Light Trails: A Sub-Wavelength Solution for Optical Networking

Michael T. Frederick, Nathan A. VanderHom and Arun K. Somani

Dependable Computing and Networking Laboratory
Department of Electrical & Computer Engineering
Iowa State University, Ames, IA, 50011, USA
E-mail: {freds, nvander, arun} @iastate.edu

AbsmcI- All-optical networks are ahle to transport data

from source to destination entirely in the optical domain. This
is a departure from current optical networks that rely on
optical-electrical-optical(OEO) conversion at each intermediate
connection node to route data properly. The opacity inherent in
traditional networks is costly in terms of the limiting bandwidth
and increasing switching complexity.
MPLS, OBS and OPS have been proposed as solutions to
realizing an all-optical network. MPLS and OBS have the
advantages of creating all-optical connections between nodes,
hot don't allow intermediate nodes to also use the wavelength.
Additionally, optical switches are constantly being reconfigured
to accommodate new connections. OPS can make switching
decisions in the optical domain, hut the technology is immature.
Light trail technology tries to avoid the pitfalls of immature
technology, the inability of intermediate nodes to use a connection
wavelength, and the constant reconfiguration of switches. A light
trail is a unidirectional optical bus between nodes that allows
intermediate nodes to access the bus. The goal is to minimize the
amount of active switching that needs to he done by allowing
intermediate nodes to use a connection that has already been
setup. Connections are not constantly being setup and torn down,
but rather exist lor as long as they are being used by any of the
nodes along their light trail.

Index Terms- Wavelength Division Multiplexing, Light 'bail,

Grooming, Optical Network Design.

One of the most pressing issues concerning current WDM

implementations is the opacity of the network. A network is
considered opaque if it requires its constituent nodes to be
aware of the underlying packet format and bit rate. The need
to handle data streams in the electrical domain with respect
to the aforementioned factors creates a huge optical-electronic
bandwidth mismatch. The bandwidth on a wavelength is 10
Gbps today and is likely to increase to 40 Gbps, while the
sub-rate traffic connections can vary from STS-1 (51.84 Mbps)
to the full wavelength capacity. Advances are being made to
make electronic switches more scalable by adding additional
ports to the switching fabric, however, as the optical data rate
increases electronic switches will be hard-pressed to keep up.
If a legacy switch operating at 2.5Gbps is present anywhere
in the network, all data passing through that switch is limited
to 2.5 Gbps. In this case, upgrading the network requires
replacing the legacy switch. With an all-optical switch, data
rate is only limited by the ability of the switch to realize a
given transmission rate.
As more advanced devices become available, such as
optical cross-connects and micro-electro-mechanical systems
( E M S ) , it is possible to design transparent optical networks
in which optical signals on an arriving wavelength can be
switched to an output link of the same wavelength without
electronic conversion, thus creating what is termed a lightpath
from souce to destination. Signals on all-optical lightpaths
can be at different bit-raws and use different forma&, as fie
signals are never terminated inside the core network. This bitand pmtocol transparency are key requirements
rate,

I. INTRODUCTION
The last decade has Seen huge
in network
as the PIiFiber optics have replaced
mary transmission medium, effectively increasing single-link
bandwidth from IO Mbps to over 160 Gbps utilizing Wavelength Division Multiplexing (WDM).As high bandwidth =-- ....._ *...-_
L"
1".
l"L"lr CILiLW"lhJ.
applications are pushing current network technologies to their
capacity limits, researchers are being forced to create highAnother problem with current WDM technology is that,
speed networks capable of supporting various bit-rate, protocol once a lightpath is established, the entire wavelength is used
and format applications in a highly scalable manner.
exclusively by the connection's 'source and destination nodeWavelength Division Multiplexing (WDM) technology was pair; no sub-wavelength sharing between nodes along the
initially developed to increase the capacity of point-to-point lightpath is allowed. Therefore, the wavelength capacity may
fiber links. WDM enabled networks allow multiple opaque be underutilized for IF'bursts unless the source and destination
point-to-point connections to be established where the optical nodes efficiently aggregate traffic.
signal must undergo optical-electronic-optical (OEO)conLight trail technology has been proposed in [9] as a solution
version at each intermediate node in the network. Network
to
the problem of all-optical network switching. The goal
designers are able to occupy multiple wavelengths leading
to increased bandwidth and fault tolerance while decreasing is to establish a connection between source and destination
nodes as a unidirectional bus (a light trail). This unidirectional
congestion and blocking.
bus will allow intermediate nodes to transmit data to any
me research reponed in lhir paper is funded in pan by the National Science other node downstream. Light trail technology avoids costly
Foundation under gmr ANI-9973102 and by Defense Advanced Research
Projects Agency and National Sccuriry Agency under grant N66Wl-W-1- OEO switching at intermediate nodes and offers complete
8949.
transparency to signal bit-rate, format, and protocol.

