Network level capacity planning with efficiently computable global optimum

Abstract

A novel, network level capacity planning model is presented with the following key properties. (1) The link level traffic and blocking model is very general. It allows multiclass traffic and the individual bandwidth demands may aggregate in complex, nonlinear ways. The conventional assumptions, such as Poisson arrivals or exponential holding times, are not required, thus allowing non-standard behaviour patterns, including self-similarity. (2) On the network level, the stochastic behavior of the links is not required to be independent. Thus, we drop the traditional link independence assumption, because it is questionable in a number of modern networking situations. (3) Surprisingly, despite the very general setting, the model yields a convex optimization problem, in which the global optimum of the capacity assignment under end-to-end blocking constraints can be found efficiently (in polynomial time).