[pooma-dev] Nirvana Queues

[Jeffrey Oldham]

On Tue, Oct 16, 2001 at 10:40:21AM -0700, Jeffrey Oldham wrote:
> 
> Scott asked about queues for large jobs.  Here is data from a WWW
> page, which I hope reflects reality.
> 
> According to http://www.acl.lanl.gov/nirvana/userguide/Queues.html:
> 
> shared
>  - 1 to 112 shared (oversubscription allowed) processors
>  - RUNLIMIT of 12 hours
>  - n02 (and with much less preference, n01)
> 
> small
>  - 1 to 128 dedicated processors
>  - RUNLIMIT of 12 hours
>  - n03 and n06 through n16
> 
> medium
>  - 129 to 1024 dedicated processors
>  - active Wednesday nights from the end of maintenance to 8AM and weekends from 6PM Friday to 8AM Monday
>  - RUNLIMIT of 12 hours
>  - n09 through n16
> 
> large
>  - 1025 to 1792 dedicated processors
>  - active Wednesday nights from the end of maintenance to 8AM and weekends from 6PM Friday to 8AM Monday
>  - RUNLIMIT of 12 hours
>  - n03 through n16

Unfortunately, I do not have permission to use 'large' or 'medium'.
Instead, I'll use scavenger:

scavenger
 - 1 to 1792 dedicated processors
 - RUNLIMIT of 1 hour
 - n03 and n06 through n16
 - scavenger allows computing once a project has exceeded its allocation, but it's still accounted for and isn't free

Thanks,
Jeffrey D. Oldham
oldham at codesourcery.com