> The server component will create a new thread as the previous one reached
> the maximum number of client you've setup. It is the OS which schedule the
> available CPUs to the threads in the ready state.
OK, but what if I want one thread per CPU?
So if I have a 4 CPU rig and 400 connections, I want to have 4 threads and
100 connections per thread.
But if the load increases to 600 connections, I want to have 4 threads and
150 connections per thread.
So far as I can tell, having one thread per CPU avoids serialisation effects
(ignoring any shared state or
synchronisation). Do you think that this might be faster than IO completion
ports (whose main benefits seem
to be keeping all CPUs loaded and avoiding context switches)?
Can this be done with UDP as well? (Direct incoming datagrams to threads in
a load-balanced manner?)
Thankyou for your time.
To unsubscribe or change your settings for TWSocket mailing list
please goto http://www.elists.org/mailman/listinfo/twsocket
Visit our website at http://www.overbyte.be