From mboxrd@z Thu Jan 1 00:00:00 1970 From: Andi Kleen Subject: Re: cat /proc/net/tcp takes 0.5 seconds on x86_64 Date: Wed, 27 Aug 2008 23:47:00 +0200 Message-ID: <20080827214700.GB26610@one.firstfloor.org> References: <200808261549.m7QFnVUN032543@bz-web1.app.phx.redhat.com> <20080826163719.GA25066@redhat.com> <87zlmyr5nz.fsf@basil.nowhere.org> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Andi Kleen , Dave Jones , netdev@vger.kernel.org, j.w.r.degoede@hhs.nl To: Trent Piepho Return-path: Received: from one.firstfloor.org ([213.235.205.2]:40414 "EHLO one.firstfloor.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754010AbYH0Vo2 (ORCPT ); Wed, 27 Aug 2008 17:44:28 -0400 Content-Disposition: inline In-Reply-To: Sender: netdev-owner@vger.kernel.org List-ID: On Wed, Aug 27, 2008 at 02:29:18PM -0700, Trent Piepho wrote: > On Wed, 27 Aug 2008, Andi Kleen wrote: > > On a 2GB Core2 system here I see a time cat /proc/net/tcp > /dev/null > > constently dropping from 0.44s to 0.4-0.8s system time with this change. > > This is with mostly empty hash tables. > > Umm, 0.4 to 0.8 seconds is on average _slower_ than 0.44 seconds. Did you > mean .04 to .08? Yes noticed that too -- i think it must have been .04-.08 and the zeros got lost somewhere. I remember that it was significantly faster. It's really a benchmark how fast an read lock aquire/release is, but even on the best systems it's quite slow compared to lockless code. -Andi -- ak@linux.intel.com