Protocols for High-Speed Networks: A Brief Retrospective Survey of High-Speed Networking Research

This paper considers high-speed networking research from a historical perspective, and in the context of the development of networks. A set of axioms guiding high-speed network research and design are first presented: Ø KNOW THE PAST; I APPLICATION PRIMACY; II HIGH PERFORMANCE PATHS; III LIMITING CONSTRAINTS; IV SYSTEMIC OPTIMISATION. A framework of network generations is used as the basis for the historical development of highspeed networking: 1st – Emergence; 2nd – Internet; 3rd – Convergence and the Web; 4th – Scale, Ubiquity, and Mobility. Each generation is described in terms of its application drivers, and important infrastructure and architectural characteristics. Woven into this historical thread are the important research thrusts and sub-disciplines of high-speed networking, and their impact on deployment of the Global Information Infrastructure. Based on this historical perspective, a set of SYSTEMIC OPTIMISATION PRINCIPLES are identified: 1 SELECTIVE OPTIMISATION; 2 RESOURCE TRADEOFFS; 3 END-TO-END ARGUMENTS; 4 PROTOCOL LAYERING; 5 STATE MANAGEMENT; 6 CONTROL MECHANISM LATENCY; 7 DISTRIBUTED DATA; 8 PROTOCOL DATA UNITS. We are now in the state where everything has some aspect of high speed networking, and nothing is only about high-speed networking. This is a doubleedged sword ― while it reflects the maturity of the discipline, it also means that very few people are looking after the performance of the entire Internet as a system of systems. Rather, performance analysis tends to be isolated to individual network components, protocols, or applications. Furthermore, the high-speed networking community is not pushing back at the multitude of deployment hacks by network and application service providers (such as middleboxes) without regard to global network performance effects. Thus, this paper argues that the high-speed networking community should have the future role of caring about high-speed network deployment on a global scale, and throughout the entire protocol stack from layers 1 through 7.