[GE users] queue definition

Daniel Templeton Dan.Templeton at Sun.COM
Thu Jul 12 16:28:59 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. ]

Iwona,

Without upgrading to 6.1, what you want is still possible, but it gets 
really messy.  Basically you'd have to create a second slots complex and 
use it to get your synchronization between your existing queues, but not 
your special queue.  The problem is that it would have to have a default 
of NONE.  Otherwise, you'll have the same problem you have now.  If it 
has a default of NONE, that means that users will have to explicitly 
request it when submitting jobs.  You can achieve that in a variety of 
ways (sge_request files, for example), but none of them are foolproof.  
It also means that your special user would have to undo whatever you did 
(e.g. -clear) to be able to submit a special job that doesn't consume an 
alternate slots complex.

Here's a wild and crazy idea.  What if you set up a second grid, just 
for that single job per host?  May or may not be reasonable, depending 
on what your environment and needs are.

Daniel

Iwona Sakrejda wrote:
> Hi,
>
> I am afraid I did not describe the situation adequately:
>
> Reuti wrote:
>>
>> Am 12.07.2007 um 05:11 schrieb Iwona Sakrejda:
>>
>>> I would like to be able to give a privileged user to start 1 job per 
>>> host and
>>> it should not matter whether other jobs are running or not on those 
>>> hosts.
>>> I thougth about defining a queue in which only this special user can 
>>> run,
>>> but I have trouble restricting it to only one job that does not use 
>>> a job slot.
>>> I have 2 slots per host and so far each submitted job consumes one 
>>> slot.
>>>
>>> I am running 6.0u11.
>>
>> if you have only one queue per node (for the common users with 2 
>> slots) and one special queue (with 1 slot), this should work without 
>> any further adjustments (i.e. there is no need to limit the slots per 
>> node in the exechost configuration). 
> I have a 2 slot limit per exec host and already 2 queues - one (let's 
> call it A)
> with 2 slots per host and the other  (B) with just one.   With a 2 
> slot limit per execution
> host I can run  AA or AB or just B and the other slow available to A 
> or both empty.
> Most of the jobs  from both the A and the B queues have hours long 
> running time.
>
> Now we engaged in some development work
> where jobs are very short and do not consume much resources but it is 
> important to run
> many of them and have short turn-around time to help debugging and 
> development.
> We want them to be able to start right away even when both slots on 
> the execution
> host are used by jobs from either A or B queues.
>
>> If you have more than one queue, I would suggest to look into SGE 6.1 
>> resource quotas, hence an upgrade maybe the easiest way to achieve 
>> this setup then.
> But I just updated to 6.0u11 last Tuesday. even if I wanted to upgrade
> again I need to post a 30 day warning. Is upgrading from 6.0 to 6.1 as 
> easy as
> applying patches within 6.0 (that worked really nice)?
>
> Is there really no way around it?
>
> Thank you...
>
> iwona
>>
>> -- 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




More information about the gridengine-users mailing list