Joint routing and rate assignment in MPLS based networks

The problem of jointly optimizing rate assignment (admission control) and routing with multiple quality of service (QoS) classes is formulated in this paper. A multiobjective lexicographical approach was used to force the higher priority classes to have proportionally smaller packet loss rates and delay variations and higher admitted rate. The lexicographic approach (called the virtual input method) uses the concept of virtual input rates that are shown to be necessary to obtain categorical differences in QoS between classes. This algorithm is then compared to an ad hoc iterative algorithm (called the iterative method) which optimizes delay and rate assignment on an alternating basis. M/M/1/b queues are used to model switching delays in the MPLS based networks. The average packet loss and switching delay in the network are shown to be closely related and the average packet delay is used in the main objective for the optimization. The joint optimization is equivalent to the optimal routing problem (ORP). Since the ORP with packet loss is non-convex, a reduced problem (R-ORP) is formulated by assuming no packet loss at intermediate switches. Numerical results show that the virtual input method produces excellent results for network wide optimal joint assignment.