Optical communications

Model
Digital Document
Publisher
Florida Atlantic University
Description
With the ever increasing demand for bandwidth intensive applications like video-ondemand,
interactive television services, high-defmition television (HDTV) and internet
telephony, the first mile network, referred to as the last mile network in the earlier stages
of its development, has proven to solve the "bottleneck" often experienced between users
and the central office. The Ethernet Passive Optical Network (EPON), developed by the
IEEE 802.3ah Task Force for Ethernet in the First Mile (EFM), resolves this bottleneck
by providing the user with a fiber optic link to the central office that has many
advantages, such as nearly infinite bandwidth, low cost, easy installation and immunity to
electromagnetic interference, and a saving of the need for powered components in the
signal path from the user to the switch. EPON s have proven to deliver the essential
services of voice, video, and data communications reliably, while at the same time
providing expected guarantees of the delivery of those services in terms of defined
Quality of Service measures (QOS). A continuous theme throughout EPON research has
been the study of efficient Dynamic Bandwidth Allocation (DBA) as a key factor in
achieving fairness in distributing bandwidth amongst remote network units.
This research investigates another factor and its effect on network performance and
service delivery: Cycle Demand Proportionality (CDP). By observing the patterns of
demand in the network on an individual network unit basis cycle after cycle, deductions
regarding load characteristics of some units over others can be made. Decisions can be make about subsequent grant allocations based on this factor and aim to achieve better
results in the process. Simulations of EPONs Lmder varying loads incorporating the usc of
CDP in conjunction with currently used DBA schemes are made and results are analyzed.
A major contribution of this research is a new bandwidth allocation algorithm that gives
improved performance in terms of packet delay versus offered load. Performance metrics
are compared against two common bandwidth allocation algorithms: Interleaved Polling
with Adaptive Cycle Time (IPACT) and Cyclic Polling (CP). Additional results also
include improvements in packet loss and throughput.
The data that represents traffic for this network has two properties, self-similarity and
long range dependency. Plotting the auto-covariance, auto-correlation, and variance for
this traffic, for various aggregation levels demonstrates these properties.