[GE users] Dynamic queue -- sge schedule policy

Reuti reuti at staff.uni-marburg.de
Mon Aug 28 23:06:10 BST 2006

    [ 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. ]


Am 28.08.2006 um 23:02 schrieb Juha Jäykkä:

>> in the scheduler. Inside the pool of machines with the same sequence
>> number, the load is still used.
> This is excellent! It is perfect for us, and it's even the simplest  
> way.
> I have another question, though. Suppose I have (cluster) queue  
> spanning
> ten hosts; the hosts have all h_vmem=1G and the queue has  
> h_vmem=4G. What
> happens if I submit 5 jobs with h_vmem=1G? From the manual pages of
> complex and queue_conf, I get the impression that the queue limit  
> limits
> the amount of h_vmem in *the whole* queue, thus making it  
> impossible to
> consume more than 4G of memory in the queue although in my example  
> this
> would mean there are 6 hosts doing nothing. Is this correct? I can not
> believe it is, but that's how I understand the manual anyway.

there are two issues:

1) The complex values are for each queue instance. They are not  
global for the whole cluster queue.

2) By default h_vmem is not consumable. This must be switched on in  
the complex definition (if you are looking for this feature). Then of  
course each host definition should get 4G.

-- Reuti

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