[GE users] qrsh preferred host

McCalla, Mac macmccalla at hess.com
Tue Feb 6 03:36:53 GMT 2007


Hi,
The seq_no parameter in queue definition supports a syntax that allows you to specify different values for individual hosts or hostgroups which have been referenced in the queue definition.  I have recently used this technique and it is working quite well.
HTH
Mac McCalla
--------------------------
Sent from my BlackBerry Wireless Handheld


-----Original Message-----
From: Bevan C. Bennett <bevan at fulcrummicro.com>
To: users at gridengine.sunsource.net <users at gridengine.sunsource.net>
Sent: Mon Feb 05 16:37:07 2007
Subject: Re: [GE users] qrsh preferred host

Joachim Gabler wrote:
> Hi Ionel,
> 
> Ionel GARDAIS wrote:
>> Hi,
>>
>> I have a bunch of workstations that are both submit and execution hosts.
>> Is it possible to tell qrsh to preferrably use the "local" machine
>> instead of one on the grid ? (as long as the machine I'm working on
>> meets the requested attributes)
>>   
> You can use a soft request on the host:
> qrsh -soft -l h=`hostname` .....

In a similar question, what's the best grid6 way to give a global "preference"
for nodes? In grid5 I used seq_no, but it doesn't appear that I can configure
seq_no on a per-system basis anymore.

The situation is that we have some nodes with 4GB of RAM, some with 8GB of RAM,
some with 16GB of RAM, and one with 32GB of RAM.

I would like jobs to preferentially utilize the smallest system that they can
fit on, so that small jobs will fill up first 4GB systems, then 8GB systems,
etc., finally getting to the 32GB system.

Unfortunately, I haven't yet been able to figure this out in the brave new world
of SGE6. I have a bit of a bad feeling it will involve a massive increase in
queue complexity. Currently we have two mutually suppressing queues:

all.q - "normal" jobs that take up one slot (one slot = one core)
unshared.q - "unfriendly" multi-threaded jobs that can take over multiple cpu
cores and should therefore run isolated from other jobs.

Then there's an 'unshared' boolean attribute that defaults to false and is
provided as true on the unshared queue.

Is there an easier way to rank node priority than to create separate sets of
queues (with different seq_nos) for each "priority set" of nodes, of is that
what we're supposed to be doing here?

---------------------------------------------------------------------
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