Howdy,
I would have to agree with Remy here. The example given doesn't really prove anything in my mind.
That's what I was thinking too as I read the article. It's not representative of the real world. But I wanted to see if I was alone in my skepticism or not. (And I remember we've discussed NIO multiple times in the past).
However:
- obviously it would be beneficial to do NB for HTTP keepalive, at the cost of performance for "small" servers (small means servers handling maybe up to a thousand users simultaneaoously; given the tendency of the other components of an enterprise app to scale to these numbers, it's not small, actually)
- NIO (blocking) would be beneficial performance wise, since classic I/O is a wrapper on top of it nowadays
Stuff to factor in the equation:
- OSes with competent scheduler designed for hundreds of thousands of threads, such as Linux 2.6 (as someone wrote, http://www.sics.se/~joe/apachevsyaws.html proves a threaded design scales with a good scheduler); the VMs will have to get a little more lighter weight, probably
- VMs like JRockit which could map many Java threads to one native thread
Rémy
--------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
