[GE users] creating a Master queue

Reuti reuti at staff.uni-marburg.de
Tue May 29 19:19:52 BST 2007


    [ 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. ]

Am 29.05.2007 um 17:10 schrieb Lönroth Erik:

> It seemed to work better, I seem to need some reading on that  
> "job_is_first_task" thing, but now I only get 7 SLAVES, do I need  
> to specify N+1 "pe slots" then, or is there some other way you  
> would normally do this?
>
>     348 0.55500 powerflow_ sssler       r     05/29/2007 17:02:56  
> master.103.q at ts103-3-0.sss.se. MASTER
>     348 0.55500 powerflow_ sssler       r     05/29/2007 17:02:56  
> short.103.q at ts103-3-0.sss.se.s SLAVE
>                                                                    
> short.103.q at ts103-3-0.sss.se.s SLAVE
>                                                                    
> short.103.q at ts103-3-0.sss.se.s SLAVE
>     348 0.55500 powerflow_ sssler       r     05/29/2007 17:02:56  
> short.103.q at ts103-3-22.sss.se. SLAVE
>                                                                    
> short.103.q at ts103-3-22.sss.se. SLAVE
>                                                                    
> short.103.q at ts103-3-22.sss.se. SLAVE
>                                                                    
> short.103.q at ts103-3-22.sss.se. SLAVE

No, if your special application needs N slots for the parallel task  
plus the master-process, then you will have to request N+1 slots from  
SGE, and use only 16 for the parallel part. But: is the master  
process really doing some work in the parallel part of your  
application - checked with e.g. top? Otherwise you have one node  
sitting there and waiting for the results, doing nothing in the  
meantime. Then it would be better to let also one slave task running  
on the system.

Additonally: this is only the output of SGE's granted slots for the  
job. It's not really showing, whether there is something running in  
this slot at all or not (the manpage of sge_pe might be a little bit  
missleading, as job_is_first_task is always checked for the output of  
"qstat -f", but only honored for a local qrsh if "control_slaves" is  
set to TRUE). Anyway: the generated nodelist is in both cases the same.

-- Reuti


>
>
> /Erik
>
> -----Original Message-----
> From: Reuti [mailto:reuti at staff.uni-marburg.de]
> Sent: den 29 maj 2007 16:54
> To: users at gridengine.sunsource.net
> Subject: Re: [GE users] creating a Master queue
>
>
> Am 29.05.2007 um 16:19 schrieb Lönroth Erik:
>
>> As you see:
>> master.102.q at ts102-2-0.sss.se. MASTER master.102.q at ts102-2-0.sss.se.
>> SLAVE
>>
>> ... Which is not what I want.
>>
>> I have set number os slots to "1" in the master queue, and number
>> of processors to "1", with no luck.
>
> The processors should be left to "UNDEFINED", unless you are on IRIX.
>
> What is the setting of "job_is_first_task" in your PE setting? It
> should be set to "TRUE" in your case.
>
> -- Reuti
>
>>
>>
>> Any help or ideas?
>>
>> /Erik
>>
>>
>>
>> -----Original Message-----
>> From: Reuti [mailto:reuti at staff.uni-marburg.de]
>> Sent: den 28 maj 2007 21:55
>> To: users at gridengine.sunsource.net
>> Subject: Re: [GE users] creating a Master queue
>>
>>
>> Hi,
>>
>> Am 28.05.2007 um 21:19 schrieb Milo:
>>
>>> I have, but I'm still confused as to how exactly it works. Thus far
>>> when I haven't explicitly specified the master queue using the
>>> -masterq argument,
>>> SGE obviously uses my only queue (all.q) for both master and slave.
>>> I've
>>> also noticed that the master node still gets a slave process
>>> assigned to it.
>>> Will this slave process still get spawned on the master queue if I
>>> use the
>>> -masterq argument to qsub?
>>
>> yes - might happen, if there are still free slots.
>>
>>> If so, do I have the ability to control this? I
>>> also don't exactly know how to actually create this new master queue
>>> with the correct settings and such. Could someone point me in the
>>> right direction?
>>
>> Just this morning, maybe it's still in your inbox:
>>
>> http://gridengine.sunsource.net/servlets/ReadMsg?
>> list=users&msgNo=20022
>>
>> -- Reuti
>>
>> ---------------------------------------------------------------------
>> 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
>

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