syscall messaging interface API

[Adrian Chadd]

In article <20030724082921.GA18216 at xxxxxxxxxxxxxxxxxxxx>, David Schultz wrote:

> Even if you did manage to come up with a program where ``syscall
> throughput'' is significant, batching the calls as Matt suggested
> would provide much better performance benefits than microoptimization.

I agree - I've been playing with insanely fast IO servers (where insanely fast
means 30k FDs pushing out a few hundred mbits of traffic) and I've hit syscall
throughput limits.

It'd be nice to attach actual buffers to kqueue messages which the OS could
fill out for you - that way you could do network IO in a pair of syscalls -
but I've been a little busy. :p



adrian

-- 
Adrian Chadd			<angryskul> learning is bad
<adrian at xxxxxxxxxxx>		  <angryskul> it just makes the people around you dumber
(angryskul == alfred at irc)	    <angryskul> :(