Skip to main content

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [List Home]
Re: [ptp-user] Bug report... FileTools.java - job submission framework questions

Hi ,

My 2 cents on the discussion on LL Resource Manager.
The LL Resource Manager currently does not display the class information.
Class refers to the list of available partitions and the number of job slots.

Without this information the user is not able to edit the submit file for Job submission.
This may be an issue if the classes are changing on a routine basis.
Many system administrators enable large queues while disabling small queues from time to time.

If the class information is not displayed to the end user, it can cause a case where the job just stays in queue and will not give an error message.
This is if the job was submitted when a class was available but used by another job and then the class structure changes.
The user should be able to identify this issue if llclass info is displayed to end user.

The Current RM model has two nodes - Machine and Queue.
It may be necessary to add a Separate node to the Resource manager called Class. Class is different from machine as Machine is more static while class is dynamic.
This class could be updated based on an event of change of class structure to avoid polling.

Best Regards,
Prashobh Balasundaram
Software Engineer
Blue Gene Team
Dublin Software Laboratories
Mobile: 086 039 8813
Internet: pbalasun@xxxxxxxxxx  
IBM Ireland Product Distribution Limited registered in Ireland with number 92815. Registered office: Oldbrook House, 24-32 Pembroke Road, Ballsbridge, Dublin 4



From: Dave Wootton <dwootton@xxxxxxxxxx>
To: PTP User list <ptp-user@xxxxxxxxxxx>
Date: 04/14/2010 08:48 PM
Subject: Re: [ptp-user] Bug report... FileTools.java - job submission framework        questions






Steve

I'm assuming you're asking about the LoadLeveler resource manager, and my responses apply to that resource manager


The jobs list is showing all jobs that it finds in the job queue for the cluster it is monitoring. There is no filtering based on userid.


When a job completes, it disappears off the job queue. I think the rationale was that on a large system the job queue would be so active that the user would have a hard time selecting and deleting jobs from the jobs view


The rationale for the LoadLeveler support in PTP was to provide integrated support for using LoadLeveler in PTP instead of requiring the user to telnet/ssh somewhere, find his application and the LoadLeveler job submission file and run llsubmit


Dave


From: "Steven R. Brandt" <sbrandt@xxxxxxxxxxx>
To: ptp-user@xxxxxxxxxxx
Date: 04/13/2010 10:15 AM
Subject: Re: [ptp-user] Bug report... FileTools.java
Sent by: ptp-user-bounces@xxxxxxxxxxx






Yes, that was the problem. The HEAD branch works. There are some other
issues ("AbstractProxyRuntimeSystem: invalid attribute for definition"
shows up about a million places in the workspace error log), but you may
be working on these.

So I have a question or two about the usage of the PTP job submission
framework.

First, the jobs list. It seems to show all running jobs, not just mine,
and it doesn't seem to be particularly aware of which job is mine. Is it
supposed to be like that?

Secondly, it doesn't seem to have a memory of which jobs it has
submitted and whether they have completed or failed. Should it? Am I
missing a configuration step?

What is the main benefit to running through PTP rather than logging in
and typing llsubmit or qsub?

Cheers,
Steve

On 04/12/2010 12:13 PM, Greg Watson wrote:
> Are you on HEAD or the 3.0 branch? Have you updated from CVS?
>
> Regards,
> Greg
>

_______________________________________________
ptp-user mailing list
ptp-user@xxxxxxxxxxx

https://dev.eclipse.org/mailman/listinfo/ptp-user

_______________________________________________
ptp-user mailing list
ptp-user@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/ptp-user



Back to the top