THE SRI VAN
AND COMPUTER INTERNETWORKING
Since the days when it was a stagecoach stop between San Francisco and Monterey, Rossetti's was a well known San Francisco mid-peninsula "watering hole" nestled in the second bank of foothills west of San Francisco Bay. In the 1975, it had a casual atmosphere and some outdoor seating-a good location for the small ceremony about to take place. No one would mind if we parked SRI's "bread truck" van alongside the courtyard and ran a few wires to one of the tables. It was far enough from SRI (Stanford Research Institute) to qualify as "remote," but close enough to have good radio contact with them through a repeater station atop a hill above Stanford.
So it was that this venue was chosen to mark the occasion of the first internet transmission on August 27, 19761. The van was an SRI outfitted mobile radio lab that contained the equipment needed to make it a portable node on the emerging Packet Radio Network (PRNET). PRNET was sponsored at SRI by ARPA (Advanced Research Projects Agency) and started in 1973 or so. Placing a terminal on one of the wooden courtyard tables and connecting it to the van, a number of SRI people who had gathered for the celebration fled a normal weekly Packet Radio Program report representing the work of all the Program's contractors-to ARPA. While the testing of such a connection had been going on for several months, this long e-mail report was, in a ceremonial sense, the first internet transmission; that is, the first formal use of the internet protocol known as "TCP."2
TCP was designed to carry information over dissimilar networks, in this case the PRNET, through a gateway at SRI, then across the ARPANET to a set of hosts distributed around the United States. This small, virtually unknown, but deliberate episode became a milestone in mobile digital radio and the flexible integration of digital communications networks. But let's back up a bit and review in more detail the emergence of internetworking and the role the SRI van played in it.
In the early 1970s, the ARPANET was growing rapidly. Universities, or their close affiliates, were the main players connecting to the network. Under inducement from the sponsors at ARPA, and through their own inventions of new and useful network services such as electronic mail, network traffic began to grow. In the meantime, the notion of a radio version of the wired ARPANET had come to Larry Roberts at ARPA. When Roberts left, first Bob Kahn and then Vint Cerf pursued that same idea at ARPA. Both Roberts and Kahn had seen the military need for a mobile, wireless version of the embryonic ARPANET. SRI and ARPA had also discussed the possibility of a transportable, possibly handheld, terminal or switching node for such a network rather than the massive, seemingly nuclear-hardened early IMPS (see back cover for more on the IMP) of the fixed network. Following that instinct, ARPA formed a team of contractors in what came to be called the Packet Radio Program. The team's mission was to create a wireless adjunct to the evolving ARPANET.
Members of the new Packet Radio Program were Bolt Beranek and Newman (BBN) in Boston, Collins Radio in Dallas, Network Analysis on Long Island, University of California Los Angeles, and SRI. Because it had a good understanding of radio and systems integration, SRI was chosen as system engineer and technical director (SETD) of the program as well as integrator for ARPA's packet radio effort, a position it maintained for over a decade.
It should be pointed out that the introduction of a radio segment t0 supplement the ARPANET came from imply following the military context in which this and a great deal of research in the United States is done. If the military were to ultimately employ this new interactive digital technology, there would have to be allowances for the military's inherent mobility and possible deployment to any point on earth. So a radio network, particularly one that served a mobile population, was needed. It turned out to be intrinsically different from the existing fixed, wired one. This clear difference, along with the need for the two networks to work well in tandem, led to the notion of a communication software structure that would effectively bind these disparate networks together as though they were one.
One technical insight needs to be inserted here to understand how disparate packet networks can easily function together. In most communications networks it is only the source and destination terminals that are visible to network users. The resources that lie in between are normally of little interest to them as long as they fulfill their role. In circuit switching, once chosen, the same physical pathway is maintained for the whole session. When circuits are leased, the connection may even be "hardwired."
In packet switching, where sub-units of single message may travel entirely different routes from source to destination, the exact role of intervening sources would not even normally be known. Thus, there arose the concept of virtual circuit," where the only defining network nodes lay at the ends and in which the intervening nodes are neither specified nor known by either network users or providers. This switching concept had been part of the basic ARPANET design and was now to be extended to this amalgam of wire and radio networks and thus to the world of internets.
