From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S965087AbWFILjp (ORCPT ); Fri, 9 Jun 2006 07:39:45 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S965233AbWFILjp (ORCPT ); Fri, 9 Jun 2006 07:39:45 -0400 Received: from ecfrec.frec.bull.fr ([129.183.4.8]:54712 "EHLO ecfrec.frec.bull.fr") by vger.kernel.org with ESMTP id S965087AbWFILjp convert rfc822-to-8bit (ORCPT ); Fri, 9 Jun 2006 07:39:45 -0400 Subject: Re: 2.6.17-rc6-rt1 From: =?ISO-8859-1?Q?S=E9bastien_Dugu=E9?= To: Mike Galbraith Cc: Ingo Molnar , linux-kernel@vger.kernel.org, Thomas Gleixner , John Stultz , Deepak Saxena In-Reply-To: <1149852951.7421.7.camel@Homer.TheSimpsons.net> References: <20060607211455.GA6132@elte.hu> <1149842550.7585.27.camel@Homer.TheSimpsons.net> <1149847951.3829.26.camel@frecb000686> <1149852951.7421.7.camel@Homer.TheSimpsons.net> Date: Fri, 09 Jun 2006 13:44:28 +0200 Message-Id: <1149853468.3829.33.camel@frecb000686> Mime-Version: 1.0 X-Mailer: Evolution 2.4.2.1 X-MIMETrack: Itemize by SMTP Server on ECN002/FR/BULL(Release 5.0.12 |February 13, 2003) at 09/06/2006 13:43:19, Serialize by Router on ECN002/FR/BULL(Release 5.0.12 |February 13, 2003) at 09/06/2006 13:43:20, Serialize complete at 09/06/2006 13:43:20 Content-Transfer-Encoding: 8BIT Content-Type: text/plain; charset=ISO-8859-15 Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org On Fri, 2006-06-09 at 13:35 +0200, Mike Galbraith wrote: > On Fri, 2006-06-09 at 12:12 +0200, Sébastien Dugué wrote: > > On Fri, 2006-06-09 at 10:42 +0200, Mike Galbraith wrote: > > > > After boot, it takes a very long time for KDE to finish loading... more > > > than five minutes for the desktop background to finally appear. Tasks > > > which are doing nothing but a ~50ms gettimeofday() select() idle loop > > > show up in top as using 1 to 3 percent cpu, though strace of these looks > > > fine. Starting any threaded app takes ages, whereas plain-jane things > > > like gcc work fine. For example, if I fire up xmms, the gui comes up > > > quickly, but it takes over three minutes from the time I poke play until > > > the first sound is emitted. Starting evolution takes even longer. > > > > > > Hoping that something might show up while running glibc-2.4 make check > > > to save me from wading through huge truckloads of strace, I tried it. > > > It repeatedly goes boom. RT29 goes boom the same way, but doesn't > > > exhibit the slow threaded app symptom. Drat. > > > > > > > Yep noticed that here too. As you pointed out it seems to be related > > to threaded apps. ls, vi, ... work fine whereas xemacs or others are > > real slow, portmapper fails to respond, ... > > > > I've got no indication in the logs that something went wrong, nor > > do I see any kernel BUG or WARNING. > > > > My box is a dual 2.8GHz HT xeon w/ 2GB mem. > > > > rt29 was fine as is 2.6.17-rc6. > > > > I feel a bit perplexed here. > > I found xmms problem. > > [pid 8498] 12:53:49.936186 socket(PF_INET, SOCK_STREAM, IPPROTO_IP > [pid 8498] 12:53:49.936551 <... socket resumed> ) = 9 <0.000301> > [pid 8498] 12:53:49.936774 fcntl64(9, F_SETFD, FD_CLOEXEC > [pid 8498] 12:53:49.937287 <... fcntl64 resumed> ) = 0 <0.000465> > [pid 8498] 12:53:49.937451 setsockopt(9, SOL_SOCKET, SO_REUSEADDR, [1], 4 > [pid 8498] 12:53:49.937630 <... setsockopt resumed> ) = 0 <0.000110> > [pid 8498] 12:53:49.937893 connect(9, {sa_family=AF_INET, sin_port=htons(16001), sin_addr=inet_addr("127.0.0.1")}, 16 > [pid 8498] 12:56:58.902958 <... connect resumed> ) = -1 ETIMEDOUT (Connection timed out) <188.964934> > > which should have been... > > [pid 7385] 13:21:38.715146 socket(PF_INET, SOCK_STREAM, IPPROTO_IP) = 9 <0.000011> > [pid 7385] 13:21:38.715192 fcntl64(9, F_SETFD, FD_CLOEXEC) = 0 <0.000007> > [pid 7385] 13:21:38.715237 setsockopt(9, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0 <0.000008> > [pid 7385] 13:21:38.715283 connect(9, {sa_family=AF_INET, sin_port=htons(16001), sin_addr=inet_addr("127.0.0.1")}, 16) = -1 ECONNREFUSED (Connection refused) <0.000060> > > So much for the easy part. > I'm starting to believe that it's network related. Pinging my box from a remote host gives a ~.3 ms round trip whereas pinging localhost gives ~500ms. Something real weird is going on here. I'm trying to dig deeper but I'm no expert here. Sébastien.