See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/247064634

Optical WDM Networks: Principles and Practice

Article · January 2000

CITATIONS READS

26 1,557

2 authors:

Rudra Dutta G.N. Rouskas

North Carolina State University North Carolina State University
207 PUBLICATIONS 3,196 CITATIONS 320 PUBLICATIONS 7,505 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by G.N. Rouskas on 26 June 2020.

The user has requested enhancement of the downloaded file.

Chapter 1

DESIGN OF LOGICAL TOPOLOGIES FOR

WAVELENGTH ROUTED NETWORKS

Rudra Dutta
Department of Computer Science
North Carolina State University
Raleigh, NC 27695-7534
rdutta@csc.ncsu.edu

George N. Rouskas
Department of Computer Science
North Carolina State University
Raleigh, NC 27695-7534
rouskas@csc.ncsu.edu
Keywords: Virtual Topology, Logical Topology, Wavelength Division Multiplexing
(WDM), Wavelength Routed Network, Lightpath, All Optical Network

Abstract
A virtual topology over a wavelength routed WAN consists of clear
optical channels between nodes called lightpaths. These carry traffic
end-to-end without electronic switching, creating an optical layer of the
topology. Virtual topology design aims at combining the best of optical
switching and electronic routing abilities. Designing a virtual topology
on a physical network consists of deciding the lightpaths to be set up in
terms of their source and destination nodes and wavelength assignment.
In this chapter we provide a complete formulation of the problem and
survey the literature on the topic. We restrict ourselves to transport
networks rather than local area networks, and static topology design as
opposed to topologies in which individual lightpaths are set up and torn
down in response to traffic demands.

1
2

1. INTRODUCTION
Wide area “All Optical Networks” with wavelength division multiplex-
ing (WDM), using wavelength routing, are considered to be candidates
for future wide area backbone networks. The ability to tap into attrac-
tive properties of optics, including the very high bandwidth potential
of optical fiber, makes these networks attractive for backbone transport
networks. At the same time, the WDM technique can be used to bridge
the mismatch between user and fiber equipment. A fuller discussion of
wide area optical networks can be found in (Green, 1992; Mukherjee,
1997; Ramaswami and Sivarajan, 1998; Green, 1996).
In recent times, there has been growing interest in virtual topology
design problems on these networks. Virtual topology design over a WDM
WAN is intended to combine the best features of optics and electronics.
The architecture uses clear channels between nodes, called lightpaths,
so named because they traverse several physical links but information
traveling on a lightpath is carried optically from end-to-end. Usually
a lightpath is implemented by choosing a path of physical links and
reserving a particular wavelength on each of these links for the lightpath.
This is known as the wavelength continuity constraint, indicating that a
lightpath consists of a single wavelength over a sequence of physical links.
Because of limitations on the number of wavelengths that can be used,
and hardware constraints at the network nodes, it is not possible to set
up a clear channel between every pair of source and destination nodes.
The particular set of lightpaths we decide to establish on a physical
network constitutes the virtual (otherwise called the logical) topology.
The tradeoff involved here is between bandwidth and electronic pro-
cessing overhead. Forming lightpaths locks up bandwidth in the corre-
sponding links on the assigned wavelength, but the traffic on the light-
path does not have to undergo optoelectronic conversion at intermediate
nodes. A good virtual topology trades some of the ample bandwidth in-
herent in the fiber to obtain a solution that is the best of both worlds.
Optical fiber can be used simply as a point-to-point link carrying
only one channel, using one wavelength. The use of WDM increases the
bandwidth available and the use of virtual topologies effects reduction of
delay, allowing more efficient use of bandwidth by appropriate routing.
Fig. 1.1 shows a simple physical network in which lightpaths, indicated
by dotted lines, have been set up to allow communication by a clear
channel between nodes which are not directly connected by a fiber link.
An attractive feature of the process of stepping up from point-to-point
fibers to WDM and then virtual topologies is that it can be undertaken in
an incremental manner with current networks (Mukherjee et al., 1996).
 Design of Logical Topologies for Wavelength Routed Networks 3

Routing
Node

Lightpath

Access Node

Figure 1.1 A WDM network. The routing nodes are interconnected by point-to-
point fiber links and may have access nodes connected to them. The dashed lines and
dotted lines show lightpaths.

The virtual topology provides a certain measure of independence from

the physical topology, because different virtual topologies can be set up
on the same physical topology, and allows us to choose a topology which
will result in greater network performance, given network conditions
such as average traffic between network nodes.
In general, virtual topology design problems can be formulated as op-
timization problems aimed at maximizing network throughput or other
performance measures of interest. Typically, the exact solution can be
shown to be NP-hard, and heuristic approaches are needed to find real-
istic good solutions. For this purpose, the problem can be decomposed
into four subproblems, which we discuss in detail in Section 3.2.
The above discussion focuses on the issue of throughput or delay op-
timization, which are related to network performance. There are at
least two other important related issues. The first is related to the cost
required to set up and operate the network, which is an important prac-
tical consideration. Thus, a particular virtual topology may result in
lower delay and higher throughput than another, but if the latter vir-
tual topology involves the use of fewer expensive network components
such as optical switches or converters, resulting in a lower overall imple-
mentation cost, then in practice it may well be chosen over the “better”
one. This issue is discussed in Section 2.
The second issue relates to the reconfigurability of optical networks
using virtual topology. Reconfigurability is seen as one of the strengths
of optical networks in general and the virtual topology approach in par-
4

ticular. A virtual topology is designed on the basis of traffic patterns and

a physical topology. Being able to implement a new virtual topology pro-
vides adaptability (when traffic patterns change), self-healing capability
(when the physical topology changes due to failure of network compo-
nents) and upgradability (when the physical topology changes due to
the addition or upgrading of network components). Thus being able to
redesign a virtual topology and configure the network to the new one
from the old one is of interest to the virtual topology problem in general,
and we have considered it within the scope of this survey.

