Design and evaluation of load balancing algorithms in P2P streaming protocols

Abstract

In this paper, we consider a central problem in P2P content distribution: given a set of neighboring peers connected to each other to exchange content, how they can optimally negotiate the rate in helping each other so as to achieve maximum overall throughput and minimize the content server’s load. We call this the “load balancing problem” of a P2P system. By providing an abstract formulation of the optimization problem, we contrast this problem with the network congestion control problem, both in terms of parallels and differences. We then proceed to study several versions and aspects of this problem: (a) request allocation, (b) neighbor selection, and (c) server load minimization. We have proposed and evaluated several practical algorithms that are discrete (window-based), distributed (without needing global information), and adaptive.