[Beowulf] Why I want a Microsoft cluster... redux

Philippe Blaise philippe.blaise at cea.fr
Fri Jan 6 00:48:11 PST 2006


Douglas Eadline wrote:

>I enjoyed the discussion brought on by Jim Lux's advocacy about Microsoft
>clusters. However, something about it bothered me. Well, I thought about
>the issue over the holidays and ended up writing about it on Cluster
>Monkey:
>
>"Why Is Cluster HPC So Hard?"
>
>I'm sure there are other opinions, so read and comment if you so choose.
>
>You can find it on the main page: http://www.clustermonkey.net
>
>
>--
>Doug
>
>
>
>_______________________________________________
>Beowulf mailing list, Beowulf at beowulf.org
>To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
>
>  
>
Very interesting.

"The /parallel programming/ problem has been around for quite a while 
and there are no signs it will be resolved any time soon. It was an 
issue before clusters came on the scene and will be a huge issue for 
multi-core systems."

I agree with you.
I have read sometimes that parallel architecture is the only way to go 
faster than the  Moore's Law, (let's recall us that this is a law coming 
from the industry).
But it is not really the case, Parallel clusters follow the Moore's Law 
: for the same price, you can have more and more of cpus in the same 
volume/pizza box/rack.
That's what we see in the top500.
Then, today, the main problem that shows up is not the choice of the OS, 
linux vs ms windows, (in fact Linux has not been designed for clusters, 
so why not ms windows ?!).
The real problem for the one who doesn't use any embarassingly parallel 
program, is clearly the parallel programming and scalability beyond  
some tens of cpus.

  Philippe Blaise
  CEA/Leti
  Nanotechnology Dpt.




More information about the Beowulf mailing list