At 15:37 23/03/2006, John Day wrote:

>I completely agree with Jefsey on this.
>But I do have one question.  If the Internet is the 3rd 
>architecture, what were the other 2?
>ARPANet and CYCLADES?

This post is long. It could be the basis of a Draft if others joined 
to gather and report the passed experiences we could jointly table on.

To be of use I thought necessary to introduce the answer with some 
references, including historic architectural experience. I tried to 
keep it limited to what I think of (key?) interest today. I tell what 
I saw/made. I took some decisions because I was in charge. I was not 
a geek but at a key position. I learned from that position. I only 
try here to pass what I learned, in the hope it may help. Until 1998 
I only known "the Internet" from outside. If I am incorrect or if I 
miss infos, thanks to correct me. History is not an easy task. If 
people who shared into some of this, I would be glad to get their own 
inputs. Thanks.

jfc

Response:

ARPANET and Cyclades were experimental networks which only developed 
locally. ARPANET in the USA and Cyclades at INRIA in France. You all 
know about ARPANET. Cyclades was conducted by Louis Pouzin. John Day 
pointed out some technical influences. From what I understand Louis 
Pouzin hired consulting services from ARPANET when he started the 
project in 1972, hence cross-polinization.

I would like to underline I talked of the "International Network" and 
therefore of architectures of global deployment and operations. In 
that case the protocol becomes potentially global. But this is a 
consequence, not the root of the architecture. What counts first are 
the purpose and the basic concepts. To quote some concepts this 
community is familiar with: network concatenation, addressing, 
naming, routing, end-to-end, multitechnology, etc  But there are 
others this community is not familiar with or has not commonly 
defined yet. This is where passed experience may help illustrating 
them, and understanding if they may serve the future.


Louis Pouzin

When considering a revamp of the network architecture, I think worth 
to spend lines on Louis Pouzin first. I wished to propose him for the 
ISOC award last year. He is the one who matches the best my own 
vision. Many supported the project, but Vint Cerf pointed out that 
his contribution was wider than the Internet, and not oriented 
towards the Internet community (the focus of this ISOC award, Vint 
created)- he proposed some other good ideas to pay Louis the tribute 
he deserves. The lack of knowledge of Louis' founding proposition may 
be our main problem today - missing the very conceptual roots of our 
technology. I first met Louis in 5/78. Vint wrote great things about 
him. We created together Eurolinc for a European Multilingual 
Internet (first Chair was Jean-Louis Grangé of his Cyclades team). He 
was quite active and influent at the WSIS. And an advisor of European 
and International instances.

Louis invented the prompt, the script, the mail (developed by Tom Van 
Vleck), the datagram as it is today (IP windowing). In our today area 
obviously Louis' most important contribution is the "catenet" 
concept. Vint applied it in http://www.isi.edu/in-notes/ien/ien48.txt 
to introduce the  Internet: "THE CATENET MODEL FOR INTERNETWORKING, 
July 1978, The Catenet Model for Internetworking - Introduction - The 
term "catenet" was introduced by L. Pouzin in 1974 in his early paper 
on packet network interconnection [1].  The U.S. DARPA research 
project on this subject has adopted the term to mean roughly "the 
collection of packet networks which are connected together.".


Brian Carpenter

Louis Pouzin's concepts were well understood by Brian Carpenter. He 
wrote a key consideration enlarging Vint's vision. I think important 
to quote it as I consider this text of IETF culture is the nearest 
from my own experience and vision. Even if Brian might dispute it.

"When the catenet concept (a network of networks) was first described 
by Cerf in 1978 [IEN 48] following an earlier suggestion by Pouzin in 
1974 [CATENET], a clear assumption was that a single logical address 
space would cover the whole catenet (or Internet as we now know it). 
This applied not only to the early TCP/IP Internet, but also to the 
Xerox PUP design, the OSI connectionless network design, XNS, and 
numerous other proprietary network architectures.

"This concept had two clear consequences - packets could flow 
essentially unaltered throughout the network, and their source and 
destination addresses could be used as unique labels for the end systems."

"The first of these consequences is not absolute.  In practice 
changes can be made to packets in transit. Some of these are 
reversible at   the destination (such as fragmentation and 
compression). Others may be irreversible (such as changing type of 
service bits or decrementing a hop limit), but do not seriously 
obstruct the end-to-end principle of Section 2.1. However, any change 
made to a packet in transit that requires per-flow state information 
to be kept at an intermediate point would violate the fate-sharing 
aspect of the end-to-end principle."

"The second consequence, using addresses as unique labels, was in a 
sense a side-effect of the catenet concept. However, it was a 
side-effect that came to be highly significant. The uniqueness and 
durability of addresses have been exploited in many ways, in 
particular by incorporating them in transport identifiers".

We discuss the transition to new technologies. I said that an 
architectural transition was the interconnection of a new community 
of users. This text gives many clues from the very "inside" of the 
current internet on how to address this situation. Let me add that, 
from my brainware + user-centric + extended services (network 
services to the content) point of view, I read catenet as "the 
networks of the network of the networks" (cf. conclusion). IMHO this 
is the - long, long delayed - evolution of the architecture we need.


