[e2e] Nature mag, DARPA, and the Internet

Cannara cannara at attglobal.net
Sat Aug 9 21:54:15 PDT 2003

Noel, I'm not blaming Arpa for the byte orientation in TCP, nor am I saying
IMPs were byte oriented, nor that the first Inet hosts were Unix boxes, nor am
I saying that Englebart's ideas from the '50s and '60s were not endorsed or
taken seriously by people like Licklider, because I was there too.  What is
true is that ARPA did not initiate, nor fund full development of, everything
that some may see now as comprising The Internet.  Much of what did get funded
we owe not just to Licklider, but to Kahn, who did years of political heavy
lifting.  Most of the really important stuff in computing and networking was
done without ARPA funding.

As Dave Crocker points out, the IMPs were the routing interfaces to hosts,
which created its own set of problems for the TCP/IP world, such as hiding the
concept of a host DLC address, so we ended up having to do nasties like ARP &
NAT because of an inadequate network-level address space and concept of a node
address.  People decry NAT and other kludges brought on by odd IP-level
decisions, but these are quite logical responses for folks who have real-world
business to do and now have only a protocol system with limited capabilities
to do it with.  Of course we won't even get into Inet insecurity and VPNs,

Terminal-to-host interactions, however, were the dominant mode of activity for
users, even if they initiated FTP file activity among hosts, the basic point
was that few folks had workstations before the mid '80s, so very few actually
did anything but initiate applications at the hosts they worked with, perhaps
miles away.  Even now, with remote workgroups, Telnet, in all its 64B per
character beauty (ok, maybe someone has set the timer well and a couple of
characters make it into a pkt), is common, inefficient traffic on business
nets.  Of course, browser-based applications are now rampant,  and we know
where they did not come from.

The fact that Telnet, etc. were byte-oriented tools, in terms of the PDUs
generated below them, obviously has many reasons, but the use of byte counting
within data PDUs for windowing was, and still is, not only unnecessary, but
inefficient and limiting in high-speed, large-packet networks serving the fast
end points we have today (sorry Vernon).  Having taught TCP/IP and other
protocols for some years and having had to solve over 1000 company's network
problems, often with TCP performance, I'm somewhat qualified to comment on the
mistake of byte orientation.  Yes, you need a byte pointer in a transport pkt,
no, you don't need byte-oriented windowing, sequencing and retransmission
control, with all the wasted bits they imply -- the Xerox, DEC, Banyan,
Novell, 3Com, yadda, yadda, protocols did not waste that space.

It's interesting that DEC LAT was mentioned, since LAT was far more efficient
than Telnet in servicing multiple end nodes.  Some concepts in LAT, as well as
other non-Inet protocols were largely ignored by Inet folks over the years, to
our ongoing detriment.  However, before The Internet, as we now live it, major
companies relied on vendor protocols, like DECnet, Vines, Netware, etc.  And,
DEC LAT, even when misused (as a WAT, popping the 80mS retrans timer), was
extremely good at allowing businesses to efficiently manage terminal-host

The bottom line is that even though most designers of what we have as Inet
protocols didn't see them as complete, finished, or optimal, we have a
bureaucratic attitude that has cemented us into a world of little protocol
development, which limits improvement and is often circumscribed in a fashion
summarized by:  "xxx friendly", etc.


PS "illerati" may be Latin, but "illiterati" is what we're aiming at. 

"J. Noel Chiappa" wrote:
>     > From: Cannara <cannara at attglobal.net>
>     > Well, there was really nothing relating the Arpanet gang and the folks,
>     > like Englebart, who invented mice, hypertext and collaborative computer
>     > workgroups
> I'm not quite sure of how the mental parenthesizing here went (perhaps you
> meant the "relating" to refer between all the above and the stuff just below),
> but Licklider definitely saw a relationship between the ARPAnet stuff and
> Englebart's stuff, while he was at ARPA and funded them.
>     > and the foundation of corporate, local networks, workstations, shared
>     > server-based printing and other behaviors in the '70s at Xerox. It's
>     > particularly funny that any link would be attempted, since the Arpanet
>     > and all the protocols mentioned were limited to a byte orientation
>     > because of what the phone co (AT&T) did in Unix and how the Internet
>     > began as a terminal-to-host system.
> I'm completely unable to comprehend, through all the numerous errors, what the
> point is here. The ARPAnet was *not* byte-oriented, and all the original Xerox
> protocols (PUP, etc) were! Did you mean that they didn't have the semantics
> for higher-level objects? And since Unix was very much a side-line item when
> TCP was developed (I remember, I was *there*, as were several other people on
> this list), you can't blame Unix for the byte (or low-level, if that's what
> you really meant) orientation of TCP.
> Sheesh. Talk about the illerati not knowing our history - it seems like a lot
> of *us* don't know our own history.
>         Noel

More information about the end2end-interest mailing list