Optimal granularity of MPLS tunnels

We consider the problem of tunneling MPLS label switched paths (LSPs) over an optical network in which λ-channels (lightpaths) of a fixed capacity are the main mechanism of transport. While any set of LSPs, each LSP having an arbitrary rate, can be tunneled over a given λ-channel as long as the aggregate rate does not exceed the channel's capacity, intuition suggests that this flexibility has a limit. In other words, we believe that it is sufficient for the network to provide a small set of fixed service levels in terms of the LSP data rate. We present an efficient dynamic programming algorithm to obtain an optimal set of service levels given (a) a fixed set of LSPs and corresponding bandwidth requests, or (b) the probability distribution of LSP bandwidth requests. The set of optimal service levels strikes a balance between the conflicting goals of simplicity and performance. Numerical results indicate that with as few as ten service levels, the performance penalty compared to a continuous-rate network is less than 10% of the total amount of bandwidth requested for a wide variety of LSP data rate distributions.