VI - How to Communicate Across Network Boundaries?
Shortly after the successful ICCC'72 conference, Bob Kahn left his job at Bolt Beranek and Newman (BBN) and went to work at IPTO. Joining IPTO as a program manager, Kahn initiated certain projects and also took over responsibility for one that had already been funded. A new initiative was to create a ground based packet radio network. An existing initiative was to create a satellite-based packet switching network. (Ronda Hauben, "The Birth of the Internet", 7)
The ground packet radio network would be of particular interest to the US Department of Defense(DOD), as it would make packet switching computer networks possible in otherwise difficult to reach areas or conditions. Kahn's objective was to create a multinode ground packet radio network (PRNET) where each node could be mobile. In parallel, he sought to create a packet satellite network (SATNET) utilizing INTELSAT satellites.(14) The goal of the packet satellite network research was to make resource sharing computer communications networking possible with different European sites. Two of the networks (PRNET and SATNET) would use radio transmission and the third network which already existed (ARPANET) used shared point to point leased lines from the telephone company. Though Kahn originally considered the possibility of seeking changes to each of the constituent networks to solve the multiple network problem, he soon recognized the advantage of an architecture that would directly accommodate a diversity of networks. While existing networks like the ARPANET would require that another network become a component of it, there was a need for an architectural conception that would allow the communicating networks to function as peers of each other, rather than requiring that any one become a component of another. There was a need to design an architecture that would be open to all networks, an architecture that Kahn called "open architecture".(15)
VII - Designing Protocols and Specifications for an Internet
Once at IPTO, Kahn invited Vinton (Vint) Cerf to collaborate with him. Kahn wanted to design an open architecture protocol and needed Cerf's knowledge of computer operating systems to do it. Other researchers were also interested. For example, at an INWG meeting in June, 1973, in New York City, Kahn and Cerf were joined by E Aupperle, R Metcalfe, R Scantlebury, D Walden and H Zimmerman. Scantlebury was from the UK and Zimmerman, from France. Others listed were members of the US network research community. The document also credits G Grossman and G LeLann for contributing after the meeting. LeLann is French. (INWG note #39 NIC # 18764, dated 9-13-73). Cerf explains that LeLann worked with Louis Pouzin at IRIA (now INRIA) and “spent 6 months
working with me and others on the design of the Internet's TCP protocol."
(Cerf, Email, April 13, 2003) Pouzin also remembers a June 1973 INWG meeting, noting that it was quite hot in NYC. (Pouzin, Email, April 28, 2003)
The INWG note #39 is a draft paper that Kahn and Cerf prepared for presentation at the September 16, 1973 INWG meeting in Brighton, England. A revised draft of the paper was published in May, 1974, titled "A Protocol for Packet Network Intercommunication" in the IEEE Transactions on Communications. The paper describes the philosophy and design for the TCP/IP protocol, though the original paper called the protocol TCP, as the IP function was originally embedded in TCP.(16)
After designing a protocol, there is a need to write specifications to implement the design.(16a) Cerf refers to the development of two versions of the specifications for TCP developed at Stanford University, one in December 1974 and a second in March 1977. Subsequently two further specifications were developed with other groups.(Cerf) Among the names of those working on the initial specifications for TCP, Cerf lists US researchers or graduate students including Y Dalal, C Sunshine, R Karp, J Estrin, and J Mathis, at Stanford, R Tomlinson and W Plummer, at BBN, R Metcalfe, D Boggs, and John Schoch, at Xerox PARC. He also lists several researchers from the UK, from UCL, F. Deignan, C J Bennett, A J Hinchley and M Gallard. Cerf also thanks G LeLann from the University of Rennes, France. Cerf writes that Dag Belsnes, from the University of Oslo, Norway provided "additional philosophical leavening which influenced the design of the protocol."(Cerf, The Final Report, IEN 151, 2)
When asked what he thought the term "philosophical leavening referred to," Belsnes responded, "I also wonder what 'philosophical leavening' is referring to. Perhaps that I always like to discuss and establish some understanding of problems."(17)
In 1973, Belsnes received a one year grant from the Norwegian Research Council. After meeting Vint Cerf at a conference in England in 1973, Belsnes contacted Cerf and was accepted to be part of the research effort at the Digital System Laboratory at Stanford University. "I got the opportunity," Belsnes writes, "to participate in his Protocol Design Group that worked on creating a specification for the Internet Transmission Control Program." Belsnes explains that among his main interests were "protocol correctness and flow congestion control." (Belsnes, Email, June 17,2002)
Creating a design and then specifications for the development of a protocol for an Internet is a significant step. It is, however, part of a larger research process. Elaborating on the value of the experimental work, Paal Spilling, another of the Norwegian Internet pioneers, writes:
"A group at Stanford University (SU) specified in detail a control
program ... the Transmission Control Program (TCP) allowing computers
in different inter-connected networks to communicate.... Although the
TCP was specified in detail, it had to be considered as a first
approach towards making a reliable process-to-process communication
tool in an internetwork environment. Experience showed that this was
the case.... The results obtained, helped in the debugging of this
first version of the TCP, and uncovered some deficiencies in its design.
