Ethernet Passive Optical Networks

In Chap. 14, we considered a static TDM scheme, in which every ONU receives a fixed timeslot. While this scheme is very simple, it has the drawback that no statistical multiplexing between the ONUs is possible.
In has been shown that network traffic exhibits a high degree of burstiness [LT+94]. Traffic aggregation does not solve the problem as the variance of aggregated traffic decreases more slowly than the variance of a conventional Poisson process (see App. A). The long-range dependence (heavy-tailness) of the arrival process creates a situation in which some timeslots overflow even under very light load, resulting in packets being delayed for several timeslot periods. It is also true that some timeslots remain underutilized (not filled completely) even if the traffic load is very high. This leads to the PON bandwidth being underutilized. A dynamic scheme which reduces the timeslot size when there is no data would allow the excess bandwidth to be used by busy ONUs.
In this chapter we consider a simple DBA algorithm that provides statistical multiplexing for the ONUs. We investigate the improvements in performance (multiplexing gain) compared to the SSA scheme. We also consider and analyze several flavors of statistical multiplexing.
The DBA algorithm considered here is based on the interleaved polling with adaptive cycle time (IPACT) method [KMP02]. In its original form, the IPACT method used control messages that were embedded in data frames or in interframe gaps using an escape code. Such control messages were transmitted...