a series of notes on network evolution -- worth reading (start at the end)

[David Farber <dave () farber net>]

Begin forwarded message:

From: "David P. Reed" <dpreed () reed com>
Date: September 12, 2009 5:48:01 PM EDT
To: Richard Bennett <richard () bennett com>
Cc: Dan Lynch <dan () lynch com>, Dave Crocker <dcrocker () bbiw net>, Dave  
Farber <dave () farber net>, ip <ip () v2 listbox com>, Gordon Peterson <gep2 () terabites com 
>, John Shoch <shoch () alloyventures com>, Harold Burstyn <burstynh () iname com 
>, Lauren Weinstein <lauren () vortex com>, Paul Robichaux <paul () robichaux net 
>, Steve Crocker <steve () shinkuro com>
Subject: Re: [IP] "the net"

Regarding my having my facts straight:

I just walked over to the shelf to the left of my computer, picked up  
my available paper copy of the ARPANET PROTOCOL HANDBOOK, NIC 7104,  
REV. January 1978.  On page 41, there is a document numbered NIC 29628  
(Dec. 13, 1977).  It is entitled "Transmission Control Protocol  
(TCP)", and it was written by Jon Postel.  The relevant fact-based  
paragraph (paragraph 2) is as follows (my italics for emphasis):

"TCP /is being used/ in the ARPA-sponsored work on packet switched  
radio communication (PRNET) and packet switched satellite  
communication (SATNET), and interconnections of these experimental  
facilities with each other and the ARPANET.  TCP implementations / 
exist/ for PDP10 TENEX systems and PDP-11 systems, and TCP  
implementations are under development for TOPS20, Multics, the  
Norwegian NORD-10 system, and UCLA-CCN's IBM-360/91. Some hosts in the  
MIT LCSNET will intercommunicate with hosts in the ARPANET using TCP."

Now I am very familiar with this, because I was pretty much the core  
software and design staff of "MIT LCSNET" at the time.   As Dave  
Farber knows, we were working (in collaboration with him) on a version  
of the token ring idea as an alternative to the "bus"-style coaxial  
Ethernet.   And all of this stuff was up and running in prototype  
form, and was interconnected with the ARPANET by gateways that did NOT  
translate to NCP, but instead overlaid NCP by using point-to-point  
nailed up virtual circuits.   We also shortly thereafter had in 545  
Tech Square and over dedicated microwave links across campus (thanks  
to Tom Knight and buddies) what was called "the ChaosNet" which again  
served as a layer 2 for TCP transport.  Chaosnet was a bus network  
somewhat like Ethernet.   Chaosnet carried its own private layer 2  
protocol, and also allowed TCP packets to be transported.  Finally, we  
acquired in early 1981 a complete Xerox Alto 4 Mb/s ethernet network  
that ran PUP and eventually a TCP implementation by Dave Clark, again  
before 1983.

The idea that TCP was "developed to replace NCP" on the ARPANET is  
just wrong.  Completely, utterly and foolishly wrong.





