>> Matt Fair <[EMAIL PROTECTED]> writes: > On a large scale, would it be worth having 200 100 MHZ computers, or > would I just be wasting floor space and should get a better processor > to use in a cluster?
As a norm, you want to avoid inter node communication as much as possible. This usually means each processing element has to be able to process as much information as possible and as the problem allows. For so called embarrassingly parallel problems you obtain much better results if you invest in some fast PEs than in a larger ammount of not so fast ones. You have to be careful here: if you go for the SMP solution, don't forget to double the ammount of RAM at each node, otherwise you might find yourself in a position where the problem size you can solve hasn't increased much. -- Marcelo | "Millennium hand and shrimp." [EMAIL PROTECTED] | -- (Terry Pratchett, Hogfather)
