Sally Floyd

Sally Floyd is a research scientist at AT&T Center for Internet Research at ICSI (ACIRI) an institute dedicated to Internet and networking issues. She is known in the industry for her work in Internet protocol design, in particular reliable multicast, congestion control (TCP), packet scheduling (RED), and protocol analysis. Sally received her B.A. in Sociology at the University of California, Berkeley, and her M.S. and Ph.D. in Computer Science at the same university.

How did you decide to study computer science? How did you become interested in the field?

After getting my B.A. in sociology, I had to figure out how to support myself; I ended up getting a two-year certificate in electronics from the local community college, and then spent ten years working in electronics and computer science. This included eight years as a computer systems engineer for the computers that run the Bay Area Rapid Transit trains. I later decided to learn some more formal computer science and applied to graduate school in UC Berkeley's Computer Science Department.

What made you decide to specialize in networking?

In graduate school I became interested in theoretical computer science. I first worked on the probabilistic analysis of algorithms, and later on computational learning theory. I was also working at LBL (Lawrence Berkeley Laboratory) one day a month and my office was across the hall from Van Jacobson, who was working on TCP congestion-control algorithms at the time. Van asked me if I would like to work over the summer doing some analysis of algorithms for a network-related problem involving the unwanted synchronization of periodic routing messages. It sounded interesting to me, so I did this for the summer.

After I finished my thesis, Van offered me a full-time job continuing the work in networking. I hadn't necessarily planned to stay in networking for ten years, but for me, network research is more satisfying than theoretical computer science. I find I am happier in the applied world, where the consequences of my work are more tangible.

What was your first job in the computer industry? What did it entail?

My first computer job was at BART (Bay Area Rapid Transit), from 1975 to 1982, working on the computers that run the BART trains. I started off as a technician, maintaining and repairing the various distributed computer systems involved in running the BART system.

These included a central computer system and distributed minicomputer system for controlling train movement; a system of DEC computers for displaying ads and train destinations on the destination signs; and a system of Modcomp computers for collecting information from the fare gates. My last few years at BART were spent on a joint BART/LBL project to design the replacement for BART's aging train-control computer system.

What is the most challenging part of your job?

The actual research is the most challenging part. Right now, that means designing and exploring a new mechanism for end-to-end congestion control, based on equation-based congestion control. This is intended not to replace TCP, but instead for unicast traffic such as some rate-adaptive realtime traffic that would prefer to avoid the dramatic rate change of reducing its sending rate in half in response to a single packet drop. Equation-based congestion control is also of interest as a potential foundation for multicast congestion control. More information is on the Web page at http://www.psc.edu/networking/tcp_friendly.html.

In what direction do you see the future of networking/ the Internet?

One possibility is that the typical congestion encountered by Internet traffic will become less severe as pricing mechanisms come into play and the available bandwidth increases faster than the demand. I view the trend as toward less severe congestion, though a medium-term future of increasing congestion punctuated by occasional congestion collapse does not seem impossible.

The future of the Internet itself, or of the Internet architecture, is not at all clear to me. There are many factors contributing to rapid change, so that it is hard to predict how the Internet or the Internet architecture will evolve, or even to predict how successfully this evolution will be able to avoid the many potential pitfalls along the way.

What people have inspired you professionally?

Richard Karp, my thesis advisor in graduate school, essentially showed me how to do research, and Van Jacobson, my "group-leader" at LBL, was responsible for my interest in networking and for much of my understanding of the Internet infrastructure. Dave Clark has inspired me through his clear view of the Internet architecture and his role in the development of that architecture through research, writing, and participation in the IETF and other public forums. Deborah Estrin has inspired me through her focus and effectiveness, and her ability to make conscious decisions of what she will work on and why. 

One of the reasons that I have enjoyed working in network research for the last ten years is that there are so many people working in the field whom I like, respect, and am inspired by. They are smart, work hard, and have a strong commitment to the development of the Internet, do impressive work, and can be good companions for a beer and a friendly disagreement (or agreement) after a day of meetings.