[GE users] Moved to users list: Re: [GE issues] Calendar and advance reservation

fredwag Frederik.Wagner at lrz.de
Tue May 18 10:25:31 BST 2010


On 05/11/2010 02:03 PM, reuti wrote:
> Hi,
> 
> Am 07.05.2010 um 10:24 schrieb fredwag:
> 
>> On 05/06/2010 07:39 PM, reuti wrote:
>>> Am 06.05.2010 um 14:24 schrieb fredwag:
>>>
>>>> On 05/06/2010 12:14 PM, Reuti wrote:
>>>>> Hi,
>>>>>
>>>>> (the issues list is mainly for the emails which are send automatically when changes to the issues database are made)
>>>>>
>>>>> Am 06.05.2010 um 09:06 schrieb fredwag:
>>>>>
>>>>>> Hi .*,
>>>>>>
>>>>>> I have a problem in SGE 6.2u1:
>>>>>>
>>>>>> We have setup for visulization system with multiple nodes and multiple
>>>>>> graphics cards per node. Through SGE a users can reserve a graphics card
>>>>>> to do their visualization job, for this an advance reservation system is
>>>>>> configured (its a batch queue, but interactive work).
>>>>>>
>>>>>> Now there is the demand for a nightly batch queue (no ar) for normal
>>>>>> dataprocessing jobs (since the machines have lots of memory and they
>>>>>> idle during the night). So I configured a 'nightly' batch queue, with a
>>>>>> calendar attached to it (4-20=off 20-4=on).
>>>>>>
>>>>>> My problem now is the following:
>>>>>> As soon there exists an advance reservation (for an already running job
>>>>>> or for the future) _no_ jobs are running in nightly queue. The scheduler
>>>>>> info gives (when the queue is by calendar enabled):
>>>>>> '(-l NONE) cannot run in queue "nightly at somehost..." because it offers
>>>>>> only qc:slots=0.000000 due to a reservation'
>>>
>>> And the AR will end before the nighly queue is active? Maybe the slots are enabled by the calendar then, but allocated to the AR.
>>
>> some more details:
>>
>> There are two hosts managed by this SGE instance: Both are member of two
>> queues:
>> 1. each has its 'graphic card reservation queue', i.e. host1_graphics
>> and host2_graphics handled by the AR mechanism, the only interesting
>> resource here is the graphics card.
>> 2. the 'nighlty' queue is available on both hosts with 1 slot each, here
>> the calendar is attached (week 4-20=off 20-4=on).
>>
>> So there should be no interference in terms of 'slots' between the two
>> queues on each host.
>> ARs are alway possible, so I could make a reservation for tomorrow in
>> the graphics queue, but clearly would still like to have jobs running in
>> the nigthly window.
>>
>> And that's exactly where the problem lies: if there is a reservation for
>> tomorrow, the nigthly jobs wont run, with the error:
>> '(-l NONE) cannot run in queue "nightly at somehost..." because it offers
>> only qc:slots=0.000000 due to a reservation'
>>
>> If there are no ARs evertythings works fine. Also in the case, when I
>> detach the calendar (so the queue is always available).
>>
>> What more information do you need?
> 
> you mean, the AR is requesting slots from the "host1_graphics" queue (one dedicated queue per machine or just "graphics" as name?). But the calendar is attached to the "nightly" queue?

exactly, there is a dedicated graphics Q per machine where the AR is
requesting slots from.
The calender is attached to the nightly one, which runs over all hosts.

Thx,
Frederik

> 
> -- Reuti
> 
> 
>> Thanks a lot for the help,
>> Frederik
>>
>>>
>>>
>>> Can you give more details about the setup?
>>>
>>> -- Reuti
>>>
>>>
>>>>>>
>>>>>> What could be the problem here?
>>>>>>
>>>>>> The jobs in nightly _are_ running, when
>>>>>> * I take away the calendar, so the job is always enabled
>>>>>> * no advance reservation exists
>>>>>>
>>>>>> I already played with the scheduler parameters Maximum
>>>>>> Reservation/Default Duration etc...
>>>>>>
>>>>>> It seems that the complex slots is somehow influenced by pending
>>>>>> reservations... but why? Is this a 'backfilling' problem? I don't know
>>>>>> where to search anymore. Or is this a bug?
>>>>>
>>>>> there one issue which I also observed with RQS:
>>>>>
>>>>> http://gridengine.sunsource.net/issues/show_bug.cgi?id=2761
>>>>>
>>>>> Are you using RQS?
>>>>
>>>> no, we are not using RQS, just an advance reservation queue and a normal
>>>> queue with calender.
>>>>
>>>> Thanks and bye,
>>>> Frederik
>>>>
>>>>>
>>>>> -- Reuti
>>>>>
>>>>>
>>>>>
>>>>>> Thanks a lot for any help!
>>>>>> Bye,
>>>>>> Frederik
>>>>>>
>>>>>> ------------------------------------------------------
>>>>>> http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=36&dsMessageId=256351
>>>>>>
>>>>>> To unsubscribe from this discussion, e-mail: [issues-unsubscribe at gridengine.sunsource.net].
>>>>>
>>>>
>>>> ------------------------------------------------------
>>>> http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=38&dsMessageId=256386
>>>>
>>>> To unsubscribe from this discussion, e-mail: [users-unsubscribe at gridengine.sunsource.net].
>>>
>>> ------------------------------------------------------
>>> http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=38&dsMessageId=256422
>>>
>>> To unsubscribe from this discussion, e-mail: [users-unsubscribe at gridengine.sunsource.net].
>>
>> ------------------------------------------------------
>> http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=38&dsMessageId=256509
>>
>> To unsubscribe from this discussion, e-mail: [users-unsubscribe at gridengine.sunsource.net].
> 
> ------------------------------------------------------
> http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=38&dsMessageId=256945
> 
> To unsubscribe from this discussion, e-mail: [users-unsubscribe at gridengine.sunsource.net].

------------------------------------------------------
http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=38&dsMessageId=257718

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



More information about the gridengine-users mailing list