Interesting article from IBM about how you might implement a servlet container using non-blocking IO, by essentially layering a non-blocking demultiplexer on top of the standard servlet input/output stream stuff.

Pleasingly the article vindicates my position during a recent pub conversation with Jon where I asserted that it should be possible to use NIO to write a better servlet container, but due to alcoholic impairment I was unable to articulate precisely how.