IEEE Communications Magazine • June 2008 138 0163-6804/08/$25.00 © 2008 IEEE
INTRODUCTION
The current migration of access networks from
digital subscriber line (xDSL) technologies
toward various fiber to the-x (FTTx) solutions
will expose the inner networks not only to a con-
tinuous increase in traffic but also to higher traf-
fic fluctuations. This will bring the contemporary
metro and core network architecture to its limit.
The main limitation of this approach is the
excessive amount of layer 2/3 (L2/L3) processing
and switching of such multi-protocol, multi-bit-
rate traffic, which can only rely on an excessive
deployment of expensive electronic hardware,
suffering from high power consumption, man-
agement complexity, and large office real estate.
The problem is further exacerbated by the fact
that existing networks are multidomain, meaning
that a large-scale network is partitioned into dif-
ferent administrative domains deploying equip-
ment from different vendors, whereas at their
(multi-granular) interfaces, inter-operability is
required to ensure effective end-to-end service
delivery. This network paradigm cannot scale
cost-effectively.
To address this bottleneck, architectures with
fewer layer transitions and reduced layer 2/3
processing leading to flatter, that is, less hierar-
chical, network structures are sought. Flat net-
works allow for enormous savings in operating
expenditures (OPEX) and capital expenditures
(CAPEX), and it is the only network architec-
ture that can deliver long-term economically
scalable growth. Along these lines, a two-tier
core network architecture is proposed where the
rigid distinctions between metro, outer core, and
inner core become blurred significantly, alleviat-
ing the processing burden, when traffic crosses
core nodes or when it traverses different
domains.
Recently, a considerable number of new
approaches for constructing a dynamic core net-
work were proposed around various optical burst
switching/optical packet switching (OBS/OPS)
solutions. In the OBS/OPS schemes, the lack of
effective optical buffering presents a hard dilem-
ma between high loss (when adopting one-way
reservation) and high delay (when using a cir-
cuit-switch such as two-way reservation). Specifi-
cally, OBS solutions with little or no buffering at
the nodes have been extensively investigated, for
example [1–3], and shown to suffer from heavy
burst loss even at low utilization levels due to
their ambitious on-the-fly act. The problems
arise from conflicts due to the temporal proper-
ties of data bursts entering the OBS nodes,
which the method can alter only poorly due to
lack of buffering. As a result, losses grow expo-
nentially with each node the traffic must cross.
In an alternative OPS-based solution
described in [4], a metropolitan area network
(MAN) consisted of a number of optical rings
where traffic was stored in buffers located in the
ABSTRACT
The explosion of current demand has brought
the contemporary multidomain core network
paradigm to its limit. In the quest for new
approaches that exploit recent developments in
optical technology, a novel network architecture
that obviates most of the expensive and loss-
prone centralized all-optical switches is described
in this work. It is based on clustered architecture
for nodes in optical networks and features a rec-
onciliation between dynamic resource allocation
and guaranteed end-to-end network perfor-
mance in a multidomain network. This article
enhances the distributed, collision-free slot
aggregation inside domains of clustered core
nodes with dynamic switching of slots/frames
between the domains. Thus, it can support
dynamic sub-wavelength allocations between
network domains, using standard burst-switching
techniques. This extends the high efficiency and
multiplexing gain into the inter-domain network
even under highly bursty traffic. It features both
low-cost optical add/drop edge nodes exploiting
WDM transmission and agile and modular cen-
tralized electro-optical switches that are present-
ed in conjunction with the overall network
architecture. Its performance exhibits very low
burst loss probability traded for a higher but tol-
erable and bounded delay.
MULTIDOMAIN OPTICAL NETWORKS:
ISSUES AND CHALLENGES
Alexandros Stavdas, Theofanis G. Orphanoudakis, Helen C. Leligou, Konstantinos Kanonakis, Chris
Matrakidis, Andreas Drakos, and John D. Angelopoulos, University of Peloponnese
Andrew Lord, British Telecommunications
Dynamic CANON: A Scalable
Multidomain Core Network