[Bioclusters] High Availability Clustering

Elia Stupka bioclusters@bioinformatics.org
Wed, 23 Jul 2003 23:33:13 +0800

> >Your experience may vary but I've found SGE and LSF to be superior to all
> >versions of PBS that I've ever used. They all will work at the end of the
> >day but some packages require more care/feeding and operational overhead
> >over time than others.

I was always of this opinion until lately I was debating this with IBM for a
cluster we are setting up and IBM was pushing for PBSPro... in the end they won
the debate because apparently NASA has been pushing a lot of improvements in
the last year or so (last time we tried PBS was more than a year ago) and they
now run it happily on a 14,000 CPU cluster in production.... I couldn't say
much to this, so they won the argument and soon I will have to find out as we
will be stressing out a cluster with PBS instead of our favourite (but
expensive) LSF...


> >
> It's probably best that we start into it slowly for now, I don't think we
> run enough jobs for that extra performance since the load on our cluster is
> pretty low.  For instance this month (highest month of usage thus far) we
> have run ~400 jobs in our PBS queue, translating into maybe 200 CPU days.
> We just don't stress it enough.
> >If you are running OpenPBS instead of at least PBSPro I'd argue
> >semi-seriously that you are silly to even be considering HA hardware
> >techniques for your cluster :)
> >
> Let me introduce you to my boss...  :)
>  -John
> _______________________________________________
> Bioclusters maillist  -  Bioclusters@bioinformatics.org
> https://bioinformatics.org/mailman/listinfo/bioclusters

Temasek Life Sciences Laboratory
1, Research Link
Singapore 117604
Tel. +65 6874 4945
Fax. +65 6872 7007

This mail was sent through Intouch: http://www.techworx.net/