1Economics of Network Pricing with Multiple ISPsSrinivas Shakkottai R. SrikantAbstract—In this paper we examine how transit and cus-tomer prices are set in a network consisting of multipleISPs. Some ISPs may be geographically co-located so thatthey compete for the same set of end users. We examinethe existence of equilibrium price strategies in this situationand show how positive profit can be achieved using threatstrategies. It is shown that if the number of ISPs compet-ing for the same customers is large then it can lead to pricewars. ISPs that are not geographically co-located may notdirectlycompete for users, but are nevertheless involvedin anon-cooperativegame of setting accessand transit prices foreach other. We study howsuch ISPs are linked economicallythrough transit ISPs by considering a multi-stage game. Wealso consider the economics of private exchange points andshow that they could become far more wide spread thenthey currently are.I. INTRODUCTIONIndex Terms— Internet economics, pricing, repeatedgames, Nash equilibrium.The Internet is a heterogeneous body of privatelyowned infrastructure. Roughly speaking, it consists oftwo types of networks: (i) densely meshed networks ingeographically localized regions which specialize in pro-viding consumers with connection points to the networkand (ii) networks traversing large geographical distanceswhich provide connectivity between the local networks[1]. All the networks are connected by means of an identi-cal (or at least inter-operable) protocol stack agreed uponby the Internet Engineering Task Force (IETF). Figure 1illustrates the Internet as it looks like today. There arelocal Internet Service Providers (ISPs) providing servicesin small regions and transit ISPs which transfer data be-tween local groups. The groups exchange data with eachother at Network Access Points (NAPs). Each transit ISPwould have a point of presence in each of the regions thatit is interested in providing transit service. A NAP may beprovided as a public resource or may be privately funded,with each constituent paying some charge to the NAP fortraffic exchange [2]. In the figure, the small “clouds” showlocal ISPs as well as the points of presence of transit ISPs.Research supported in part by DARPA grant F30602-00-2-0542 andNSF grant ANI-0312976Department of Electrical and Computer Engineering and Coordi-nated Science Laboratory, University of Illinois at Urbana-Champaign,Emails: (sshakkot,rsrikant)@uiuc.eduThe large transit clouds represent multiple transit (eitherin parallel or in series) ISPs linking the regions. The NAPsare shown as routers linked by a common media.Local ISPsLocal ISPsLocal ISPsMultiple Transit ISPsMultiple Transit ISPsNAPNAPNAPFig. 1. The structure of the Internet today consisting of local andtransit ISPs.A topic of current interest is the development of goodpricing models for the Internet. Most ISPs currently em-ploy flat rate pricing, i.e., end-users pay a fixed sum ofmoney every month for unlimited access (most broadbandproviders and dial-up services use this scheme). Others,following the telephony model, price the time spent con-nected to the Internet (some dial-up services are pricedthis way). They charge an hourly rate to end-users. Stillothers charge based on actual bytes transferred (manyAustralia/New Zealand ISPs use this scheme [3]). Pricingmodels for Internet economics have been discussed in [1],[4], [5]. The authors discuss how pricing may be used toprovide different qualities of service and argue that differ-entiated pricing for different types of data transfers mightbe a good approach.While some aspects of the traditional telephony model[6] may provide a starting point for economic analysisof the Internet, one must keep in mind that flows on theInternet are usually biased very heavily in one direction,whereas telephone calls impose equal loads in both direc-tions. The transfer of data in the Internet may be assumedto be unidirectional in most cases. We assume that trafficoriginates at websites and terminates at end-users. In [7]this model is used to find a Nash equilibrium solution forprices charged to websites and end-users.2There is also another difference in economic interactionbetween players on the Internet as compared to the tele-phony model. In the telephony model a single party, usu-ally the origin, funds the transport of traffic on the com-plete path from sender to receiver. Thus, each receivingISP in the path of the flow charges each transmitting ISPa termination charge. This is called bilateral settlement.A bilateral settlement in which two ISPs reciprocally settheir termination charges to zero is called a peering ar-rangement.In the Internet, there is also a hierarchy of providers[1]. The sender does not fund the end-to-end transportof traffic, only a part of the path is sender funded. Oncethe traffic passes beyond the sending ISP’s service fundeddomain, the receiver implicitly assumes funding respon-sibility for the traffic and the second part of the completetransport path is funded by the receiver. Thus, the set ofconnectivity paths within the Internet can be seen as a col-lection of path pairs, where the sender funds the initialpath component and the receiver funds the second termi-nating path component. The hierarchical model of eco-nomic interaction is illustrated in Figure 2 for a particulardirection of traffic. It is seen that higher tiers charge re-gardless of traffic direction. The higher level ISP is said toprovide a transit service to the lower levels. However, thehierarchy is not rigid. Traffic could be exchanged withinlocal groups within a region. ISPs in different regionscould also exchange traffic at private exchanges. Com-panies such as Equinix [8] provide the infrastructure forthe establishment of such exchanges. These are indicatedin Figure 2 by a circle containing a ‘P’. Thus, traffic neednot traverse the whole of the hierarchy. It is for this reasonthat only the Tier-I ISPs (which do not pay transit chargesto any other ISPs) and the local groups (which potentiallyhave to pay for all traffic to and from their infrastructureand compete for the same customers) are well defined.Understanding economic interactions between the differ-ent groups is the main focus of the paper.Several models have been proposed to study peeringversus transit relations [9], [10], [11], [12], [2]. A problemwith some of these models is that they do not take into ac-count the fact that the
View Full Document
Unlocking...