Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!seismo!rutgers!ucla-cs!zen!ucbvax!decvax!eagle_snax!geoff
From: geoff@eagle_snax.UUCP ( R.H. coast near the top)
Newsgroups: comp.dcom.lans
Subject: Re: Smart Ethernet boards
Message-ID: <146@eagle_snax.UUCP>
Date: Wed, 8-Jul-87 08:00:09 EDT
Article-I.D.: eagle_sn.146
Posted: Wed Jul  8 08:00:09 1987
Date-Received: Sat, 11-Jul-87 03:55:35 EDT
References: <283@sering.cwi.nl> <8212@utzoo.UUCP> <8255@utzoo.UUCP> <2302@ames.arpa>
Lines: 56
Summary: Please distinguish between client/server and between UDP/TCP.

In article <2302@ames.arpa>, lamaster@pioneer.arpa (Hugh LaMaster) writes:
> In article <8255@utzoo.UUCP> henry@utzoo.UUCP (Henry Spencer) writes:
> 
> >> If the on-board TCP/IP imposes more load on the CPU than an off-board
> >> TCP/IP, then something is radically wrong with the design...
> 
> I do not think that this is correct.  It takes a lot of CPU cycles to build a
> packet.  If your system is limited to one CPU, it may make a lot of sense to
> offload tcp/ip somehow, especially if you are using mostly terminal servers.

All of this originally started with smart boards for PC's under DOS, so let's
not forget that for the average PC DOS client the metric we care
about is not CPU load but response time. In the case of telnet, everything
is pretty well serialized (except for really demon typists :-), so
adding the overhead of (1) managing the CPU-adapter interface protocol
and (2) an extra copy [difficult to avoid] really can slow things
down.

While it's appropriate to focus on TCP, we shouldn't ignore UDP-based
services like NFS : a smart-board version of PC-NFS working with downloaded
protocols would be _significantly_ slower than the dumb board version.

BTW, check out the comment by Bob Metcalfe in the latest PC Week, in
which he says that the benefits of a smart card have never really been
demonstrated. I know 3Com offers both dumb and smart cards, but I suspect
that's because some people will always buy something "smart" in preference
to something "dumb"....

> 
> >....  And TCP/IP is not inherently a
> >very costly protocol to implement off-board.
> 
This is consistent with our experience in PC-NFS.

[long discussion about 4.2BSD telnetd, etc., all very true but irrelevant
to the PC]

> 
> >Given that we have a similar CPU and a large amount of memory, why not go
> >for a dumb Ethernet board and a dual-CPU main processor?  That's probably
> >a better use of the second CPU.

Distinguish also between the theoretical and the here-and-now. All of the
current smart cards use 80186 or (slow) 68000 processors. This is
a sufficiently price-sensitive area that it would be difficult to
market a really powerful design except for server applications.
Meanwhile almost all AT class CPUs are significantly faster
than the smart card processor.
> 
>   Hugh LaMaster, m/s 233-9,  UUCP {seismo,topaz,lll-crg,ucbvax}!

-- 
"You want a disclaimer form? Next window, please..."

Geoff Arnold, Sun Microsystems East Coast Division (home of PC-NFS)
UUCP: {ihnp4,decwrl,...}!sun!garnold  ARPA: garnold@sun.com