Abstract: The existing research on Bandwidth Management or Dynamic Bandwidth Allocation (DBA) algorithms for standardized single channel EPONs as well as emerging multiple channel EPONs are discussed and classified into a clear taxonomy. The Bandwidth Man agement problem for EPONs is extended to multiple channels and partitioned into grant sizing and grant scheduling. Grant scheduling is further broken into two dimensions: (1) time and (2) channel assignment. Using static channel assignment results in a simple algorithm but it is not efficient in its utilization of the multiple channels due to the very dynamic nature of bandwidth demands in an access network. Using dynamic channel assignment is facilitated more efficiently through a multidimensional scheduling approach. The multiple channel EPON scheduling problem is modeled using standard scheduling theory notation. The multiple transmission channels are the machines to be scheduled, the Optical Network Unit (ONU) transmission grants are the jobs, and the sizes of the grants indicate the processing time of the jobs. Further, the scheduling is layered into a scheduling framework and scheduling policy. A scheduling framework determines when the Optical Line Terminal (OLT) produces a schedule as ONU transmission grant requests arrive. Once a set of ONU transmission grants to be scheduled are determined by the scheduling framework, the scheduling policy schedules the grants in time and channel. Extensive simulation analysis shows that the scheduling framework has the largest impact on average queueing delay and channel utilization. An offline scheduling framework is not work conserving and consequently wastes channel resources while waiting for all ONU REPORT messages before making access decisions. An online scheduling framework, although work conserving, does not provide the best performance since scheduling decisions are made with the information contained in a single ONU REPORT. The novel online JIT scheduling framework is also work conserving but can allow for increased scheduling control by allowing the channel availability to drive the scheduling process. Therefore, multiple ONU REPORRs can be considered together when making scheduling decisions. This results in better average queueing delay performance under certain conditions and a more effective method of providing differential ONU treatment.