[ih] [addendum, lost email of Jack Haverty, for the record] Re: Baran and arbitrary reliability from arbitrarily unreliable components

Matthias Bärwolff mbaer at cs.tu-berlin.de
Tue Feb 16 14:31:44 PST 2010


I have just realized that a previous email from Jack Haverty in a March
2009 thread on this list did not actually make it to the list, but only
to me -- god knows why. here it is, for the record:

Date: 	Thursday, March 19, 2009 9:08:26 PM 
Subject: 	Re: [ih] Baran and arbitrary reliability from arbitrarily unreliable 
components


	

Hmmm.  I tried to send this yesterday, but since it hasn't come back to me yet 
it must have been drawn into one of those Internet black holes.  Apologies if 
you get it twice.... /Jack

=======================================================

Greetings everyone.  Good to hear voices from the past, and it's a great 
discussion -- made me feel guilty enough to sit down and type in my 2 
cents..maybe making amends just a little bit for all those papers that never got 
written back in the 70s/80s.  If you're not interested, just hit delete now....

Matthias' original question concerned whether and to what extent prior academic 
work influenced the Internet.  That got me thinking about the question more 
generally - namely what *did* influence the zillions of decisions made by people 
building the Internet.  I've always been accused of being a "big picture" kind 
of guy...

I agree wholeheartedly with the previous comments about the difficulty of 
determining whether or not someone's work or publication influenced someone 
else.   Even if you could somehow prove that someone subscribed to a journal, or 
actually read a paper, or even understood it, he or she may not have agreed with 
it - and simply ignored it as noise.  I'm equally suspect of anyone who claims 
to have invented any particular mechanism or approach.   It's simply too hard, 
in my experience at least, to be sure that an idea that pops into your head is 
original, and not a product of your subconscious working on something you saw, 
or heard, or read, days or months before.

Still, it's interesting to try to deduce where things have come from, especially 
things as pervasive, influential, and surprising as "The Internet".  So here goes...

I think the first issue to nail down is exactly what you mean by "The 
Internet".  In my experience, it means different things to different people, 
depending on facts like how and when you first encountered "it", what you were 
doing at the time, and how long it took you to recognize it as something new and 
important to you.  For some people, the Internet is email.  Others equate it 
with the Web.  To some it began with the Arpanet, which was "Release 1".   Or it 
may have begin with CSNET and/or NSFNET, which enabled a lot of people to 
"connect" for the first time.   Did any particular piece of prior work influence 
those efforts?  I think you'd have to ask the people involved in building them.

So, although it's difficult to determine what influenced someone's decisions, 
it's perhaps more reasonable to ask what influenced your own decisions.  It's 
not so easy to answer that question, but at least you're asking an "eye 
witness".  So, since I was involved in various aspects of what many people might 
call "The Internet" for much of the late 70s and 80s, I've been asking myself 
that question.  What influenced me?

First, the definition, or better put, my own definition -- The Internet.   I 
think I personally focus on the "Inter" part of "Internet".  There have been 
many networks over the years.   In the 70s/80s, there were a variety of computer 
networks.  Mujltidrop lines and terminal networks had been around for a while; I 
remember using one around the late 60s, which connected MIT to Xerox.  IBM had 
SNA.   ARPA had ARPANet.  Other technologies were being created to form networks 
of computers, e.g., the "Packet Radio" networks for tactical military use.  
Commercial X.25 networks were seeking customers for the telephone companies of 
the world.   Tymnet, Compuserve, and others were connecting users to applications.

All of those networks used a variety of technologies, but had some common 
characteristics:  each was composed of uniform components (no SNA nodes in the 
ARPANet, for example); each was owned and operated by a single owner; and except 
for very special ad-hoc linkages, each was isolated from the others.

Where the ARPANet focused on connecting dissimilar computers, my personal 
definition of the Internet focuses on "Inter"connection of dissimilar "net"works. 

In particular, ARPA had this problem of building different kinds of networks, 
with a need for computers on one network to talk to ones on the other.  As far 
as I remember, at least in the late 70s (when I first encountered "The 
Internet") there wasn't any project specifically to "build the Internet".  
Instead, there were a variety of projects to build different kinds of networks, 
and applications that would run over those networks, and every project had the 
requirement to interconnect with the others to make those applications 
possible.  There were a variety of scenarios that had to work.  E.G., someone in 
a helicopter, obviously on a radio-based network, interacting with someone on 
another continent at headquarters, obviously not on the same radio network.  The 
aggregate of those projects' teams were the de facto Internet team - which in 
fact started as two separate teams: the "TCP" and the "IP" working groups, but 
was pretty quickly merged.

