On Thu, 22 Dec 2005 21:16:11 -0000 (GMT) "christopher baus" <christopher@baus.net> wrote:
Interesting. Could you give an example where select() is faster than epoll()? I suspect it is with smaller numbers of file descriptors.
Actually, I do not remember it being a smaller number of FDs. I believe I remember epoll outperforming select() only when there were a few file descriptors ready. In tests of lots of FDs, I think I remember select() actually outperforming epoll() when lots of the file descriptors were ready for input at the same time. IIRC, epoll was FABULOUS with lots of FDs, with infrequent activity. However, with the same number of FDs, select() actually outperformed epoll() when there was data ready to be read on most of the file descritors. Again, this is from memory of work done a long time ago, but I bet if you wrote a quick example of what I describe, you will find similar results. If not, I'll give you your money back on the price you paid me for my opinion ;-)