Least measured cost routing in VP-based ATM networks

We consider the routing problem in a virtual path based ATM network with multiple classes of traffic. The traffic classes are assumed to have identical end-to-end quality of service (QOS) requirement but at different data rates. The concept of effective bandwidth is used to simplify call admission control and to transform the VP-based ATM routing problem into an equivalent multi-rate circuit-switched network routing problem. The Markov decision process (MDP) is known to provide an optimal solution but is computationally infeasible to implement in any realistic network. We propose to measure the relative costs in a Markov decision process directly to simplify the required computation and yet achieve near-optimal routing. The performance of the proposed algorithm is compared with least loaded routing (LLR) and minimum free capacity routing (MFCR). Simulation results show that the Markov decision process with measured relative cost outperforms other routing algorithms.