Bayesian Estimation and Prediction-Based Dynamic Bandwidth Allocation Algorithm for Sleep/Doze-Mode Passive Optical Networks

In this paper, we propose a new Bayesian estimation and prediction-based Just-In-Time dynamic bandwidth allocation algorithm. The algorithm exploits the sleep and doze capabilities of a 10-Gbps vertical-cavity surface-emitting laser optical network unit (ONU) for improved energy-efficiency across all network loads. The main objectives of the proposed algorithm are to: (a) effectively estimate the average interarrival time of packets for sleep and doze control; (b) transition the ONUs from sleep or doze modes into active mode Just-In-Time to receive packets from the optical line terminal (OLT); and (c) minimize the additional queuing delays introduced by the sleep- and doze-mode operations. The algorithm uses Bayesian estimation to effectively estimate the average interarrival time of packets at the ONUs. The OLT determines the sleep or doze time of an ONU based on the estimated average interarrival time information. Multipoint control protocol messages such as GATE and REPORT are used to transition the ONUs from sleep or doze mode into active mode Just-In-Time to receive packets from the OLT. The prediction mechanism of the proposed algorithm, deployed at the OLT, predicts the number of packets accumulated during the sleep or doze time and minimizes the additional queuing delay introduced by the sleep- and doze-mode operations. Simulation results indicate that the proposed algorithm results in 63% of energy-savings, compared to an always active ONU without sleep/doze operation. Our results further show that using the prediction mechanism, delay values can be reduced by 13% at low network loads with only an increase of 0.01% in average power consumption at the ONU.