Some of these could be taken care of rather easily, while others were
subjects for further investigations."(Spilling, Proposal to Nato)
Kahn had recognized the need to include at least three different kinds of packet switching networks to test if the protocol created for intercommunication among dissimilar networks would be adequate. If a prototype has only two different entities, it is difficult to tell what is particular about each and what is general about the two. With three or more dissimilar networks as part of a prototype, it is possible to identify what is general to them all despite the dissimilar nature of each.
In June 1973, a TIP was installed at Kjeller, Norway for the NDRE researchers. By the end of July 1973, the UCL TIP in the UK was also passing packets between the US and UK. These packets went from the US via satellite to the Tanum Earth Station in Sweden, via land and underwater lines to NORSAR in Kjeller, Norway, and then to London in the UK. Kirstein and Kenny provide a diagram of the relation between the
UK TIP, the Norwegian TIP and the US ARPANET.(18)
Kirstein writes that one of the significant activities in the early work to develop the Internet was "an early protocol experiment in late 1974 between a junior assistant professor at Stanford (Vint Cerf) and a visiting scholar from Norway at UCL (Paal Spilling) of the Proposed Transmission Control Protocol." Spilling visiting UCL from NDRE, worked with Kirstein's research group. Judy Estrin was a graduate student working with Vint Cerf at Stanford. Estrin and Spilling "did what was probably the first TCP tests with each other. They were independent implementations," Kirstein explains. (Kirstein, Email, May 20, 2002.) Describing this research, Spilling elaborates, "As I remember the fellows at the Stanford side may have been Judy Estrin and Jim Mathis. At the UCL side were Frank Deignan, Andrew Hinchley and me. Frank was the implementer. It was extremely exciting to observe packets coming from Stanford and after an initial debugging being accepted and processed by Frank's implementation of TCP. One critical problem I can remember was that the TCP checksum was applied slightly differently at Stanford and at UCL." (Spilling, Email, August 1, 2002)
Kirstein describes how the British government became more supportive of his research by 1975. He writes:
"The British authorities became increasingly positive from 1975. I had
set up a management committee to oversee the use of the Arpanet link.
This included representatives from the British Post Office, the Ministry
of Defence, the Science Research Council and the Department of Industry.
They had to approve all requests for usage. From 1976, there was increasing
pressure for using the emerging X.25 infrastructure (International Packet
Switched Service - IPSS) as an alternative to SATNET. First this involved a
commercial 9.6 Kbps line from about 1978 between UCL and BBN; here it was
necessary to arrange the link so that no commercial charges would arise to
BBN and DARPA. Later, I think it was around 1980, a 64 Kbps IPSS link was
provided also free of charge by the British Post Office. This link existed
until around 1984, and allowed much fuller research into multiple routes
with different capacity, charging and access control considerations. The
IPSS link was always using IP; for this reason the multiple use of the
commercial use and SATNET was an important landmark into the use of
interconnected networks. It was their existence which allowed UCL to adopt
a phased approach to the adoption of the Internet Protocol. We first proved
it on the IPSS link without affecting NCP traffic on SATNET; this needed
NCP-TCP relays at UCL and BBN. We could then move it onto SATNET, without
impacting too drastically our service traffic -- which could use the IPSS
route in an emergency. Finally, when the ARPANET had moved to Internet
Protocols, we could abandon our relays in BBN and also leave SATNET; all
the traffic could use IP/X.25 over IPSS. It is the phased nature of this
transition which explains why UCL finally left SATNET (see below) after the
Norwegians -- though they used IP for service traffic much earlier.
By the time we got to around 1983, complete alternate mail nets, like
UUCP and Bitnet started coming into being. The various gateways these
provided gave a much richer topology. When the DNS was added, its
impact on the international infrastructure was not realised at first.
When we introduced blocking on some of our IPSS routes, we suddenly
realised the magnitude of international traffic that was passing
over the UK-US routes originating from these other networks. It was
then that the work on peering and service agreements took on a new
urgency for these data networks." (Kirstein, Email, October 8, 2002)
|
