Ex-Post Internet Charging

Pricing Internet bandwidth is of growing importance as the Internet grows and matures. While many residential consumers have "flat rate" Internet pricing, the pricing policies facing campus, regional, and metropolitan-area networks are becoming more complex. While it may be possible to price every bit or packet of data, this pricing policy requires significantly additional accounting and billing overhead to make it practical. Instead, many Internet Service Providers (ISPs) are using a form of aggregate statistics to determine prices. This class of pricing policies---wherein the pricing algorithm is determined ex-ante but the charge is calculated after the fact---we call "ex-post" because the charges are determined after the traffic has been sent. In this paper, we construct a framework that identifies the mechanisms and goals of Internet pricing. We use this framework to argue that charging models that have better user accountability and are likely to be implemented could achieve these goals through better network management. Expanding on the potential for ex-post charging, we construct a general form of an ex-post charging model. This model determines prices using metrics describing both the utilization and burstiness of network traffic. We then provide examples of the general form of the expost pricing model using three different ways of characterizing utilization and burstiness: 1) effective bandwidth, 2) token buckets, and 3) using first and second moments of the traffic distribution and the moving average of the traffic distribution. Finally, we discuss the implications of ex-post Internet charging. We discuss the granularity or sampling rate that is reasonable given the cost of collecting traffic data and the need for accurate representation of the traffic using aggregate statistics. We also explore the possibility of new business opportunities for ISPs to use ex-post Internet pricing to better manage their Internet Protocol (IP) networks and compete in the market.