From mboxrd@z Thu Jan 1 00:00:00 1970 From: =?ISO-8859-1?Q?D=E2niel?= Fraga Subject: Re: [PATCH] tcp FRTO: in-order-only "TCP proxy" fragility workaround Date: Tue, 26 Aug 2008 14:18:12 -0300 Message-ID: <20080826141812.589848a0@tux> References: <20080819213417.45133573@tux> <20080822183224.2d52f16c@tux> <20080822.143709.65615512.davem@davemloft.net> <20080823111446.06a350a2@tux> <20080824163843.33b4f890@tux> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: David Miller , thomas.jarosch@intra2net.com, billfink@mindspring.com, Netdev , Patrick Hardy , netfilter-devel@vger.kernel.org, kadlec@blackhole.kfki.hu To: "Ilpo =?ISO-8859-1?Q?J=E4rvinen?=" Return-path: Received: from yx-out-2324.google.com ([74.125.44.29]:56265 "EHLO yx-out-2324.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753751AbYHZRSW (ORCPT ); Tue, 26 Aug 2008 13:18:22 -0400 Received: by yx-out-2324.google.com with SMTP id 8so1192498yxm.1 for ; Tue, 26 Aug 2008 10:18:21 -0700 (PDT) In-Reply-To: Sender: netdev-owner@vger.kernel.org List-ID: On Tue, 26 Aug 2008 17:10:46 +0300 (EEST) "Ilpo J=E4rvinen" wrote: > There is more than one TCP flow in your workload btw (so using=20 > "connection" is a bit more blurry from my/TCP's pov). Some stall and = never=20 > finish, some get immediately through without any stalling and proceed= ok.=20 > So far I've not seen any cases with mixed behavior. Interesting. > It could be userspace related thing. Hmmm. I'll try to report this to the dovecot and inn lists. > It seems that there could well be more than one problem, with symptom= s=20 > similar enough that they're hard to distinguish without a packet trac= e. Yes, exactly! I think the same. > Did it solve in this particular case? At least for 995 nothing=20 Yes. nmap -sS always solves the problem. Very strange. nmap -sS for me is kind of brute force attempt to restablish the normal behaviour of the server...=20 Anyway, I disabled htb and frto and everything is fine for now. I'll keep investigating this. > ListenOverflows might explain this (it can't be ListenDrops since it'= s=20 > equal to ListenOverflows and both get incremented on overflow). Are y= ou=20 > perhaps short on workers at the userspace server? It would be nice to= =20 I use dovecot por mail. I'll post on the dovecot list. If it's an userspace issue, better. > capture those mibs often enough (eg., once per 1s with timestamps) du= ring=20 > the stall to see what actually gets incremented during the event beca= use=20 > there's currently so much haystack that finding the needle gets impos= sible=20 > (ListenOverflows 47410) :-). Also, the corresponding tcpdump would be= =20 > needed to match the events. Ok. If I had more useful information, I'll reply. Thank you very much! --=20