[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Debian Beowulf (and intent to package xmpi and scalapack)



On Tue, Dec 15, 1998 at 04:42:15PM -0500, Camm Maguire wrote:
> 
> 3) setting up a web site with benchmarks, comparing with
> our SGI Power Challenge, and discussing our choice of Debian. (2-9
> times faster, more than 10 times cheaper)
> 
   STREAM benchmarks would be interesting.  I find our Octane is about twice
the bandwidth of our dual PII/266.  The POWER Challenge has an older, less
scalable memory system and a new PII should have a considerable faster
memory bus, especially when measured in aggregate over a cluster.
Management-wise a POWER Challenge is easier than a cluster though- it's
either all up or all down.  You never get load imbalances like you do in a
cluster (running jobs can't shift to different nodes). As your cluster
ismade of identical nodes this doesn't matter much to you, but my users are
very disappointed when their long jobs end up on the slowest machine in the
cluster.

> 4) Perhaps writing and contributing a Beowulf administration package.
> I was thinking about something like procmeter, running in gtk over the
> whole cluster.  Maybe even some transparent way to do package
> management over the cluster.

   The dqs source package has a number of X11 monitoring programs (in
XSRC/).  I've never used them myself as most of my cluster nodes have people
sitting at them who stand up and walk over to "alert" me when things break
down.  dqs is unfortunately not quite DFSG- I already failed to convince the
authors.  Perhaps if you can get to Florida and get them all drunk..


> 
> 5) Perhaps contributing a kernel module package which enables
> temperature sensing (lm-sensors).
> 
> There is so much which we need to do anyway, and which can be usefully
> IMHO shared with the Debian Project, that its hard to know where to
> begin.  I would greatly appreciate ideas about what others would like
> to see in Debian for Beowulves; it seems that this is an area in which
> the distribution could excel.

   If you don't need parallel jobs and user-level queue suspensions (for
COWs like mine, useless on headless beowulves) you might be able to use
Generic NQS (GNQS), which is GPL and unpackaged.


-- 
Dr. Drake Diedrich, Research Officer - Computing, (02)6279-8302
John Curtin School of Medical Research, Australian National University 0200
Replies to other than Drake.Diedrich@anu.edu.au will be routed off-planet


Reply to: