Fraser Research Annual Reports

Annual Report 2004


155 Mbit/sec to Every Home in America

More than 50 million U.S. citizens use Internet service today, yet for most people communication is still provided by legacy networks - phone wires or TV cable. Their experience will be very different when they use a new infrastructure designed holistically with the new world in view. Only a few people have experienced a web that responds in less time than a human can detect, and with image quality that matches the print in a glossy magazine. One taste ensures addiction. The new communications will be immediate, always available and everywhere they need it.

The Internet has grown from nothing to span the world. The achievement has been outstanding, but it is only a start. There are backbone and premises networks but the most expensive part has yet to be built. The need now is for access networks that connect homes and small businesses everywhere to the backbone. This new construction will be by far the biggest part of the total infrastructure cost. Rather than build upon a problematic foundation conceived 35 years ago, it would be best at this time to address the problems of the present Internet, so that the new always on, everywhere available communication service is safe, reliable and easy to use.

A Packet Switch to Serve One Million Households

A topdown study of a network for 100 million households suggests a much simpler network topology than is presently evolving for the Internet. Just as Federal Express discovered that computer networks and air transport enable a more efficient centralized architecture for package delivery, so it is apparent that fiber and the practicality of very high capacity switches point the way to a more efficient communications infrastructure. The question until now has been whether switches of sufficient capacity can be constructed. This talk seeks to show that indeed they can.

Clos and Benes developed the principles which enabled large telephone switches to be made from many small switches. In recent years the study of large packet switches has found inspiration in the work of Clos and Benes even though their analytic results do not apply in a packet switched world. The problem is that bad traffic patterns cause points of severe congestion and it is hard to manage large and diverse traffic flows without triggering significant side-effects, like loss of proper sequence in the delivered data.

Quite by accident an organizing principle for large switching networks has been found. Simulation results confirm the possibilities. In this talk the underlying principle will be explained and there will be discussion of simulation results for a machine to serve 256,000 homes. Simple calculations lead to the belief that switches with capacity in excess of 1 Pb/sec are practical


Architecture and Algorithms for a Reliable Access Network. Patzer, A. Submitted for Degree of Master of Science, Princeton University, January 2004.