[Bioclusters] Re: BLAST job on SGE

Chris Dagdigian dag at sonsorol.org
Fri Dec 3 14:47:58 EST 2004


This is the default behavior of Grid Engine and works well on busy or 
contested clusters where you want to pack jobs together on the same 
machine while leaving others free to pick up different jobs from others.

This is less ideal on idle clusters or clusters being used by small 
usergroups. Here you want more of a 'round robin' style of allocation.

There are a number of ways you can alter this behavior including playing 
tricks with np_load_average and adjusting the value of queue_sort.

Let me look into this for v5.3 and 6.0 and I'll post some options back 
to this list.

-Chris




> We are running SGE on an Apple cluster. When an array of BLAST jobs are 
> submitted, it schedules every two jobs to one compute node, regardless of how 
> many compute nodes are available or how many jobs. Any one knows how to adjust 
> it to schedule one job to one node each time? 


-- 
Chris Dagdigian, <dag at sonsorol.org>
BioTeam  - Independent life science IT & informatics consulting
Office: 617-665-6088, Mobile: 617-877-5498, Fax: 425-699-0193
PGP KeyID: 83D4310E iChat/AIM: bioteamdag  Web: http://bioteam.net


More information about the Bioclusters mailing list