The obvious determinant of "dissimilar" networks  was that they used different 
technologies.  ARPANet, SATNET, PRNet, etc. all had their own unique hardware, 
software, algorithms, and architectures.   I think at first that's how we 
thought of the problem.  TCP/IP was the "glue" that allowed communication to be 
accomplished through a series of dissimilar steps.

But there was another, and I believe in retrospect even more important, nature 
of being "dissimilar", namely organizational.  In the 70s, I think networks were 
pretty consistently homogeneously managed.  I'm sure someone will correct me if 
that's wrong...  But as I remember, networks were owned by someone, either a 
government agency, or a company, or a utility like a PTT.  It was also likely 
that all of the components were the same, or at least from the same vendor.   
Even the ARPANet, which interconnected dissimilar computers, was built and 
operated by one company.  Someone was in charge - not zero, not two or more; 
some ONE.   That was simply the state of the art at the time.

In retrospect from 2009, the Internet is a bonanza of heterogeneity.  No one 
owns it, and it's not clear why it works or whose hand is on the wheel.   
Virtually anyone can buy some hardware and become an ISP, hooking yet more 
things into "The Internet".

Since that aspect of organizational dissimilarity wasn't prominent in the very 
early days of interconnecting ARPANet with PRNet, I started wondering how did we 
get from there to here.  As was pointed out previously, CSNET and NSFNET 
arguably led the way in making the Internet organizationally heterogeneous.  But 
what made it possible for CSNET and NSFNET to do that...?

If I remember correctly, at the time the ARPANet had been very successful, and 
had become more of an operational communications utility than a research 
project.  Lots of organizations and universities wanted to connect, but couldn't 
because either it was very expensive (IMPs, IMP ports and circuits cost $$s), or 
because they weren't ARPA contractors and the ARPANet was restricted to support 
ARPA business.   There was a pent-up demand to "get on the net".

Going back to my earlier comment above, I've been thinking about what influenced 
me personally at that time in the work I was doing.  I was at BBN, managing a 
bunch of projects which were arguably part of "the Internet", for DARPA as well 
as other government agencies, all involved with either building parts of "the 
Internet", or deploying clones of it into other organizations (Yes, there are 
"Internets").  One of the DARPA projects was the building and operating of the 
"core gateways", which were the ones which interconnected various long-haul 
networks among others.   So I think it's reasonable to consider whatever was 
influencing me in working on those projects was influencing "the Internet" too.

At the time, there was a lot of pressure to deploy a functional Internet - one 
which could support successful demonstrations of the kinds of mostly 
government-oriented applications that would cause the "operational" government 
gang to keep the research funding going and the funds going to ARPA (and then to 
us and others).    That focus, coupled with the fact that BBN was not a 
university, led to a noticeable bias toward engineering rather than science.  
Getting it to work, using proven techniques rather than academic ideas, became 
the primary goal.   Writing papers, presenting at conferences, trying new 
interesting approaches, and other such science-oriented activities got pushed to 
the back burner.  As I remember also, all through that period most of the 
interesting discussion and argument happened on the various mailing lists, or at 
the quarterly Internet meetings, rather than in the traditional journals -- at 
least from my personal perspective at the time.  Sadly, I suspect all of that 
has been lost.

The focus on engineering and using proven techniques to get things to work led 
naturally to what I think was the primary influence on the Internet -- the 
ARPANet, which was arguably "working" well, and more specifically the *internal* 
unpublicized mechanisms of the ARPANet which had evolved over a decade of its 
operating experience.  Although there was considerable material in publication 
about the more mathematical aspects of the ARPANet, especially the routing 
algorithms, there was a rich soup of engineering "best practices" embedded in 
the IMP code, the NOC tools, policies, and procedures, and the management 
approaches.  If I remember correctly, there were something like a thousand 
separate parameters in the code that could be "tweaked" to deal with different 
situations that had come up over time.  As far as I know, little if any of this 
was ever documented in any detail in any public way.  It was pretty boring after 
all.  Some of it was simply to be experienced - standing as a fly on the wall of 
the NOC during some network crisis was very educational.

The "Internet guys" resided at BBN literally just around the corner from the 
"ARPANet group".   This was purposeful, and I fought several battles over the 
years within the BBN management to keep it that way.  As a result, a lot of the 
internal decisions associated with the "core gateways" were to steal, adapt, and 
integrate ARPANet mechanisms and philosophies into the nascent Internet.  This 
wasn't really intentional, it just happened because of the proximity - 
"technology transfer" at its best.

That ARPANet influence would naturally have evolved the core gateways and "the 
Internet" into a homogeneous network like the ARPANet.   All gateways would be 
built by BBN, all software running the same release (and obviously the same 
algorithms etc.), all operated by the same single centralized entity.

