Load Balancing by MPLS in Differentiated Services Networks

Multi Protocol Label Switching (MPLS) as a new technique assigns a short label to each packet at the ingress node of MPLS network and packets are forwarded according to these labels. The capability of MPLS of explicit routing as well as of splitting of the traffic on several paths allows load balancing. The report concentrates on two previously known approximations of the minimum-delay algorithm presented by Gallager. The first approximation defines the paths using the LP-optimization and after that allocates traffic using the NLP-optimization. The second approximation divides traffic into parts and routes them consecutively using Dijkstra’s algorithm. Using these load balancing algorithms from the literature as a starting point, the main goal of this report is to develop optimization algorithms that differentiate classes in terms of mean delay. The differentiation is achieved by the use of both routing and WFQ-scheduling. Both optimal and approximative algorithms are developed for the joint optimization of the WFQ-weights and routing. The load balancing algorithms and the algorithms to differentiate classes are implemented and tested. As a result it is found that the use of the approximations simplifies the optimization problem but still provides results that are near to optimal. The algorithms that try to differentiate traffic classes by both routing and WFQ-scheduling provide the best performance.