Python company worth $100M

[Bill Tutt]

> From: Mike Fletcher [mailto:mcfletch at vrtelecom.com]
> 
> 
> Nope, don't think you're missing anything. If I recall correctly, the
> asyncore docs even mention their assumption that the 
> processing to be done
> per call is trivial.  Since all other channels are "blocked" while the
> compute-heavy call is processing, the system will suffer performance
> degradation for all other clients.  Thread-pools seem to be a decent
> approach for longer-running calls.
> 

Only if they're synchronous operations, the really crazy people move
completly to an asynchronous finite state machine, have one thread per CPU
(and that thread only runs on that CPU), and schedule their work in chunks
so that one hunk of code can process multiple "work packets" at once,
thereby playing nice with the memory caches. There are apps that need this
performance and actually do this btw. Not sure if you can schedule a thread
to sit on one CPU on anything besides NT though. I'd be interested to know
otherwise...

Bill