Hello!

At the moment pound uses threads and blocking reads to communicate
with its clients and its backends. IMHO the use of epoll would speed
up pound by far.
Is there any reason for not using epoll?

Best wishes

Bernhard Grün, Germany

On Sun, 2006-09-10 at 17:25 +0200, Bernhard Grün wrote:[...]

1. epoll() is Linux specific. We want no unportable code in Pound.

2. it requires a very different model (async routines). Pound is
multi-threaded.

Re. point 2: I am aware that many people are reluctant to go to
multi-threaded models - they are complex, and many of us object to the
thread creation overhead. On the other hand it is the only model that
scales well on multi-processor architectures, and with all new chips
moving in that direction I believe it to be a win in the long run.[...]

[...]
Would you like to consider to reuse anything from the following approaches?
- http://www.monkey.org/~provos/libevent/
- http://liboop.org/why
- http://apr.apache.org/docs/apr/group__apr__poll.html
[...]
I see the possibility to mix and match several techniques.
Example:
http://httpd.apache.org/docs/2.2/mod/event.html

Regards,
Markus