Re: Paul Graham's PyCon Keynote & The Programmer's Apprentice

[Steve Dekorte <address@hidden>]

On Monday, March 31, 2003, at 02:42 PM, Andrew McKinlay wrote:
> How about Hillis's Connection Machine and its Lisp?

I think it had the idea of combining data/computation parallelism right 
but it was done on a macroscopic level - so it ended up just being a 
bunch of PCs on a custom network. And the software parallelism was 
either vector or processor oriented, which limits it's usefulness. I 
think it would be more interesting to explore the question of how 
fine-grain can combined data/computation parallelism be taken. Could we 
give every number it's own local ALU? If not, how about every 100 
numbers? etc. Do we find we still need CPUs at all if start heading 
down that path?

Cheers,
Steve
Io, a small language: http://www.iolanguage.com/