the load adjustments work. It is
>>a scheduler internal feature and has nothing to do with the rest of
>>system (that is why the configuration parameters are part of the
>>configuration ;-) ).
>>The job_load_adjustments mean, that when a job is dispatched to a
>>it will add that amount of load to that host (in your example:
>>would increase by 1.0). Since it will take some time, before the job
>>is started
>>and the load sensor reports the new load, one can simulate that load
>>the help of the load adjustments. 
>>The decay time is the time interval, after which the simulated load is
>>gone and
>>one is sure, that the job is started and the real load is correct. 
>>The purpose of this feature is to prevent an overload situation on the
>>hosts and
>>to ensure, that a queue will not go into alarm state based on the jobs
>>by the scheduler. 
>>To ensure, that the load_adjustments are working correct, take your
>>example and
>>assign that queue 10 slots. Submit 10 jobs into the system and you
>>will see, that
>>only one will be started.
>Thanks for the response.  After my original post, I noticed the load
>adjustments showing up if I ran 'qstat -j' on a job.  I was just
>confused since the load adjustment didn't show up anywhere else.
>I still think the adjusted values should shows up as an alarm state if
>they essentially cause one.
>If I were to set a load_threshold to 'mem_free=-900MB', will that
>actually affect people who submit jobs with '-l mem_free=XXX' ?  Due to
>the fact that setting that load threshold doesn't change the output of
>'qhost -F mem_free', I would think not, but now I'm not sure.

Well, yes and no. The load thresholds are evaluated against load sensors
that is
reported by the hosts. But if you set mem_free to 900MB and the host has
800 MB available, the queue instances on the host will go into alarm
state, and no
further jobs will be dispatched, regardless of what the jobs request.

If the host has 1 GB available, the queue will not be in alarm state.
But a job, that
requests -l mem_free=1.1 GB will not be dispatched to that host, since
it has not
enough memory available.


>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