1981 Reports

Selfish Optimization in Computer Networks

Yemini, Yechiam

This paper describes two applications of decentralized (Pareto) optimization to problems of computer communication networks. The first application is to develop a generalized principle for optimality of multi-hop broadcast channel access schemes. The second application is to decentralized flow-control in fixed virtual-circuit networks (e.g., SNA) using power maximization as the performance index. The decentralized approach to optimum network behavior yields, among other results, characterization of fair global objective functions, and optimal decentralized greedy network control algorithms. The main conclusion of this paper is that Pareto-optimality methods can be successfully used to develop optimal decentralized behavior algorithms where a centralized approach is (sometimes provably) not applicable.