On 09/12/2009 05:06 PM, Richard Bennett wrote:
> David, you don't quite have your facts straight. TCP didn't run *on  
> top* of NCP on the ARPANET, it was a replacement for NCP. NCP was  
> the early host-host protocol for ARPANET, a transport layer, not a  
> network layer. The Host-IMP protocol and IMP-IMP protocols were  
> outside the scope of NCP.
>
> It turns out that BBN's IMP-IMP code had a datagram mode in addition  
> to the more commonly used connection mode, as did ARPANET's progeny,  
> X.25. I wonder if the implementation of IP over ARPANET actually  
> used it. The ex-BBNers don't seem to know.
>
> RB
>
> David P. Reed wrote:
> > I cannot begin to understand why the "flag day" was important as  
> > anything other than a symbol.  Everyone had already made the  
> > transition earlier, and there was no dramatic "cutover".  The  
> > reason is simple: TCP ran on *top* of NCP when it traversed the  
> > ARPANET.  And by this time, the NCP network may have been very  
> > important to those who were still connected only via IMPs, but  
> > guess what?  Many, many computers were connected by ethernets, etc.  
> > which NEVER ran NCP. Ever.  They used TCP over 802, based on ARP,  
> > etc.
> >
> > So Richard, you imagine a world that didn't actually exist.  In the  
> > early 1980's I was managing the Dover printers, which talked to ITS  
> > machines and so forth, and to SAIL, and to TENEX machines around  
> > the country using TCP, etc.   The ITS and TENEX and Multics systems  
> > had implementations of file sharing based on private file sharing  
> > protocols that ran on TCP, providing a internet-wide file system -  
> > and the only role NCP played was as one of many "layer 2" transport  
> > subnets on some of the paths that had been built in the ARPANET.   
> > At that time there were many cross-country links that were over non- 
> > NCP paths.
> >
> > So it was only the *laggards* who were somehow lost in the past who  
> > were still using Telnet over NCP by 1983.
> >
> > The result of looking for an imaginary "cutover" in the flag day of  
> > the time is to completely distort history.  It's  a nice date, and  
> > some work was done, but no one noticed much, because by then NCP  
> > was pretty much unimportant.
> >
> > On 09/11/2009 05:16 PM, Richard Bennett wrote:
> > > OK, same apps modulo the Small Matter of Programming to use the  
> > > sockets API instead of whatever facility NCP provided. I suppose  
> > > the point is that from the end user's perspective, nothing  
> > > happened on New Year's Day 1983 except a few programmers missed  
> > > some bowl games as a consequence of being otherwise engaged and  
> > > telnet got a little more sluggish. Seems like ftp should have been  
> > > faster, however.
> > >
> > > Dan Lynch wrote:
> > > > Oh Richard, you jest.  I was sorta responsible for getting all the
> > > > applications converted to run on TCP/IP across the research  
> > > > world. That was
> > > > a generally unfunded activity, but it had to be done anyway.   
> > > > Telnet, FTP,
> > > > various email systems (varying a lot at the client level, but a  
> > > > single
> > > > protocol (SMTP) at the server level) were all rewritten  
> > > > extensively to run
> > > > on TCP/IP.  The user saw little performance change.  Making  
> > > > Telnet just as
> > > > responsive as before was challenging due to buffer management  
> > > > differences in
> > > > NCP vs.TCP.   Lots of other details I have forgotten in 27 years...
> > > > Dan
> > > >
> > > >
> > > >
> > > > On 9/11/09 1:43 PM, "Richard Bennett" <richard () bennett com> wrote:
> > > >
> > > >
> > > > > I'm curious about something, being a latecomer to The 'Net  
> > > > > (1986). After
> > > > > the Flag Day transition, users continued to run the same ftp,  
> > > > > telnet,
> > > > > email applications as before, and the BBN network continued to  
> > > > > operate
> > > > > as before, with the same IMP-IMP and IMP-Host protocols as ever.  
> > > > > Was
> > > > > there any perceptible difference to the end user from the  
> > > > > transition
> > > > > from NCP to TCP? Did the apps run any faster, was telnet more
> > > > > responsive, did file transfers finish quicker?
> > > > >
> > > > > I can imagine that the pipelining in TCP might have sped things  
> > > > > up a
> > > > > bit, but can also see that it might not have had much effect as  
