[GE users] global consumables problem

Shannon V. Davidson svdavidson at swbell.net
Tue Jun 22 06:22:41 BST 2004


Robert,

Robert Konecny wrote:

>Thanks Shannon! That did the trick - although I'm not sure I understand 
>why changing the default value changes the consumable behavior so 
>dramatically.
>  
>

The default value means that every job automatically requests this 
amount of the consumable resource.  So, if the default is 1 for a given 
consumable resource, then every job requests that resource.

>However, now I have a different issue now: I have set ncpu_user1=48 but
>any time I submit a job which sets ncpu_user1 to 8 or more I get this from
>'qstat -j':
>
>...
>

If you are running parallel jobs (qsub -pe option), the requested amount 
is automatically multiplied times the number of requested slots, so you 
can simply request 1 in your qsub request.

You might want to use qstat -F ncpu_user1,ncpu_user2 to see how much of 
the consumable resource is available after jobs are scheduled.

Shannon


>
>(-l h_rt=24:00:00,ncpu_user1=8) cannot run globally because it offers only 
>gc:ncpu_user1=48.000000
>
>It doesn't make sense to me. When ncpu_user1 is between 1-7 job runs just 
>fine.
>
>Again, I would appreciate any help since I'm obviously SGE challenged ...
>
>Thanks!
>
>robert
>
>On Mon, Jun 21, 2004 at 07:49:48PM -0500, Shannon V. Davidson wrote:
>  
>
>>Robert,
>>
>>I believe you'll have to set the default value in the complex 
>>configuration to 0 and then explicitly request the resource in the qsub 
>>wrapper using -l ncpu_$USER=1.
>>
>>Shannon
>>
>>
>>Robert Konecny wrote:
>>
>>    
>>
>>>Hi,
>>>
>>>I need to set up a limit for maximum number of processors (slots) a single
>>>user can be using at one time (which is different from 'maxujobs'). So I'm
>>>setting up per user consumable ncpu_$USER as outlined here
>>>http://gridengine.sunsource.net/servlets/ReadMsg?msgId=11702&listName=users
>>>
>>>My current configuration is:
>>>
>>># qconf -sc global
>>>
>>>#name            shortcut   type   value  relop requestable consumable 
>>>default
>>>#---------------------------------------------------------------------------
>>>ncpu_user1     ncpu_user1 INT    0          <=    YES         YES    1
>>>ncpu_user2     ncpu_user2 INT    0           <=    YES         YES     1
>>>...
>>>
>>># qconf -se global
>>>
>>>hostname                   global
>>>load_scaling               NONE
>>>complex_list               NONE
>>>complex_values             ncpu_user1=4,ncpu_user2=4,...
>>>load_values                NONE
>>>processors                 0
>>>user_lists                 NONE
>>>xuser_lists                NONE
>>>projects                   NONE
>>>xprojects                  NONE
>>>usage_scaling              NONE
>>>resource_capability_factor 0.000000
>>>
>>>
>>>and I have a qsub wrapper which sets npcu_$USER to appropriate value (1 
>>>for a serial job, N for parallel job running on N processors).
>>>
>>>However, when user1 submits 4 jobs on 1 CPU/slot each (using qsub -l
>>>ncpu_user1=1 script.sge) all ncpu_* counters are subtracted, not just
>>>ncpu_user1 as I would expect.
>>>
>>>Is there something wrong with my setup? I would appreciate any help with
>>>this.
>>>
>>>thanks!
>>>
>>>robert
>>>
>>>
>>>---------------------------------------------------------------------
>>>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