[pphaneuf]

Right on. Myself, I always figured epoll was a bit slower than poll when most/all of the fds are ready all the time, from reading the code.

Now, there's an experiment to figure out what the threshold is, and frankly, it's a bit lower than I had hoped (I would have guessed 0.9, 0.8 at worst), which is exactly why having the actual data is awesome.