Scope A similar virtual topology design problem exists for broadcast

optical networks, used as LANs, also known as multihop networks. Vir-
tual topology design for these networks is a different problem. One
of the reasons is that with a broadcast medium, the physical topology
does not constrain the virtual topologies that can be implemented. An-
other reason is that since each lightpath in the network needs a unique
wavelength, there is no possibility of wavelength reuse as with WDM
WANs. A survey of these problems for multihop networks can be found
in (Labourdette, 1998).
The virtual topology design problem outlined has been formulated in
terms of static traffic demands. That is, the bandwidth demand from one
node to another is considered to be known when designing the virtual
topology. This is distinct from topology design problems for networks in
which we are interested in estimating and obtaining optimum blocking
probabilities under dynamic traffic demands, that is, calls which are
established and terminated on demand (Chlamtac et al., 1993). In the
present case, if the traffic pattern changes significantly, it would act
as the input data for a new virtual topology design. The old virtual
topology would be reconfigured to the new virtual topology, a topic we
discuss in Section 5.
The advantages of optical technology lie in switching and transmis-
sion, not processing or storage. Thus, electronic switching and trans-
mission are more suitable in access networks where the bandwidth re-
quirements are low and processing requirements (as in routing or con-
solidating) are relatively high. The virtual topology design problem is
accordingly defined on transport networks only, not access networks.
The rest of this chapter is organized as follows. In Section 2, the
architecture of wavelength routed WANs is described and notations per-
taining to these are introduced. Section 3 describes approaches related
to network performance optimization, including mathematical formula-
tions and algorithms. Some particular approaches not conforming to
 Design of Logical Topologies for Wavelength Routed Networks 5

any of these categories are described in Section 4. Section 5 addresses

the reconfiguration issue. Section 6 concludes the chapter.

2. ARCHITECTURE AND NOTATIONS

Wavelength Division Multiplexing (WDM) refers to the use of distinct
wavelengths over an optical fiber to implement separate channels. An
optical fiber can carry several channels in parallel, each on a particular
wavelength. An add/drop multiplexer (ADM) is an optical system that
is used to modify the flow of traffic through a fiber at a routing node
(Gerstel et al., 1998). An ADM passes traffic on certain wavelengths
through without interruption or optoelectronic conversions, while other
wavelengths are added or dropped, carrying traffic originating or ter-
minating at the node. A Wavelength Router (WR) is a more powerful
system than an ADM. It takes in a signal at each of the wavelengths
at an input port, and routes it to a particular output port, indepen-
dent of the other wavelengths (Chlamtac et al., 1993; Ramaswami and
Sivarajan, 1996). A WR with N input and N output ports capable of
handling W wavelengths can be thought of as W independent N × N
switches. These switches have to be preceded by a wavelength demul-
tiplexer and followed by a wavelength multiplexer to implement a WR.
They are sometimes also called Wavelength Routing Switches (WRS)
or wavelength crossconnects. A Wavelength Converter is an optical de-
vice that can be used in an optical router, to convert the wavelength a
channel is being carried on without intermediate optoelectronic conver-
sion (Ramaswami and Sasaki, 1997). Wavelength conversion allows a
clear optical channel to be carried on different wavelengths on different
physical links. Different levels of wavelength conversion capability are
possible. Full wavelength conversion capability implies that any input
wavelength may be converted to any other wavelength. Limited wave-
length conversion denotes that each input wavelength may be converted
to any of a specific set of wavelengths, which is not the set of all wave-
lengths for at least one input wavelength. If a node has limited or full
wavelength conversion capability, then the conversion to be effected can
be configured as part of the virtual topology design.
The advantage of wavelength conversion is that the virtual topology
that can be implemented is less constrained, since the wavelength con-
tinuity constraint is removed. Thus wavelength use is more efficient.
However, the use of converters increases cost, as well as the complexity
of the problem. The cost can be decreased by using limited conversion
rather than full conversion, and assuming a small number of converters
rather than conversion capability in every node. But these assumptions
6

introduce the problems of specifying the nature of the limited conversion

and placement of converters in the network, which greatly increase the
difficulty of topology design.
The virtual topology designed and implemented on a physical network
not only determines the performance of the network in terms of metrics
like throughput, but also carries a cost associated with the virtual topol-
ogy, determined by how many and what network components are used to
implement that virtual topology. Attempting to model the network cost
is a topic related to the virtual topology design problem. The primary
goal of such studies is to provide a measure of the relative impact of var-
ious system components on system cost, and hence provide guidelines
for economically efficient virtual topology design, rather than actually
determine the cost of implementing a virtual topology. Comparatively
few studies have been undertaken in this area. Guidelines that result
from such studies may relate to choosing some initial parameters for
the virtual topology, as suggested in (Banerjee and Mukherjee, 1997),
or may be integrated into the optimization procedure to find the virtual
topology. The latter approach is taken in (Chen and Banerjee, 1995),
where a heuristic is designed for the topology design problem with a goal
of maximizing wavelength utilization in the Wavelength Routers, which
would certainly have an impact on the cost of the virtual topology.

2.1 NOTATIONS
In this section, we define some terminology and notations and in-
troduce some concepts which will be used in the following sections, and
which are common to most formulations of the virtual topology problem.

Physical Topology A graph Gp (V, Ep ) in which each node in the

network is a vertex, and each fiber optic link between two nodes is an
arc. Each fiber link is also called a physical link, or sometimes just a
link. The graph is usually assumed to be undirected, because each fiber
link is assumed to be bidirectional. There is a weight associated with
each of the arcs which is usually the fiber distance or propagation delay
over the corresponding fiber.
Lightpath A lightpath is a clear optical channel between two nodes.
That is, traffic on a lightpath does not get converted into electronic
forms at any intermediate nodes, but remains and is routed as an optical
signal throughout. With the usual wavelength continuity constraint,
the lightpath becomes a sequence of physical links forming a path from
source to destination, along with a single wavelength which is set aside
on each of these links for this lightpath.
 Design of Logical Topologies for Wavelength Routed Networks 7

Virtual Topology A graph Gv (V, Ev ) in which the set of nodes is the

same as that of the physical topology graph, and each lightpath is an arc.
It is also called the logical topology, and the lightpaths are also called
logical links. This graph is assumed to be directed, since a lightpath
may exist from node A to node B while there is none from node B to
node A. This graph is also weighted, with the lightpath distance of each
lightpath (see below) acting as the weight of the corresponding arc.
Link Indicator Whether a physical link exists in the physical topol-
ogy from a node l to another node m, denoted by p lm which is 1 if such
a link exists in the physical topology and 0 if not.
Lightpath Indicator Whether a lightpath exists from a node i to
another node j, denoted by bij which is 1 if such a lightpath exists in
the virtual topology and 0 if not.
Lightpath Distance The propagation delay over a lightpath, denoted
by dij for the lightpath from node i to node j. It is the sum of the
propagation delays over the physical links which make up the lightpath
in the virtual topology.
Physical Degree The physical degree of a node is the number of
physical links that directly connect that node to other nodes.
Virtual Degree The virtual (or logical) degree of a node is the num-
ber of lightpaths connecting that node to other nodes. The number of
lightpaths originating and terminating at a node may be different, and
we denote them by virtual out-degree and virtual in-degree respectively.
We speak simply of the virtual degree if these are assumed to be equal,
as they often are. If this degree is assumed to be same for all nodes of
the network, then this is called the virtual degree of the network. The
virtual degree is determined in part by the physical degree, but is also
affected by the consideration of what volume of electronic switching can
be done at a node (Ramaswami and Sivarajan, 1996).
Physical Hops The number of physical links that make up a lightpath
is called the physical hop length of that lightpath.
Logical Hops The number of lightpaths a given traffic packet has
to traverse, in order to reach from source to destination node over a
particular virtual topology, is called the virtual or logical hop length of
the path from that source to that destination in that virtual topology.
Traffic Matrix A matrix which specifies the average traffic between
every pair of nodes in the physical topology. If there are N nodes in the
network, the traffic matrix is an N × N matrix Λ = [λ (sd) ], where λ(sd)
is the average traffic from node s to node d in some suitable units, such
8

as arriving packets per second, or a quantized bandwidth requirement.