It was the clear differences between the wire-based ARPANET and the radio-based packet radio (and eventually satellite networks) that led Kahn, then heading the networking efforts at ARPA, and Cerf at Stanford University, to design the first end-to-end protocol that would span dissimilar packet networks. The essence of such a construct began to emerge when Kahn addressed the problem posed by these dissimilar networks at a seminar held by Cerf in the summer of 1973.3 After some airing in the Internet community, the rudimentary elements of such a protocol came together for them on an October 1973 weekend at the Palo Alto Rickey's Hotel.4 They published the design in May 1974,5 and named it the TCP, or Transmission Control Protocol. With some modifications, it is still in use as the basis for transport in the worldwide Internet.
Following the introduction of TCP, ARPA contracted for three separate implementations: Stanford University, BBN, and University College in London. The first, clearly "buggy" specification to appear was in December of 1974 when Stanford produced RFC 675. BBN had an in-house version working reliably about a year later and began exchanging TCP traffic with Stanford on an intranet basis. Jim Mathis, a student of Cerf's at Stanford, started to implement their protocol in 1975. He came to SRI in the summer of 1976, where he completed a version that would run on the much more modest hosts of the packet radio network (Digital Equipment Corporation LSI-11 microcomputers). In the meantime, Cerf, now a program manager at ARPA, was trying his best to inculcate the Department of Defense with the virtues of packet switching and TCP for their future data networks.
As a part of this emerging digital radio network, SRI foresaw the need for a mobile laboratory. A lot of design work lay ahead regarding the notions of nodal power and reach, the size of packets and the functions they were to perform, and the routing and reliability strategies in a network characterized by packet loss rates much higher than that seen on wire-based networks. Then there were the critical choices of radio frequencies and the signal processing strategies for the propagation and noise environments in which such a packet-switched radio network would operate. Since computers are notoriously intolerant of errors, how could a vulnerable radio environment be made to transport perfect data?
The SRI van was first used to characterize the radio frequency channel on which a packet radio system would be expected to operate. This was to be a fault-tolerant, dynamically-adaptable network. And so, a tough urban setting, with its shielding, reflective buildings, and electrical noise, was chosen. Radio modulation was designed that was tolerant of multipath distortion and noise. Packets were encoded for error detection and re-transmission when received inaccurately. Noise and the propagation patterns were characterized. When it came time to transport information across the packet radio network, a subnet was installed in the Bay Area and the van became a mobile node in that network. The PRNET became a self-organizing network, with addressing and routing, capable of accommodating the transmission challenges imposed by mobile users. It was the first mobile packet network.
Given the difficulty of the radio environment, a couple of interesting demonstrations were often used at the time to illustrate the robustness of this new concept of networking. To illustrate the flow of traffic between a terminal in the mobile van and some distant network host, a character generator would grind out continuous alphanumeric sequences that formed patterns on a CRT in which errors would be obvious. While moving at high speed in the SRI van, the signal would sometimes be interrupted due to shielding of the radio signal (as when going beneath an underpass). The flow would stop momentarily but no errors were observed. Error- detecting cyclic redundancy checks, applied at the end of each transmitted packet, were used to verify reception accuracy. These checks plus the end-to-end ordering and re-transmission properties of TCP would not permit delivery of altered packets even though packets were frequently lost! Another similar procedure was to withdraw the synthesizer card from the packet radio. This would terminate the character flow, but re-inserting it would start it again. Thus, traffic would stop, then resume, but no errors were ever observed. Those demonstrations were splendid evidence that each packet could have sanctity, even in a tough environment of intermittent propagation and noise. This was an exciting consequence and certainly foreign to those circuit-oriented engineers who saw mobile digital radio systems as some sort of oxymoron.
The first testing of TCP across dissimilar networks started in the summer of 1976. The first trials stayed one radio hop from the Packet Radio station (the PRNET's controlling node) where the bidirectional ARPANET gateway software, built by Ginny Strazisar at BBN, was located. During July and August the SRI team tested and tuned Mathis' version of TCP for better accuracy and speed. It was in August of 1976 that a terminal, attached to an LSI-11 "host" running TCP that was in turn attached to the PRNET, proceeded through a gateway to first access an ARPANET host. For the first time, at least in a ceremonial sense, dissimilar networks were bridged by TCP, thus clearly creating a two-network internet connection. That specific network configuration is shown in the figure at the top right, which is copied from a packet radio progress report written at that time.' The occasion was the aforementioned distribution via TCP of the normal, long weekly Packet Radio Program report. Other two-network TCP connections would soon follow.7
(top) Diagram of the first two-network Internet transmission on August 27,
1976. (Original illustration from an SRI technical report "Progress Report on
Packet Radio Experimental Network" published in September 1977.)
