QoS Routing and Management in Backbone Networks

Introduction

The Internet supports a growing number of services ranging from legacy web browsing to the more recent online gaming or voice over IP. Among these services, some emerging network applications, such as IP television or video on demand, have stringent requirements in terms of guaranteed endtoend performance. These requirements translate into multiple quality of service (QoS) constraints related to, for example, delay, jitter, bandwidth, and packet loss. Network operators use several QoS mechanisms to protect critical applications and to ensure that proper resources are allocated for each service. Soldatos et al. (2005) classify these mechanisms depending on their operation level. For example, scheduling and queue management operate at packet scale in the Differentiated Services (DiffServ) architecture described in (Blake et al., 1998), whereas resource reservation and signaling operate at flow scale in the Integrated Services (IntServ) architecture standardized in (Braden, Clark, & Shenker, 1994), and Traffic Engineering (TE) operates at network scale. In particular, QoS aware routing is essential when ensuring that each flow follows a path with appropriate performance. For instance, delay- and losssensitive traffic should typically be routed on paths with short delays and low packetloss rates.

The level of QoS experienced by end-users depends on the performance of the endto-end path traversed by service flows. Thus, the successful delivery of services with stringent QoS requirements, like realtime multimedia services, requires QoS to be enforced on all the network segments crossed by service flows. However, varied problems must be addressed depending on the considered segment, and thus, network operators deploy diverse QoS techniques. For example, bandwidth is usually a scarce resource in access networks; hence QoS mechanisms are required in this segment to guarantee that appropriate resources are available for every service. In backbone networks, network operators commonly favor practices related to overprovisioning over sophisticated QoS techniques. However, overprovisioning requires a continuous increase in network capacity in order to cope with traffic evolution or the required level of QoS. As a result, network operators are increasingly interested in efficient network management techniques in backbone networks to delay their investments while maintaining QoS. The present chapter describes such techniques and their application in inter-domain networking.

An Autonomous System (AS) is a set of routers and hosts administrated with homogeneous policies by a network operator. An operator can divide its network into several ASes for administrative reasons. ASes and smaller routing areas are often named domains. The Internet relies on the interconnection of operator's backbone networks, that is, of their domains. The endtoend QoS performance experienced by flows crossing several domains depends on how the packets are processed inside each traversed domain. In particular, efficient traffic management is required inside every domain to protect critical services as well as to control network costs. Numerous studies propose solutions for QoS routing and management inside a single domain, based on enhancements of IP routing or on the MultiProtocol Label Switching (MPLS) architecture. Wang et al. (2008) present a good overview of these solutions.

It is important to note that, in the Internet, connectivity for flows crossing several domains usually relies on the cooperation of interconnected entities with competing interests. In particular, QoS policies often differ from one AS to another and every AS typically takes unilateral routing and networkmanagement decisions, even though these decisions may directly affect other domains. This absence of coordination complicates the provision by an operator of performance guarantees for packets once they leave its own network. Therefore, interdomain traffic flows require mechanisms for QoS routing and traffic management that involve several domains. Specifically, recent work calls for more collaboration among domains, for example in (Mahajan, Wetherall, & Anderson, 2004; J. P. Vasseur, Zhang, Bitar, & Le Roux, 2009), to enhance network performance and to provide support for delivering endtoend QoS. Therefore, the main purpose of this chapter is to describe the emerging management models for the future Internet and, in particular, their application for interdomain traffic management.