This matrix provides in numerical terms the nature of how the total
network traffic is distributed between different source-destination node
pairs, that is, the pattern of the network traffic.
Virtual Traffic Load When a virtual topology is established on a
physical topology, the traffic from each source node to destination node
must be routed over some lightpath. The aggregate traffic resulting over
a lightpath is the load offered to that logical link. If a lightpath exists
from node i to node j, the load offered to that lightpath is denoted by
λij . The component of this load due to traffic from source node s to
(sd)
destination node d is denoted by λij . The maximum of the logical
loads is called the congestion, and denoted by λ max = maxi,j λij .

2.2 ARCHITECTURE
In this section we characterize in more detail the WDM wavelength
routed network we have been describing above, and which Fig. 1.1 il-
lustrates. The network consists of several routing nodes which are con-
nected to each other by point-to-point optical fibers, specified by the
physical topology. Each of the routing nodes may have access nodes
connected to it. For the purposes of virtual topology design, however,
only the aggregate traffic between routing nodes is important. Thus we
can assume that each routing node has exactly one access node con-
nected to it. We concentrate on the routing nodes and refer to them
simply as nodes. The traffic matrix specifies the aggregate traffic from
every node to each of the other nodes.
The fiber links connecting the nodes each support a specific number of
wavelengths, say W . Each of the nodes is equipped with a WR capable
of routing these W wavelengths. In general, no wavelength conversion
capability is assumed to exist at any of the nodes.
Lightpaths are set up on the physical topology, creating the virtual
topology. A lightpath is set up by configuring the source and destination
nodes to originate and terminate a specific wavelength, then choosing a
path from the source to destination node and configuring the WR at each
intermediate node on that path to forward that wavelength optically
to the next node. Two lightpaths that share a physical link must be
assigned different wavelengths. The total number of wavelengths used
on all links must be W or less. It is usually assumed that the numbers of
lightpaths terminating and originating at each node are equal, and this
number is same for each node. Thus the network is usually assumed to
have a unique logical degree.
 Design of Logical Topologies for Wavelength Routed Networks 9

Traffic is routed from each source to destination node over a single

lightpath if one exists for that source and destination, or a sequence of
more than one lightpaths or logical hops. It is usually assumed to sim-
plify the optimization problem that traffic for a single source-destination
pair may be bifurcated over different virtual routes. The aim of creating
the virtual topology is to ensure that more traffic can be carried with
fewer optoelectronic conversions along the way. The extreme case of
this would be if a lightpath could be set up from each source to each
destination; however, the number of wavelengths available is usually too
limited to allow this. At the other extreme is a virtual topology which is
identical to the physical topology, so that optoelectronic conversion oc-
curs at every intermediate node. With reasonable and achievable virtual
topologies, the number of optoelectronic conversions should not be very
large. Together with the fact that in high speed wide area networks the
propagation delay dominates over the queueing delay (as long as links
are not loaded close to capacity), queueing delays are typically neglected
in the problem formulation (Ramaswami and Sivarajan, 1996).
The goal of the virtual topology design process is usually to optimize
some network performance metric. Thus, a particular formulation of the
problem may seek to minimize network congestion, or minimize average
packet delay. In the optimization, usually the number of wavelengths
available is taken as a constraint. If both minimizations are desired, then
one of them is usually expressed as a constraint by relating it to a known
physical network characteristic. In general both are important because
too little emphasis placed on the congestion aspect usually results in
a virtual topology very similar to the physical topology, and too little
emphasis placed on the delay aspect can result in virtual topologies
which bear little relation to the physical topology, with long lightpaths
that increase delay (Ramaswami and Sivarajan, 1996).

3. PERFORMANCE OPTIMIZATION
In this section we provide an exact formulation of the virtual topology
design problem using the packet traffic approach, and discuss specific
techniques and heuristics used to solve it.

3.1 FORMULATION
The exact formulation of the virtual topology problem is usually given
as a Mixed Integer Linear Program. The formulation provided here fol-
lows closely that in (Krishnaswamy and Sivarajan, 1998), and also those
in (Ramaswami and Sivarajan, 1996; Mukherjee et al., 1994; Mukherjee
10

et al., 1996). The symbols and terminology are as defined in Section 2.1.
New terminology is defined as necessary.

Additional Definitions Let H = [hij ] be the allowed physical hop

matrix, where hij denotes the maximum number of physical hops a
lightpath from node i to node j is allowed to take. This hop matrix is
one of the ways to characterize the bounds which lightpaths in the vir-
(k)
tual topology must be within. Let cij be the lightpath wavelength
(k)
indicator, i.e., cij is 1 if a lightpath from node i to node j uses the
(k)
wavelength k, 0 otherwise. Let cij (l, m) be the link-lightpath wave-
length indicator, to indicate whether the lightpath from node i to node
j uses the wavelength k and passes through the physical link from node
l to node m. Let ∆l denote the logical degree of the virtual topology.