-~

~~

0-7803-8375-3/04/$20.000 2004 IEEE

175

In the next section will discuss grooming and how it can be

implemented on all optical networks with the use of light trails.

11. CURRENT
SOLUTIONS
A. Multiprotocol Lnbel Switching

Although not a new technology, Multiprotocol Label

Switching (MPLS) has been gaining attention as a way to
offer Quality of Service (QoS) in next generation IP-based
optical networks. MPLS was originally designed to provide a
way of grouping individual traffic flows together for transport
over a core MPLS-enabled network. In this fashion, various
traffic flows are grouped together to provide varying service
level agreements.
The basic idea in MPLS is to add a routing label that MPLS
enabled routers can easily understand and route. thus creating
a label switched path from source to destination in which
intermediate routing decisions can be made more quickly
based upon the label.
The all-optical Wavelength Selective Cross-connect (WXC)
is the first attempt to make all optical switching decisions where the label is actually the incoming wavelength
(MPLambdaS). As an optical signal enters a switch, the
wavelength is used to determine the correct output port. Two
types of WXC are available for use in an all optical transport
network.
Lnmbda Switch Capable (LSC) - Forwarding occurs by
switching a lightpath on the basis of its wavelength. As
a pure analog device, the LSC interface is not capable of
receiving control messages in-band with the data. Data
forwarding decisions are made based upon the incoming
wavelength.
Fiber Swifch Capable (FSSJ - Does not recognize bits
or have any concept of wavelength. Out-of-band control
information is used to configure each switch en route to
the destination.
B. Optical Burst Switching

Optical Burst Switching (OBS) is similar to MPLambdaS in

that a path is established between two edge nodes for a given
flow. The difference between OBS and MPLambdaS lies in
how the path is established. In OBS [l] [2] [31, a control
packet is sent ahead of the flow that sets up the connection by
configuring each of the switches along the path. Data follows
this packet. delayed by a guard time. never having to leave
in the optical domain. OBS is an attempt to offer burst level
communication at the optical layer by pre-allocating resources
for the duration of an IP traffic burst. Consequently, OBS is
able to provide burst level granularity entirely in the optical
domain. However, in order to provide reasonable network
utilization, current OBS solutions require very high-speed
switches to lower the ratio between burst length and set up
time. Additionally, because resources must be reserved along
the entire bunt path, these resources are rendered unusable to
other connections, contributing to an-overall underutilization
of the network. The ability to accommodate sub-rate IP bunts
is a major obstacle in realizing all-optical networks.
Although MPLS and OBS offer network designers the
ability to utilize multiple wavelength routes without expensive
OEO conversion, networks are still unable to provide sub-rate
multiplexing or traffic grooming along the wavelength path.

C. Grooming
Grooming is a technique currently used ti3 more efficiently
utilize fiber resources. As mentioned earlier, sub-rate optical
traffic connections can vary from OC-1 to full wavelength
capacity and, although WDM nctworks can utilize multiple
wavelengths for each connection, a cost is associated with
adding additional wavelengths due to the need for additional
transmitters and receivers. Thus, it is cost effective to combine
sub-wavelength connections to more efficiently utilize each
wavelength, whenever possible, i n the form of traffic grooming.
WDM grooming networks can be classified into two categories 141: dedicated-wavelength grooming @WG) networks
and shared-wavelength grooming (SWG) networks. In a DWG
network, the source-destination node pairs are connected by
lightpaths shared by connections between the pair. In a SWG
network, the lightpath can be shared by connections from different s-d pairs, The performancg: of SWG networks depends
on the efficient merging of fractional wavelength requirements
into full or almost-full wavelength requirements. WDM networks can be classified into various categories depending on
the level of grooming capability they provide.
ADM-constrained grooming node - The node can multiplex and demultiplex low-.rate traffic only on dropped
wavelengths at an add-drop multiplexer (ADM).
Wavelength continuity consrrained gmoming node - The
node can switch connections acmss different lightpaths.
but cannot switch between (differentwavelengths.
Full grooming node - The node can switch connections
in any permutation from one wavelength to another,
Grooming creates a unit of capacity smaller than an entire
wavelength in a WDM network [!il. Instead of lower rate traffic
monopolizing the use of an entire wavelengfh, multiple lowerrate traffic streams can be multiplexed on the same wavelength,
and the capacity more effectivel:y utilized.

