[GE users] Slot-wise suspend on subordinate

pollinger harald.pollinger at sun.com
Tue Jul 28 15:21:19 BST 2009


reuti wrote:
> Am 28.07.2009 um 11:54 schrieb pollinger:
> 
>> fx wrote:
>>> pollinger <harald.pollinger at sun.com> writes:
>>>
>>>> 3.2 Functionality
>>>>
>>>>     Like with queue-instance-wise suspend on subordinate, both  
>>>> the subordinate
>>>>     and the superordinate queue instance must be on the same host.
>>> I meant to ask before:  what's the reason for that restriction in the
>>> queue-instance-wise case?
>> The goal of the suspend on subordinate was in the first place to free
>> host resources for high priority jobs. I know there are possibly  
>> other,
>> not host bound resources, that could be freed by suspend on  
>> subordinate.
> 
> Well, CPUs (i.e. slots) are freed. But consumed resources aren't  
> freed by a suspension. They are still occupied when a job gets  
> suspended.
> 
> This would be a completely new project to free for example licenses  
> or consumed virtual_free by a suspension and give them back to the  
> system I think. Maybe this should be called hibernate job (with new  
> options to qmod) then and have similar entries like the  
> subordination. The algorithm is much more complex, as first the job  
> must be stopped, then the resources given back and a new job can  
> consume the freed resources. And the un-hibernate of the first job  
> must be forbidden, as long as the new jobs runs...

I'd call this "preemption". For consumables that are managed by SGE this 
will work. For other consumables like a license, it requires either the 
application to free the license when it gets a "you will be preempted" 
signal, or the license manager to trust SGE and free the license.

I will add it to the spec, it will likely become a RFE.

Regards,
Harald

> 
> -- Reuti
> 
> 
>> I don't want to change the scheduler completely, so I will stick to  
>> this
>> behaviour, at least for the first version of the slot-wise suspend on
>> subordinate.
>>
>> Regards,
>> Harald
>>
>> ------------------------------------------------------
>> http://gridengine.sunsource.net/ds/viewMessage.do? 
>> dsForumId=38&dsMessageId=209870
>>
>> To unsubscribe from this discussion, e-mail: [users- 
>> unsubscribe at gridengine.sunsource.net].
> 
> ------------------------------------------------------
> http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=38&dsMessageId=209873
> 
> To unsubscribe from this discussion, e-mail: [users-unsubscribe at gridengine.sunsource.net].


-- 
Sun Microsystems GmbH         Harald Pollinger
Dr.-Leo-Ritter-Str. 7         Sun Grid Engine Engineering
D-93049 Regensburg            Phone: +49 (0)941 3075-209  (x60209)
Germany                       Fax: +49 (0)941 3075-222  (x60222)
http://www.sun.com/gridware
mailto:harald.pollinger at sun.com
Sitz der Gesellschaft:
Sun Microsystems GmbH, Sonnenallee 1, D-85551 Kirchheim-Heimstetten
Amtsgericht Muenchen: HRB 161028
Geschaeftsfuehrer: Thomas Schroeder, Wolfgang Engels, Wolf Frenkel
Vorsitzender des Aufsichtsrates: Martin Haering

------------------------------------------------------
http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=38&dsMessageId=209906

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



More information about the gridengine-users mailing list