Design of next-generation subscriber access systems based on ethernet passive optical networks (EPON).

摘要

Compared to metropolitan area networks (MANs), subscriber access networks serve a relatively small number of users and therefore are very cost-sensitive. The cost issues have prevented the deployment of optical access networks, creating a bottleneck between high-capacity local area networks (LANs) and MANS.; This dissertation investigates Ethernet Passive Optical Networks (EPON), an emerging optical architecture, optimized for deployment in subscriber access networks. EPONs combine a point-to-multipoint fiber topology, variable-sized Ethernet packets, and a centralized in-band scheduling protocol to deliver subscriber's traffic at low operational cost while guaranteeing service-level agreements and ensuring fairness among the subscribers.; However, EPONs present network designers with several challenges. EPON properties such as significant queue switch-over overhead, large control-plane delay, and limited control-plane bandwidth do not allow easy adaptation of existing scheduling algorithms. To efficiently allocate bandwidth in the presence of the above constraints, we propose a new algorithm called Interleaved Polling with Adaptive Cycle Time (IPACT). IPACT provides rate-proportional fair service and solves the issues of switch-over overhead and control-plane delay by interleaving the grant/request cycles of different nodes.; EPON is expected to be a truly converged network, supporting voice communications, video, real-time transactions, and data traffic. To enable this multitude of applications, we extend the IPACT protocol to support multiple classes of service. We identify network conditions resulting in anomalous network behavior where the queuing delay for some traffic classes increases when the network load decreases (a phenomenon we call light-load penalty). We suggest two optimization schemes, which eliminate the light-load penalty.; To operate in densely-populated areas, EPON must be scalable with the number of subscribers and maintain fairness among all the subscribers. Existing hierarchical scheduling protocols are scalable, but do not provide fairness across all the subscribers. Existing single-level protocols provide fairness, but are not scalable. We present a novel framework called Fair Queuing with Service Envelopes (FQSE), which successfully achieves both goals: it uses hierarchical control messaging to achieve scalability and it employs a concept of a service envelope to achieve farness across the entire subscriber population.