D. Optical Packet Switching

An OPS is the true optical equivalent of an electronic packet
switch [61, reading the embedded label and making a switching
decision based on the information contained within, entirely in
the optical domain. The major advantage of OPS is its flexible
and efficient bandwidth usage, which enables the support of
a variety of services. Pure OPS technology in which packet
header recognition and control are performed in all-optical
domain is still many years away, and may never become
reality.
An OPS,with electronic head'tr processing and control. is a
more realistic near term solution;. A practical OPS experiment
has already been performed und'tr the European ACT KEOPS
(KFiys to Optical Packet Switching) project [7][8].In KEOPS,
the header is sent with data (payload) but at lower bit rate,
and the header processing is still in electrical domain. This
potentially requires massive optical buffering at the input

176

Fig. 1. A light vail form node 1 10 node 4

port to allow the header processing circuits to finish the job.

Similarly, an optical buffer is required at the output side to
avoid packet loss. At present. the buffering technology is also
not mature and has to overcome a number of technological
constraints, including the large and varying size of optical
buffering and high speed header processing.
The next generation all optical transport network should
be able to make dynamic switching decisions in the optical
domain, allow traffic grooming at intermediate nodes and
utilize multiple wavelengths. The result will be a highly
utilized bit-rate and protocol transparent network. The next
section explains the light trail protocol and the benefits of
such a network.
111. LlGHT TRAIL TECHNOLOGY
A recent solution, introduced in [91. offers some remedies
to the shortcomings of the previously described protocols with
the use of what are termed lighr :rails. The concept of a light
trail is proposed to enable communication at the optical layer
by establishing an optical path or trail between source and
destination nodes, while allowing intermediate nodes along the
trail access to the same wavelength without the establishment
of an additional connection.
A light trail can be viewed as a unidirectional optical
bus between the source and the destination nodes, with the
characteristic that intermediate nodes can also access this bus.
This is in major contrast to a lightpath or burst-level path
provisioned under conventional architectures such as standard
WDM or OBS. The main goal is to minimize the amount
of active switching needed by allowing intermediate nodes
along a lightpath to use a connection that has already been
established. In this way, connections are not constantly being
established and terminated, but rather exist for as long as
they are being used by any of the nodes along the light trail.
Such a network can reduce path setup time and improve path
utilization.
An example light trail is shown in Fig. I at the level of the
shutter configurations.-The total number of connections in a
trail of length n that can be supported is
as long as the
aggregate bandwidth does not exceed the fiber capacity.
A typical node in a light trail network is shown in Fig. 2. A
modified version of this node, known as a light bus node,
is shown in Fig. 3. Although it is omitted in the picture
of the light trail node, the same electrical signaling channel
controls the optical shutters. Each node consists of wavelength
multiplexing and demultiplexing units, variable optical taps to

(9,

enable the node to receive incoming data as well as pass it

on, optical shutters to control the initiation and termination of
light trails, and combiners to allow the node to source its own
data.

(n-scr -it&
.p;itCa.

oo.bllu*

Fig. 3. Light bus "Ode ai the wawlengfh level

Each node on the light trail taps a sufficient amount (based

upon its location in the light trail) of optical power from the
incoming signal for local processing. The remaining optical
power is sent through the optical shutter where optical routing
decisions can be made based upon the shutter configuration.
The node processes the tapped optical signal, determines if the
data is destined for them, and if so, utilizes it. If the node finds
that the optical bus free,it can transmit data to any downstream
node.
Collisions occurring between data transmitted from and
upstream node and a downstream node beginning transmission can be handled through the control channel. or through
avoidance by using a light bus node (Fig. 3). In the case of the
light bus node, the incoming signal may be routed through an
optical delay line equal to the length of the maximum frame
size. The purpose of the delay loop is to prevent collisions
on the bus between data transmitted from an upstream node
and data sourced from the current node. If the delay line is
free a node can source its own data. However. if an upstream

