On Mon, 2011-09-12 at 09:52 -0700, Rajiv Mordani wrote:
> Hi All,
> Please review and provide input on the initial draft proposal - It
> is similar to what is there in Grizzly and also Tomcat AIO. It does
> not expose any of the NIO constructs per say but I think that we don't
> necessarily need to expose them. Please let me know what you'll think.
I did the proposed design (and implementation) in JBoss a few years ago,
it was a nice experiment and I don't plan to remove it, but:
- It doesn't actually help many people right now. The complexity in the
application is (much) higher, so until frameworks do it automagically,
it's not going to be used.
- It mandates the use of NIO (or worse ;) ) for the implementation,
taking away the "it just works" factor of using a default based on
java.io.
- Use will remain at 95% blocking IO, which means java.io is the most
efficient for the IO. The savings due to low thread counts are not
enough to compensate in some cases.
So it's a nice improvement, with a number of caveats.
--
Remy Maucherat <rmaucher_at_redhat.com>
Red Hat Inc