eosterweil at hotmail.com
Thu Dec 14 15:44:21 PST 2000
This is a case where you're going to have to look at your problem as more of
a parallel program than a distributed program. You're processing sounds
like it is very tightly coupled (or course grained) in nature. You might
consider following a parallel programming methodology and re-examine you
partitioning scheme. There are a lot of clever ways to parallelize
seemingly atomic processing. There is a book called "Parallel Computer
Architecture : A Hardware/Software Approach" by David E. Culler, Jaswinder
Pal Singh, Anoop Gupta that has a VERY good section on how to parallelize
very centralized algorithms. In addition they have some very informative
case studies in there that can give examples of this. I highly recommend
>From: Mofeed Shahin <shahin at labf.org>
>Reply-To: shahin at labf.org
>To: beowulf at beowulf.org
>Subject: Distributed Algorithm.
>Date: Thu, 14 Dec 2000 19:31:26 +0000
>I am currently working on a program which is simulating the movements of
>number of nodes.
>The problem I am having is that it doesn't take long to calculate the next
>position of Node n1, n2, n3, etc...... I need all the locations for time t,
>before I can start calculation the locations for t+1.
>And therefore the bottleneck becomes the IO bandwidth between the cluster
>processing units, and not CPU.
>Is there a way to get rid of the bottleneck? (ie change of algorithm?)
>Beowulf mailing list
>Beowulf at beowulf.org
Get more from the Web. FREE MSN Explorer download : http://explorer.msn.com
More information about the Beowulf