Tymnet

Now, I can address the question. The International Network really 
started a few years before Vint's IEN 48 proposition. Larry Roberts 
had created a commercial network service named Telenet, using BBN 
machines. They had applied to the FCC for a public operator licence. 
To force an FCC decision they threaten to sue those who operated a 
public packet switch service without a licence. There were two of 
them: Tymnet and Uninet.

Tymnet had issued the first packet-switch service bill (we would name 
it an ISP today) on Feb, 28th, 1972. It was the affiliated network of 
Tymshare Inc., the third US timesharing service, with affiliates in 
UK, France, Belgium, Germany and Japan. At that time it offered 
domestic and international access to more than one hundred 
information databases, banks, corporations and specialised services. 
Its CEO, Bill Combs, and Tom O'Rourke, Tymshare Chair, found its was 
a good idea. They also applied for a licence. FCC gave both a 
temporary nationwide Value-Added-Network licence. Tymnet used names 
and addresses: FCC also gave Tymnet responsibility for the namespace. 
Tymnet negotiated interconnections - along with FCC and ITU rules - 
with the US International Record Carriers (first ITT, RCA and WUI) 
and to transfer its foreign users and services to the foreign monopolies.

Bob Tréhin had to "sovereignise parts of an existing network" (1977). 
This was very much like what need to do today after, Tunis and China 
started with Chinese Names, without balkanising the Internet. He was 
assisted by Joe Rinde who designed the Tymnet core system. Bob 
introduced the "root name" principle 
(http://jefsey.com/rootname.pdf). This compartimented the network 
into "externets" ("external network look-alike, possibly within the 
same network, possibly through a virtual gateway" - the term came 
later on, when discussing Internet 'external networks' interconnects 
- RFC 920). When we met Louis Pouzin to hook Cyclades in 5/78, the 
International Network "catenated" already the French, British, 
German, Belgium, Swiss, Italian, Spanish, Dutch public externets. 
They had a local gateways, their own registry, their independent 
management, stats, accounting, etc. an interconnect (through Tymnet 
and the IRCs) to Telenet and Uninet networks. We also supported the 
ESA (European Space Agency) private net. ARPANET was indirectly 
connected in several places, including in Versailles for Ichbiah's 
development of ADA during the next year.

The initial Tymnet I, architecture was replaced on the end of the 70s 
by Tymnet II and ISIS a networked OS offering huge possibilities. 
Tymnet from then on opened every new US international relation and 
most multinational relations until 1986. We operated 90% of the 
International traffic (foreign end) and 100% of the US international 
traffic. I opened the first foreign/foreign relation 
(Belgium/Netherlands) and hooked the first large Tymnet International 
private network (at Philips - where "com" and "net" where created and 
then pasted everywhere). Bob Tréhin implemented the Swedish Public 
Network, while Telenet built PSS the UK network. Germany then opened 
their Northern Telecom Datex-P. The French Transpac soon became the 
largest network in the world, quickly reaching one million users, and 
then 80% of the French people by mid-80s, with Minitel. We teamed 
with many for experimentation in the US and in Paris. Web like 
applications were developed using stop-TV, local TV, etc. Tymshare 
hired Doug Engelbart (I tried to implement his "Augment" solutions on 
my international new machines, but we found that networking "extends" 
services [and errors :-)], far far more than it "augments" people's 
IQ. The next generation systems we experimented included local proxy 
databases for image, text, etc. and content oriented extended services.

Tymnet III was specified from this accumulated field proven 
experience. The Extended Network System model was worked on. I have 
not  read yet a proposition which offered something which was not 
easy with Tymnet III, ISIS II and the new generation of Tymnet 
Engines under test. Tymnet having been purchased by McDonnell 
Douglas, we progressively disbanded. The status of the International 
Network, mainly under Tymnet technology, when the Internet was 
created is documented at http://intlnet.org/intlhist.htm . A service 
description is at http://intlnet.org/intl84.htm (my draft for the 
document we distributed). Tymnet brought the support of global 
communications from nil to millions of users.


OSI

Tymnet technology supported smart interfaces or services in the 
network nodes (named "slots"). Located at the "edge" as we would say 
today, they permitted to support any other technology (DECNET, SNA, 
SWIFT, OSI, etc ..). Along with one of the possibilities described in 
RFC 2775, it transmitted protocols metadata separately from data. 
Tymnet had introduced the virtual circuit concept [as a secure "pipe" 
- no spoofing permitted]: it was actually a reliable virtual circuit 
switch. It used internode multiplexed packets: offering impressive 
end to end  keyboard echoing at no extra load/cost. There was a 
proportional decrease of the resulting overhead with the traffic 
load. Liaisons were fast and efficient (for the time). This permitted 
to translate protocols, and to provide TCP/IP-X.25 liaisons for 
example. The initial architecture was not datagram oriented (Tymnet 
was however intensively used for credit card transactions). The 
architectural evolution underway would have progressively lead to 
their full support, while keeping the speed, security and surety benefits.

