[GE users] multiple jobs on a same node while other nodes are idle

templedf dan.templeton at sun.com
Mon Jan 4 14:25:48 GMT 2010


Did you set it to "0.5" or "np_load_avg=0.5"?

Daniel

gascan211 wrote:
> Daniel,
>
> Thanks for info. I changed load_avg back to 0.5
> and I still see the fill-up behavior. Not sure, if you have read Micheal's reply regarding this issue, he notices similar behavior when multiple jobs are submitted at same time. But if I submit jobs at different interval its submitting to different nodes as wanted. And we think its a bug with 6u1 as he doesn't see this in earlier versions. It would be great if you can also share your thoughts to that email chain.
>
> Thanks,
> Kumar 
>
>   
>> I realized yesterday in the shower that I got my wires crossed when I 
>> told you how to set the job load adjustments. :)  What I was trying to 
>> say is that if it is set to NONE, then you get fill-up behavior.  If you 
>> set it to something like np_load_avg=0.5, then you should see jobs 
>> spread out across your hosts.  (I got job_load_adjustments mashed up 
>> with load_report_time in my head, hence the 0:0:0 thing.)  What you have 
>> now as your job_load_adjustments is equivalent to NONE, which would 
>> explain the fill-up behavior.
>>
>> So, just to be clear, you should have something like:
>>
>> job_load_adjustments              np_load_avg=0.5
>>
>>
>> Daniel
>>
>> gascan211 wrote:
>>     
>>> Thanks Daniel, Rayson
>>>
>>> I tried all the options you guys suggested, still behaving hosts fill up instead of least used fill up. I am attaching my scheduler config, may be you can suggest anything else.
>>>
>>> Thanks,
>>> Kumar
>>>
>>> Also tried with job_load_adjustments              np_load_avg=0.000000
>>>
>>> algorithm                         default
>>> schedule_interval                 0:0:5
>>> maxujobs                          0
>>> queue_sort_method                 load
>>> job_load_adjustments              0:0:0=0.000000
>>> load_adjustment_decay_time        0:7:30
>>> load_formula                      -slots
>>> schedd_job_info                   false
>>> flush_submit_sec                  0
>>> flush_finish_sec                  0
>>> params                            none
>>> reprioritize_interval             0:0:0
>>> halftime                          168
>>> usage_weight_list                 cpu=1.000000,mem=0.000000,io=0.000000
>>> compensation_factor               5.000000
>>> weight_user                       0.250000
>>> weight_project                    0.250000
>>> weight_department                 0.250000
>>> weight_job                        0.250000
>>> weight_tickets_functional         0
>>> weight_tickets_share              0
>>> share_override_tickets            TRUE
>>> share_functional_shares           TRUE
>>> max_functional_jobs_to_schedule   200
>>> report_pjob_tickets               TRUE
>>> max_pending_tasks_per_job         50
>>> halflife_decay_list               none
>>> policy_hierarchy                  OFS
>>> weight_ticket                     0.500000
>>> weight_waiting_time               0.278000
>>> weight_deadline                   3600000.000000
>>> weight_urgency                    0.500000
>>> weight_priority                   0.000000
>>> max_reservation                   0
>>> default_duration                  INFINITY
>>>
>>> ------------------------------------------------------
>>> http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=38&dsMessageId=235078
>>>
>>> To unsubscribe from this discussion, e-mail: [users-unsubscribe at gridengine.sunsource.net].
>>>
>>>       
>
> ------------------------------------------------------
> http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=38&dsMessageId=235306
>
> To unsubscribe from this discussion, e-mail: [users-unsubscribe at gridengine.sunsource.net].
>

------------------------------------------------------
http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=38&dsMessageId=236352

To unsubscribe from this discussion, e-mail: [users-unsubscribe at gridengine.sunsource.net].



More information about the gridengine-users mailing list