BGP-based Interdomain Traffic Engineering

Preamble In a few years, the Internet has rapidly evolved from a research network serving a handful of users to a huge interconnection of about 350 million hosts (June 2005 [ISC05]). In this way, the Internet is the largest distributed system ever built. The Internet is organized in a multitude of administratively independent networks called domains or Autonomous Systems (AS). For example, an AS can be an In-ternet Service Provider (ISP), a University campus or a corporate network. At the time of this writing, there are more than 21,000 ASs in the Internet [Hus06]. Over this huge infrastructure, there is a growing trend to deploy new applications such as the transmission of Voice or Video over IP and new services such as Virtual Private Networks (VPNs). These new applications and services require better or strict guarantees of quality while the Internet has been designed to provide a best-effort service. This evolution puts a lot of pressure on the ISPs for whom the ability to offer better than best-effort service or support tight Service Level Agreements (SLAs) [FE04] is a key differentiating factor. Network engineers rely on Traffic Engineering (TE) to adapt the configuration of their network in order to support the evolution of the traffic demand and the customer SLAs. Traffic Engineering is defined by the IETF TE Working Group as the process of evaluating and enhancing operational IP networks performance [ACE + 02]. The objectives of Traffic Engineering can be summarized in avoiding congestion, providing resilience and supporting Quality of Service (QoS). Most of the Traffic Engineering complexity comes from hop-by-hop destination-based IP forwarding, i.e. each router on the path selects the next router to forward the packet to based on the packet destination only. There is no way to explicitly determine the path followed by IP datagrams to reach their destination. Moreover, the routing decisions are taken in a distributed manner by each hop along the path. One of the main difficulties of Traffic Engineering comes thus from routing. Inside a single domain, routing is done thanks to link-state protocols such as IS-IS or OSPF. From the perspective of Traffic Engineering, these protocols have the twofold advantage of propagating information on the whole topology and optimizing a single global objective (least-cost-path). Intradomain Traffic Engineering is a well understood problem and solutions exist [FT00, FRT02]. In contrast, when Traffic Engineering has to be performed over the boundaries of multiple …