MDP routing for multi-rate loss networks

Markov decision theory has been successfully used in adaptive routing in tr aditional circuit-switched networks. When extending Markov decision-based r outing algorithms to future Broadband Integrated Service Digital Networks ( B-ISDNs), the required computational complexity becomes extremely high. In this paper, we propose an approach towards adaptive routing in multi-rate n etworks using a Markov decision theoretic framework which maintains low com putational complexity while still providing quite good routing information. In this approach, each link, based on PASCAL distribution, is modeled as a one-dimensional birth-death process to reduce the state space size and a p olicy iteration method from Markov decision theory is iteratively applied t o achieve better network performance. Our results show that routing algorit hms based on this approach yield better performance than least-load path ro uting (LLP) without incurring any significant increase in computational com plexity. (C) 2000 Elsevier Science B.V. All rights reserved.