Optical Burst Switching - A Novel Paradigm for Gigabit WDM Optical Networks

			Chunming Qiao, et al.
                   Department of ECE, SUNY at Buffalo
                            Amherst, NY 14260 

	The expected growth in broadband multimedia services gives
rise to the need to transport user data using high-bandwidth
fiber-optical channels. A Wavelength Division Multiplexed (WDM)
all-optical or almost all-optical network offering multi-Gigabit rate
per wavelength channel may soon become economical as an underlying
backbone in wide area networks. Given the limited number of
wavelengths available and the limitation of the optical memory
technology, a challenge is how to combine the best of the
coarse-grained optical circuit switching (e.g. via wavelength routing)
and the fine-grained optical packet/cell switching while avoiding
their shortcomings in order to support multimedia traffic, which is
often bursty and requires low latency, economically and efficiently in
an optical network.
	Optical burst switching (OBS) is proposed as a novel paradigm
to provide high-end users or applications with on-demand sessions
having a high bit-rate, a low latency and a relatively short duration.
OBS uses one-way reservation so that a burst of user data (e.g. IP
packets) can be sent without having to have a dedicated wavelength
path a priori.  Instead, a control (set-up) packet is sent first to
reserve the wavelength channel, which is followed by the burst after
an offset time. In this way, OBS not only avoids the long end-to-end
setup delay, but also increases the utilization of the ultra-high
speed optical channels.  
	In addition, OBS facilitates multicasting in the optical layer
because of the limited degree of opaqueness provided by the control
packet and more importantly, because of the much better sharing of the
scarce multicasting and wavelength resources. Overall, OBS enables a
flexible, efficient and bandwidth-abundant fiber-optic network
infrastructure capable of providing ubiquitous services to IP and
other existing (e.g. ATM, SONET) and future protocols.
	A key design issue is to reduce the burst dropping probability
under high traffic load, which is especially important for
multicasting traffic.  Several protocol approaches can be taken. They
include, for example, assigning a higher priority or an additional
offset time (accordingly, a higher success probability) to a burst
whose destination is further away.  Note that, using an offset time
longer than that minimally required is necessary to facilitate
deflection-routing. For a multicasting request, approaches allowing
for non-simultaneous serving (or fan-out splitting) of its group
members will greatly increase the success probability. They can be
even more effective when coupled with a routing algorithm that uses
the nodes capable of O/E and E/O conversions as the branching (or
ramifying) points in the multicasting tree so that retransmissions can
be performed not only by the source but by intermediate nodes as well.
	Our focus will be on the design and evaluation of advanced OBS
signaling protocols capable of efficient bandwidth allocation,
intelligent optical buffer management, adaptive routing (e.g.
deflection and priority-based routing), and multicasting. Our approach
differs from other Terabit/s burst switching protocols that either
switch bursts electronically, or send a burst together with the
control packet as in packet-switching, thereby requiring that the
burst be buffered optically while the control packet is being
processed at each intermediate node.