(bottom) Diagram of the first three-network Internet transmission on November 27, 1977, comprised of three physical and four logical networks, the ARPANET being used twice. (Original illustration from an SRI technical report "Progress Report on Packet Radio Experimental Network" published in February 1978.)
Within a year, and fulfilling the assumed need for a network of global reach, ARPA moved to include its third packet network, one that was satellite-based. It was then time to demonstrate all three networks together. On November 22, 1977, what has come to be more generally regarded as the first internet transmission occurred between the SRI mobile packet radio van and a host computer at USC by way of London! The route is shown on the bottom right.8
So internetworking was born of necessity, to demonstrate at ARPA that the innovations of packet switching were indeed relevant to the military's mode of operation. No matter where deployed, they could move about as needed and still be tethered to the powerful computing hosts kept safely away from the fighting. The robustness of the networks, be they fixed or mobile, was, of course, not just a military feature. Packet switching was sensible from the point of view of high network utilization and for offering a soft failure in the presence of moderate network congestion or even limited node failure. To be sure, the PRNET was a collective effort of many people, just as were the first workings of the internet. But the SRI van, purchased by SRI as a piece of capital equipment and designed to be used in a wide variety of experimental roles, found its major role in these first internetworking experiments.
The possible importance of "the van" began to surface sometime in 1996 when an IEEE Spectrum editor called and mentioned that Vint Cart had said in an interview that "SRI was the site of the first internet transmission." I said I would look into it and began digging through old PRNET documentation to verify a couple of events that I vaguely remembered.
After the November 1977 date was confirmed and defined accurately and had been promulgated a bit, the next call came from The Computer Museum History Center (now the Computer History Museum) about celebrating the 20th anniversary of the Internet at the Supercomputer Conference in San Jose in 1997. I offered that the van was still at SRI but had languished unused for perhaps 10 years on the back lot. When it was clear there was interest in putting it on the convention floor, Don Alves of SRI and I began the job of getting it running, dressing it up as best we could, trying to replenish the almost non-existent radio and internet equipment that had been in it, getting it re-licensed, and coaxing it to San Jose. While not beautiful, it did seem to carry some symbolism for many who saw it. So, rather than returning it to certain deterioration and scrap, SRI offered it to the Museum, where it lives today.
Don Nielson has been at Stanford Research Institute, now SRI
International, for 40 of its 55 year history. During the events associated
with the Internet transmissions mentioned above, he was the SRI
principal investigator for ARPA in the early stages of the packet radio
program. While that program was unfolding, he became director of SRI's
Telecommunication sciences Center (1975), the center at SRI for
computer networking. To better align its work with the future of computing,
this group was permitted by SRI to join the Computed science Division,
which Don came to head from 1983 until his retirement as an SRI vice
president in 1998. Since than he has been writing a book an SRI's major
innovations, from which this segment about the SRI van was drawn.
THE SRI VAN AND EARLY PACKET SPEECH
When the ARPANET was perhaps five years old and before the development of internet protocols, Bob Kahn at ARPA set a group of contractors exploring how the new network could handle normal telephone traffic. Given the initial focus on reliable data transmission, it was not clear whether the variability in interpacket delay would permit the smooth flow required by a voice call. In 1974, Kahn Initiated the Network Speech Compression Program because of the narrow bandwidth of the initial circuits comprising the net. This program resulted in the choice of some compression algorithms and these were first tried over the ARPANET. In 1976, SRI's Earl Craighill and Tom Magill, both of whom had been working on the speech program, convinced ARPA to let them try speech on the Bay Area PRNET. By this time the internet protocol, TCP, was also being tested and so speech experiments began also on an internet basis.
Because the SRI van was an easily outfitted facility and already had packet radio and internet equipment installed, it became the first mobile node for packet speech experiments. In addition to the challenges of mobile data transport, transporting natural-sounding speech focused on the importance of delay variance. Innovations were needed in variable rate encoding, new buffering strategies, and rapid rerouting of packets whenever the route in use failed. All these were to help smooth the flow of speech. Importantly, these requirements for packet speech influenced the decomposition of the protocol into reliable or guaranteed (TCP) and non- guaranteed (IP) services. Thus, internet speech connections were being conducted as early as 1977-1978, about the same time as the Internet itself was becoming a reality.