[GE users] Suspending jobs based on license load not cpu load

elrond2k Are.Aarseth at atmel.com
Tue Jan 19 09:50:51 GMT 2010


Ok, I can see a problem in starting another grid job to take the license of the first job. But in my case I am more interested in giving back licenses to the pool so that interactive jobs outside the grid can run. And as such I do not need to worry about complexes not released.

In my case the grid compete for licenses on a global pool, and I try at anytime to keep at least X licenses free for interactive jobs. I did a modification to how qlicserver handles negative limits to make this happen. This works well as long as the jobs on the grid are relative short. But if the jobs are long then the interactive jobs can eat the X free licenses before a grid job completes. And then you have angry mails incoming asking you to fix it asap :)

So my idea was to suspend using qmod when the license complex is less than 0 for more than a Y minutes. 

-----Original Message-----
From: olesen [mailto:Mark.Olesen at emconTechnologies.com] 
Sent: 18. januar 2010 15:10
To: users at gridengine.sunsource.net
Subject: Re: [GE users] Suspending jobs based on license load not cpu load

On Mon, 2010-01-18 at 14:47 +0100, elrond2k wrote:
> Does anyone know if there is an existing way of suspending jobs based
> on the licenses they are using as opposed to the server load.

>  
> We have a setup with Grid + flex-grid to monitor licenses. And in some
> cases where it would be useful if we could suspend jobs temporarily to
> give licenses back to the pool if there are no licenses free. At least
> some tools support releasing licenses on ctrl-z, so a suspend / resume
> should work here, and it is also possible to add a lmremove in the
> suspend method I suppose. But I can't see exactly how to tackle this
> problem other than having an external daemon like qlicserver that has
> the power to suspend/resume... Any ideas?


Simply triggering a suspend is probably not the biggest issue. The real
issue is managing the resources to avoid any problems.
Take a read of this thread to see what I mean.

http://markmail.org/thread/xiafku4hibowmqgs

I don't see a simple solution.

/mark


This e-mail message and any attachments may contain legally privileged, confidential or proprietary Information, or information otherwise protected by law of EMCON Technologies, its affiliates, or third parties. This notice serves as marking of its "Confidential" status as defined in any confidentiality agreements concerning the sender and recipient. If you are not the intended recipient(s), or the employee or agent responsible for delivery of this message to the intended recipient(s), you are hereby notified that any dissemination, distribution or copying of this e-mail message is strictly prohibited. 
If you have received this message in error, please immediately notify the sender and delete this e-mail message from your computer.

------------------------------------------------------
http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=38&dsMessageId=239536

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

------------------------------------------------------
http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=38&dsMessageId=239704

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



More information about the gridengine-users mailing list