On Sat, Jun 11, 2011 at 1:59 PM, Martin Sustrik <[email protected]> wrote:

> Ok. I should have not mentioned the second caveat. The real point is that
> epoll is meant to reduce complexity of polling from O(n) -- where n is the
> size of pollset -- to O(1). If we use epoll to implement zmq_poll(), we get
> complexity of O(n log n) which is actually worse than O(n) that you get
> using poll or select.

Yes, if you use epoll you'd want a new API, not zmq_poll. Like I said,
if you use a reactor on top of zmq_poll, it should be transparent to
applications. In any case, using zmq_poll for anything complex is a
pain, it gets rapidly messy to keep track of timers and events.

Do we have figures for the cost of zmq_poll? I'm kind of sceptical of
any optimization that's not based on solid numbers.

-Pieter
_______________________________________________
zeromq-dev mailing list
[email protected]
http://lists.zeromq.org/mailman/listinfo/zeromq-dev

Reply via email to