From owner-cvs-all Wed Jul 17 19:28:17 2002 Delivered-To: cvs-all@freebsd.org Received: from mx1.FreeBSD.org (mx1.FreeBSD.org [216.136.204.125]) by hub.freebsd.org (Postfix) with ESMTP id 2CA2D37B400; Wed, 17 Jul 2002 19:28:10 -0700 (PDT) Received: from patrocles.silby.com (d105.as8.nwbl0.wi.voyager.net [169.207.132.105]) by mx1.FreeBSD.org (Postfix) with ESMTP id A872043E42; Wed, 17 Jul 2002 19:28:08 -0700 (PDT) (envelope-from silby@silby.com) Received: from patrocles.silby.com (localhost [127.0.0.1]) by patrocles.silby.com (8.12.4/8.12.4) with ESMTP id g6I2WTcv083824; Wed, 17 Jul 2002 21:32:29 -0500 (CDT) (envelope-from silby@silby.com) Received: from localhost (silby@localhost) by patrocles.silby.com (8.12.5/8.12.5/Submit) with ESMTP id g6I2WTYf083821; Wed, 17 Jul 2002 21:32:29 -0500 (CDT) X-Authentication-Warning: patrocles.silby.com: silby owned process doing -bs Date: Wed, 17 Jul 2002 21:32:29 -0500 (CDT) From: Mike Silbersack To: Matthew Dillon Cc: cvs-committers@FreeBSD.ORG, Subject: Re: cvs commit: src/sys/netinet tcp_timer.h In-Reply-To: <200207180106.g6I16Dqd004375@apollo.backplane.com> Message-ID: <20020717211230.G83269-100000@patrocles.silby.com> MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: owner-cvs-all@FreeBSD.ORG Precedence: bulk List-ID: List-Archive: (Web Archive) List-Help: (List Instructions) List-Subscribe: List-Unsubscribe: X-Loop: FreeBSD.ORG On Wed, 17 Jul 2002, Matthew Dillon wrote: > : > :1. The value should be scaled by hz so that it does not change with HZ > :settings. (That's not central the problem, though.) > > The value should be scaled for algorithmic stability. It should not > be set at a particular latency because then you are making some > severe assumptions in regards not only to the network latency and > bandwidth, but also in regards to the type of applications running > over it. > > I have one person who is running filesystem-like transactional code > over the network. One second of latency screws things up in a major > way for him. Algorithmic stability or not, you must make this a constant factor, NOT something that changes with a HZ setting. Having a machine change TCP behavior as HZ is ranged from 100 to 10000 is NOT something which should happen! Your three tick setting is also making an assumption about the relation between bandwidth and latency. If I have 100 machines feeding through a gigabit link, the rtt may well be < 1ms. However, the available bandwidth for each machine is far less than one gigabit. Hence, flooding the network with retransmissions every 3ms is still very unfriendly. Another potentially bad sideeffect is that with low RTTs, connections may timeout excessively fast, causing many reset connections just because a router went down for 15-30 seconds. This reduces the reliability of TCP. > On a gigabit network a 1 second retransmission timeout is simply > absurd. Period. Yes, but something approaching 1ms could potentially cause more congestion and be worse. I checked, and linux has already moved to a 200ms RTO. That also seems reasonable, and I don't see a problem with setting that as our minimum RTO as well. Mike "Silby" Silbersack To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe cvs-all" in the body of the message