Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!watmath!clyde!burl!ulysses!harpo!ihnp4!drutx!houxe!hogpc!houti!ariel!vax135!floyd!cmcl2!seismo!hao!hplabs!hpda!fortune!wdl1!jbn
From: jbn@wdl1.UUCP
Newsgroups: net.lan
Subject: Comments on IP/TCP on Ethernet
Message-ID: <243@wdl1.UUCP>
Date: Mon, 14-May-84 21:14:27 EDT
Article-I.D.: wdl1.243
Posted: Mon May 14 21:14:27 1984
Date-Received: Thu, 17-May-84 03:33:06 EDT
Lines: 60


    Some advice on IP/TCP networks:

    1.  Trailer encapsulation (a la 4.2BSD) is nonstandard.  So is 
	the use of IP datagrams in excess of 576 bytes.  You are only
	entitled to use either if you have ``a priori information that
	the receiving host is able to accept them''.  TOPS-20, for example,
	won't.

    2.  Ethernet address resolution protocol, (Plummer's) is becoming
	accepted as the means for finding Ethernet addresses from IP
	addresses.  Undocumented, but increasingly common, is arranging for
	gateways on an Ethernet to answer address resolution polls for
	anything they can reach.  This means that ARP is all the routing
	a host on the Ethernet needs.  The case of two gateways on an
	Ethernet still presents problems.

    3.  There are no really satisfactory gateway implementations that one
	can buy.  There are a number of university-developed gateways, but
	no products as yet.  We are developing two, but they aren't available
	yet.

    4.  Comments on implementations:

	    -  3COM's UNET was the first commercial implementation; it
	       has been withdrawn from the market.  We use a much-modified
	       version of UNET on some of our systems.

	    -  COMPION's implementation for VMS is very slow, especially
	       for TELNET.  Compion isn't very interested in the business,
	       as far as we can tell; we are an unhappy customer.

	    -  The Wollongong implementation for VMS is a port of the 
	       4.2BSD one for UNIX.  It appears to be about as good as
	       the 4.2BSD, and better than Compion's, but we have not tested
	       it ourselves.

	    -  DEC's TOPS-20 implementation is buggy and poorly supported.
	       Plan on investing systems programming time.  Contact Mark
	       Crispin at Stanford (MRC@SCORE) for details and fixes.

	    -  4.2BSD's implementation is fair; there are known bugs in
	       retransmit timing and mixed-speed nets have problems with
	       gateway congestion.

	    -  SUN has the standard 4.2BSD bugs.

	    -  Apollo seems to have done a nice job; we tested UTAH-APOLLO
	       and didn't find anything wrong, a statement we seldom make.
	
            -  Perkin-Elmer still uses UNET.

	    -  Communications Machinery and Excelan have boards which
	       support Ethernet, IP, and TCP.  We haven't tested these
	       yet.

	    -  Network Research Corporation offers a version of Fusion,
	       their networking package, which uses IP/TCP.  This is 
	       presently available for some M68000 machines, and is
	       supposedly becoming available for the IBM PC and its clones.