A distributed algorithm for weighted max-min fairness in MPLS networks

We propose a novel distributed algorithm to achieve a weighted max-min sharing of the network capacity. We present the Weight Proportional Max-Min policy (WPMM) that supports a minimal rate requirement and an optional maximal rate constraint and allocates network bandwidth among all aggregates based on a weight associated with each one. Our algorithm achieves this policy for IP/MPLS networks using the RSVP-TE signalling protocol. It uses per-LSP accounting in each node to keep track of the state information of each LSP. It uses a novel explicit bottleneck link strategy and a different control architecture in which we update the control packet in the forward path. Simulations show that these two elements improve substantially the convergence time compared to algorithms designed for ATM networks.