IEEE Optical Communications • August 2004
S8
0163-6804/04/$20.00 © 2004 IEEE
INTRODUCTION
Access networks connect business and residential subscribers
to the central offices of service providers, which in turn are
connected to metropolitan area networks (MANs) or wide
area networks (WANs). Access networks are commonly
referred to as the last mile or first mile; the latter term empha-
sizes their importance to subscribers. In today’s access net-
works, telephone companies deploy digital subscriber loop
(xDSL) technologies, and cable companies deploy cable
modems. Typically, these access networks are hybrid fiber
coax (HFC) systems with an optical-fiber-based feeder net-
work between central office and remote node, and an electri-
cal distribution network between remote node and subscribers.
These access technologies are unable to provide enough band-
width to current high-speed Gigabit Ethernet local area net-
works (LANs) and evolving services (e.g., distributed gaming
or video on demand). Future first-mile solutions have to not
only provide more bandwidth but also meet the cost sensitivity
constraints of access networks arising from the small number
of cost sharing subscribers.
In so-called FTTx access networks the copper-based distri-
bution part of access networks is replaced with optical fiber;
for example, fiber to the curb (FTTC) or home (FTTH). In
doing so, the capacity of access networks is sufficiently
increased to provide broadband services to subscribers. Due to
the cost sensitivity of access networks, these all-optical FTTx
systems are typically unpowered and consist of passive optical
components (e.g.,, splitters and couplers). Accordingly, they
are called passive optical networks (PONs). PONs attracted a
great deal of attention well before the Internet spurred band-
width growth. The Full Service Access Network (FSAN)
Group (http://www.fsanweb.org) was the driving force behind
the International Telecommunication Union — Telecommuni-
cation Standardization Sector (ITU-T) G.983 broadband PON
(BPON) specification, which defines a PON with asynchronous
transfer mode (ATM) as its native protocol data unit (PDU).
ATM suffers from several shortcomings. Due to the segmenta-
tion of variable-size IP packets into fixed-size cells, ATM
imposes a cell tax overhead. Moreover, a single corrupted or
dropped ATM cell requires retransmission of the entire IP
packet even though the remaining ATM cells belonging to the
corresponding IP packet are received correctly, resulting in
wasted bandwidth and processing resources. Finally, ATM
equipment (switches, network cards) is quite expensive.
Recently, Ethernet PONs (EPONs) have gained a great
amount of interest in both industry and academia as a promis-
ing cost-effective solution for next-generation broadband
access networks, as illustrated by the formation of several fora
and working groups, including the EPON Forum
(http://www.ieeecommunities.org/epon), the Ethernet in the
First Mile Alliance (http://www.efmalliance.org), and the
IEEE 802.3ah working group (http://www.ieee802.org/3/efm).
EPONs carry data encapsulated in Ethernet frames, which
makes it easy to carry IP packets and eases interoperability
with installed Ethernet LANs. EPONs represent the conver-
gence of low-cost Ethernet equipment (switches, network
interface cards [NICs]) and low-cost fiber architectures. Fur-
thermore, given the fact that more than 90 percent of today’s
data traffic originates from and terminates in Ethernet LANs,
EPONs appear to be a natural candidate for future first-mile
solutions. The main standardization body behind EPON is the
Michael P. McGarry, Arizona State University
Martin Maier, Centre Tecnològic de Telecomunicacions de Catalunya
Martin Reisslein , Arizona State University
ABSTRACT
Optical networks are poised to dominate the access net-
work space in coming years. Ethernet passive optical net-
works, which leverage the ubiquity of Ethernet at subscriber
locations, seem destined for success in the optical access
network. In this article we first provide a brief introduction
to Ethernet passive optical networks, followed by a discus-
sion of the problem of dynamic bandwidth allocation. We
then introduce a framework for classifying dynamic band-
width allocation schemes and provide a comprehensive sur-
vey of the dynamic bandwidth allocation methods proposed
to date. We conclude with a side by side comparison of the
schemes based on their most prominent characteristics, and
outline future developments of dynamic bandwidth alloca-
tion schemes.
E
THERNET PONS: A SURVEY OF D
YNAMIC
B
ANDWIDTH ALLOCATION (DBA) ALGORITHMS