[GE users] Queue configuration question.

Reuti reuti at staff.uni-marburg.de
Tue Apr 11 06:49:23 BST 2006


Am 10.04.2006 um 22:44 schrieb Baudilio Tejerina:

> Hi,
>
>
> On Apr 10, 2006, at 2:12 PM, Reuti wrote:
>
>> Hi,
>>
>> Am 10.04.2006 um 19:08 schrieb Baudilio Tejerina:
>>
>>> Hi.
>>>
>>> In a clusters composed by single- and dual-processors machines, I  
>>> have defined queues associated with each individual computer so  
>>> that 'hostlist' contains just the name of machine (not a group)  
>>> and the 'slots' contains the number of CPUs that machine may  
>>> provide. For instance:
>>>
>>> qconf -sq buckyball
>>>
>>> qname                 buckyball
>>> hostlist                 buckyball
>>> seq_no                0
>>> ....
>>> slots                 1,[buckyball=2]
>>> ....
>>>
>>> Likewise, if the machine has a single CPU
>>>
>>> slots                 1,[zinc=1]
>>>
>>>
>>>
>>> It happens that if I submit a job requesting 2 CPUs, GE allocates  
>>> one from "buckyball" and the other from "zinc".
>>>
>>
>> this is the intended behavior. You can:
>>
>> a) define one PE for each queue with similar names and request the  
>> PE with a wildcard
>>
>> b) use allocation rule $pe_slots, this way you could still stay  
>> with one queue for all machines, as all slots have to be on one  
>> machine
>>
>>
>>>
>>> How can I have the job be allocated in a machine with two  
>>> processors and never in the way it is now?
>>> Related to this, I'd like to configure the queues so that if a  
>>> user requests more that 2 CPUs, the job gets always rejected.
>>>
>>
>> What is the definition of your PE now? Maybe you need a second PS  
>> with a different allocation rule.
>>
>
> I only have one PE defined, 'make', as follows:
>
> pe_name           make
> slots             999
> user_lists        NONE
> xuser_lists       NONE
> start_proc_args   NONE
> stop_proc_args    NONE
> allocation_rule   $round_robin
> control_slaves    TRUE
> job_is_first_task FALSE
> urgency_slots     min
>
> The option b) you point out seems more general  than a), doesn't it?

a) can be used to bind jobs to hostgroups, i.e. one hostgroup, queue  
and PE for each hosttype

> So, should I modify 'make' by substituting $round_robin by  
> $pe_slots or create a new PE with that allocation rule in it?

In your case I would create a new PE, which reflects the usage of  
this PE, e.g. smp or openmp named.

-- Reuti


>
> Thanks Reuti.
>
>
> Baudilio
>
>
>> -- Reuti
>>
>>
>>>
>>> I've been reading  the documentation but I couldn't find  
>>> anything. In fact, I had thought that the configuration of 'one  
>>> queue for one node' would work, but apparently is not sufficient.
>>>
>>> Baudilio
>>>
>>> -------------------------------------------------------------------- 
>>> -
>>> 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