ABSTRACT
This chapter reviews and compares the standard definitions of fairness for a communication network. The max-min fairness is seen to be exactly a realization of the aforementioned two principles of Rawls's Justice theory. Max-min fairness is used widely in networking areas for the purpose of bandwidth allocation, with the aims at allocating as much as possible to poor users, while not unnecessarily wasting resources. The chapter presents a general and centralized algorithm with linear programming computing complexity, termed as max-min programming (MP), for computing the max-min fair allocation in all cases where it exists. It discusses the MP algorithm, which finds the max-min fair vector on any feasible set, if it exists. The chapter proposes efficient off-line optimal algorithms and heuristic-based greedy online algorithms to solve the above problems. It discusses experimental results on FAST transmission control protocol (TCP) by the Linux prototype and compared its performance with TCP Reno, high-speed TCP (HSTCP), and scalable TCP (STCP).
