[GE users] question for the allocation rules

Ryoji Takahashi ryoji at cct.lsu.edu
Mon Jul 28 17:36:24 BST 2008


    [ 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. ]

thanks,

(i didn't realized ... the guy who was asking same question was our 
formal manager of our cluster ...)

Reuti wrote:
> Hi,
>
> Am 28.07.2008 um 18:21 schrieb Ryoji Takahashi:
>
>> Hi,
>> We are facing weird problem by $fill_up rules ...
>>
>> setting is following ...
>>
>> pe_name           mvapichfill
>> slots             96
>> user_lists        NONE
>> xuser_lists       NONE
>> start_proc_args   /home/sge/mvapich/startmpi.sh -catch_rsh $pe_hostfile
>> stop_proc_args    /home/sge/mvapich/stopmpi.sh
>> allocation_rule   $fill_up
>> control_slaves    TRUE
>> job_is_first_task FALSE
>> urgency_slots     min
>>
>> our machine is quad core, each nodes has 8 processors.  When we 
>> submit two 4 processors jobs,
>> $fill_up assigns from 0 to 3 for the two jobs ... two jobs collide at 
>> the same processors.
>> it looks like two jobs are filled out from 0 to 3.
>>
>> what could be solution for this?
>
> this was recently on the Beowulf list. Please have a look at this thread:
>
> http://www.beowulf.org/archive/2008-July/022484.html
>
> -- Reuti
>
>
>> thanks,
>>
>>>
>>>
>>> I would recommend picking a more flexible allocation rule:
>>>
>>> - If you want to force your jobs to only stay within one physical 
>>> 8-core chassis then use "$pe_slots"
>>>
>>> - If you want to pack the parallel job on as few machines as 
>>> possible, choose "$fill_up"
>>>
>>> - If you want to spread the job across as many machines as possible, 
>>> choose "$round_robin"
>>>
>>>
>>> -Chris
>>>
>>>
>>>
>>>
>>>
>>> On Jul 17, 2008, at 4:04 PM, Ryoji Takahashi wrote:
>>>
>>>> pe_name           mvapich8
>>>> slots             96
>>>> user_lists        NONE
>>>> xuser_lists       NONE
>>>> start_proc_args   /home/sge/mvapich/startmpi.sh -catch_rsh 
>>>> $pe_hostfile
>>>> stop_proc_args    /home/sge/mvapich/stopmpi.sh
>>>> allocation_rule   8
>>>> control_slaves    TRUE
>>>> job_is_first_task FALSE
>>>> urgency_slots     min
>>>>
>>>> ...
>>>> our cluster's architecture is quad core ... so admin guy seems to 
>>>> make strict to 8 processors per nodes.
>>>> however, fill out rules solve this problem.
>>>>
>>>> thanks.
>>>
>>>
>>> ---------------------------------------------------------------------
>>> To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
>>> For additional commands, e-mail: users-help at gridengine.sunsource.net
>>>
>>
>>
>> ---------------------------------------------------------------------
>> To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
>> For additional commands, e-mail: users-help at gridengine.sunsource.net
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
> For additional commands, e-mail: users-help at gridengine.sunsource.net
>


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