A Linear Algorithm for Reliable Predictive Network Control
Richard Schoeffauer, Gerhard Wunder · 2018
This paper introduces a predictive and reliable control approach for network scheduling and routing for arbitrary topologies in discrete time, packet level formulation. We assume existence of a Markov chain, governing the behavior of the network links, so that our controller can optimize its decisions over a finite time horizon. Furthermore, we define a notion of reliability in order to facilitate meaningful forecasts of individual packet delays which can be used by the recipient network agents. We make a point to formulate the optimization problem as a linear program for the purpose of its fast execution in practice. With extended simulations, we demonstrate the gains in performance over the well known MaxWeight policy.