[GE users] power off idle nodes

reuti reuti at staff.uni-marburg.de
Mon Feb 8 22:17:20 GMT 2010


    [ The following text is in the "utf-8" character set. ]
    [ Your display is set for the "ISO-8859-10" character set.  ]
    [ Some characters may be displayed incorrectly. ]

Hi Richard,

Am 08.02.2010 um 15:56 schrieb rhierlmeier:

> Hi Reuti,
>
> On 02/08/10 13:56, reuti wrote:
>> Am 08.02.2010 um 11:31 schrieb marcingoldyn:
>>
>>> We have Service Domain Manager with Power Saving adapter.
>>
>> I'm also thinking of implementing some power saving feature. How
>> sophisticated is this setup using SDM? I mean: suppose one machine is
>> switched off, then some jobs are submitted. But as these jobs require
>> more than one mache each, switching on the one switched-off machine
>> is useless. Will it discover that at least one other machine needs to
>> get idle before any further action should happen and switch it on at
>> that time?
>>
>
> Correct me, may be I am wrong, but I have the impression that with  
> your last
> email it was the first time you had a question to the mailing  
> list ;). I am glad
> to answer your question.

no, there were some in the past. But maybe they slipped through  
general attention:

http://gridengine.sunsource.net/ds/viewMessage.do? 
dsMessageId=212283&dsForumId=38


> SDM does it as follows:
>
> In SDM a service representing a the GE cluster exists. Inside of  
> this service a
> MaxPendingJobsSLO exists. This SLO observes the pending queue the  
> GE cluster. If
> it finds any job (matching some job criteria) it sends a resource  
> request to the
> SDM system. The resource request contains the information what  
> resources (CPU,
> memory, ...) and how many are needed.
> The MaxPendingJobsSLO knows how many slots the pending jobs need.  
> It knows PEs,
> arrays etc. and creates out of the pending jobs a proper resource  
> request.

Great.

> SDM asks other services whether they can fulfill this resource  
> request. The
> power saving spare pool can be such a service. Resources which are  
> assigned to
> this service are powered off. If there is a demand for a powered  
> off resources
> it powers on the hosts. Once the hosts are available SDM moves the  
> resource from
> the power saving spare_pool to the requesting service.
>
> In combination with a normal spare_pool you can implement a caching  
> strategy to
> reduce the number of power cycles for the machines. To many power  
> cycles reduce
> the lifetime of your systems.

Well, I never touched Hedeby/SDM for now, as I have not so many  
clusters to move resources around. Shutting down parts of the cluster  
automatically is of course something I'm interested in.

-- Reuti


> Richard
>
>
>> A simple approach could also be to look at the 15 minutes avg. load
>> (3rd value in `uptime`) and issue an `ipmitool ... chassis power
>> soft` when there is no load and no job having this node assigned any
>> longer; and issuing `ipmitool ... chassis power on` when there are
>> waiting jobs.
>>
>
>
>
>
>> -- Reuti
>>
>>
>>> Marcin
>>>
>>>
>>> W dniu 2010-02-08 11:03, cgull pisze:
>>>> Hi,
>>>>
>>>> We would like to automatically switch off machines when they have
>>>> been idle for a time approx an hour.
>>>> Is there anyway through Sun Grid Engine to do this, or do we need
>>>> to create our own scripts to do this?
>>>> Maybe we need to install something like Hedeby to work with SGE to
>>>> do this?
>>>>
>>>> Any thoughts on the best way to achieve this would be appreciated.
>>>>
>>>> Thanks for your time in advance.
>>>>
>>>> ------------------------------------------------------
>>>> http://gridengine.sunsource.net/ds/viewMessage.do?
>>>> dsForumId=38&dsMessageId=243926
>>>>
>>>> To unsubscribe from this discussion, e-mail: [users-
>>>> unsubscribe at gridengine.sunsource.net].
>>>>
>>> ------------------------------------------------------
>>> http://gridengine.sunsource.net/ds/viewMessage.do?
>>> dsForumId=38&dsMessageId=243927
>>>
>>> To unsubscribe from this discussion, e-mail: [users-
>>> unsubscribe at gridengine.sunsource.net].
>>
>> ------------------------------------------------------
>> http://gridengine.sunsource.net/ds/viewMessage.do? 
>> dsForumId=38&dsMessageId=243938
>>
>> To unsubscribe from this discussion, e-mail: [users- 
>> unsubscribe at gridengine.sunsource.net].
>
>
> -- 
> - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -  
> - - - -
> Richard Hierlmeier           Phone: ++49 (0)941 3075-223
> Software Engineering         Fax:   ++49 (0)941 3075-222
> Sun Microsystems GmbH
> Dr.-Leo-Ritter-Str. 7	     mailto: richard.hierlmeier at sun.com
> D-93049 Regensburg           http://www.sun.com/grid
>
> Sitz der Gesellschaft:
> Sun Microsystems GmbH, Sonnenallee 1, D-85551 Kirchheim-Heimstetten
> Amtsgericht München: HRB 161028
> Geschäftsführer: Thomas Schröder, Wolfgang Engels
> Vorsitzender des Aufsichtsrates: Martin Häring
>
> ------------------------------------------------------
> http://gridengine.sunsource.net/ds/viewMessage.do? 
> dsForumId=38&dsMessageId=243949
>
> To unsubscribe from this discussion, e-mail: [users- 
> unsubscribe at gridengine.sunsource.net].
>

------------------------------------------------------
http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=38&dsMessageId=243994

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



More information about the gridengine-users mailing list