Re: the benifits
On Thu, Jul 12, 2001 at 09:18:46AM -0400, Kurt Keville wrote:
> My rule of thumb is that if a smaller, faster solution costs more than
> twice as much as 2 other solutions, don't do it. In other words, don't
> violate Moore's Law to accommodate Amdahl's Law (n.b. as far as gross
> overgeneralizations go this seems to hold up pretty well). This is why I
> would rather buy 2 Athlon boxes than 1 (equivalent) Athlon MP box. CPU
> density is not necessarily a goal of the Beowulf archetype and DC real
> estate is one of the "free" variables in the equation... high speed
> interconnects don't change the cost structure as much as they used to; you
> can get Gigabit cards for $100 now....
Don't forget the additional costs of double the number of uniprocessor
boxes in terms of power, air conditioning, and maintenance. All of
these make CPU density more important again.
On small processor farms, this might not be so important, but once you
get into the realm of thousands of CPUs it makes an enormous difference.
Tim.
Reply to: