From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S263715AbTDNUwv (for ); Mon, 14 Apr 2003 16:52:51 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S263720AbTDNUwv (for ); Mon, 14 Apr 2003 16:52:51 -0400 Received: from 64-60-75-69.cust.telepacific.net ([64.60.75.69]:45329 "EHLO racerx.ixiacom.com") by vger.kernel.org with ESMTP id S263715AbTDNUwt (for ); Mon, 14 Apr 2003 16:52:49 -0400 Message-ID: <3E9B1F33.3060701@ixiacom.com> Date: Mon, 14 Apr 2003 13:50:59 -0700 From: Dan Kegel User-Agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.0.0) Gecko/20020615 Debian/1.0.0-3 MIME-Version: 1.0 To: linux-kernel@vger.kernel.org, provos@citi.umich.edu Subject: re: epoll support in libevent-0.7 Content-Type: text/plain; charset=us-ascii; format=flowed Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org Niels Provos wrote: > I just released a new version of libevent that supports Linux new > epoll mechanism; see > > http://www.monkey.org/~provos/libevent/ > > The page contains some performance comparisons for different event > notification mechanisms. Very cool. This will probably be quite useful for many users. BTW, Niels should perhaps mention that only the new, level-triggered version of sys_epoll is supported. > The library supports platform independent high performance network > applications. It chooses the fastest notification mechansims > supported by the operating system. That may not be *quite* true. Davide and I feel the edge-triggered version of sys_epoll can be slightly faster than the level-triggered version. Niels disagrees, but presumably a benchmark could be arranged to settle the question. Of course, after Davide and I win the argument :-), Niels will say "but the level-triggered version is easier to use", and he'll be mostly right. I'm stuck in /dev/epoll-land for a while, but one of these days, maybe I'll do that benchmark. - Dan