Path: utzoo!attcan!uunet!lll-winken!lll-tis!helios.ee.lbl.gov!pasteur!ucbvax!TWG.COM!dcrocker
From: dcrocker@TWG.COM (Dave Crocker)
Newsgroups: comp.protocols.tcp-ip.ibmpc
Subject: Re:  Novell and TCP/IP
Message-ID: <8808241146.aa00303@Louie.UDEL.EDU>
Date: 24 Aug 88 15:43:00 GMT
Sender: daemon@ucbvax.BERKELEY.EDU
Organization: The Internet
Lines: 27

The clarification about Micom/Interlan put their protocols -- on the gateway
pc vs. on a board attached to the gateway -- is, of course, correct.  Micom
uses an on-board processor with the protocols running on the card, attached to
the gateway pc and the ethernet.  As I understand their implementation, they
provide a berkeley-like sockets interface at the various PCs on the Novell
network.  (The pcs talk with the gateway over Netbios Session connections.)

Leo McLaughlin's note was trying to distinguish the pc/gateway differences
between the Micom and Wollongong products.  Relative to that topic, Micom's
approach does put the protocols "on the gateway".  I.e., the intelligent
card is part of the gateway.  The primary point was that the Wollongong
approach puts the protocols, particularly TCP, on the individual users'
pcs and the gateway is simply a pure IP router, albeit with an unusual
link-level protocol (currently netbios datagrams and soon Novell's ipx)
for one of its sub-networks.

The technical effects of our architecture are to make each pc a true
tcp/ip host, remove any limitation to the number of pcs that may be
simultaneously active -- as I understand it, the Micom approach limits you
to 32, whereas the Wollongong approach is limited only by the computational
and data-copying performance of the gateway pc -- and gives true end-to-end
reliability, since the tcp checksum is at the user's host, rather than at
the gateway.  There may also be some per-connection performance differences,
but we have not certified this, yet.

Dave Crocker
VP, Engineering
The Wollongong Group