CPUs for a Beowulf
James.P.Lux at jpl.nasa.gov
Mon Sep 8 12:30:27 EDT 2003
RGB's book at the Duke Brahma site (I'm sure he'll post the URL) covers
some of these tradeoffs..
You've posed an interesting question, because it's in the generic "what's
the best way to get lowest dollars per instruction executed" way, but
It's that sticky word "performance" which is the problem. Wwe can all
agree on what dollars mean and how much they are worth now, and in the
future. However, performance is different things to different people.
Is time worth money? (that is, is there a "wall clock time" as well as a
CPU cycles aspect... Older computers are cheaper in terms of executing a
particular number of instructions, but consume more support infrastructure
(cooling, staff time, etc.) because, if nothing else, they have to run
Is capital cost important, and, are intermediate results of interest..
There's a well known example where you have a "really big problem" that you
could either spend some money now, and compute for the next two years, or
save your money, wait a year, buy the (twice as fast) computers for the
same price and do the computation then, finishing at the same time. Of
course, you don't get any results during the first year, and for many
applications, the partial results early are used to guide the later work.
What's your particular labor/hardware maintenance/capital investment
tradeoff.. If you have copious FREE and SKILLED labor, the trade is
different... Likewise, if you have a "hard" reliability requirement and
can't tolerate partial (or complete) downtime, the trade is different.
Cluster computing, in some forms, also lends itself to "stealth,
below-the-funding-watchdog-radar" work. You can buy, borrow, collect,
etc., CPUs and gradually add them to an ever growing configuration. I
notice though, that the whole cluster computing thing is a validated way to
work, these days, and anyone with real work to do, and the budget to pay
for it, just goes out and builds a real Beowulf.
At 09:58 AM 9/8/2003 -0400, Robert Kane wrote:
> If anyone doesn't mind, may I ask a few questions. When given a
>specific application for which a cluster is being built it should be
>relatively simple to look are the requirements of the problem and the
>available hardware, and then determine which hardware solution is best
>for the problem. However, if the cluster is being built as a general
>purpose cluster for research, things become a bit more difficult, as (as
>I far as I can tell) there is no one answer. But, if anyone has any
>insight into the following problems it would be greatly appreciated.
>1. Single versus Dual CPUs?
> Both of these choices have their pros and cons and are each best
>suited for different types of problems. Given that the cluster will be
>used for a variety of problems, is there one which would be a better
>choice? Is there a particular configuration for which the majority of
>problems will run better? Is there a solution that on average provides
>more performance per dollar?
>2. CPU Type
> Intel and AMD's new 64-bit processors are finally beginning to become
>more common it appears. And from what I've seen the benchmarks are
>rather impressive. However, there seems to be a significant price
>increase going from previous generation chips (ie Xeon) to the new
>64-bit chips. In general is the increased performance worth the money
>invested, or would a larger number of slower chips be effective
>cost/performance wise? Apart from the increased electricial, A/C costs
>Thank you for any information concerning these issues, whether
>information be answers or links to good resources,
>Beowulf mailing list, Beowulf at beowulf.org
>To change your subscription (digest mode or unsubscribe) visit
James Lux, P.E.
Spacecraft Telecommunications Section
Jet Propulsion Laboratory, Mail Stop 161-213
4800 Oak Grove Drive
Pasadena CA 91109
Beowulf mailing list, Beowulf at beowulf.org
To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
More information about the Beowulf