A new predictive dynamic priority scheduling in Ethernet passive optical networks (EPONs)

Efficient uplink scheduling in Ethernet passive optical networks (EPONs) is very important for maximizing the network capacity while maintaining the required quality of service (QoS). Several variants of dynamic bandwidth resource allocation have been proposed in recent research literature. However, the available techniques do not fully exploit the elastic properties of the user traffic. In this paper, we explore optimal predictive resource allocation strategies by exploiting the elasticity of QoS-constrained traffic and using the knowledge of traffic patterns of different service classes. We propose a predictive dynamic uplink bandwidth allocation scheme that offers lower access delay and packet loss rate, yet achieves a higher overall network throughput. We formulate a model for determining the traffic burstiness-dependent optimum prediction order that would enhance the quality of prediction with a minimum possible prediction-related processing overhead. We then demonstrate that, in a multi-class access scheduling, with respect to the conventional dynamic allocation strategies, our priority scheduling with judicious prediction of individual traffic classes can enhance the system performance significantly. Our analytic observations are supported by extensive simulation results.