[GE users] queue wait policy

reuti reuti at staff.uni-marburg.de
Fri Nov 7 14:25:57 GMT 2008


Am 07.11.2008 um 13:13 schrieb Mag Gam:

> I have implemented this fair scheduling process by looking at this:
> http://gridengine.info/2006/01/17/easy-setup-of-equal-user- 
> fairshare-policy
> However, I want to be 100% certain I am not mixing it up with other
> parameters such has having share tree and functional, how can I check
> that?
> Also, is the policy  Reuti describes "Share Tree" or "Functional"? I
> prefer share tree because I need a half life factor of 48 hours.

it's just Functional to balance the slots among all users equally  
without any history. This is always a good starting point. When of  
course one user arrives early in the morning he could flood the cluster.

What you might want in addition, is to limit each user to use only 5  
slots at all which can be done by an RQS

limit users {*} to slots=5

see: man sge_resource_quota

How did you come to the 48 hrs half life time for a share tree setup?

-- Reuti

> On Thu, Oct 16, 2008 at 8:55 AM, Mag Gam <magawake at gmail.com> wrote:
>> Thankyou very much for the fast responses.
>> I will try the methods you mentioned but is there a easy way to
>> confirm if my fair scheduling is enabled and working? I have done the
>> 'qconf', but basically if a professor comes and asks me "prove it the
>> fair scheduling is working?"
>> TIA
>> On Thu, Oct 16, 2008 at 9:41 AM, Chris Dagdigian  
>> <dag at sonsorol.org> wrote:
>>> I'd recommend setting up a fairshare policy that treats all users  
>>> equally:
>>> http://gridengine.info/2006/01/17/easy-setup-of-equal-user- 
>>> fairshare-policy
>>> By default Grid Engine will fill all available job slots if there  
>>> are
>>> appropriate jobs in the pending queue. Unless you go out of your way
>>> configuring policies or resource quotas you are not going to see  
>>> job slots
>>> left free for other users. SGE does almost all of its resource  
>>> allocation
>>> work by sorting an resorting the pending list -- it will not mess  
>>> with
>>> running jobs or keep slots open unless you specifically configure  
>>> this.
>>> The fairshare policy above will almost do what you are requesting  
>>> except for
>>> the fact that you will have to wait for a few of User A jobs' to run
>>> successfully and drain out of the system before you see User B  
>>> and C jobs
>>> starting to get dispatched. Eventually you will see User A,B,C jobs
>>> reflected equally across the nodes but again - this is done by  
>>> waiting for
>>> job slots to free up so that User B and C jobs can be taken from  
>>> the top of
>>> the pending list.
>>> Regards,
>>> Chris
>>> On Oct 16, 2008, at 9:31 AM, Mag Gam wrote:
>>>> Hello All:
>>>> We have implemented Sun Grid Engine at our University's lab  
>>>> which has
>>>> about 70 CPUs. I have an interesting scnerio:
>>>> userA submits 10 jobs  at 5:00AM
>>>> userB submits 10 jobs  at 5:10AM
>>>> userC submits 10 jobs at 5:20 AM
>>>> These 30 jobs go into the queue and its a FIFO operation. It seems
>>>> like we have to wait for userA's job to finish first, then  
>>>> userB's job
>>>> and finally userC.
>>>> What we would like to see is:
>>>> UserA takes up 5 slots (the remainder go into queue wait) and when
>>>> userB submits her job at 5:10AM we want her 5 jobs to start
>>>> immediately and place the remainder 5 in a queue. Same goes for  
>>>> userC.
>>>> Basically, we don't want userB, and userC to wait for userA to  
>>>> finish
>>>> since userA's job take a very long time to complete.
>>>> Any thoughts or tweaks I can look into?
>>>> TIA
>>>> ------------------------------------------------------------------- 
>>>> --
>>>> 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
> ------------------------------------------------------
> http://gridengine.sunsource.net/ds/viewMessage.do? 
> dsForumId=38&dsMessageId=88295
> To unsubscribe from this discussion, e-mail: [users- 
> unsubscribe at gridengine.sunsource.net].


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

More information about the gridengine-users mailing list