Hi Mike,

On Apr 13, 2012, at 2:59 PM, Starch, Michael D (388L) wrote:

> Cynthia,
> 
> Capacity is what you set it to be. Load is then used to determine how jobs
> can run on the machine.  To the best of my knowledge OODT does not monitor
> cpu, memory or anything else (correct me someone if it has changes since
> snapshot 0.2).

Well I just want to clarify. By itself, Apache OODT provides the facilities
for these interfaces to be written (see the Resource Manager object
model and user guide [1]). However, out of the box implementations
of extension points that do this (like I referenced in my prior email to 
Cynthia [2]) don't exist yet.


>  It runs as many jobs on a node as it can until the sum of
> the loads of the jobs is the capacity (or as close to it as possible
> before it goes over the capacity).
> 
> On our project we have set the capacity to be the amount of memory on the
> machine, and the load to the amount of memory the job uses.  This is
> because we find our biggest concern with our jobs is the threat of
> becoming memory-bound and then swapping memory to disk.  Thus we use these
> numbers to reflect the capacity and usage of memory.

+1 this is a fine way to use the notion.

> 
> If you are concerned about being cpu-bound set capacity to be the number
> of cores. And load to the number of cores a job will take.

+1.

> 
> Our version is a bit older but I hope this helps.

Thanks Mike!

Cheers,
Chris

[1] http://oodt.apache.org/components/maven/resource/user/
[2] http://s.apache.org/Fhw

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
Chris Mattmann, Ph.D.
Senior Computer Scientist
NASA Jet Propulsion Laboratory Pasadena, CA 91109 USA
Office: 171-266B, Mailstop: 171-246
Email: [email protected]
WWW:   http://sunset.usc.edu/~mattmann/
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
Adjunct Assistant Professor, Computer Science Department
University of Southern California, Los Angeles, CA 90089 USA
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

Reply via email to