The contents of this legacy page are no longer maintained nor supported, and are made available only for historical purposes.

Interconnection Economics

From 2010-2013, CAIDA performed a study of the economics of Internet interconnection, supported by the NSF grant CNS-1017064, "NetSE-Econ: The economics of transit and peering interconnections in the Internet".

    Significant Results: (cumulative findings)

  • Modeling the transition from a transit hierarchy to a peering mesh (published at ACM CoNEXT, December 2010)

    We developed ITER, an agent-based network formation model. We parameterized ITER using recently reported trends in three factors -- the fraction of traffic sourced by large content providers, the geographical presence of content providers, and peering openness -- which results in two different internetworks which we term the 'hierarchical' and 'flat' Internet. We showed that the recently observed changes in these three factors can transform the Internet ecosystem from a multi-tier hierarchy that relies mostly on transit links to a dense mesh of horizontal interconnections that relies mostly on peering links. Traffic in the 'flat' Internet follows shorter routing paths, especially when each path is weighted by its traffic volume. In the flat Internet, however, strategic peering becomes more important for small and large transit providers; both can be profitable by peering selectively with the largest content providers.

  • A Value-based framework for Internet peering agreements (published at the International Teletraffic Congress (ITC), Sep. 2010)

    We presented a quantitative framework for the creation and evaluation of settlement-free and paid-peering relationships. Our framework is based on the 'value' of a peering link for the participating networks. This value represents the monetary benefit from the peering link, i.e., the amount of money a network saves, that it would otherwise pay to its transit providers. We proposed a paid-peering price that is fair, optimal, and stable, as long as the aggregate value of the peering link for the two networks is positive.

  • Measuring/modeling interdomain traffic matrices

    We used passive flow data from GEANT, the largest academic/research backbone in Europe to measure the statistical properties of the interdomain traffic matrix (ITM). Using this data, we directly measured elements of the ITM routed via the GEANT network. We found that the distribution of traffic sourced by ASes is heavy-tailed, and is either Pareto or LogNormal depending on the source AS. We found significant correlations across different rows of the ITM, mostly due to a few highly popular prefixes. (published at IFIP Networking, Prague, Czech Republic, May 2012.)

    We developed ITMgen, a simple tool for generating synthetic but representative ITMs. ITMgen works at the level of connections, taking into account the relative sizes of ASes, their popularity with respect to various applications, and the relation between forward and reverse traffic for different application types. The necessary parameters for integrating application types and the distribution of the content popularity can be realistically estimated by combining public sources like Alexa that capture traffic trends at a macro level with local traffic sampling (NetFlow, DPI). ITMgen can synthesize ITMs that match real-world measurements closer than the the current state of the art. (published at IEEE International Conference on Communications (ICC), June 2013)

  • A cost model for network traffic (published at ACM SIGCOMM Computer Communications Review, Jan 2012)

    We developed a holistic cost model that operators can use to help evaluate the costs of various routing and peering decisions. Using real traffic data from a large carrier network, we showed how network operators can use this cost model to significantly reduce the cost of carrying traffic in their networks. We showed that adjusting the routing for a small fraction of total flows (and total traffic volume) significantly reduces cost in many cases. We also showed how operators can use the cost model both to evaluate potential peering arrangements and for other network operations problems. 5) Twelve years in the evolution of the Internet ecosystem (published at the IEEE/ACM Transactions on Networking, October 2011)

    We studied the evolution of the Internet ecosystem during the last twelve years (1998-2010). We proposed a method to classify ASes into four types (large transit providers, small transit providers, content/access/hosting providers, and enterprise networks) depending on their function and business type. We also considered the semantics of inter-AS links, in terms of customer-provider (CP) versus peering (PP) relations, and distinguished between the customer, provider and peering role of an AS in each relation. We measured the evolution of the AS-level topology over the last 12 years in terms of growth and rewiring, four distinct economic/business classes of ASes, and customer-provider links. Our findings highlight some important trends, trend shifts, and sketch what the Internet may be heading towards, in terms of topological and economic organization.

  • Modeling peering strategy selection by autonomous networks

    We developed GENESIS, an agent-based network formation model for the Internet. GENESIS captures key factors that affect network formation dynamics: highly skewed traffic matrix, policy-based routing, geographic co-location constraints, and the costs of transit/peering agreements. GENESIS is a computational model that simulates the network formation process and allows us to actually compute distinct equilibria and to also examine the behavior of sample paths that do not converge. We found that such oscillatory sample paths occur in about 10% of the runs, and they always involve tier-1 ASes, resembling the tier-1 peering disputes often seen in practice. GENESIS results in many distinct equilibria that are highly sensitive to initial conditions and the order in which ASes (agents) act. This result implies that we cannot predict the properties of an individual AS in the Internet. However, properties of the global network or of certain classes of ASes are predictable. We found that the peering openness that maximizes the fitness of different network classes (tier-1, tier-2 and tier-3 providers) closely matches that seen in real-world peering policies. (published at IEEE Infocom, Orlando FL, March 2012)

    We measured peering strategies announced by networks participating in PeeringDB, showing that about 70% of the transit providers in that database use an Open peering strategy. We used GENESIS simulations and game-theoretic analysis to explain why transit providers gravitate towards Open peering even though that may be detrimental to their economic fitness. We also identified classes of transit providers that would do better (or worse) under the Open peering regime. Finally, we examine the impact of an Open peering variant where ASes do not peer with direct customers of their existing peers. We find that this simple mechanism that introduces some co-ordination between transit providers produces a cumulative fitness for transit providers comparable to the case where transit providers do not engage in Open peering. (Extended abstract with results from this work published at NetEcon, March 2012. Full paper published at IEEE Infocom 2014.)

  • Characterizing Internet peering policies (published at ACM SIGCOMM Computer Communications Review, Apr 2014)

    We mined data from PeeringDB, an online database where participating networks contribute information about their peering policies, traffic volumes and presence at various geographic locations. Using BGP data to cross-validate three years of PeeringDB snapshots we have archived, we find that PeeringDB membership is reasonably representative of transit, content, and access networks in the larger Internet. We found strong correlations among different measures of network size -- BGP-advertised address space, PeeringDB-reported traffic volume and presence at peering facilities, and between these size measures and advertised peering policies. Historical snapshots of the PeeringDB database revealed evolutionary trends of the peering ecosystem, including geographic expansion and contraction by players, increases and decreases in traffic volume, and shifts toward more restrictive peering.