Mailing List Archives
Public Access
|
|
|
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Condor-users] Condor performance problem
- Date: Tue, 16 Jan 2007 09:48:03 -0500
- From: Jeff Linderoth <jtl3@xxxxxxxxxx>
- Subject: Re: [Condor-users] Condor performance problem
Hello Hao,
> However there are still some comments abouts such performance tests:
> "Dr. David Jones at UCL conducted some tests of the Condor system.The
> system had 80 free processors but Jones didn't manage to get more than
> 10-15 running at any one time. The management simply made it
> impossible to use the resources effectively. The reason
> here was quite simple. Jones used a series of relatively short (5-10
> minutes) jobs. The transport and
> assignment service was too slow in comparison with these small tasks.
> Even if previous submitted
> tasks are completed, new tasks might be still in the queue waiting for
> scheduling or hold up in
> transportation. Consequently he could never make full use of 80
> processors although most of them
> are free. This performance problem is fatal for those applications
> requiring real time execution as
> we discuss here. Because of inefficient resource allocation, Condor
> could not satisfy the
> requirement for high response performance."
>
> Does you guys have any thoughts about those tests and comments ?
Do you have a specific citation for this quote that you could provide?
For running many relatively short jobs through Condor, I would recommend
using MW.
http://www.cs.wisc.edu/condor/mw/
Using MW, we have successfully performed large computations, where jobs
were only a few seconds, and scaled Condor configurations to hundreds
(in some cases thousands) of processors.
I hope this helps. Let me know if you would like help or more specific
information about MW.
Cheers,
-Jeff
--
------------------------------------------------------------
Jeff Linderoth O: 610-758-4879
Asst. Professor
Industrial and Systems Engineering jtl3@xxxxxxxxxx
Lehigh University www.lehigh.edu/~jtl3