IEEE Communications Magazine • October 2007 94 0163-6804/07/$20.00 © 2007 IEEE This work was supported in part by Cisco Systems under their URP program. 1 The term routing domain, as used in this article, refers to a network under a single administra- tion with common poli- cies. A routing domain, therefore, may be an autonomous system or a routing area within the autonomous system. INTRODUCTION The emergence of mission-critical and other multimedia applications such as voice over IP (VoIP), videoconferencing, e-commerce, and vir- tual private networks (VPNs) has translated into stringent real-time quality of service (QoS) requirements for carrier networks. A key factor in meeting the QoS requirements of such appli- cations is the ability to route traffic along explicit paths computed through constraint-based rout- ing. The destination-based shortest path routing paradigm employed in IP routing does not sup- port routing network traffic along explicit paths. However, the emergence of label switching paradigms such as multiprotocol label switching (MPLS) and generalized MPLS (GMPLS) has overcome this limitation by presenting the ability to establish a label switched path (LSP) between two points in an IP network. This ability to do traffic engineering (TE) using MPLS maintains the flexibility and simplicity of an IP network while exploiting the advantages of an asyn- chronous transfer mode (ATM)-like connection- oriented network. Ingress routers of an MPLS network classify packets into forwarding equivalence classes (FECs) and encapsulate them with labels before forwarding them along precomputed paths. The path a packet takes as a result of a series of label switching operations in an MPLS network is called a label switched path (LSP). For provid- ing connectivity from ingress to egress, an LSP traverses a sequence of links that could be either physical links between adjacent nodes or logical links between two nodes that may or may not be adjacent. LSPs may be computed by using a con- strained shortest path first (CSPF) algorithm that essentially finds a shortest path between two network nodes subject to constraints such as maximum delay, minimum available band- width, and resource class affinity. Thus, the constraints dictate how the traffic should be engineered through the network; for this rea- son, the paths computed under the constraints are called traffic engineered (TE) paths. For the computation of a TE path an LSP would traverse, a CSPF algorithm uses TE informa- tion, such as the remaining or reserved band- width along a logical or physical link, advertised throughout the network. Resources along a computed TE path are reserved during label distribution, using protocols such as Constraint- Based Routing Label Distribution Protocol (CR-LDP) and Resource Reservation Protocol with TE (RSVP-TE) [3]. Existing solutions for traffic engineering in MPLS and GMPLS networks are mostly limited ABSTRACT For label switched networks, such as MPLS and GMPLS, most existing traffic engineering solutions work in a single routing domain. These solutions do not work when a route from the ingress node to the egress node leaves the rout- ing area or autonomous system of the ingress node. In such cases, the path computation prob- lem becomes complicated because of the unavail- ability of the complete routing information throughout the network. This is because service providers usually choose not to leak routing information beyond the routing area or AS for scalability constraints and confidentiality con- cerns. This article serves two purposes. First, it provides a description of the existing and ongo- ing work in interdomain TE within the IETF. This information is currently found in various Internet drafts and has not yet been collectively presented in a single document. To this end, a summary of both existing path computation architectures — PCE-based and per-domain — is provided. Second, it compares two per-domain path computation schemes in terms of the total number of LSPs successfully placed and average number of domains crossed, without assuming availability of complete topology information. We notice that the two per-domain path compu- tation schemes, proposed in [1, 2], have compa- rable path computation complexities and setup latencies. ADVANCES IN CONTROL AND MANAGEMENT OF CONNECTION-ORIENTED NETWORKS Faisal Aslam, Zartash Afzal Uzmi, and Adrian Farrel Interdomain Path Computation: Challenges and Solutions for Label Switched Networks