[e2e] It's all my fault
day at std.com
Thu May 17 21:07:23 PDT 2007
At 20:55 -0400 2007/05/17, Noel Chiappa wrote:
> > From: Vadim Antonov <avg at kotovnik.com>
> > There was no market for IP routers... why?
>Because in the beginning (late 70's) there was no Internet to connect to.
>There's no economic incentive for someone to go out and buy/install/use a
>wide-area networking protocol when there's no wide-area network to connect
>to. Until the government jump-started it by setting up the initial Internet,
>there was no movement at all to an inter-organization network.
>Local-area networking protocols did get designed and sold (e.g. Banyan,
>Novell) for intra-organization use, but they weren't designed for
>inter-organization use (as TCP/IP was, from the start - albeit not as well as
>it ought to have been).
>In fact, what private industry was doing at that point in time was 180
>degrees out of phase with the goal of ubiquitous inter-organizational
>networking, actually. Every computer company (IBM, DEC, etc) had their own
>protocol family (SNA, DECNet, etc), and their idea was to lock their
>customers in to their (incompatible) protocol family. Letting everyone talk
>to everyone was absolutely not in their minds. Not only that, they didn't
>even want everything speaking over an industry standard protocol (so that you
>could easily add in boxes from vendor B, in a company which heretofore had
>all boxes from company A).
>Which is not to say we might not have gotten there eventually. But the
>governments intervention definitely sped things up greatly.
There were several companies I could name that wanted to and built
prototype routers in the early 80s, but their marketing dept "knew"
there was no market for them and wouldn't even try to sell them.
Cisco got it going because there was enough critical mass of Internet
nodes in the Bay Area that weren't just universities that they had a
100 customers before they needed to go for funding. Several groups
saw the market coming and knew it was there, but as usual marketing
only 20-20 hindsight and 1-1 foresight.
> > the principle of storing and forwarding chunks of discrete data was in
> > commercial use long before Baran's work - it was common since Victorian
> > times in telegraph networks.
>What about the postal system? That stores and forward discrete chunks too.
>But I reject those predecessors as being that meaningful because what was
>unique about Baran's work was that he came up with the idea of breaking up
>user's messages into smaller pieces, and forwarding the pieces independently
>- something nobody before him had thought of.
You know I am still not sure of this. I have read Baran's reports
and I can't tell if he is describing packet switching as in the
ARPANet or packet switching as in the CYCLADES. Given that
everything Baran was involved in after the reports is more the former
than the latter, I am tending to the conclusion that Baran and Davies
independently invented packet switching. (NPLnet was definitely more
like ARPANet.) But the kind of connectionless networking and clean
separation between Network and Transport seems to have come from
Pouzin. CYCLADES had a very clean distinction between CIGALE and TS
which the ARPANet did not have. The ARPANet was more like X.25 than
>And if you think it's that obvious, try reading Kleinrock's contemporaneous
>work on queing theory - it's all in terms of complete messages. Lots of great
>ideas are "obvious" in hindsight.
I agree. Kleinrock's thesis is definitely in the "beads-on-a-string"
mindset. Picking up on what I said above, it is also interesting
that everything Larry Roberts did after the ARPANet was more in the
connection-oriented packet switching mode than the connectionless.
But this kind of transition in thought is common when a paradigm
shift occurs. Some never leave the old paradigm, the very early ones
are usually a foot in both camps because they are still sorting it
out, and then the just early may be in the new model.
> > May it be because there was no Intel with 8086 and no IBM with PCs, and
> > no Bill Gates with Windows, and no Apple?
>Actually, the initial breakout of Internet growth occurred before all those
>events. I started selling IP routers when basically everyone I sold to was
>using time-sharing systems running on DEC hardware.
Noel is right. It was the rise of workstations and LANs, not the PC
that drove it.
>The PC and uSoft drove the later rapid growth, yes, but that was later.
> > In fact, the Internet was impossible without transistors and
> > minicomputers; and their availability is what make Internet possible.
