From owner-freebsd-isp Mon Dec 28 13:26:23 1998 Return-Path: Received: (from majordom@localhost) by hub.freebsd.org (8.8.8/8.8.8) id NAA22785 for freebsd-isp-outgoing; Mon, 28 Dec 1998 13:26:23 -0800 (PST) (envelope-from owner-freebsd-isp@FreeBSD.ORG) Received: from korin.warman.org.pl (korin.nask.waw.pl [195.187.243.10]) by hub.freebsd.org (8.8.8/8.8.8) with ESMTP id NAA22778 for ; Mon, 28 Dec 1998 13:26:20 -0800 (PST) (envelope-from abial@nask.pl) Received: from localhost (abial@localhost) by korin.warman.org.pl (8.9.1/8.8.5) with SMTP id WAA26601; Mon, 28 Dec 1998 22:30:45 +0100 (CET) X-Authentication-Warning: korin.warman.org.pl: abial owned process doing -bs Date: Mon, 28 Dec 1998 22:30:42 +0100 (CET) From: Andrzej Bialecki X-Sender: abial@korin.warman.org.pl To: Javier Henderson cc: Dennis , Alan Batie , isp@FreeBSD.ORG Subject: Re: ATM WAN interface In-Reply-To: <199812281955.LAA26758@kjsl.com> Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: owner-freebsd-isp@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.org On Mon, 28 Dec 1998, Javier Henderson wrote: > Dennis writes: > > > Note that ATM has a LOT of overhead (like 30%). so ATM over a DS3 > > is not nearly the bandwidth of using straight HSSI or PTP. ATM is meant > > as a medium that can be switched at high speed, but as a PTP mechanism > > it is very poor. > > I've often wondered how well IP would do, throughput-wise, > over ATM with its 53 byte cells. "Packet fragmentation overhead" comes > to mind. I've never seen actual performance figures, however. Quite bad.. :-/ Let's do some math: 1) Suppose we have a 256-byte IP packet (which is quite close to average packet length on the net). This is usually wrapped into LLC/SNAP (8 bytes), then passed to AAL5 (which adds 8-byte trailer), and finally broken into 48-byte pieces of ATM payload (with 5-byte overhead). This gives us 6 cells to transmit, and total of 318 bytes. The overhead is around 24%. Of course this is only the overhead of ATM. 2) Suppose we have an interactive session, where 1-10 bytes are transmitted as a TCP payload. This gives us a 41-50 byte IP packets. Let's add LLC/SNAP and AAL5 (16 bytes, total 57-66 bytes), then we break it into cells (2 cells), and the total sum of bytes to transmit is 106. The overhead is 86% (for 41-byte IP packets), and 61% for 50-byte packet. The overhead counted against the TCP payload is quite horrendous.. :-)) So, the beauty of ATM is its ability to transmit _different_ types of data over the same link (such as IP, circuit emulation, video etc..), and not its efficiency when it comes to throughput - because this is where it really sucks... Hope this helps a bit. Andrzej Bialecki -------------------- ++-------++ ------------------------------------- ||PicoBSD|| FreeBSD in your pocket? Go and see: Research & Academic |+-------+| "Small & Embedded FreeBSD" Network in Poland | |TT~~~| | http://www.freebsd.org/~picobsd/ -------------------- ~-+==---+-+ ------------------------------------- To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-isp" in the body of the message