> > > > > the
> > > > > actual transit network was still the same ole connection- 
> > > > > oriented system
> > > > > as before.
> > > > >
> > > > > RB
> > > > >
> > > > > Dave CROCKER wrote:
> > > > >
> > > > > > Dan,
> > > > > >
> > > > > > Thanks for reminding us of the earlier name.
> > > > > >
> > > > > > It occurs to me that, in fact, the Arpanet was at the heart of an
> > > > > > actual small-i internet significantly before the introduction of
> > > > > > TCP/IP into production operation and the January 1983  
> > > > > > switchover to
> > > > > > using it.
> > > > > >
> > > > > > In the 70s and very early 80's Xerox PARC, Berkeley and CSNet  
> > > > > > were all
> > > > > > providing email gatewaying -- that is, relaying amongst  
> > > > > > heterogeneous
> > > > > > systems -- to accomplish a seamless email internet.  Someone on a
> > > > > > CSNet Phonenet site could exchange email with someone on the  
> > > > > > PARC XNS
> > > > > > service quite handily, via the Arpanet, with no TCP/IP anywhere  
> > > > > > in view.b
> > > > > >
> > > > > > Not only did this create a perception of using a single,  
> > > > > > integrated
> > > > > > service, I claim it relied on the technical underpinnings of  
> > > > > > the same
> > > > > > kind of architectural overlay that was the essence of the IP  
> > > > > > model for
> > > > > > internetworking.
> > > > > >
> > > > > > This complexity and subtlety of service variations was why I  
> > > > > > later
> > > > > > wrote RFC 1775, To Be "On" the Internet.
> > > > > >
> > > > > > d/
> > > > > >
> > > > > > Dan Lynch wrote:
> > > > > >
> > > > > > > I do believe that it was even called the Catanet at the  
> > > > > > > beginning as in
> > > > > > > conCatenation of networks when we thought of a single network  
> > > > > > > as a
> > > > > > > set of
> > > > > > > data links.
> > > > > > ...
> > > > > >
> > > > > > > On 9/11/09 7:46 AM, "Dave Crocker" <dcrocker () bbiw net> wrote:
> > > > > > >
> > > > > > > > David Farber wrote:
> > > > > > > >
> > > > > > > > > From: Steve Crocker <steve () shinkuro com>
> > > > > > > > >        At the upper layers,
> > > > > > > > > the user community, the open protocol architecture and the
> > > > > > > > > expansion of
> > > > > > > > > applications continued more or less continuously.  From this
> > > > > > > > > perspective, the Arpanet and Internet are part of a  
> > > > > > > > > continuous arc.
> > > > > > > > It's worth stressing that this continuity is a technical  
> > > > > > > > reality,
> > > > > > > > not just a
> > > > > > > > (possibly indelicate) user perception.
> > > > > > > >
> > > > > > > > The same Telnet, FTP and email services have been in  
> > > > > > > > operation since
> > > > > > > > the early
> > > > > > > > '70s.  Calling it "the net" is a (possibly delicate) way to  
> > > > > > > > assuage the
> > > > > > > > sensitivities some of us might have about basic differences  
> > > > > > > > in the
> > > > > > > > lower
> > > > > > > > layers. It lets us, as speakers, reserve "Internet" for its
> > > > > > > > lower-layer precision.
> > > > > > > >
> > > > > > > > But we probably should not get any more upset about having this
> > > > > > > > continuously
> > > > > > > > available user environment popularly be called "The Internet"  
> > > > > > > > than
> > > > > > > > we are in
> > > > > > > > having an internet packet relaying device now be called a  
> > > > > > > > router,
> > > > > > > > rather than
> > > > > > > > its original "gateway".
> > > > > > > >
> > > > > > > > Highly precise historical distinctions have their place, but  
> > > > > > > > the
> > > > > > > > fact that
> > > > > > > > popular discourse finds them an excessive burden does not
> > > > > > > > automatically make
> > > > > > > > that discourse incorrect.
> > > >
> > > >
> > > >
> > > > Tel. 707-967-0203   Cell  650-776-7313
> > > > My assistant is Dori Kirk   Tel. 707-255-7094  dori () lynch com
> > >
> > > --
> > > Richard Bennett
> > > Research Fellow
> > > Information Technology and Innovation Foundation
> > > Washington, DC
>
> --
> Richard Bennett
> Research Fellow
> Information Technology and Innovation Foundation
> Washington, DC




-------------------------------------------
Archives: https://www.listbox.com/member/archive/247/=now
RSS Feed: https://www.listbox.com/member/archive/rss/247/
Powered by Listbox: http://www.listbox.com