undergrad senior project idea, help

Dean Johnson dtj at uberh4x0r.org
Fri Aug 30 12:26:56 PDT 2002

On Thu, 2002-08-29 at 17:21, Mark Hahn wrote:
> I think this is critically important.  undergrad projects are supposed
> to be not mere "recitations" - these days, building a cluster and 
> running povray really fast is just that.  projects that make a real
> contribution are the ones that lead to glowing faculty letters,
> mentions in the local newspaper, etc.
> so, find someone who can use the horsepower, scare up the money,
> and improve the world a little bit.

Sometimes simple awareness is a real contribution, atleast in my
experience. There seem to be a lot of people out there that have what
they consider to be Cray-sized problems, or atleast aspirations of
Cray-sized problems, but not Cray-sized budgets. Its really fun to watch
the wheels turn when they find out just how economically their
Cray-sized appetites can often be satiated with clusters. Simply showing
them a small, loud, and very hot room full of computers isn't terribly
illuminating, but very complex pretty povray pictures generated very
fast can show the power at a very intuitive level, even if it doesn't
directly relate to their field of study. I recall what a profound moment
it was when I first witnessed Doom being played on a piddly PC, even tho
I didn't play computer games hardly at all.

> I guess that people would also be impressed with awareness of reliability.
> after all, any sufficiently large compute cluster project 
> must in large part also be a project in high-availability.
> I'm guessing that 30 whitebox PCs are getting close to the point of 
> "cpu-fan collapse", where the whole is just not useful to anyone because
> it doesn't stay up long enough.  I guess that counts as another argument
> against the undergrad-povray-project thing.

For most uninitiated folks, in my experience, reliability doesn't sell
well. White box quality is "good enough" for nearly everyone. What sells
better is "If one craps out, I can just run down the Best Buy and get
another". Thats were you get the knowing nods. If you have baby-sat a
few hundred boxes, then you can appreciate reliability, but most
people's experience doesn't scale that high.


More information about the Beowulf mailing list