From owner-freebsd-net Mon Aug 12 12:25:54 2002 Delivered-To: freebsd-net@freebsd.org Received: from mx1.FreeBSD.org (mx1.FreeBSD.org [216.136.204.125]) by hub.freebsd.org (Postfix) with ESMTP id 31EE337B400 for ; Mon, 12 Aug 2002 12:25:50 -0700 (PDT) Received: from web10409.mail.yahoo.com (web10409.mail.yahoo.com [216.136.128.122]) by mx1.FreeBSD.org (Postfix) with SMTP id EDF4A43E65 for ; Mon, 12 Aug 2002 12:25:49 -0700 (PDT) (envelope-from opolyakov@yahoo.com) Message-ID: <20020812192549.24783.qmail@web10409.mail.yahoo.com> Received: from [67.112.212.200] by web10409.mail.yahoo.com via HTTP; Mon, 12 Aug 2002 12:25:49 PDT Date: Mon, 12 Aug 2002 12:25:49 -0700 (PDT) From: Oleg Polyakov Subject: Re: Initial congestion window increase To: Mike Silbersack Cc: freebsd-net@freebsd.org In-Reply-To: <20020810215044.O62906-100000@patrocles.silby.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Sender: owner-freebsd-net@FreeBSD.ORG Precedence: bulk List-ID: List-Archive: (Web Archive) List-Help: (List Instructions) List-Subscribe: List-Unsubscribe: X-Loop: FreeBSD.org --- Mike Silbersack wrote: > > On Fri, 9 Aug 2002, Oleg Polyakov wrote: > > > Here is a patch allowing to increase TCP's initial > congestion > > window up to 4 mss but less then 4380 bytes as specified in > > experimental RFC 2414 and draft-ietf-tsvwg-initwin-04.txt. > > It doesn't touch idle congestion window as per draft. > > Despite this change being in an RFC, I'm not sure that it's > really worth > implementing. While increasing the slowstart flightsize might > do wonders > for benchmarks of short connections, the actual effect on real > world tests > seems much more murky. Yes, real world is murky;) RFC2414 references RFC2415 - Simulation Studies of Increased Initial TCP Window Size which shows significantly increased throughput and decreased delays and some increased drop rate - around 1% more for connections with initial window=3. It also references RFC2416 - When TCP Starts Up With Four Packets Into Only Three Buffers where experiment shows decreased delays and "no harm" for transmission. > I believe that there _is_ an argument for using mss * 2 as the > default > flightsize, however. Supposedly, some OSes using delayed ACKs > will delay > the first ack, causing a 200ms delay which can really slow > down the > transfer of small web pages / etc. If you can find (tcpdump) > evidence to > back this up, I could agree with raising the value to 2 * mss. > Beyond > that, however, seems like a cheap way to inflate benchmarks > and cause > congestion. Well, if we have congestion with cwnd = 2 * mss, we have it regardless of initial window. Just in case of initial window = 2 * mss we have congestion 1 rtt earlier... By the way looking into tcpdump revealed we have initial window = 2 * mss in case of slowstart_flightsize = 1. I've seen it on FreeBSD 4.5 and 4.6 . What is this - an ancient bug in BSD stack? If I set slowstart_flightsize to 0 then tcpdump shows we starting with initial window 1... > Mike "Silby" Silbersack > > __________________________________________________ Do You Yahoo!? HotJobs - Search Thousands of New Jobs http://www.hotjobs.com To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-net" in the body of the message