Computational performance bounds for Markov chains with applications

Abstract

For Markov chains exhibiting translation invariance of their transition probabilities on polyhedra covering the state space, we develop computational performance bounds for key measures of system performance. Duality allows us to obtain linear programming performance bounds. The Markov chains considered can be used to model multiclass queueing networks operating under affine index policies, a class of policies which subsume many that have been proposed.