1 77

node has transmitted data unbeknownst10 the current node, the

upstream data will pass through a delay, enabling the current
node to finish its transmission, thus preventing collision. The
light bus node is just one proposal for dealing with potential
collisions on the bus.
The protocol utilizes an out of hand communication channel
that controls the optical shutter at each node. The signaling
channel also carries information about all the light trails in
the network, and is responsible for provisioning connections
within existing light-trails. Notice that nodes 1 and 4 have
closed optical shutters, while nodes 2 and 3 have open ones.
This isolates this particular wavelength from the rest of the
network and establishes the unidirectional bus known as a light
trail.
IV. PERFORMANCE
EVALUATION
In order to assess the performance of a light trail network,
it was tested against 3 different routing protocols. The first is
standard lightpath routing, where a connection is established
on a wavelength between the source and destination nodes.
No other connections can use the same wavelength, regardless
of its utilization. The second routing strategy tested was
sourcddestination lightpath routing (SD routing). It performs
like the previously described lightpath routing, with the exception that the source and destination nodes can dynamically
aggregate traffic to improve wavelength utilization. The third
routing strategy that light trail performance is contrasted
against is node level grooming. Node level grooming allows
all connections received by a node to be aggregated together
so long as they are transmitted on the same link upon leaving
the node.
Three different topologies were used to assess the performance of light trail routing. A 14 node, 23 link NSFNET
an I 1 node, 22 link NJLATA network; and a 9 node, 18
link 3x3 mesh toms were used as topologies. Each link in
an arbitrary network consists of a two unidirectional links
(in opposite directions) with 16 wavelengths per link and a
capacity of 4 per wavelength. In essence, each link consists
of 32 wavelengths (I6 in each direction between 2 arbitrary
nodes), with a capacity of 4 on each wavelength.

v.

Fig. 4. Blacking Probability vs. Link hiad

trails and SD routing performed about 3x worse than nodelevel grooming even under heavy loading. Considering the
advantage of having unresmcted grooming, the difference in
performance between light trails and node-level grooming is
relatively low.
Additionally, the SD routing can be viewed as the best
case performance for light trail routing. In a network where
grooming is constrained, the best performance should occur
when all traffic sharing common :source and destination nodes
is aggregated together. This would have the effect of maximizing bandwidth utilization. Light trails also ;allow intermediate
nodes to access the bus as source and idestination nodes,
thus allowing some bandwidth to go unused. However, the
performance difference between SD routing and light trail
routing is small.
Fig. 5 displays the results for average light trail length. It is
interesting to note that the averagq: path length for connections
in the network was longer than the average path length for
the light trail servicing them. This is a result of longer light
trails having the ability to be mo1.e fully utilized. A light trail
of length 1 hop can only accommodate connections having
the same source and destination nodes, whereas a light trail
of length n can accommodate (I: different sourcddestination
pairs. Consequently, many short light trails are established
and are undemtilized. while fewer longer light trails are
established, each heavily utilized.
+Ps4"U

+*.*E"y

RESULTSA N D DISCUSSION

+*_,&U

Fig. 4 shows the blocking probability associated with using

each of the 4 routing strategies for all topologies. In all cases,
node-level grooming outperformed all other routing metrics.
This is expected because node-level grooming offers the most
flexibility in capacity usage. The worst performing routing
strategy is, as expected, no grooming. No grooming uses an
entire wavelengtti between source and destination nodes for
the duration of the connection, allowing no capacity sharing
for sub-lambda traffic. SD routing and light trails performed
almost the same in all cases, with SD routing performing
slightly better. Other research performed in [IO] confirms that
no-grooming WDM is outperformed by light trails.
Node-level and no groomed routing serve as upper and
lower bounds for assessing light trail performance. Light

i 5. ~ A~~~~~
.
Light T ~ kngul
~ I "S.

ink

baa

Light trail capacity utilization indicates that utilization increases as load increases (Fig. 6). In general, utilization ran at
about 30 - 35% for all topologies for similar blocking probability performance. SD routing, on ibe other hand, actually had a
lower connection utilization than light trails. Part of the reason
for this might be the lower blwking probability associated

178

with SD routing. However, Fig. 7 displays the average number

of connections per trail or connection, and indicates that far
fewer connections use each sourceldestination connection than
use each light trail. This, more than likely, accounts for the far
more underutilized sourcddestination connections witnessed
in Fig. 6.

Average Light Trail Utilization vs. Link Load

Fig. 6

Increasing the number of connections per trail has the effect

of minimizing the frequency of establishment or termination of
light trails, which also minimizes the amount of switch reconfiguration that needs to be done. Less switch reconfiguration
also leads to less out-of-band control signaling that needs to
occur to form and terminate light trails. This is one of the
major advantages to light trail technology.

