Hi guys,
we are having a lot of discussion about the way to process chains, FSM
and other high level concerns.
This is *really* interesting.
Although, this is just part of the problem.
Julien has already coded a rough semi-working soltuion using a very
basic chain, with some HTTP protocol codec implemented. I do think we
should use this experiment not as a base for any Chain implementation,
but as a base for a performance benchmark.
I think that we need to determinate what is the best approach regarding
the core system (ie, Acceptor and Processor). How many threads ? How
many selectors ? How to spread events (OP_READ, OP_WRITE) on selectors ?
Etc.
This can be seen as an orthogonal work, but a very important one. We can
have the best possible chain API, but if we have some kind of bottleneck
in the front end, we are dead anyway.
Once we have some good metrics on this part, changing the internal chain
will be really easy, IMO.
Just my morning random thougths...
--
Regards,
Cordialement,
Emmanuel Lécharny
www.iktek.com
