Dynamic Preemption Call Admission Control Scheme Based on Markov Decision Process in Traffic Groomed Optical Networks

Abstract

In optical grooming networks, the capacity fairness issue can be resolved by utilizing a call admission control (CAC) mechanism. Existing CAC schemes are generally based on one of four different techniques, namely, static bandwidth reservation, static threshold setting, mathematical statistics, and Markov decision processing without buffer implementation (NB). However, irrespective of the technique used, a trade-off exists between the network fairness and the network throughput. In our previous work, a conditional-preemption CAC (CP-CAC) mechanism was proposed to increase the network throughput while simultaneously maintaining the fairness. However, a CP-CAC mechanism considers only the blocking probability at particular instants of preemption. This paper proposes the use of a dynamic-preemption call admission control scheme (DP-CAC) to decide whether or not to preempt existing requests based on the optimal policy derived from a Markov decision process. Similar to CP-CAC, the DP-CAC method is also based on a dynamic threshold setting concept and is implemented using a single connection buffer and an associated set of virtual indicators. The simulation results show that compared to the CP-CAC mechanism, the proposed DP-CAC further improves the network throughput without sacrificing the fairness. Additionally, the average waiting time induced by the buffer implementation for DP-CAC is just 0.23 time units shorter compared to 0.25 for CP-CAC. Finally, it is shown that the proposed method also ensures fairness in a variety of common network topologies including 6 × 6mesh-torus, NSF, and Cost239.