Objective: Subject to the constraints below, minimize the congestion

of the network, that is,
min λmax (1.1)

Degree Constraints
X
bij ≤ ∆l , ∀i (1.2)
j
X
bji ≤ ∆l , ∀i (1.3)
j

Traffic Constraints

λij ≤ λmax , ∀(i, j) (1.4)

X (sd)
λij = λij , ∀(i, j) (1.5)
sd
(sd)
λij ≤ bij λ(sd) , ∀(i, j), (s, d) (1.6)
 
(sd) ,
 λ
 s=i 

X (sd) X (sd)
λij − λji = −λ(sd) , d=i ∀(s, d) (1.7)
j j  0,

s 6= i, d 6= i 


Wavelength Constraints
W −1
X (k)
cij = bij , ∀(i, j) (1.8)
k=0
 Design of Logical Topologies for Wavelength Routed Networks 11
(k) (k)
cij (l, m) ≤ cij , ∀(i, j), (l, m), k (1.9)
X (k)
cij (l, m) ≤ 1, ∀(l, m), k (1.10)
ij

PW −1 P (k) PW −1 P (k)
k=0 l cij (l, m)plm − k=0 l cij (m, l)pml
 bij , m=j
 
 (1.11)
= −bij , m = i ∀(i, j), m
0, m=6 i, m 6= j
 

Hop Constraints
X (k)
cij (l, m) ≤ hij , ∀(i, j), k (1.12)
lm

Discussion Most of the above constraints are self-explanatory. Many

of them enforce the consistency between the various parameters and
variables of the formulation. Constraint (1.7) asserts the conservation of
traffic at lightpath endpoints. Expression (1.11) asserts the conservation
of every wavelength at every physical node for each lightpath.
The parameters, or inputs, to the formulation are the traffic matrix
Λ, the hop bound matrix H, the number of wavelengths in a fiber W ,
the desired logical degree ∆l , and the details of the physical topology
graph. The variables, whose values at optimum are the “output” of the
MILP, relate to the virtual topology graph, wavelength assignment in
the virtual topology, and the traffic routing over the virtual topology.
The lightpath indicators bij provide the virtual topology graph. The
lightpath wavelength and link-lightpath wavelength indicators provide
the wavelength assignments to the lightpaths in the virtual topology
and also the physical links implementing each lightpath. Lastly, the
(sd)
virtual traffic load variables λij and λij provide the routing of the
traffic between each source and destination on the virtual topology. This
formulation allows for no more than one lightpath from one node to
another.
Formulations of this problem are possible that address only some and
not all of these aspects. In Section 3.2 we discuss such approaches. Even
when all these aspects are addressed, or the same aspect is addressed,
different formulations of the problem are possible. Specific formulations
can be found in the literature but are not discussed here.
This formulation gets quickly intractable with the size of the network.
One of the ways it can be made more tractable is to aggregate traffic
12

from a given source node to all destination nodes, that is, not formulate
the problem in terms of the traffic components between each source-
destination pair λ(sd) , but traffic components for each source node λ (s)
only. This results in a more tractable formulation because the num-
ber of variables and constraints is smaller, otherwise the formulation is
similar. Of course, a solution to such an aggregation does not provide a
complete solution, moreover there may be no corresponding complete so-
lution. However, the aggregate problem, being less constrained than the
original one, helps set achievability bounds on the full problem, such as
lower bounds on the achievable congestion (Ramaswami and Sivarajan,
1996; Krishnaswamy and Sivarajan, 1998). Bounds which can be calcu-
lated with significantly lower computational costs than solving the full
problem are useful in evaluating heuristics, as discussed in Section 3.2.
Usually, such an aggregate formulation is used after relaxing the MILP
above into an LP, that is, allowing the lightpath, lightpath wavelength
and link-lightpath wavelength indicator variables to take up values from
the continuous interval [0, 1] rather than constraining them to be bi-
nary variables. The relaxation, like the aggregate formulation, results
in a less constrained formulation, When the MILP is relaxed, an extra
“cutting plane” constraint is introduced (Ramaswami and Sivarajan,
1996; Krishnaswamy and Sivarajan, 1998), to ensure that the definition
of congestion remains consistent with the MILP formulation when traffic
components may be weighted with the “fractional lightpaths” that the
relaxation introduces.

3.2 HEURISTICS
The problem, whose exact formulation is given in Section 3.1, and
some of its subproblems are known to be NP-hard (Chlamtac et al., 1992;
Mukherjee et al., 1994; Krishnaswamy and Sivarajan, 1998; Banerjee and
Mukherjee, 1996). Thus for networks of moderately large sizes it is not
practical to attempt to solve this problem exactly. Heuristics to obtain
good approximations are needed. In the rest of this section we discuss
heuristic approaches to the virtual topology design problem or to related
subproblems.

Subproblems The full virtual topology design problem can be ap-

proximately decomposed into four subproblems. The decomposition is
approximate or inexact. Solving the subproblems in sequence and com-
bining the solutions may not result in the optimal solution for the fully
integrated problem. It is also possible that some later subproblem may
have no solution given the solution obtained for an earlier subproblem,
so no solution at all to the original problem may be obtained. Although
 Design of Logical Topologies for Wavelength Routed Networks 13

this decomposition follows (Mukherjee et al., 1996), it is also consistent

with the decompositions of (Ramaswami and Sivarajan, 1996; Mukherjee
et al., 1994; Krishnaswamy and Sivarajan, 1998; Banerjee and Mukher-
jee, 1996). The subproblems are as follows.
1. Topology Subproblem: Determine the virtual topology to be
imposed on the physical topology, that is determine the lightpaths
in terms of their source and destination nodes.
2. Lightpath Routing Subproblem: Determine the physical links
which each lightpath consists of, that is route the lightpaths over
the physical topology.
3. Wavelength Assignment Subproblem: Determine the wave-
length each lightpath uses, that is assign a wavelength to each
lightpath in the virtual topology so that wavelength restrictions
are obeyed for each physical link.
4. Traffic Routing Subproblem: Route packet traffic between
source and destination nodes over the virtual topology obtained.
In terms of the formulation provided in Section 3.1, the topology sub-
problem consists of determining the values of the lightpath indicator
variables bij , the lightpath routing subproblem consists of determining
(k)
the values of the variables cij (l, m), the wavelength assignment sub-
(k)
problem consists of determining the values of the variables c ij , and
the traffic routing subproblem consists of determining the values of the
(sd)
variables λij . It may be noted that the above description of the light-
(k)
path routing subproblem is approximate, since determining c ij would
(k)
be redundant after determining cij (l, m).
The traffic routing subproblem may appear to be not essential to the
virtual topology design issue. Indeed, once the virtual topology is fixed
by solving the first three subproblems, the traffic routing subproblem is
the known one of routing traffic over a given topology, for which many
algorithms exist. However, it is included in the list of subproblems since
in the exact formulation it is an integral part of the problem to determine
how traffic flows over the virtual topology being designed, as it should
be to optimize network performance metrics.
As we remarked above, the decomposition into subproblems is inex-
act. Exact solution of all the subproblems is also not possible since some
of the subproblems are NP-hard as well. Heuristics must be employed
to obtain good solutions to the subproblems. This also leads to the pos-
sibility of obtaining no solution to the full problem. Some constraints
14

are usually relaxed so that at least some solution is obtained from the
heuristics, which can be then tested for near optimality using achievabil-
ity bounds as we discuss in the following section. One of the constraints
which is commonly relaxed is that of the maximum number of wave-
lengths that can be carried by a fiber. Sanity checks must be performed
at the end to verify that the solution obtained is feasible.
The virtual topology problem can be decomposed into different sub-
problems than the ones we list above. Such different decompositions
are used in many of the studies we survey. However, we consider the
above decomposition to be reasonable and fairly consistent with any
others proposed in the literature we survey, and we shall refer only to
this decomposition while discussing such studies.

