Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP Path: utzoo!linus!philabs!cmcl2!harvard!husc6!panda!genrad!decvax!ittatc!dcdwest!sdcsvax!sdcrdcf!hplabs!hpda!hpisoa2!hpitg!gipsy!ch@gipsy From: ch@gipsy Newsgroups: net.lan Subject: Re: Orphaned Response Message-ID: <263@gipsy> Date: Wed, 30-Apr-86 02:07:00 EDT Article-I.D.: gipsy.263 Posted: Wed Apr 30 02:07:00 1986 Date-Received: Wed, 14-May-86 01:48:41 EDT References: <424@gould9> Lines: 35 We have been running ISO and TCP-Ip in parallel for some time now at INRIA. This will be reported in the next USENIX conference. Each has pro and contras, so lets state the advantages: For TCP-IP: Already established, comes with a lot of applications likes FTP, TELNET, RLOGIN, SMTP, RSH, etc. Available today on a large range of computer. This is not the case of ISO: the only standard application available today is X400, and in a certain sense the Teletex. For OSI: Getting agreed upon by a large number of manufacturer. Technically much better than TCP-IP, which hardly matches the services of the OSI transport, but does none of the functionalities of the session or presentation layers. We are running OSI on X25 and on Ethernet. We use X25 layer 3 on Ethernet, according to DIS8881, above IEEE802.2 LLC1. We establish X400 connections on an operational basis we other systems in UK and Germany; note that the OSI was independantly developped on these systems. It is extremely easy to set up a random connection using the OSI protocol and X25 networks, whilst to do the same with IP you have to formally interconnect networks. From an architectural point of view, ISO gives you the choice between connection oriented and connectionless networks. In the latter case of ISO-IP & TC-4, you have exactly the same repartition of function has with IP & TCP, with one advantage to IP due to short headers: the IP adresses are 32 bits, while the ISO NSAP addresses can be 20 bytes. On the other hand, there is a possibility with TC4 to negociate the usage of transport level checksum on a per-connection basis. We have measured almost the same throughputs for TCP and the OSI session, in file transfer applications. The OSI code is slightly smaller than the TCP code in the kernel (both are implemented as "sockets" protocols). Christian Huitema ()