Abstract

Internet is composed of the inter-connection of tens of thousands of independent networks called Autonomous Systems (ASes). ASes that only provide connectivity to the other ASes are called Internet Service Providers (ISPs). The vast majority of ASes in the Internet are multi-homed with several ISPs to reduce their costs and improve their capacity and robustness. Multi-homing increases the path diversity and thus might enable ASes to choose better quality paths. Unfortunately, the current Internet design does not allow to optimize routing for performance. Indeed, the routes are propagated gradually from AS to AS. If an AS has several routes for a given destination, it can only advertise and use its best one. In addition, the current Internet routing suffers from scalability issues. As a consequence, ASes lack of control on the traffic entering their network.

In this thesis, we propose new techniques to enable performance based interdomain incoming traffic engineering. These techniques allow any AS to control its incoming traffic to follow paths that optimize the efficiency (e.g., minimize the delay or maximize the bandwidth). For that purpose, we first introduce a system that allows an AS to control its incoming traffic in a scalable way. This system relies on the Locator/ID Separation Protocol (LISP). We improve the scalability and the robustness of the LISP control-plane with LISP-Tree. Afterward, we present a system that is able to rank the paths in order to determine the best path according to any arbitrary performance criterion. We call this system IDIPS (for ISP-Driven Informed Path Selection). Finally, we show how to combine LISP, LISP-Tree, and IDIPS to achieve performance based interdomain incoming traffic engineering.