A framework for resource management of VPLS connections over MPLS core networks

Virtual Private LAN Service (VPLS) is a Layer 2 (L2) service that emulates Ethernet LAN across a WAN. It differs from conventional L2 Virtual Private Networks (VPNs) by connecting customers through a multipoint network. One solution to provide the service is through an Multi-Protocol Label Switching (MPLS) core that supports Traffic Engineering (TE). However, TE functionality is limited to Point-to-Point (P2P) paths, while supporting VPLS requires point-to-multipoint traffic forwarding. This thesis presents an integrated framework for resource management of VPLS connections over MPLS networks. The framework consist of a group of interdependent algorithms to manage resources over all network levels. At the network level, we propose a single path and multi-path critical link Point-to-MultiPoint (P2MP) routing algorithms. The algorithms compute paths such that the links that are important for routing the traffic of other source-destinations sets are avoided as much as possible. Evaluating the performance of the algorithms revealed that they are capable of doubling the number of admitted requests of two outstanding algorithms reported in the literature.; At the path level, we propose a delay partitioning algorithm that simultaneously partitions multiple end-to-end Quality of Service (QoS) requirements. The proposed algorithm, Distribution Based Delay Partitioning (DBDP), optimally partitions the end-to-end delay and the logarithm of the end-to-end violation probability into the link delays and the logarithm of the link violation probabilities, all in a manner that minimizes the upper bound of the violation probability. Extensive simulation shows that DBDP is able to admit around two times the number of connections admitted by two QoS partitioning algorithms used for comparison.; At the link level, we design two Measurement Based Admission Control (MBAC) algorithms to reserve link resources. The algorithms reserve resources based on the measurement of delay rather than measurement of bandwidth, which has been the norm in existing algorithms. As a result, the algorithms are capable of meeting the distinct delay violation probability requirements of each VPLS connection. Simulation reveals that the proposed algorithms increase the number of admitted connections by about two to three times that of deterministic algorithms.