Re: [Linux-decnet-user] LATD 0.8

[Paul K. <pk...@xe...>]

>>>>> "Patrick" == Patrick Caulfield <pa...@pa...> writes:

 Patrick> On Mon, Jun 26, 2000 at 11:51:52PM +0100, Kenn Humborg
 Patrick> wrote:
 >> On Sun, Jun 25, 2000 at 01:04:38PM +0100, Patrick Caulfield wrote:
 >> > I am pleased to annouce the release of LATD 0.8.  > > This
 >> release is feature-complete for 1.0 so please report any bugs you
 >> find - > the plan is to do a bug-fix 0.9 and then release a 1.0
 >> shortly afterwards > depending on how many problems people find.
 >> 
 >> I've been wondering...  LAT looks like a fairly simple protocol.
 >> The most 'innovative' element seems to be the way it reduce LAN
 >> traffic my aggregating users' keystrokes for sessions going to the
 >> same host (if I understand it correctly).
 >> 
 >> Why were they so insistent on keeping it proprietary?
 >> 
 >> Patrick, have you discovered any deeper meaning to all of this?

 Patrick> Not really. I have been searching the patent archives for
 Patrick> DEC's terminal server entries and one is just a description
 Patrick> of a terminal server in general and one concerns the "slot"
 Patrick> system in use by LAT. Of course it had to be a simple
 Patrick> protocol because there was not enough memory or CPU power in
 Patrick> a DECserver 100 to fit a complicated one into !

 Patrick> I can understand why it was kept proprietary - it was just
 Patrick> the prevailing mood at the time. But the patents make no
 Patrick> more sense to me that any other of the "obvious" patents
 Patrick> that the US office grants so freely :-(

I can see a bunch of points to comment on here.  Take these from one
who was near the origin of all this stuff but not involved first-hand.

LAT started as an advanced development exercise, driven by
dissatisfaction with the high complexity and low performance of
CTERM.  A consideration was the target platform: a device called
"Pluto" (yes, quite a dog...) which was a 19 inch rack, about 15
inches high, with a PDP11/23 processor, an Ethernet NIC, and assorted
other stuff.  Bruce Mann said "there has to be a better way".

Part of why LAT looks the way it does is a religious conviction that
everything should be done in one layer.  I don't agree with that one;
if LAT had run over UDP that would have been a good thing.  Then
again, internet protocols hadn't become the world-conquering force yet
at that time as they are now.

LAT slots are there because a typical terminal server message
(especially server to host) only contains about one or two characters
of payload.  With a multiport server (64 ports was considered a very
sensible target) muxing several input streams into a single packet
significantly improves things, because you get more payload per packet
interrupt.  And packet interrupts are expensive, always have been --
especially back then in the early 1980s.

Patents: I will be the first to agree about junk patents but I don't
think that applies here.  There certainly are more impressive patents,
but I think calling this notion "non-obvious and innovative" is
reasonable enough (again, remember, early 1980s...)

Finally, why is the protocol proprietary?  I have no idea.  Possibly
because that was the norm at the time.  DECnet was quite exceptional
by the rules of the late 1970s, early 1980s, by having open specs and
free permission for anyone to implement.  I think the decision went
"LAT isn't part of DECnet (correct) so we don't have to open it up, so
let's not."

DECservers were very popular and rather profitable little boxes, and I
suspect the decision may have helped DEC's bottom line for a number of
years.  In the long run, it hurt because everyone else moved to
Telnet, even though Telnet is drastically less efficient.

Incidentally, the slot concept of LAT re-materialized in ATM AAL2,
used with voice over ATM.  I don't know if the creators of that spec
realize they copied it, or acknowledged the prior art...

	paul