Bounds To evaluate an approximate solution produced by a heuristic,

we would like to know how close the obtained solution is to the optimal
one. Since we are using the heuristic because of the very reason that the
optimal solution cannot be obtained in the first place, we must resort
to comparing the solution obtained with known bounds on the optimal
solutions derived from theoretical considerations. These are the achiev-
ability bounds we have mentioned before (they are bounds on what can
be achieved in principle) and we discuss them below.

Lower Bounds on Congestion The goal of virtual topology design

is often to minimize network congestion, as in our formulation in Sec-
tion 3.1. A lower bound on the congestion obtained from theoretical
considerations allows us to know that an even smaller value of congestion
cannot be achieved by any solution, and helps us evaluate the solution
produced by some heuristic. We discuss several lower bounds on con-
gestion below. Our discussion follows closely that of (Ramaswami and
Sivarajan, 1996), and also that of (Krishnaswamy and Sivarajan, 1998),
as well as literature on virtual topology problems in broadcast LAN sce-
narios as referred to in (Ramaswami and Sivarajan, 1996; Labourdette,
1998). More details can be found in these sources.

Physical topology independent bound: This bound utilizes the

fact that the load on each logical link would be the same, and this
would be the congestion, if the total traffic in the network were equally
distributed among all the lightpaths. The value of this congestion would
then act as a lower bound on any virtual topology that could be designed
for the network under the given traffic conditions. This bound takes into
account the total traffic demand, but not the distribution of total traffic
among the different source-destination pairs (that is, the traffic pattern).
 Design of Logical Topologies for Wavelength Routed Networks 15

As such, it assumes that traffic for any source-destination pair can be

assigned to any lightpath in the virtual topology, and hence, it ignores
the physical topology.
Let H be the traffic weighted average number of logical hops in the
virtual topology. If El denotes the number of lightpaths in the virtual
topology and r denotes the total arrival rate of packets to the network,
then it is easy to see that λmax ≥ rH /El , and setting a lower bound on
H results in a lower bound on the congestion. For the traffic weighted
number of hops to be minimum, source-destination pairs with the largest
amount of traffic must be connected by a small number of logical hops.
Since there can be a maximum of N ∆l node pairs connected by a single
logical hop, we assume these are exactly the node pairs with the largest
traffic between them, and similarly for two, three, and larger number of
logical hops. The traffic weighted average number of logical hops in this
case is a lower bound. This is given by the expression H ≥ k kSk where
P

Sk is the sum of the traffic fractions (with respect to total network traffic
r) which are assumed to be carried in k hops. These traffic fractions can
be determined as follows: arrange the traffic fractions in descending
order of magnitude, and divide them in blocks, the i-th block being
made up of N ∆il successive fractions in that list. Thus the first block
consists of the first N ∆l traffic fractions, the second block consists of
the next N ∆2l traffic fractions, and so on. Then the sum S k is the sum
of the traffic fractions which form the k-th block.

Minimum flow tree bound: This bound is derived from similar con-
siderations as above, but on a per node basis. In this bound, we take
into account the restriction that each source node can only source ∆ l
lightpaths altogether, in addition to the considerations above. Thus this
is a stronger bound. The traffic weighted average number of logical hops
H is bounded by assuming that each node is connected by one logical
hop to the ∆l nodes to which it has the largest amounts of traffic, by
two hops to the ∆2l nodes to which it has the next largest amounts of
traffic, and so on. We omit the derivation and exact expression of this
bound, which can be found in (Ramaswami and Sivarajan, 1996).

Iterative bound: This type of bound is developed in (Ramaswami

and Sivarajan, 1996; Krishnaswamy and Sivarajan, 1998) by aggregating
and then relaxing the MILP formulation and solving it as mentioned in
Section 3.1. The additional constraint imposed on the relaxed aggregate
formulation is that the congestion be higher than a lower bound on
the congestion known a priori, such as the minimum flow tree bound
discussed above. To improve the tightness, the value obtained for the
16

congestion by solving the relaxed aggregate LP can be used as a new

value of the a priori bound and the LP solved again to yield a further
tightened bound on the congestion. This iterative process can be carried
out repeatedly to improve the tightness of the bound.

Independent topologies bound: This bound is proposed in (Baner-

jee et al., 1997) as another method of taking into account the physical
topology in computing a bound on the congestion. Successive topologies
are derived to maximize one-hop, two-hop, three-hop traffic etc., and
these topologies are not allowed to constrain each other. Finally the
resulting congestion is read out as a bound. The authors note that this
bound is only a little tighter than the flow tree based bound if traffic is
uniform, but becomes much tighter for highly nonuniform traffic.

Lower Bounds on the Number of Wavelengths It is usually nec-

essary in virtual topology design to complete the design using as few
distinct wavelengths as possible, since in practice there is a limit on the
number of wavelengths a fiber can carry. This limit may be known and
introduced in the exact formulation as in the formulation of Section 3.1,
but such a limit is often not included in heuristic approaches. A lower
bound on the number of wavelength needed for a particular problem is
then useful in evaluating the solution provided by the heuristic. Also, in
the presence of practical limitations, an easy-to-compute lower bound on
the number of wavelengths can provide a quick negative answer to the
question of whether a virtual topology design problem is at all feasible
or not. Two such bounds can be found in (Ramaswami and Sivarajan,
1996). The first bound is derived from the simple consideration that the
node with the minimum physical degree δ p must source ∆l lightpaths.
Then the number of wavelengths required is bounded from below by
W ≥ d∆l /δp e. The second bound is derived by assuming that each node
sources lightpaths to exactly those nodes it can reach with the minimum
number of physical hops. We omit the derivation and exact expression
of this bound, which can be found in (Ramaswami and Sivarajan, 1996).
Bounds on the number of wavelengths required can also be found un-
der specific assumptions regarding the solution to the different subprob-
lems. Two such bounds are demonstrated in (Chlamtac et al., 1993), one
based on the assumption that the virtual topology being implemented is
a hypercube and a specific node mapping algorithm is used, the other re-
lating the number of wavelengths needed for topologies with and without
the wavelength continuity constraint.
 Design of Logical Topologies for Wavelength Routed Networks 17

