Algorithms for Traffic Engineering

Traffic Engineering in IP networks with MPLS In the past, routing protocols for IP networks (e.g. BGP, IS-IS, OSPF) used the shortest path to forward the traffic on network links, each of which is assigned by a routing weight. This may lead to an unbalanced distribution of the traffic in the network with high utilization of some links while others remain underutilized. Network planing and an adaptive provisioning of bandwidth to each link can improve the situation only to a limited extend, since there are constraints on the network topology and since dynamically changing network loads cannot be adequately handled in this way. Multiprotocol label switching (MPLS) [6] has been introduced by the IETF with one of its main goals to support traffic engineering for an appropriate distribution of the traffic over a service provider network. Approaches for traffic engineering in pure IP networks have also been studied by fitting the routing weights such that shortest paths meet a predefined distribution of the traffic on the links, see e.g. [2]. MPLS facilitates traffic engineering by explicitly providing one or several paths for each traffic demand from a source to a destination in a concept of label switched paths (LSP) and enables to measure the traffic flowing on each LSP [6]. Thus the traffic matrix of source-to-destination flows in a network becomes available from a full mesh of LSPs, which is more difficult to obtain by IP level measurement.