Fastest server I've ever seen was at Microsoft (it's not shipping). It's all
in ASM and the outgoing response is generated in the network card's
interrupt handler and transmitted before EOIing the interrupt. The TCP/IP
stack and the server are integrated as device drivers. At heavy loads the
network card is polled for more incoming requests from inside the interrupt
context. This results in a single interrupt supporting thousands of requests
(of course your machine looks like it is frozen). All data transfer to/from
the net card is DMA based.
The design will beat TUX. But who cares, it only really matters for
benchmarks.
Jon Smirl
[EMAIL PROTECTED]