118 connection agreements changed dramatically at the time of the privatization of the Internet, and that at the same time concerns regarding the possibility of anticompetitive interconnection agreements started to be intensely analyzed by competition authorities and competition economists. The costs of providing Internet transport services include the access costs to network resources of the physical layer as well as the costs of switches and routers, the costs for transmission software and the costs for employed staff. These costs are driven by the geographic extent of the network as well as by the bandwidth of the links making up the network.103 Most of these costs are long-run variable costs. The short-run marginal costs for any particular product or service provided over a given infrastructure are close to zero. As is typical for network services, most of the costs involved in Internet transport services are also overhead costs, meaning that they cannot be allocated to the incremental costs of particular products and services. The pricing for Internet backbone services therefore necessarily does not reflect shortrun marginal costs or even long-run incremental costs of the service. In general, the price of a particular product must cover at least the long-run incremental costs of this product. If these are not covered then, from an economic point of view, the product should not be produced. In addition, the entire set of products and services offered must cover all overhead costs of production, that is, all costs which cannot be allotted to the incremental costs of a particular product or service. To cover their considerable overhead costs, ISPs must use pricing strategies that calculate mark-ups on the incremental costs, which allocate the overhead costs to particular products and services according to the price elasticity of demand for these products and services. The elasticity of demand for Internet backbone services depends on the possibilities for substitution. To offer universal connectivity, a network provider can combine the components 1) own network services, 2) network services from peering partners, and 3) network services from transit partners. These components are interchangeable to a degree and the amount used will depend on the costs of each of these services. With network interconnection, an ISP can avoid building out its own network to particular regions and customer groups, instead profiting from the network investments made by the interconnection partners. The following two subsections look at the pricing of transit and peering interconnection respectively. 7.2.2 The price for transit interconnection The main difference between interconnection by a transit contract and interconnection by peering is the degree of coverage of the Internet offered by either transit (complete coverage) or peering (only the direct customers and transit customers of 103 Transmission links can be leased. Leased lines are priced by their length and by the capacity of the pipe. The larger the extent of the network, the more switches and routers are needed. The costs for employees also rise with the geographical extent of the network. 119 the peering partner are reached).104 Furthermore, in a transit relationship one party pays the other party for delivery of its data traffic from and to the rest of the Internet. Transit services can be bought from transit givers at any available point of network interconnection. Transit fees must cover at least the costs of the network resources into which a transit provider has invested to be able to offer transit services in addition to the interconnection fees which the transit giver pays to third parties for the termination of the transit takers traffic. A transit giver must further try to cover some of its overhead costs by a mark-up on the incremental costs of providing the transit service. In practice, transit fees are typically two-part tariffs. A flat-fee is charged, which varies depending on the bandwidth of the pipe connecting the two networks and the arranged peak throughput of data on this pipe. A variable fee is charged for traffic in excess of this agreed level, generally charged on the basis of Megabits per second (Mbps). The transit giver therefore has the opportunity to price-differentiate in the market for Internet backbone services. A transit taker will pay a lower average price if more traffic is sent via a particular interconnection and if the amount of traffic sent over this interconnection is correctly predicted beforehand. For inelastic demand, often characterized by a short-term need to shift traffic to a new transit provider, the average price paid will be higher. Such price differences should not be taken as evidence of significant market power by the transit giver. The need to cover the substantial overhead costs in this market force the transit giver to find ways of implementing surcharges on marginal costs, that can cover the overhead costs of production. 7.2.3 The implicit price for peered interconnection Peering generally involves no monetary compensation for using the peering partner’s network. There is, however, an “implicit price for peered interconnection” (Elixmann and Scanlan, 2002: 47), namely the cost of providing the reciprocal service for one’s peering partner. In order to understand which interconnection services ISPs consider equal, one must understand how traffic exchange among peering partners is organized. Generally, peering partners interconnect their networks at several dispersed geographic locations. The practice which decides where traffic is handed off to the peering partner has tellingly been called “hot potato routing” (Kende, 2000: 5ff.). Traffic is passed on to the peering partner at the nearest point of exchange to the origin of the communication.105 The bits of data are then transported to the receiving user on the receiving user’s network. Figure 7.1 illustrates this principle. ISP 1 and ISP 2 are interconnected by Routers R1, R2, and R3 at three different geographical locations. To deliver a data stream from a user connected to ISP 1 via Host 1 (H1) to the user connected to ISP 2 via 104 See section 3.5.1 above. 105 This convention also makes sense, considering that the physical geographic location of the receiving host is known only to the home network of the receiving host.