Re: [patch] total-epoll r2 ...

[Davide Libenzi]

On Sat, 2 Nov 2002, Davide Libenzi wrote:

>
> The new epoll approach to event delivery seems stable on my machine an
> result about performance look pretty good. The changes that the new design
> introduce are :
>
> *) Full support for all devices that support ->f_op->poll()
>
> *) Multi-threaded support ( if you really like them )
>
> *) The function epoll_ctl(EP_CTL_ADD) drops an event if conditions are met
>
> *) Less intrusive design by hooking directly the poll support
>
> *) The patch is smaller, and this is a good news
>
> *) Looks even faster to me
>
> *) The function epoll_create(int size) has been changed. Not the "size"
> 	parameter is just an hint to the kernel about how to dimension its
> 	own internal data structures. In theory you can call
> 	epoll_create(200) and stock 10000 fds inside epoll

*) The old /dev/epoll access has been dropped with this release.

- Davide