Lee Center for Advanced Networking

John Doyle is a professor of control and dynamical systems, electrical engineering, and bioengineering. Together with Steve Low, he is designing a new theoretical foundation for large-scale communication networks, called Highly Optimized Tolerance (HOT), as well as an analytical theory of protocol design.

Doyle says that HOT applies to all complex systems, not just communication networks. All complex systems, whether communication networks or biological systems, share certain qualities. For example, the biological concept of “convergent evolution” describes how two unrelated lineages of organisms can evolve similar traits. For example, birds, bats, and many groups of insects all have wings.

In the same way, says Doyle, unrelated communications networks can evolve similar overall architectures, composed of elaborate hierarchies of protocols and layers, even when the simplest component parts are entirely different. These features developed from the interplay among complexity, robustness, modularity, feedback, and fragility—factors that are common to all complex systems.

HOT places networking in a larger context. In the future, the global communication network will be made up of smaller networks, such as transportation, energy, and finance. The HOT framework is beginning to create a unified theory of complex systems with particular focus on issues such as robustness, scalability, and computability.

Network complexity, says Doyle, is an adaptive response to uncertainty in a system’s environment. That is, Internet protocols are complex because it makes them robust to uncertainty. But this same complexity can leave systems vulnerable to rare or undersigned-for perturbations. Doyle and Low believe this robust-yet-fragile character is the most important and universal feature of complex systems. They hope that HOT will provide a framework for understanding both the successes and shortcomings of existing Internet technology and guide rational design for ubiquitous networking unimpeded by evolutionary dead ends.