Has anyone yet tried if NIO is doing any better in Tiger? Haven't been able to get any info if they have done some overhauling to the libs.
Trying out some benchmarks out here myself, will post info about it soon, but would be great to know if someone's already tried it out.
I've run a basic connection acceptance test on NIO. Tried it out on Linux as well as Windows. Seems like there is some problem with the select() implementation. Basically the CPU usage is 100% during the test. Managed to reach over 100K concurrent connections, but even at idle there is CPU usage of 30% (Athlon XP 2400+ 2.0 Ghz). Memory usage is also much much higher than a java.net.ServerSocket based server. The NIO client seems to work fine.
Will be making a test suite and posting it up someplace soon so others can try it out as well.
I have put together a simple NIO service system that acts as a router for xml data ( rss feed etc ) depending on the preferences set by connected users.
I am wanting to load test my app and so I am intersted in how I can implement a NIO based load client.
Would it be possible for you to send my you test harness?
I came across your post from last year and was wondering if you made any progress with NIO on JDK 1.5? I am just starting out on a project that will do something simialar to what yours does -- parse XML. What became of your XML-parsing system? I would appreciate any hints and/or source code.
Alot of the NIO work in 5.0 just didn't happen. That said there is a good NIO demo in the JDK instead. From my own experiments you have to do more tuning with Java's select() than with the C library version.