Top percentile network pricing and the economics of multi-homing

Top-percentile pricing is a relatively new and increasingly popular pricing policy used by network providers to charge service providers. In contrast to fixed cost pricing and to pure per-usage pricing, top-percentile pricing has not been studied. Thus the efficient design and operation of networks under top-percentile pricing is not well understood yet. This work studies top-percentile pricing and provides an analysis of the expected costs it inflicts on a service provider. In particular we use our analysis framework to investigate the popular multi-homing architecture in which an Internet Service Provider (ISP) connects to the Internet via multiplicity of network providers. An ISP that uses multi-homing is subject to extra charges due to the use of multiple networks. Important questions that are faced by such an ISP are what is an efficient routing strategy (as to reduce costs) and how large the costs are. We provide a general formulation of this problem as well as its probabilistic analysis, and derive the expected cost faced by the ISP. We numerically examine several typical scenarios and demonstrate that despite the fact that this pricing aims at the peak traffic of the ISP (similarly to fixed cost), the expected bandwidth cost of multi-homing is not much higher than that of single-homing.