Heuristic Approaches and Techniques In the design of heuristics

or approximate solutions to the virtual topology problem, emphasis is
placed on different aspects of the problems by different authors. In the
majority of the literature, heuristics are designed for only some and not
all the subproblems. Some assumption regarding the nature of the vir-
tual topology to be implemented is often a starting point for heuristic
methods. Below we discuss heuristics found in the literature surveyed
under three different categories. In the first, it is assumed that the vir-
tual topology to be implemented is a well-known regular topology, such
as a hypercube or a shufflenet. In the second, the lightpaths of the vir-
tual topology are assumed to be already known in terms of sources and
destinations for each instance of the problem, and the lightpath routing
and wavelength assignment subproblems are addressed. No particular
assumption is made regarding the virtual topology in the last category.
Some of the interest in the study of regular topologies in the context
of virtual topologies for WANs came from the assumption that to some
extent the virtual topology could dictate the physical topology, that is,
fibers could be laid to supplement a physical topology before implement-
ing a virtual topology. As more and more fiber has been laid in practice
and has become part of single wavelength optical networks utilizing the
fibers as point-to-point links, the concern has shifted to extracting more
utilization out of these fibers using WDM and virtual topologies, rather
than having to lay more fibers. Thus studies relating to arbitrary phys-
ical topologies have attracted more interest in recent times.

Regular Topologies Regular topologies such as hypercubes or shuf-

flenets have several advantages as virtual topologies. They are well un-
derstood, and results regarding bounds and averages are comparatively
easier to derive. Routing of traffic on a regular topology is usually also
simpler and results are available in the literature, so the traffic rout-
ing subproblem usually becomes trivial. Also, regular topologies possess
inherent load balancing characteristics.
Once a regular topology is decided upon as the one to be implemented
as a virtual topology, it remains to decide which physical node will re-
alize each given node in the regular topology (this will be referred to
as the node mapping subproblem) and which sequence of physical links
between two physical nodes will be used to realize each given edge in the
regular topology, that is, lightpath (this will be called the path mapping
subproblem). This procedure is also called embedding a regular topol-
ogy in the physical topology. In terms of the subproblems introduced in
Section 3.2, the choice of the regular topology together with the node
mapping problem make up the virtual topology subproblem, and the
18

path mapping problem corresponds to the lightpath routing subprob-

lem. Obviously, the number of nodes in the (regular) virtual topology
may not be chosen with complete freedom, instead it must obey the con-
straints of the regular topology. In case the physical topology has a few
nodes less than the regular topology, this can usually be circumvented by
adding fictitious nodes to it before embedding (Mukherjee et al., 1994).
If a few more nodes are present in the physical topology, then some of
the ones with less traffic may be combined for the purpose of embedding,
though this introduces further approximations to the virtual topology
solutions. In general, the node mapping and path mapping problems
leave out of consideration the network traffic pattern, and utilize met-
rics such as fiber distance and wavelength reuse to route lightpaths over
the physical topology. Thus there is a tacit assumption of a uniform or
close to uniform traffic pattern in the use of regular topologies as virtual
topologies. The mappings must also be free of wavelength clashes and
must obey any predefined limit on the number of wavelengths. These
problems are themselves known or conjectured to be NP-hard (Chlam-
tac et al., 1993; Mukherjee et al., 1994), hence heuristics are needed for
them.
We omit a detailed discussion of the studies available in the literature
on this topic and only mention a few of them. In (Chlamtac et al., 1993),
the physical topology is first mapped into an “equivalent string” preserv-
ing some characteristics of the topology, and then the selected regular
topology is embedded into this string. Three different regular topolo-
gies are compared for their suitability as virtual topologies in (Marsan
et al., 1993). The comparison is based on the number of logical hops
between nodes and the number of wavelengths required. In (Mukherjee
et al., 1994), heuristics are developed for the node mapping problem for
regular topologies (specifically, hypercubes), and the solution is carried
through to path mapping as well as wavelength assignment. Thus, the
node mapping part of the virtual topology subproblem, the lightpath
routing, and the lightpath wavelength assignment subproblems are ad-
dressed. A greedy algorithm and a simulated annealing algorithm are
specified for obtaining an initial mapping and then refining the mapping
based on considerations of minimizing overall network message delay. In
(Mukherjee et al., 1996), a very similar simulated annealing heuristic
is presented, with the difference that the traffic routing subproblem is
assumed to be solved using the flow deviation method. The flow devia-
tion method is a good heuristic alternative to an exactly optimal linear
programming routing flow solution. The literature in which it was de-
veloped is referred to in (Mukherjee et al., 1996) and also (Labourdette,
1998). This method starts from an initial flow assignment, and iter-
 Design of Logical Topologies for Wavelength Routed Networks 19

atively deviates flows over alternate paths, avoiding links carrying the
largest amounts of traffic.

Pre-specified Topologies We now discuss studies which focus on the

lightpath routing subproblem, and possibly the wavelength assignment
and traffic routing subproblems. In other words, the virtual topology in
terms of a list of lightpaths with their source and destination nodes is
supposed to be given for each instance of the problem.
The traffic pattern in the network would certainly have been taken
into account when the lightpaths were decided upon. Because of this,
in some approaches we discuss below, the traffic pattern is not taken
into account, though it is also possible to utilize the traffic pattern in-
formation in routing lightpaths. The lightpath routing and wavelength
assignment subproblems can then be viewed as having goals defined
purely in terms of the lightpaths, such as minimization of the number
of distinct wavelengths needed.
We mention a few of the studies in the literature, omitting detailed
discussion as before. In (Chlamtac et al., 1992), not only the source and
destination, but also the routing of the lightpaths is assumed to be given.
That is, the lightpath wavelength assignment subproblem is addressed,
and it is called the Static Lightpath Establishment (SLE) problem. It
is proved that SLE as stated is equivalent to the n-graph-colorability
problem, and hence NP-complete. A greedy heuristic algorithm to as-
sign wavelengths to a given set of lightpaths with the aim of using as few
wavelengths as possible is presented. The study presented in (Chen and
Banerjee, 1995) assumes that the virtual topology subproblem has been
solved and the set of lightpaths to be established is available in terms
of the source and destination nodes of the lightpaths. The objective for
the routing and wavelength assignment problem presented is to maxi-
mize wavelength utilization at the switches. This objective is presented
in terms of the utilization of the Wavelength Routers at each network
node. To formally define the problem, the concept of a “Latin Square”
is introduced. Lightpath routing and wavelength assignment is posed in
terms of completion of partial Latin Squares. Two heuristic algorithms
are presented to complete partial Latin Squares at individual network
nodes, and then a scheme is specified to use these in combination to
solve the lightpath routing and wavelength assignment problem at the
network level. In (Banerjee and Mukherjee, 1996), the virtual topol-
ogy is assumed to be given in terms of a list of lightpaths with their
source and destination nodes for each instance of the problem. The
lightpath routing problem is formulated in terms of lightpath traffic as
a multicommodity flow problem which is known to be NP-complete. It
20

is suggested that the problem size can be reduced considerably by cus-

tomizing the formulation for each instance of the problem, by pruning
the search tree for lightpath routes and relaxing integer constraints. The
study employs known heuristic methods, including randomized rounding
and graph coloring, with provably good characteristics, to address the
lightpath routing and wavelength assignment problems.

