A Fair and Dynamic Load-Balancing Mechanism

Abstract

The current data network scenario makes Traffic Engineering (TE) a very challenging task. The ever growing access rates and new applications running on end-hosts result in more variable and unpredictable traffic patterns. By providing origin-destination (OD) pairs with several possible paths, load-balancing has proven itself an excellent tool to face this uncertainty. Most previous proposals defined the load-balancing problem as minimizing a certain network cost function of the link's usage, assuming users would obtain a good performance as a consequence. Since the network operator is interested in the communication between the OD nodes, we propose instead to state the load-balancing problem in their terms. We define a certain utility function of the OD's perceived performance and maximize the sum over all OD pairs. The solution to the resulting optimization problem can be obtained by a distributed algorithm, whose design we outline. By means of extensive simulations with real networks and traffic matrices, we show that our approach results in more available bandwidth for OD pairs and a similar or decreased maximum link utilization than previously proposed load-balancing schemes. Packet-level simulations verify the algorithm's good performance in the presence of delayed and inexact measurements.