> I'll be working on those patches.... but the spec talks about the
> servlet container, not the connector, right? So if Tomcat has a
> container which relies on NIO it would be ok, right? And the container
> is where NIO would make the most difference.
Don't you mean the CONNECTOR in those last two sentences?
> Servlets still need to run in plain old threads, but static content
> (which is where Tomcat is weakest) are what would benefit from NIO.
Hence the reason for using apache as your front-end. If your static content
isn't under a filtered domain, and therefore does not need to be seen by
tomcat, it needn't go to tomcat.
> In fact, I think that a pure Java web server using NIO properly
> (ie, select loop, not threads) may be able to outperform Apache
> (which uses threads, not a select loop). Maybe I will try to
> write a little select loop http, and then use it as a container.
This was, in fact, the test case for java.nio performance testing last
summer. I don't want to reiterate the entirety of the findings, but I'll
see if they are being published to the web. Summary is that switching to
java.nio didn't create miracles.
Personally, I like java.nio. And it might be worthwhile to see a connector
that uses java.nio. But you'll find that you've still got real work to do.
--- Noel
--
To unsubscribe, e-mail: <mailto:[EMAIL PROTECTED]>
For additional commands, e-mail: <mailto:[EMAIL PROTECTED]>
