[Beowulf] Grants for Beowulf Clusters

[Robert G. Brown]

On Fri, 11 Mar 2005, Timo Mechler wrote:

> Hi all,
> 
> I'm wondering what kind of success rate people are having with obtaining
> grants for Beowulf type Linux Clusters (for example, from the National
> Science Foundation).  Let me give you a little bit more info as to why I'm
> asking this:  I'm a junior undergraduate at a small liberal arts college
> in Iowa (~2600 students), and have solely been pursuing Beowulf clusters
> for well over a year now.  I believe strongly that even though that my
> school is small, several departments on campus could benefit from the use
> of a beowulf cluster in the research that does go on.  I've been using
> older, slower machines as a proof of concept for now.  Ideally, we would
> want a faster beowulf system eventually that offers significant
> improvements over anything desktop pc's have to offer nowadays.  Being
> that money is an issue at smaller schools, is there any I could obtain a
> grant for a beowulf cluster?  If so, besides the NSF, what would be some
> other sources to apply to?  Since some of you guys on this come from big
> companies or Univesities, I would appreciate any insight and suggestions
> you can give me.  All input is appreciated.  Thanks in advance!

Nearly all the university clusters in the world (with a few exceptions
like my or Jeff Layton's home clusters:-) are purchased with grant money
of one sort or another, so "success" or not it's the only game in town.
Businesses, of course, often pay out of pocket, but that's the game in
the big city (so to speak).

MOST of the university clusters are likely sponsored to do some specific
piece of grant-funded research.  That is, I want to do Monte Carlo
research into the dynamic and static critical properties of continuous
Heisenberg ferromagnets, so I write a proposal to do this research that
contains a hardware budget for the cluster upon which I plan to do it.

However, the NSF has LOTS of grants for different categories, including
grants to stimulate and improve undergraduate educational experiences or
to build shared infrastructure.  Few of them are likely to be available
to an undergraduate, though.

What I'd recommend is that you find a local faculty person (or three)
that you can convince to share your vision and that might "need" the
cluster either to teach students or to do some actual research (or
both).  Perhaps one from computer science, one from physics, one from
chemistry or biology.  See if all of you together can write an
institutional grant proposal for a startup cluster.  Note that this need
not be terribly expensive or a lot of money -- you can build a perfectly
reasonable starter cluster for around $20-25K if you can get the school
to pick up the tab for power and cooling (likely to be a few thousand a
year).  If the cluster is thoroughly firewalled from the university
network backbone, students could probably install it and administer it
and manage it and write applications for it for the participating
faculty or for course credit.  It would be fun.

Note that you CAN get started for a lot less money than even this.
Barebones compute nodes can be had for almost nothing.  Doug Eadline and
Jeff Layton recently demonstrated this rather spectacularly here:

  http://www.clusterworld.com/value_cluster.shtml

This is basically an 8 node cluster made with all new components for
$2500.  While they are experts (and hence can actually ride the bleeding
edge of what will work for the least possible amount of money) this does
indicate what can be done.  More reasonably, you can get started for
something like $500 for miscellaneous stuff (shelving, network switch
and cabling, $1000 for a fairly nice desktop for a front end,
fileserver, head node, and somewhere between $300 and $600 per node,
where the range in prices reflects the amount of memory and hard disk
and kind of network per node.  Check out pricewatch -- some of the
barebones, no-OS systems are less than $300 even for e.g. AMD-64s (not
exactly turtle-slow, that is:-).

What this means is that NSF or not you CAN build a student cluster at
your school.  You can damn near fund it with a bake sale or other
simple/fun fundraisers.  This kind of petty cash one can often convince
the school to pony up out of their petty cash budget, or you can start a
"cluster computing club" and (maybe) get a few thou out of your school's
clubs and activities budget.  Or you and seven friends can each
contribute $500 tax-deductable dollars (yes, sigh, from your parents).
Or you can walk main street and knock on doors and ask local businesses
to fund it, or...

Anyway, you get the point.  You don't need NSF money to get started, and
in fact it would be a lot easier for you to build a small cluster to get
started first and THEN look for that $20K grant from NSF to make it into
a middling big cluster!

   rgb

> 
> Best Regards,
> 
> -Timo Mechler
> 
> 

-- 
Robert G. Brown	                       http://www.phy.duke.edu/~rgb/
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567  Fax: 919-660-2525     email:rgb at phy.duke.edu