[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: [Condor-users] Extending Condor's scheduling? (for sched.researchpurposes)

David brings up an interesting point here: access to Condor's scheduling
algorithms via an API would be a nice thing. I certainly would have
looked to tweaking the scheduler to be non-fair share if an interface
had been availble. You wouldn't have to provide source code necessarily,
just a way to replace the library that does the scheduling for the
negotiator and matching deamon and then some documentation.


> -----Original Message-----
> From: condor-users-bounces@xxxxxxxxxxx 
> [mailto:condor-users-bounces@xxxxxxxxxxx] On Behalf Of David Petrou
> Sent: October 15, 2004 5:58 PM
> To: condor-users@xxxxxxxxxxx
> Subject: [Condor-users] Extending Condor's scheduling? (for 
> sched. researchpurposes)
> Hello.  My name is David Petrou and I've designed novel scheduling 
> algorithms for speculative tasks for my thesis at CMU.  (pls. reply 
> offline if interested in details.)
> Simulation results are promising.  I would like to make a 
> proof-of-concept implementation for Condor to demonstrate these ideas 
> in the real world.
> I am wondering what the easiest way would be to take over Condor's 
> scheduling policy.  I understand how user priorities strive for 
> long-term fair-share among users.  So, it seems that if I can hijack 
> Condor's computation of priorities, I could determine how things are 
> scheduled among users.  Further, if I could do the same for job 
> priority, it seems like I'd have control among a user's jobs as well.
> Alternatively, I am wondering if there is some way to extend Condor's 
> scheduling, perhaps via a library, so that I can bypass priorities 
> altogether and simply have Condor call my own functions when 
> trying to 
> determine what job should run.
> Someone must have tried extending Condor's scheduling before, but I'm 
> coming up empty.  Pointers to a fast-path to integrating my own 
> scheduling logic into Condor would be greatly appreciated, 
> acknowledged, and research results shared.
> Thank you very much,
> david
> p.s.: please cc: dpetrou@xxxxxxxxxxx on replies.
> _______________________________________________
> Condor-users mailing list
> Condor-users@xxxxxxxxxxx 
> http://lists.cs.wisc.edu/mailman/listinfo/condor-users