Optimized Traffic Load Distribution in MPLS Networks

Multiprotocol label switching (MPLS) has been developed by the IETF in order to support traffic engineering for a balanced and higher network-wide utilization of resources as a main objective. Meanwhile MPLS has been extended towards a generalized common control plane (GMPLS) for resource provisioning including optical networks. Traffic demands for each source to destination pair can be directed through a MPLS network on predefined paths (LSP: label switched path), whereas routing in IP networks is based on the shortest path first principle and does not allow to establish direct control of the load balance with a per flow or per demand granularity. We investigate optimization algorithms that compute a LSP design for a given traffic matrix with regard to the following goals:(i)Minimize the maximum link utilization corresponding to the considered path design and(ii)minimize the length of paths without affecting the first goal (i).We evaluate the obtainable traffic engineering gain using a heuristic algorithm, whose performance is compared to bounds obtained by linear programming and the max-flow-min-cut principle. The implications of MPLS traffic engineering in the context of network planning, failure recovery and quality of service (QoS) provisioning are outlined.