[Mauiusers] Could maui ignore the ppn in qsub

Bas van der Vlies basv at sara.nl
Fri Feb 11 09:44:51 MST 2005


Bas van der Vlies wrote:
> Ake wrote:
> 
>> On Fri, Feb 11, 2005 at 11:22:32AM +0100, Bas van der Vlies wrote:
>>
>>> I know a put the question wrong, but when job 1508 is run it will 
>>> aquire the whole node and the MAXPS will be recalculated to 7200, 
>>> because the user has the whole node with 2 processors in it.  Now the 
>>> user has also reach the MAXPS limit as maui concerns. That is why i 
>>> asked can we add an CONFIG that MAXPS will skip the multiplication 
>>> factor. Maybe MAXPS can see if SINGLEJOB is set, we forgot about the 
>>> ppn factor for the calculation of requested time.
>>
>>
>>
>> I don't think MAXPS will be recalculated for that ppn=1 job, neither
>> will the actual usage numbers.
>>
> I know for sure that when the jobs runs the Used amount of time will 
> exceed the MAXPS factor. We now have limited the user for 600 hours.
> 
> I user can subit an job, eg:
>     qsub -I -lnodes=60:ppn=1 -lwalltime=10:00:00
> 
> This will run and he gets 60 nodes. Our nodes have two procs. When this 
> job run his MAXPS used factor will be 60*10*2 = 1200 hours what is much 
> more then MAXPS LIMIT and he has to wait a long time before other jobs 
> can run
> 
> That is why i asked if we could ignore the ppn factor for calculating 
> the requeted time if we only schedule one job per node.
> 

I have adjust the calculation of MAXPS and SINGLEJOB settings for maui. 
The time calculation is: (for maui is set SINGLEJOB )
	nodes * walltime

instead of: (this is used for the other maui settings )
	tasks * walltime

The reason why is that if an user submits an job it acquire the whole 
node. eg:
	max_walltime = 3600 ( 1 hour)
         MAXPS        = 7200 ( 2 hour)

qsub -I -lnodes=2 -lwalltime=3600 ( will run)

qsub -I -lnodes=2 -lwalltime=3600 ( blocks saying that user has used 4 *
                                     3600 = 14400 ( 4 hours)

qsub -I -lnodes=2:ppn=2 -lwalltime=3600 ( will be blocked for ever and
                                      ever, because it request 14400 sec)

This because each node has 2 processors an when it runs maui sees this 
as 4 tasks. With this patch we have the following situation, it skips 
the ppn parameter for maui setting SINGLEJOB:

qsub -I -lnodes=2 -lwalltime=3600 ( will run)

qsub -I -lnodes=2 -lwalltime=3600 ( blocks saying that user has used 2 *
                                     3600 = 7200 )

qsub -I -lnodes=2:ppn=2 -lwalltime=3600  ( blocks with same message as
                   above what is correct and will also run, because
                   requested value is <= MAXPS )



-- 
--
********************************************************************
*                                                                  *
*  Bas van der Vlies                     e-mail: basv at sara.nl      *
*  SARA - Academic Computing Services    phone:  +31 20 592 8012   *
*  Kruislaan 415                         fax:    +31 20 6683167    *
*  1098 SJ Amsterdam                                               *
*                                                                  *
********************************************************************


More information about the mauiusers mailing list