"Rousseau, John" <[EMAIL PROTECTED]> writes:
>
> A performance critical application server that needs to handle
> hundreds of simultaneous requests and wants to keep a pool of
> threads around so as to not re-instantiate a client thread per
> incoming request.
This is the scenario we discuss in a paper that I recently wrote,
entitled "A Design Framework for Highly Concurrent Systems". You can
get it from:
http://www.cs.berkeley.edu/~mdw/papers/events.pdf
We detail the problems with threads-based and event-driven programming
for building highly-concurrent servers exactly of the sort you
describe, and describe a design based on a hybrid between events and
threads which allows you to get high performance and get the benefit of
straightforward threaded programming. Your suggestion to use thread pools
above is telling, since that's core to our model.
Anyway, I'd be interested to hear any comments about this if you
get a chance to check it out.
Matt Welsh, UC Berkeley
----------------------------------------------------------------------
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
