Forum OpenACS Q&A: Anyone have experience with Beowulf clusters?

Collapse
1: Anyone have experience with Beowulf clusters?
Posted by Louis Gabriel on 09/07/01 05:25 PM
I know it is probably off topic, but I think you all are really a
great bunch of knowledgeable people.  Hope it's OK to ask this here.
Heck, maybe it is on topic if one used a Beowulf cluster for web
services (some requiring lots of calcs in a short time?).  Talk about
a load balancer!  *wink*

I've been reading up on this topic.  It seems really interesting.
Plus, a lot of stuff, like data mining, protein folding and other
areas requiring huge computational power are likely to be hot areas
for years to come.

What are your thoughts and experiences regarding Beowulf clusters?

Seems like an area a geek could **really** sink his or her teeth into!

Take care,

Louis

Collapse
2: Response to Anyone have experience with Beowulf clusters? (response to 1)
Posted by S. Y. on 09/07/01 06:54 PM

This topic was discussed over the the aD web/db bboard in April 1999. Basically, there are certain situations where having clusters of cheap Linux boxes, etc. make sense: a Web-based USENET server (like Dejanew, now Google Groups). Google, as most people here probably already know, runs on thousands of Linux boxes to build their search indexes.

Beowulf clusters are most effective when the tasks can be chopped up in nice little discrete and easily manageable units and there isn't much data dependency, and you need mostly CPU cycles.

Summary: it depends on the task you're doing.

Note: I personally have never worked on a web site that ever required more than one CPU, nor do I have any experience with Beowulf clusters. I do have lots of experience with SMP and cc-NUMA boxes (a.k.a., "The Big Iron"), mostly dealing with rendering computer animation though.