>Yeah, but they are so basic you could just as well make the same statement,
>but replace "transistors and minicomputers" with "electricity and quantum
Agreed. It is interesting if one looks at "data comm" and how SNA
was defined such that it didn't tread on the phone company and the
phone company didn't tread on it. Networking happened when
minicomputers come along and are cheap enough that they can be
dedicated to "non-productive" work, i.e. not running user
applications but just supporting them. This puts computer and phone
companies in direct competition. The trouble was that peer layered
Internet architecture with an end-to-end Transport layer was a major
threat to both "monopolies:" The peer architecture prevented a
transition path for IBM and the Transport layer relegated the phone
companies' core business to a commodity.
> > .. the market for "the Internet" was created not by availability of
> > TCP/IP networks, but by the lowly BBSes, e-mail, and USENET. Neither of
> > which depended on anything developed by DARPA-funded research.
>An interesting list. Let's look at a few in more detail.
>Intra-machine e-mail (the first form to appear) first appeared (as far as we
>can tell) on SDC's Q32 and MIT's CTSS - both created with government funding.
>Inter-machine e-mail appears to have originated with the ARPAnet.
Agreed and I was telecommuting using the Net and email by 1976.
>USENET ran on Unix systems. Unix was heavily influenced by the time Thompson
>and Richie spent working on the Multics project - another DARPA project.
>(Although it turns out the strongest technical influence on Unix was actually
>the earlier CTSS system, but that was also government funded.)
I have always described UNIX as the amount of Multics one could get
on a PDP-11/45. I always thought we should have redone the exercise
for 68000s! ;-)
>Not that it's all government: e.g. the personal computer, the heart of most
>BBS'es, only took off after the creation of the micro-computer, the Intel
>4004 being the first. (Although some BBS's existed before that, running on
>mini-computers.) But to say government had no role is not correct.
If government had not been grossly overprovisioning it and turning a
blind eye to the "abusive" use of its resources, there would never
have been an Internet. I hate to think what would have happened if
at anytime from 1970 to 1990 if some crusading journalist had figured
out all of the non-DoD activities going on the ARPANet/Internet and
done an expose! The things we were doing! The waste of tax payer
dollars! The nice thing about high tech then was they didn't really
> > Claiming the impossibility of the Internet without the contribution of
> > a particular research based on the ideas which were known for over a
> > hundred years and only waited availaibility of technology to get
> > implemented in automated form is, well, stretching the truth very
> > thinly.
>I never said it was "impossibl[e]". I merely said that government funding had
>a role in making it happen, and further (see above) probably made it happen a
>lot faster than it would have otherwise.
I think it nearly was impossible. First, corporations don't do the
research that is as far out as the ARPANet was from showing a benefit
or is as risky. (Actually governments don't any more either for the
most part.) Second, the mind set of corporations would not have
taken the risk on something that risky. Third, you should remember
as I said above, the ARPANet was more like X.25 than IP. But it was
grossly overprovisioned compared to the other experimental networks
that you couldn't see the limitations. (NPL and CYCLADES had 9.6K
lines whereas the ARPANet was 56K.)
>Speaking of "a lot faster", it's interesting to look at history and see how
>technological development/deployment speeds up during wartime - which is, of
>course, all government funding. The US Civil War (which greatly accelerated
>the deployment of telegraphs and railroads), World War I (the aeroplane),
>World War II (modern electronics). Now there's an interesting conjunction:
>war and technological development...
I have another one: intellectual flowerings, the kind of period of
new thinking, happens after mass extinctions. Renaissance comes
after the Black Death, the Enlightenment after the 30 years war, etc.
> > What *is* a well-established fact is that monopoly in every area of
> > human endeavour leads to stagnation.
>I think most of us generally agree with you on that.
> > it is a vastly different network, too. With tons of band-aids and
> > workarounds and ugly hacks needed to keep it running despite
> > short-sighted decisions made by the original designers.
>Again, I think most of us would agree that the original design had some
>significant areas where is was not developed enough. (Why that happened is a
>whole separate discussion.)
Agreed. Basically what we have is an unfinished demo.
>However, the important observation is that some of us think that innovations
>like moving the path-selection out of the network core, and other similar
>things, is a good way to get rid of some of those "band-aids and workarounds
>and ugly hacks" - provided that we can make them economically viable, of
It seems that there is a large group think headed in that direction.
Is it another imitation of the Faber Marching Band? We will see.
More information about the end2end-interest