Self-adaptive dynamic bandwidth allocation for GPON

Upstream dynamic bandwidth allocation for Gigabit Passive Optical Networks (GPONs) is an important design problem since the shared uplink carries many bursty streams with different quality of service (QoS) requirements, and hardware constraints necessitate infrequent update processing. We present self-adaptive and efficient bandwidth allocation solutions, where the optical line terminal (OLT) allocates bandwidth among the optical network units (ONUs) in an efficient and responsive manner. In status reporting implementations, where the ONUs report their buffer status, we present algorithms which maximize throughput and minimize delay based on adaptive learning of traffic demands. For ONUs that do not report status information, and thus are simpler and less expensive, we develop a method in which the OLT locally observes only a sample path of the bandwidth used by the ONUs. Infinitesimal perturbation analysis and stochastic fluid models are used to learn the gradients of the ONUs' buffer contents, which are then applied in an adaptive optimization of the bandwidth allocation.