> With small changes to the existing model, we can get significant
> speedups. A problem I see with ioevents is it's a quite different
> model.
Oh definately different. But it seems worth it based on real-data and
real problems I've observed with Apache. There's some nice papers talking
about a webserver written on NT which uses this alternate model of i/o.
See <http://www.cs.wustl.edu/~jxh/research/research.html#papers>.
It's still a few months down the road before I can play with this stuff in
Apache... we have to go through 1.3 betas now. But I'm excited that there
appears to be a POSIX method of accomplishing the same thing as NT
completion ports. I'd much rather code it to POSIX.
Dean