On Sun, Oct 24, 2010 at 11:42:45PM -0400, Markus Mottl wrote:
> On Sun, Oct 24, 2010 at 18:50, Jérémie Dimino <jeremie@dimino.org> wrote:
> > I made an implementation of lwt using libev [1]. I tested it with
> > ocsigen and ab but the result was always a bit better with select than
> > with epoll. That is why i did not replace select by libev in the main
> > branch. In fact i never found the source of any benchmark comparing
> > select to epoll on the web.
> 
> The performance of select was also usually slightly better in my
> experiments than with epoll for at least a few tens of descriptors.
> It really depends on what your requirements are.  If you are facing
> hundreds or even thousands of connections, you'll probably want to
> consider epoll.  select does not scale well.

I think if you really had 10,000 clients per server you'd probably
want to consider your whole architecture.  Putting nginx or a very
cut-down Apache on the front and memcached between the webserver and
the database.

Rich.

-- 
Richard Jones
Red Hat