Re: [PATCH] epoll more scalable than poll

[Jamie Lokier]

Davide Libenzi wrote:
> IMHO sys_epoll is going to be a replacement for rt-signals, because it
> scales better, it collapses events and does not have the overflowing queue
> problem.

Scalability is also solved by the signal-per-fd patch, as you know.
The main advantage of epoll is that it's lighter weight than rt-signals.

(IMHO signal-per-fd really ought to be included in 2.6 _anyway_, regardless
of any better mechanism for reading events.)

> The sys_epoll interface was coded to use the existing infrastructure w/out
> adding any legacy code added to suite the implementation. Basically,
> besides the few lines added to fs/pipe.c to support pipes ( rt-signal did
> not support them ), the hook lays inside sk_wake_async().

I agree that was the way to do it for 2.4 and earlier - you have to
work with a range of kernels, and minimum impact.

But now in 2.5 it's appropriate to implement whatever's _right_.

Time for me to take the big plunge and try a 2.5 kernel on my IDE
laptop, I guess :-)

-- Jamie