IE

light trail to serve as a bus controller. The complete solution

will provide connection-orientedservicewhile offering various
service level agreements and QoS for data traffic on the same
network and providing sub-lambda granularity. Other future
work includes dynamically modifying light trails by either
extending or trimming them to accommodate new connections.
Current optical networks perform costly OEO conversion
at each node in order to route connections. The advantages
of being able to avoid the conversion stage are significant.
All-optical switching should be cheaper as there is no need
for large quantities of expensive high-speed electronics, also
reducing physical switch size. Finally, optical switches are
more scalable because they do not need to be replaced the
bandwidth of the optical fiber is changed. All optical switches
are bit-rate, format, and protocol independent as well, because
they only switch light signals and don't need to interpret them.
Proposed solutions for all-optical networks such as MPLS
and OBS monopolize a wavelength throughout connection
duration. They also do not lake advantage of switches that
are already configured for ensuing connections. OPS is a
technology that is not mature, and most likely will not be in the
near future. Standard WDM networks don't efficiently manage
fiber optic bandwidth, as the wavelength is the smallest unit
of capacity.
Light trail technology altempts to address the shortcomings
of theses various technologies by establishing a unidirectional
light bus from source to destination. The goals are to allow
intermediate connection nodes to also use the bus, as well as
maximize the reuse of already configured optical switches. In
this fashion, passive switching is maximized and a network
transparent to bit-rate, format, and protocol is established.

REFERENCES

IO

I$

I,

I/
I
Ulle.I,EMFl

LI

[I] M. Yo0 and C. Qiao, "Optical burst switching (OBS) - a new paradigm
for and optical Internet:' Journd oJHigh Speed N e w o r b (JHSNJ,voI 8.
no. 1. pp. 69-84. 1999.
121 Ge. F. Callegati and L. Tamil. "On optical burst switching and self-similar
udffic:' IEEE Cornmmication Lmers, vol. 4, no. 3, pp. 98-IM). March

s6

2ooO.

Fig. 7

Average Number of Connections per Light Trail vs. Link Load

VI. CONCLUSION
Although light trail technology provides for an all optical
transport path between communicating nodes, the media access control (MAC) that allows intermediate nodes access to
the light trail does not provide any fair usage mechanism.
Consequently. intermediate nodes wishing to use their light
trail may be starved of resources and forced to create an
overlapping light trail-adding to the complexity of the network. The current MAC mechanism is similar to a Carrier
Sense Multiple Access (CSMA) scheme in which intermediate
nodes are not allowed to transmit if the lightpath is sensed
to be occupied. With such high-speed communication, it is
necessary to implement a scheduling mechanism to arbitrate
bandwidth in a fair usage manner.
The utilization of an MAC mechanism will be examined,
and may incorporate be similar the Robust Dynamic and
Fair Network WEN) [ I l l , or require the end-node of the

[3l S. Verma. H. Chaskar and R. Ravikanfh, "Optical bunt switching: A

viable solution for Imbit IP backbone:' IEEE N e w w k . pp. 48-53.
NovemberlDeccmba 2wO.
141 1. Yam, J. l a c y and D. EveriR "Blocking in mvltiwavelengtb TDM
network:' Pmcredings OJ the 41h I n r c ~ i o M ConJercnce
I
on Tefecomniunicarion Sysrems, Modeling, and Analysis, March 1996, pp. 535-541.
[SI R. Snniwan and A. K. So&.
"Request-specific routing in WDM
gmoming "elworks:' IEEE Iruemiond Conference on Communicaionr.
vol. 5, pp. 2876-2880, 2001.
[6] M. Mahany. D. Simeonidou. D. Hunler and A. Tzanakaki. 'me application of optical packel switching in htun communication networks:'
IEEE Cormmicarion Mopmine, pp. 128-135. March 2001.
171 P. Gambini et al.. 'Tranrparent optical packef switching: network archilectun and demonsvatom in fhe KEOPS project,'' IEEE Journnl on Selected
area^ in Convnunicolions. vol. 16. no.1. pp. 1245-1219. September 1998.
181 C. Guillemot et al.. 'Transparent optical packet switching: The Eumpean
ACTS KEOPS project approach:' Joumnl o/Lighwme Technology. vol.
16. no. t2. pp. 2117.2134, December 1998.
[91 1. Chlamtac and A. Gumaste. 'Light-mils: A solution to IF' eenoic
communication in the opdcal domain:' Online publicalion: February 17.
2003,Springer-Valag Berlin Heidelberg ZW3.
[IO] 1. Fang, W.He and A. K. So-,
IF' Traffic gmoming in light vail
optical networks, submitted to IEEE JSAC 2 W 4
[ I l l Nalhan Vanderhom and A m K. S a d , "RDbUSt. Dynamic and Fair
Network A High-speed Option for Meoopoliw Ana Networks". Pmcerdings oJthe 7th lFlP Working Con/errnce on Opricol Ncwork Design
and Modeling,ONDM-2003. F e h a r y 2003.

179