Fwd: Grid Engine: ongoing and future work

Donald Becker becker at scyld.com
Mon Oct 1 09:07:07 PDT 2001


On Mon, 1 Oct 2001, Greg Lindahl wrote:
> On Mon, Oct 01, 2001 at 01:26:25PM +0200, hanzl at noel.feld.cvut.cz wrote:
> 
> > OK, OpenPBS too runs on linux but on Scyld it is not easy ('mom' on
> > every node does not suit it well, running scripts on nodes is not easy
> > - I think people work on it, but certainly it takes more than just
> > minor modifications).
> 
> Actually, it would be interesting to have a discussion about how one
> would port OpenPBS or SGE to Scyld.

Note that PBSPro is available for Scyld.
    http://www.pbspro.com/news_0106.html

> I think it should look like the SGI O2k version, with 1 mom on the
> master. This mom would query Scyld to find out the resources
> available, and report that as its # of cpus. When a job starts, it's
> mom who picks which cpus it uses, perhaps using the Scyld resource
> allocation layer. And then that information needs to get communicated
> to beompi.

That's one way (IMHO, the best way) to do support PBS when you use PBS
for the whole site.  The downside is that you can't use the
sophisticated scheduler to handle the details of node allocation.

I haven't yet looked at SGE..

Donald Becker				becker at scyld.com
Scyld Computing Corporation		http://www.scyld.com
410 Severn Ave. Suite 210		Second Generation Beowulf Clusters
Annapolis MD 21403			410-990-9993





More information about the Beowulf mailing list