Arbitrary Topologies There are various studies proposing heuristic

methods for arbitrary virtual topologies. These studies address the vir-
tual topology subproblem itself, as well as some or all of the subsequent
subproblems of virtual topology design. Most of these methods take into
account the effect of the network traffic pattern, since arbitrary virtual
topologies are usually called for in response to non-uniform traffic pat-
terns and irregular physical topologies. Some of the heuristics proposed
are similar to each other.
In (Zhang and Acampora, 1995), the problem is looked upon as the
establishment of an optical connection graph over a WAN based on the
average traffic demand, and then using demand based routing on this
connection graph, that is, dynamic virtual circuits, which allocate whole
lightpaths at a time. The connection graph subproblem presented is
therefore identical to the first three subproblems of the virtual topology
problem as presented in Section 3.2. The problem is formulated as a
nonlinear integer programming problem, and an approximate decompo-
sition is presented. The heuristic algorithm is then presented, which
is based on a greedy approach, and attempts to utilize the number of
wavelengths used for a maximum number of lightpaths.
Several different heuristics are presented in (Ramaswami and Sivara-
jan, 1996), including one which also attempts to create lightpaths be-
tween nodes in order of decreasing traffic demands. Lightpaths are es-
tablished between the nodes that have the maximum amount of traffic
between them. If all traffic is accounted for but each node does not have
the required degree, the rest of the lightpaths are placed at random obey-
ing the constraints. A modified version of this heuristic is also presented
in which a pair of lightpaths in opposite directions is initially set up for
each physical edge, then the original algorithm is exactly followed. This
ensures that traffic can always be routed on the shortest physical path
between any two nodes and hence can satisfy any physically realizable
delay constraints. Another heuristic depends on the iterative bound de-
veloped in this study by relaxing the MILP formulation as described in
Section 3.2, and rounding off the lightpath indicators. Finally, a heuris-
tic is presented that does not take into account the traffic pattern at
 Design of Logical Topologies for Wavelength Routed Networks 21

all, but concentrates on creating lightpaths that use only a few physical
edges, since this should conserve wavelengths.
A similar heuristic maximizing one logical hop traffic is briefly de-
scribed in (Banerjee and Mukherjee, 1997), but a heuristic with the
opposite objective is also suggested. This heuristic aims at maximiz-
ing multihop traffic, since concentrating only on single hop traffic can
lead to congestion due to multihop traffic. Some results are provided
in which the two approaches appear to perform very similarly to each
other. Details of wavelength assignment are not discussed.
The study in (Banerjee et al., 1997) also suggests that attempts to
maximize one logical hop traffic concentrate on the comparatively larger
traffic components, and may cause the smaller traffic components to be
routed unreasonably and cause congestion on some physical links. A
scheme involving mapping the network to a bipartite graph is specified
to avoid unbalanced loading. A known graph algorithm is then specified
as a heuristic for wavelength assignment.
In (Krishnaswamy and Sivarajan, 1998), a heuristic algorithm follow-
ing the LP relaxation heuristic from (Ramaswami and Sivarajan, 1996),
but more complete, is presented. The lightpath wavelength indicator
variables are also rounded and then a least resistance algorithm followed
to choose a single routing for each lightpath. A final phase of the design
eliminates wavelength clashes.

4. RELATED APPROACHES
In this section we discuss some techniques and algorithms that are
different from those described in Section 3.2, but which are related to
the problem of virtual topology design for wavelength routed networks.

Incremental Benefit Analysis: In (Mukherjee et al., 1996), a study

of the incremental benefits of introducing a virtual topology over optical
WANs is undertaken. A realistic traffic pattern is obtained from the T1
NSFNET backbone data of January 1992. Three schemes are applied on
this data to scale up the traffic pattern. The first scheme merely used
efficient routing to establish a baseline and the other two used WDM
and virtual topology with WDM respectively. The most dramatic result
was in the increase of the scaleup factor, from 49 and 57 in the first
two schemes to 106 in the third, and the link utilization which went
from 32% and 23% in the minimum loaded link to 71% in the last one
(the maximum link load remained 99% in all three schemes). Thus
this analysis provides demonstration of the benefits of implementing a
virtual topology, as well as the incremental nature in which it may be
undertaken.
22

Limited Conversion: The motivation for the study presented in (Ra-

maswami and Sasaki, 1997) is the lower cost associated with limited
conversion of wavelengths at Wavelength Routers as opposed to full
conversion, as we remarked in Section 2. The wavelength assignment
subproblem is the focus of this study. Several results are obtained in
theoretical terms about ring networks with specific wavelength conver-
sion capabilities. These results are followed by constructive proofs rather
than simply existence proofs, so that a blueprint is provided for the ac-
tual construction of such ring networks. Some results are derived for
more general physical network topologies.

Traffic Grooming: As we have remarked in Section 1, each lightpath

has a high bandwidth and it may not be possible for single users to uti-
lize this bandwidth. Lightpaths must be viewed as transport channels in
the backbone network, in which traffic from multiple user applications is
multiplexed in by access networks. In a sense, this is the justification for
including the traffic routing subproblem in the virtual topology design
problem, since traffic for individual applications must be routed onto the
virtual topology provided, so that lightpaths carry traffic obtained by ag-
gregating lower speed traffic streams. The pattern of multiplexing traffic
onto lightpaths affects the efficiency of optical forwarding of information
through Wavelength Routers, since all information in an entire lightpath
will need to undergo optoelectronic conversion and electronic routing at
an intermediate node if even one lower speed traffic stream from that
lightpath has to be terminated at the intermediate node. This also re-
flects in the cost (in numbers and capabilities) of network components
needed. (Gerstel et al., 1998) addresses some of these issues. Different
ring architectures are specified and compared on the basis of results de-
rived regarding the average number of transceivers at the nodes, number
of wavelengths, average number of physical hops and characterization of
traffic patterns on which they perform best. Similar issues arising in
arbitrary physical topologies and extension to actual grooming methods
would appear to be areas worth further investigation.

