Re: how can i start a cluster proyect?
On Sun, Apr 22, 2007 at 01:21:03AM -0400, Joksan Nempu Fernandez wrote:
> Im new to this and i want to know how can i make a cluster?
> Where i can find the steps to this, how to use pmi or pvm, etc
Look through the archives of this list. Read for a while, "listening on
the list" to get a flavour of what is talked about. After a couple
of weeks, you may start to understand some of what you are reading.
Think about some of the following general questions to start you off.
1. Every cluster is different: everybody's problem set is different
but some things are common - does any of what I'm reading from
other people apply to me?
2. What (engineering/maths/computer science) problems am I trying
to solve? Can I share them with the list? How good are my C coding
3. Do these problems scale? Can I make a simplified problem / "toy"
to try out ideas? Can I beg/borrow/steal a couple of machines to build
a four node cluster or something similarly small :)
That's the first set: now come the set where you're trying to sort
out what resources you will need:
4. Am I solving problems for me or for a community of users?
5. Remembering physical and engineering constraints: do I have adequate
power/cooling/clean air? How reliable are the machines I'm thinking
of using - has anybody else used similar machines? [Do I have funding?]
If you've done 2 and 3 above or talked to other researchers.
6. I've classified my problem/profiled my code: it's embarrasingly
parallel/has serial bottlenecks/chokes on lack of memory.
What do I do now?
Robert Brown (rgb), a physicist at Duke University, has a good page
detailing some of the background including an (on-going) book available
as a PDF. I'm sure he won't mind me posting the current URL to save you
a Google trip :)
I know this is a lot of questions but stopping to think about them
will help you narrow down your field. Everyone has to start on the
process somewhere and carry on learning from the people on the list.
[Unfortunately, the members of this list are mostly busy people and
don't have time to write CS homework assignments/term papers/
undergraduate dissertations either to order or for free - they
appreciate sensible questions which show a little research,
background reading or willingness to learn :) ]
All the best,
> Joksan Nempu Fernandez
> Estudiante Ingeniería Civil en Computación e Informática.
> Encargado del Laboratorio de Sistemas Distribuidos.