AIM INTERVIEWS

Who are you?

I'm a scientist and engineer with a B.E.E. degree from City College of NY in January, 1960, MS and PhD degrees from Princeton University in 1962 and 1964 respectively. Worked at Bell Labs from 1/60 – 9/60 and summers of 1961 and 1962. Was on the MIT faculty from 1964-1966 then took a leave of absence to work at BBN, joined DARPA in late October, 1972 and stayed until 9/85 when I founded CNRI, which I still run.

You spent part of your early career as a member of the ARPANET team at the BBN IMP in Cambridge in 1966. Why did you decide to work on ARPANET? Was there any inspiration?

My career started (briefly) at Bell Labs and then (after my PhD) at MIT. I started working on computer networking at BBN before knowing what ARPA was (they added the D later) and shared with them my early work on the subject. When I learned that they were interested in funding an actual network to link computers, I wrote the technical part of the BBN proposal and we won the contract. Although I had planned to go back to MIT, I stayed on at BBN to help build the network and then decided instead to join DARPA. I didn't decide to work on ARPANET – that was the name given to the system we built after it was deployed.

Your work on TCP helped solve the problem of connecting ARPANET to other wireless networks like PRNET and SATNET. Why was this an important problem to solve at the time and what was the basic idea behind TCP that made it an ideal solution? How was it later decided to split the protocol into TCP & IP?

Both PRNET and SATNET were designed as computer networks, which meant they could connect "directly" to computers or to other computer networks. By directly, I meant either that literally or via a router to intermediate. We used the term "gateway" (between two networks) to refer to what is known today as a router.

At the time the PRNET and SATNET efforts began, all the networked computers that DARPA was involved with were on the ARPANET. The only way a user or network connected to either PRNET or SATNET could generate or access any realistic computation result was for those networks to interact with the ARPANET. The Internet effort solved that problem. The basic internet architecture was elegant enough to have lasted over 50 years to date (and continuing) despite enormous increases in computer speed, network speed and amount of affordable memory over that time period.

The split was created to enable certain real-time applications for which the reliability of TCP, as actually implemented, introduced too much delay due to error detection and retransmission. There were several alternatives, but the simplest was to split out the delivery part from the reliability part and let the receiver decide how to proceed with the packets that did arrive (and in consideration of others that were missing). In many instances, there was no need for reassembly of a sequence of packets into a larger message or to have every packet arrive for the application (such as speech) to be able to continue even if every single one of the relevant packets had not been received.

After your first TCP demonstration with Vint Cerf in 1977, what did you believe would happen for the world if TCP succeeded? Did you foresee something like the World Wide Web emerging from it?

The inspiration for what became TCP was the recognition that we would need a more sophisticated protocol than NCP in a world of multiple networks some of which are wireless. I saw the Internet as a research project originally that might provide a useful infrastructure for the research community. Applications like the Web only emerged years later.

While at DARPA in the 1980s, you were involved with the US government's Strategic Computing Initiative, which heavily invested in developing early AI research, chip design and supercomputing. What did your role entail? Was there a particular technology or research area at the time that you wish had received more support and funding?

Involved with is too benign; it was my idea and I worked with others at DARPA and in the government to create and run the program, at least initially. Read the book called "Strategic Computing" which was published by MIT Press. It will give you a sense of my role there. It caused industry to become involved with AI in a serious way for the first time and helped to create a number of high performance computer systems that make AI applications run in close to real-time. We did what was deemed appropriate for the funding provided. With more funding, we would have done more, but there were no essential areas that did not get addressed initially.

Are there any people or projects you particularly admire today?

Lots (and even more from history), but I prefer not to generate a list as I'll undoubtedly leave some important folks off.

What's a message you have for the world?

Information infrastructure, such as the Internet, is essential for progress in the world we live in. The Internet (and ARPANET before it) were early examples, followed by the web – but they exist to make use of and build upon. The Digital Object Architecture is one example of such new infrastructure.