Interconnected communication networks provisioned selfishly.

摘要

The Internet is modeled as a collection of Internet Service Providers (ISPs) that exchange IP traffic through transit and peering agreements, using the traditional models for multi-commodity flow networks. Each ISP provisions the cheapest possible network to meet an exogenous level of demand. Transit prices and the price of bandwidth are also exogenous. In this context, a non-cooperative game among ISPs arises whose Nash equilibrium characterizes the topology of the resulting network. This network is at least as expensive to provision as the network that ISPs would provision if they cooperated to reduce overall provisioning costs still using transit and peering agreements to interconnect, which we call the optimal network. The extra provisioning cost of a Nash network relative to the cost to provision any optimal network constitutes an inefficiency cost. We call cost of anarchy to an upper bound on the inefficiency cost over all possible Nash networks.; We compute the cost of anarchy and we show how a lower and tight bound to it relates to the level of economies of scale in the price of bandwidth. Currently this lower bound is at 25%. We show that if only peering agreements are allowed there is no inefficiency cost, but when transit agreements are available inefficiency cost arises. On the other hand, transit agreements reduce overall provisioning costs because allow for aggregating traffic better. We provide an example in which there is a Nash network strictly more expensive than any optimal network even when ISPs choose transit prices, which shows that the market for provisioning interconnected communication networks is inefficient. We also show that there exists a set of prices of transit that make at least some optimal networks Nash networks, but it is not guaranteed that ISPs will choose these prices and even if they do so it is still possible that the network ends up in a Nash configuration that is more expensive to provision than any of the optimal networks. We conclude with a discussion of the difficulties that the regulator would face if it were to intervene to help mitigate the sources of these inefficiencies.