Re: LAM/MPI Parallel processing

[Simon Valiquette <Simon.Valiquette@example.com>]

Austin Kurahone wrote:
> 
>   Supprising that something like this just came up, I just got back from a
>   summer of hacking this stuff among otherthings in the US.
> 
>   Minor nitpick. Beowulf is the project at the CESDIS. Cluster would be a more
>   appropriate naming for the technology that you describe.

  I know about it, but so many people call a Beowulf any dedicated
cluster of computers made for high-performance computing tasks that have
their own private networking.  Maybe it's just a kind of technical
language evolution.

>   >     By the way, the structure using a master that just dispatch jobs to
>   > slave computers are not suitable for me, because each nodes have to
>   > communicate with some others nodes, and that the amount of data is too
>   > big (say, easilly over one GB for a single simulation).
>
>   Ouch, that's a lot of data.  What kind of networking setup are you using.
>   Unless you've got really expensive interfaces Gig/E or Myrinet, I'd reconsider
>   doing something like this with a cluster...

  I'm thinking about putting two 100Mb/s network cards in each computer
and using two special switching Hub.  Their is many possible
configurations and i'm not sure yet which one I will use.

  Using 1 Gb/s network card are a possibility, but I prefer to avoid
using that.  In fact, my problem is mainly money related.  Otherwise, we
would just run it on a supercomputer.  But many others company don't
have the budget to buy time at 300$/h on a Cray class supercomputer.

Simon