We therefore soon became the main (and quasi only) international 
implementors of OSI of the time. First X.25 and then X.75 once we had 
stabilised common parameters with PTTs through an operator/user club. 
We established X.75 real or virtual links to all the operators and 
X.25/PAD local support, operating a catalog of more than 40 X.25/75 
brands. This lead me to create and deploy (with Dominique Marchand) 
the X.121 addressing scheme of all "our" countries externets - 
everyone but an handful. We treated them as numeric addresses, 
numbers or names, depending on the used architecture (OSI, 
Telephone/Teletext/Fax  or Tymnet).

Tymnet was later on purchased by BT and standardised to X.75. OSI 
brought the International Network to one hundred million users. 
Mostly in Europe and Japan. It supported Videotext, etc.


Internet

End 1983 the DoD asked Tymnet to interface the ARPANET Internet. This 
was for international connections, local virtual links to some 
Universities and remote public access and out-dial (Tymnet at that 
time had 500 access/out-dial (to private hosts/PC) nodes nationwide). 
We announced TCP/IP support in Dec. 83. The technical dialog was with 
our "Federal" office (probably until mid-84?). I supervised all the 
interconnects being responsible for International operations for 
naming and addressing compatibility.

We had two main issues: naming and tunnelling (I made the mistake not 
to press for addressing, otherwise we would be IPv6 for a long).

1. the target was to provide a tunnel with Internet foreign 
communities (standard "virtual circuit" - or a "pipe") at better 
price than a dedicated line or/and as a back-up for dedicated lines. 
This was called "refilling". It was also to provide access to DoD and 
State Department personnel when travelling, and to Embassies. 
Refilling had been actively explored by Jack McDonnell (Chair of TNS 
Inc.) and was mostly used by private international telex operators 
(another architecture partly supported). For this reason, refilers 
were named after ISO 3166 Alpha2 [telex codes] (Public services 
registries were named after ISO 3166 Alpha3 [Radio codes]). This 
created the ccTLDs as documented in RFC 920.

IP clusters were easily supported this way when introduced in 1986 - 
if I am correct - in Wien. The path was TCP/IP to the Gateway, Tymnet 
Protocol to the X.25/PAD service of Radio Austria

2. International naming was then left to right, one single string 
(root-name [TLD] + network-name [domain name] + local-name [3rd+LD]). 
This was less flexible than the right to left doted format. But it 
permitted a direct support of X.121 (DNIC+Numbers) and Out-dial 
(local telephone numbers). Internet was reluctant to change their 
inner naming to the international naming: the  number of networks 
they had to concatenate was large, and this made the caller to decide 
the routing (via Internet or via a gateway when this was possible). 
So they came from "name" to "name.arpa" to train everyone. Then to 
"name.com" through the Gateway. We first only checked if there was a 
"." in the name to accept the call, reversing the order. But this 
created a problem with IP addresses. We asked that names would come 
from the Internet with a final dot ("dot-root") and checked it. We 
also decided that all numeric names would not be changed.

Internet has now brought the International Network to one billion 
user. We now look for the approach which will rise usage to 10 
billions people and stations.


Some resulting experience

Among other (this is just a rushed mail) I think this shows from experience:

- the user is the decision maker. The QoS they get and the respect of 
the user (operator) needs/decision/interests makes the adherence and 
the decision. When we talk of market acceptance, we only say "if 
users think they get enough for their money". If they poorly 
understand a brilliant concept, it will not fly. What flies is are 
rustic, simple, secure solutions which works and people understand. 
In fact, the network is not what the designers designed, but what the 
users think/want it is. If both do not match, forget about it!

- This is important because it means that the IAB/IETF is _not_ in 
charge of the future network (which should be backward compatible) 
and has the divine right to design it, based upon a TCP/IP+. It means 
that we all are in competition to deliver something appealing to the 
next generation people. There is no exclusive of situation. If I 
fought (and globally won) toughly the RFC 3066 Bis proposition to get 
a consensus I was denied and they now challenge, it was because it 
concerns the core of the Multilingual Generalised NGN (MGN). A 
proposition that people will _refuse_, killing the chances of any 
IETF new architecture proposition.

- packet massaging on the plug to plug interoperability path is not a 
violation of the "end to end" concept, nor of the [more interesting 
to the user] brain to brain interintelligibility one.

- we need to go one step further, accomplishing the whole catenet vision.
   o "the networks" is the hardware oriented infrastructure which 
supports the convergence.
   o "the network of the networks" is the software oriented 
superstructure we live with.
   o "the networks of the network of the networks" is the brainware 
oriented metastructure where the network architecture evolution has to deliver.

etc.

_______________________________________________
Architecture-discuss mailing list
Architecture-discuss@ietf.org
https://www1.ietf.org/mailman/listinfo/architecture-discuss