1420 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 19, NO. 10, OCTOBER 2001
WDM Packet Routing for High-Capacity Data
Networks
Qimin Yang, Student Member, IEEE, Keren Bergman, Member, IEEE, Member, OSA, Gary D. Hughes, and
Frederick G. Johnson
Abstract—We present experimental and numerical studies of
a novel packet-switch architecture, the data vortex, designed for
large-scale photonic interconnections. The selfrouting multihop
packet switch efficiently scales to large port counts ( k) while
maintaining low latencies, a narrow latency distribution, and high
throughput. To facilitate optical implementation, the data-vortex
architecture employs a novel hierarchical topology, traffic control,
and synchronous timing that act to reduce the necessary routing
logic operations and buffering. As a result of this architecture, all
routing decisions for the data packets are based on a single logic
operation at each node. The routing is further simplified by the
employment of wavelength division multiplexing (WDM)-encoded
header bits, which enable packet-header processing by simple
wavelength filtering. The packet payload remains in the optical
domain as it propagates through the data-vortex switch fabric,
exploiting the transparency and high bandwidths achievable
in fiber optic transmission. In this paper, we discuss numerical
simulations of the data-vortex performance and report results
from an experimental investigation of multihop WDM packet
routing in a recirculating test bed.
Index Terms—Optical networks, optical packet switch, wave-
length division multiplexing (WDM) optical packet routing.
I. INTRODUCTION
W
ITH THE ever-growing demand for higher data rates
and a wide variety of services, a fully transparent
optical-switch network element presents an attractive way to
overcome the electronic speed bottleneck. However, all-optical
implementations encounter considerable challenges in the
processing and buffering of optical data, which are essential
to ensuring switch performance and to providing the neces-
sary services and protections. Therefore, most optical-switch
architectures actually employ hybrid optical-to-electronic tech-
nologies where the strengths of optics such as transparency and
Manuscript received January 23, 2001; revised May 30, 2001. This work is
sponsored by the Defense Advanced Research Projects Agency and by the Na-
tional Security Agency through an agreement with the National Aeronautics
and Space Administration under Contract 960 199, and by the National Science
Foundation under contract ECS98-00 401.
Q. Yang is with the Electrical Engineering Department, Princeton University,
Princeton, NJ 08544 USA (e-mail: qyang@ee.princeton.edu).
K. Bergman was with the Electrical Engineering Department, Princeton Uni-
versity, Princeton, NJ 08544 USA. She is now with Tellium, Inc., Oceanport,
NJ 07757 USA.
G. D. Hughes is with the Laboratory of Physical Science, College Park, MD
20742 USA.
F. G. Johnson was with the Laboratory of Physical Science, College Park,
MD 20742 USA. He is now with Little Optics, Inc., Annapolis Junction, MD
20701 USA.
Publisher Item Identifier S 0733-8724(01)08794-1.
high bandwidth can be fully exploited while the weaknesses
can be bypassed in the electronic domain. Nevertheless, issues
such as contention resolution, buffering, scalability, and latency
remain the key challenges in photonic packet switching [1]–[4].
To achieve high throughputs, a packet switch generally pro-
cesses multiple packets simultaneously. Therefore, whether the
architecture is based on a single hop or multiple hops, contention
is inevitable when multiple packets are competing for the same
output port. To solve the contention problem, either buffering
or deflection techniques can be implemented. Simple deflection
methods without buffers (hot-potato routing) usually introduce
severe performance penalties in throughput, latency, and latency
distribution [5]. Current optical buffering techniques configured
as optical fiber delay lines in a traveling or a recirculating geom-
etry do not have random access capability [6], [7]. The straight
delay-line buffers can introduce large timing errors and the re-
circulating rings are generally bulky and expensive. Therefore,
it is still impractical to employ optical buffers within a switching
network element, especially for very large-scale switches.
In recent research, the use of wavelength division mul-
tiplexing (WDM) in packet switches has been explored to
achieve greater flexibility. Technologies such as wavelength
routing (WR) and wavelength conversion (WC) are proposed
to solve the difficulty of optical buffering [8]–[12]. These
techniques are very attractive in maintaining the throughput
performance as well as the optical data transparency without
resorting to optical buffers. The wavelengths are exploited as
logical buffers within the existing architectures. Generally,
however, the additional WR and WC devices add performance
penalties and make such systems complex and costly.
In this paper, we present experimental and numerical studies
of a new architecture, the data vortex, as a candidate for
large-scale packet switching [13]. The multihop packet switch
tightly couples the deflection method and a virtual buffering
mechanism to achieve hardware simplicity, scalability, and high
throughput. The timing and control algorithm permits only one
packet to be processed at each node in a given clock frame and,
therefore, the need to process contention resolution is elimi-
nated. The wavelength domain is additionally used to enhance
the throughput and to simplify the routing strategy. Numerical
studies of the traffic flow within the switch architecture have
shown that the data vortex can efficiently scale to greater than
10 000 ports while maintaining a low packet switching latency
and a narrow latency distribution.
The rest of the paper is organized as follows. Section II pro-
vides an overview of the data-vortex switch architecture, and in
Section III, the numerical studies of the system performance are
0733–8724/01$10.00 © 2001 IEEE