Speaking of "inventing" things, I might feel inclined to claim responsibility, 
or blame, for causing a mutation of the Internet.  Originally, the idea was that 
the Internet interconnected networks, and the boxes that accomplished that were 
called "gateways".   I pushed the notion that a wire, i.e., a leased line or 
virtual circuit, was simply a very degenerate type of network, which could 
connect to only two "hosts" -- "this end" and "that end".   So, if you put a 
wire from a port on one gateway to a port on another, and treated it as just a 
very simple "network", the overall Internet would still work just fine.  I must 
have given the slideshow presentation about this concept at least a hundred 
times.  Of course it meant that you would no longer need those ARPANet IMPs to 
plug your gateway into.  You could just unplug the long-haul leased line from 
the IMP port, plug it into the gateway port instead, and the Internet would 
simply take over the task of routing traffic across that line.  We actually did 
exactly that, to make sure it worked.   At that point the boxes we called 
"gateways" became "routers" and took on the role of switching.  I don't think 
the BBN management ever understood the implications of that concept on future 
sales of packet switches, despite my slideshow efforts.  Of course, at the time, 
the "conventional wisdom" was that the Internet was just a research project, and 
the "real" system was going to come from the PTTs, CCITT, ISO, and other 
professionals, with X.25 and X.75 playing the lead role to upstage TCP and IP.

Anyway, that environment would have led to an Internet that looked a lot like 
the ARPANet, probably ending up as a homogeneous network of routers 
interconnecting mostly Ethernets, thanks to Bob Metcalfe.   There was a lot of 
momentum along that trajectory.  But Bob Kahn stepped in the way....

This is one case where I remember quite clearly how the Internet was 
"influenced" to change directions radically.  Again, all of this is from my 
personal perspective, with my personal definition of "the Internet".   Others 
undoubtedly had different experiences.

I was at one of innumerable meetings.  Sorry, I can't remember where or when.  
It was probably in DC, where I spent a lot of time, but my gut feeling tells me 
it was the European Internet meeting, maybe in Munich.  Anyway,... Bob and I 
were hanging on the same subway strap, with the usual group of a dozen or two 
people heading out to find dinner.  Bob wanted to talk about the Internet 
architecture, and in particular the core gateways.  He managed over the 
squealing of the car's wheels to overcome my skepticism and make it clear that 
it would be a good idea to figure out how to make it possible for gateways not 
built by BBN to be full participants in the system of gateways.  I don't know 
whether this was motivated by political pressures to enable CSNET/NSFNET, or 
some technical considerations, or by the ARPA charter to focus on new technology 
and new ideas, rather than replicating the old ones.  But he convinced me, and I 
went away with a new direction, and a harder task to make something work using 
an unproven approach.

Back at BBN, the challenge was not only to figure out how to make a stable 
heterogeneous Internet, but also how to convince the people on the project that 
it was a good idea to let other people build gateways and hook them up to "our" 
system.   Fortunately the meetings of the TCP and IP working groups were great 
training for this kind of work.   I recruited one of the best thinkers from the 
ARPANet crowd - Dr. Eric Rosen.  He and I sat down for several multi-hour 
brainstorming sessions, and came up with the notion of "autonomous systems", 
which were sets of routers owned/managed by a single organization, and 
interconnected with other such systems to form the overall Internet.  EGP (which 
I think evolved into BGP) and the concept of IGP (which basically means whatever 
mechanisms are used among the routers inside their own closed system)  made it 
possible to use different approaches within different ASes.   This led to RFC 
827 and a bunch of others in the early 80s.

To me, this is what diverted the Internet from homogeneous to heterogeneous in 
nature and enable the "Inter"connection of different organizations -- even if 
they used the same network products, they retained control over their own 
assets.  In retrospect, given the experiences I later had in various commercial 
"user" situations, this characteristic was crucial to making the Internet 
reliable and successful.

Of course, this wasn't a new idea.  Mainframes and timesharing were succumbing 
to the onslaught of PCs - which allowed you to own and control your own 
computing resources, rather than being at the mercy of the clowns in the "Data 
Center".   People like to control what's important to them.

Bob Kahn had it just right.  Maybe that subway strap should be bronzed and put 
in the Smithsonian.  Of course, adding in all of Vint's "subway straps" would 
require another wing.

These have been just a few examples, of the influences I felt while doing what I 
was doing to help build what I think of as the Internet.  Whew, that should be 
enough caveats.

