Inter-domain traffic management in an evolving Internet peering ecosystem

Abstract

The operators of the Autonomous Systems (ASes) composing the Internet must deal with a constant traffic growth, while striving to reduce the overall cost-per-bit and keep an acceptable quality of service. These challenges have motivated ASes to evolve their infrastructure from basic interconnectivity strategies, using a couple transit providers and a few settlement-free peerings, to employ geographically scoped transit services (e.g. partial transit) and multiplying their peering efforts. Internet Exchange Points (IXPs), facilities allowing the establishment of sessions to multiple networks using the same infrastructure. IXPs have hence become central entities of the Internet. Although the benefits of a diverse interconnection strategy are manifold, it also encumbers the inter-domain traffic engineering process, and potentially increases the effects of incompatible interests with neighboring ASes. To efficiently manage the inter-domain traffic under such challenges, operators should rely on monitoring systems and computer supported decisions. This thesis explores the IXP-centric inter-domain environment, the managing obstacles arising from it, and proposes mechanisms for operators to tackle them. The thesis is divided in two parts. The first part examines and measures the global characteristics of the inter-domain ecosystem. We characterize several IXPs around the world, comparing them in terms of their number of members and the properties of the traffic they exchange. After highlighting the problems arising from the member overlapping among IXPs, we introduce remote peering, an interconnection service that facilitates the connection to multiple IXPs. We describe this service and measure its adoption in the Internet. In the second part of the thesis, we take the position of the network operators. We detail the challenges surrounding the control of inter-domain traffic in the Internet environment previously described, and introduce an operational framework aimed at facilitating its management. Subsequently, we examine methods that peering coordinators and network engineers can use to plan their infrastructure investments, by quantifying the benefits of new interconnections. Finally, we delve into the effects of conflicting business objectives among ASes. These conflicts can result in traffic distributions which do not satisfy the (business) interests of one or more ASes. We describe these dissatisfactions, differentiating their impact on the ingress and egress traffic of a single AS. Furthermore, we develop a warning system that operators can use to detect and rank those conflicts. We test our warning system using data from two real networks, where we discover a large number of traffic flows that do not satisfy the interest of operators, thus stressing the need to identify the ones having a larger impact on their network.