ERCIM News No.19 - October 1995

Time to Get to Thinking

by David Farber

Over the past four years, the United States has undertaken a joint industrial, university, and governmental research initiative designed to study the impact of gigabit networking on the future of networking, networking applications, and computer architecture. This study has led to the formation of five testbeds, each exploring different aspects of the emerging technology as well as motivating several non US experiments The experiment is now drawing to a close, at least in its first phase, so it is reasonable to ask what we've learned and what the implications are for the future.

First, and maybe foremost, we have shown that industry and academia can work together for their mutual benefit. In the case of the U.S. testbeds, the largest percentage of the resources and manpower were contributed to the effort by industry without government support or tax incentives. Universitites, while supported by the government, found themselves as partners in pushing the frontiers of research. Students and engineers worked hand in hand in each others laboratories. This model of collaborative research, which was so effective here, should be applicable to other frontier activities.

But what did we learn from the research activities themselves? Gigabit speeds have raised a whole new set of very difficult technical issues. Designing and building switching devices and interface devices which can operate at these speeds is not simple. It pushes both hardware design and VLSI technology to their limits. As a result, it has been necessary to take innovative architectural approaches to even hope to achieve speeds nearing a gigabit.

Perhaps most interesting, though, is the conclusion that many of the ideas developed over the past twenty years in computer architecture, operating system design, and networking protocols seem to be ineffectual when applied to such high speeds. It is worth observing that these communication speeds are of the same order of magnitude as the main memory bus speeds of modern workstations. Thus it is not surprising that we have run into problems. When streams of data arrive at memory speeds, it becomes difficult, given the protocol systems currently in use, to get the data into memory, to allow the processor enough processing bandwidth to examine the data and move it, and still to have processing power left over for other tasks. I will not elaborate in this piece on the solution I and others have proposed for this problem. But basically, the solution revolves around the creation of a geographically dispersed distributed machine, the components of which would be interconnected by high-speed networks. This approach has been well documented.

What is more important than a particular solution is the challenge of facing a future in which gigabit speed networking will be considered slow, in which our communication infrastructure will consist of multi-gigabit, low error, high-latency networks, in which our processing units, while growing faster, will not keep up with increasing communication speeds. It is too easy to just remove a few instructions, hack a few cures, and show that one can operate not too badly at current speeds of communication. Perhaps this is equivalent to saying, "let the next generation solve the problem." I believe that there is a challenge facing the computer communication field of at least the same magnitude as the challenge the field faced in the very early days of networking. Attacking this problem will require the talents of people from every area of both the computer and communications fields--people willing to experiment and willing to face the same set of challenges those in the fifties faced with the then-new computers.

In 1996, we will be celebrating the 50th anniversary of the eniac computer, developed at the University of Pennsylvania. The children of the eniac have transformed our society in many ways, both for better and for worse. As we turn to the next fifty years, we are facing an era in which the convergence of computers and communications will be the key technological innovation. The impact of this development on our technology and our society will most likely be considerably greater than that of the previous fifty years of evolution. It is time to start thinking and working and innovating, so that in 2046 we can look back at these fifty years as a time of insight and advance even greater than that of the last fifty.

Please contatct:
David Farber - University of Pennsylvania