Getting back to Matthias' original question...  I'd have to say that Pouzin's or 
Baran's ideas and papers, or anyone else's papers for that matter, didn't have 
much direct influence on the Internet mechanisms that I was associated with.  
The culture wasn't very scientific/academic oriented; magazines like Data 
Communications were popular, and maybe some IEEE journals, but they were 
peripheral.  The science didn't apply very readily to the Internet environment 
where nice well-understood, easily modelled wires were replaced by a gaggle of 
networks with unknown, unpredictable, and variable behavior.   So, the design 
and implementation decisions were made by the practical needs of trying to make 
something that worked, even if we couldn't explain why.  It was very much an 
experimental, engineering kind of world - watch what's happening, try something 
that seems likely to help, repeat until done; this was largely what had been 
going on in the ARPANet for a while at that point to refine the basic mechanisms 
of packet switching.   The ARPANet was probably the biggest influence on the 
Internet at that time (again, from my personal perspective).   To the extent 
that the ARPANet mechanisms were influenced by earlier academic work, that 
influence may have flowed forward to the Internet.  You'd have to ask the 
ARPANet guys.

I think it's also true that a big influence on decisions is simply the personal 
experience of the people involved.  In my youth I had a model train set, and 
learned something about how railroads worked.  If you think of the movement of a 
commodity like coal, or wheat, or logs, there's lots of analogies to "packet 
switching".  Logs or lumps of coal are like bits, cars ("wagons" in Europe) are 
like packets, freight yards are like routers.   In the late nineteenth and early 
twentieth centuries, railroads built by different companies used different sizes 
of track, so that cargo had to be moved from one car to another at their 
junctions - not unlike what a "gateway" had to do?  Railroads encountered all 
sorts of problems, many with blatant relationships to our Internet world 
(congestion control comes to mind), and they developed elaborate technologies to 
deal with it, such as a series of signalling systems and "routing protocols" - 
all well before Eniac (or pick your favorite "first computer").

With personal experiences like that, what you know from those "technology 
architectures" can easily influence decisions made in a new situation.  In the 
Internet meetings in the early 80s, concepts like "expressway routing" were 
discussed at length, and people pretty much knew the issues from their human 
experiences with the highway systems.  Their experiences must have influenced 
their Internet work too, probably more than just adopting highway terminology.

So, if the notion of "packet switching" has a long history, how do you tell 
whether you're influenced by a recent example, or one from centuries ago.  Roman 
generals used to dispatch several runners with copies of a message to assure 
that their reports got back to Rome.  They used to split messages up and send 
pieces by different messengers, over different routes, to avoid interception by 
the enemy.   I don't think they called it "fragmentation and reassembly" but 
it's the same idea.  I knew this history from high school -- did it influence me 
in my later work on the Internet?  Did it influence Pouzin and others?

Enough philosophy, and enough words.  If you've gotten this far, thanks for 
listening, and I hope I've helped answer the original question at least a little 
bit.

To me, the history of the Internet is the aggregate of all of these kinds of 
anecdotes by all the people involved in whatever they individually think of as 
"the Internet".  Unlike other huge projects, like putting a man on the moon, 
there was no "Internet project", and no crisp milestones -- like when it 
started, or when it was completed.  Are we done yet?  There weren't a whole crew 
of journalists documenting it along the way, since it was supposed to be 
ephemeral and wasn't considered important enough to warrant the attention.  But 
it's a fascinating story.

Just for completeness, my personal view of the Internet was driven by my 
involvement, so here's a brief summary for the record.  In the mid 70s, I was a 
user of the ARPANet, building various applications that ran over the net - in 
particular email.  From 1978 through to 1990 I was at BBN, doing a variety of 
things over time, almost all somehow associated with the Internet.  I started by 
building TCP for Unix, ended up managing a gaggle of research projects, and 
later created a "system engineering" organization to help real users (government 
and early commercial adopters) deploy and use their own Internets.  In 1990 I 
went west to Oracle as "Internet Architect", which among other things involved 
getting the corporate internet up and running, getting products to be 
Internet-capable, and helping Oracle's customers (which is just about everybody) 
understand the new-fangled Internet stuff and get it to work, since none of the 
Internet industry's products seemed to come with a "How to Run and Manage an 
Internet" or a "How to Use Your New Internet for Fun and Especially Profit"  
pamphlet.  So, my personal perspective on "the Internet" is an amalgam of all 
those situations.

The Internet is like the elephant in those proverbial stories about the elephant 
and the blind man.  What you think it is depends on your perspective as you 
encounter it.

A few years ago, some organization (Smithsonian I believe) ran a project to 
capture as much as possible of the experiences of World War II veterans, in 
order to document the "inside" of the war experience while the opportunity still 
existed.  Perhaps, in its role as organizer of the world's information (which 
continually amazes me), Google might take on such a project for the 
Internet...?   Vint, did you make it this far through my rambling......?

/Jack Haverty
Point Arena, CA
- where we still don't have cell service, and the phones only do 24kb/sec, but 
fortunately there are satellites.....and I'm not on the hot seat at the 
operations center!


-- 
Matthias Bärwolff
www.bärwolff.de




More information about the Internet-history mailing list