Discounted integral priority routing for data networks

Abstract

A Discounted Integral Priority (DIP) packet routing algorithm is presented. The method is derived for the network flow model of packet routing used for the derivation of backpressure type methods. Unlike backpressure type methods, DIP routing is designed to reduce the queue lengths rather than simply stabilize them. Our work leverages time discounted integral control to generate an adaptive packet routing algorithm which significantly outperforms its optimization motivated counterparts. Connections are drawn with stochastic heavy ball methods which allow implementation of a decaying stepsize. Stability proofs are presented for a stochastic heavy ball variant of the Discounted Integral Priority routing algorithm with a decaying step size. Our numerical experiments implement Discounted Integral Priority Routing with a unit step size and demonstrate fast convergence and significantly smaller steady state queue backlogs as compared with Soft Backpressure and Accelerated Backpressure.