MACOS X VERSION PLEASE.

Bruce Sterling needs to start running faster; the future is catching up.

In related news, Hank Dietz at U Kentucky uses genetic algorithms to wire his Beowulf clusters; the result is super-fast and super-messy: http://www.aggregate.org/KASY0/

GA's are so old school, and limited; much too biased by the initial problem design. This problem seems like a natural match for genetic programming instead. GP for when you want a real evolutionary struggle.

Yeah, the only problem with this is that, while GA's are good at approximating global minima/maxima, they're not really good for local optimization... "Super-messy" is about the right word for it. GA's might get you to 90 or 95% of peak efficiency -- and when that's good enough, great! -- but their stochastic nature more or less implies that you'll only rarely get a genuinely optimal solution out of them.