Generalized Lightpaths: In (Sahasrabuddhe and Mukherjee, 1999),

the concept of a lightpath is generalized into that of a lighttree, which,
like a lightpath, is a clear channel implemented with a single wavelength
with a given source node. But unlike the lightpath, a lighttree has mul-
tiple destination nodes, thus a lighttree is a point-to-multipoint channel.
It is emphasized that topologies using lighttrees would be more optimal
for any given situation since the lighttree is a more general construct,
and there may be possibilities of optical multicast.
 Design of Logical Topologies for Wavelength Routed Networks 23

As we already know, optimal solutions are not practically obtainable,

and with a more general construct and hence a much larger search space
this is going to be even more true. Heuristic solutions will have to
be designed to obtain good solutions, and must be tailored to suit the
larger search space. With unicast traffic problems, the lighttree approach
trades off more bandwidth to further improve delay, congestion, and
physical hop characteristics than the lightpath approach. This is the
tradeoff we mentioned in Section 1. The challenge in this case will be
to design heuristics that can cope with the increased complexity of the
problem and yet produce solutions in which a good tradeoff is achieved.

5. RECONFIGURATION ISSUES
As we have already remarked, the problem of reconfiguring a network
from one virtual topology to another is a related problem to virtual
topology design. Two possible approaches to this problem are discussed
in this section.

Cost Approach In this approach, it is assumed that the current vir-

tual topology as well as the new virtual topology that the network must
be reconfigured to are known, together with the physical topology de-
tails. The concern is to minimize the cost of the reconfiguration. The
cost can be expressed in terms of the number of Wavelength Routers
that need to have their optical switching reprogrammed, or the total
number of optical switchings that need to be changed to implement the
new lightpaths and eliminate old ones. These metrics are appropriate
since they reflect the amount of time the network must be taken off line
to make the changes, as well as the reprogramming effort for the recon-
figuration. Other similar metrics may also be applicable. It may be the
case that the network cannot be taken off line at all, but that a succes-
sion of intermediate virtual topologies have to be designed to eliminate
single, or groups of, routers which can be reconfigured and put back in
operation. Much more complicated metrics reflecting total time taken to
reconfigure as well as the effort to redesign the intermediate topologies
need to be developed in this case.
We have not found any study of these reconfiguration problems in the
literature for wavelength routed WANs, though studies involving the re-
configuration of virtual topologies for broadcast LANs exist, as detailed
in the survey of related literature carried out in (Labourdette, 1998).
These studies involve link-exchange and branch-exchange techniques to
minimize the cost of converting one virtual topology into another, and
similar methods may be possible for wavelength routed network which
are the topic of this survey.
24

Optimization Approach Another approach is to assume that only

the current virtual topology is given, together with the changed traffic
pattern and/or physical topology that makes reconfiguration necessary.
This is the approach taken in (Banerjee and Mukherjee, 1997). The re-
configuration algorithm proposed involves solving the new virtual topol-
ogy problem on its own without reference to the current virtual topology
to obtain a new optimal solution, with a new optimal value for the ob-
jective function which is noted. The virtual topology design problem is
then reformulated with an additional constraint that constrains the old
objective function to this noted value, and a new objective function that
involves minimizing the number of lightpaths that must be either added
or removed.
While this method is guaranteed to find a solution that results in
a virtual topology that is optimal for the new conditions, it does not
achieve a balance between finding an optimal new virtual topology and
one that involves as little change from the old one as possible. It is
possible that a very costly reconfiguration will be undertaken for only a
slight gain in network performance. More balanced formulations of this
problem may be possible, and heuristics designed on such formulations
are likely to perform better in practice.

6. CONCLUDING REMARKS
The problem of virtual topology design for wide area wavelength
routed optical networks covers a considerable area, and many approaches
to this and related problems have been taken in the literature. The light-
paths of a virtual topology are set up to trade off the ample bandwidth
available in the fiber with the optoelectronic conversion and electronic
processing at intermediate nodes. Exact formulations of the problem are
known to be computationally intractable, so heuristics for determining
and implementing a virtual topology have been proposed. Most heuris-
tics attempt to address parts of the problem rather than the whole, by
decomposing the problem approximately into subproblems, or address
special cases of network topology.
Virtual topology design is a growing research area. New areas of inves-
tigation include extending results obtained for special cases to broader
context, and extending the freedom allowed in formulating the problem
to take advantage of improving equipment capabilities.
References

Banerjee, D. and Mukherjee, B. (1996). A practical approach for routing

and wavelength assignment in large wavelength-routed optical net-
works. IEEE JSAC, 14(5):903–908.
Banerjee, D. and Mukherjee, B. (1997). Wavelength-routed optical net-
works: Linear formulation, resource budgeting tradeoffs, and a recon-
figuration study. Proc. IEEE INFOCOM, pages 269–276.
Banerjee, S., Yoo, J., and Chen, C. (1997). Design of wavelength-routed
optical networks for packet switched traffic. Journal of Lightwave
Technology, 15(9):1636–1646.
Chen, C. and Banerjee, S. (1995). Optical switch configuration and light-
path assignment in wavelength routing multihop lightwave networks.
Proc. IEEE INFOCOM, pages 1300–1307.
Chlamtac, I., Ganz, A., and Karmi, G. (1992). Lightpath communica-
tions: An approach to high bandwidth optical wans. IEEE Trans.
Communications, 40(7):1171–1182.
Chlamtac, I., Ganz, A., and Karmi, G. (1993). Lightnets: Topologies
for high-speed optical networks. Journal of Lightwave Technology,
11(5/6):951–961.
Gerstel, O., Ramaswami, R., and Sasaki, G. (1998). Cost effective traffic
grooming in WDM rings. Proc. IEEE INFOCOM, pages 69-77.
Green, P. (1992). Fiber optic network. Englewood Cliffs, NJ Prentice
Hall.
Green, P. (1996). Optical networking update. IEEE JSAC, 14(5):764–
779.
Krishnaswamy, R. and Sivarajan, K. (1998). Design of logical topologies:
a linear formulation for wavelength routed optical networks with no
wavelength changers. Proc. IEEE INFOCOM, pages 919–927.

25
 26

Labourdette, J.-F. P. (1998). Traffic optimization and reconfiguration

management of multiwavelength multihop broadcast lightwave net-
works. Computer Networks and ISDN Systems, 30(9-10):981–998.
Marsan, M., Bianco, A., Leonardi, E., and Neri, F. (1993). Topologies
for wavelength-routing all-optical networks. IEEE/ACM Trans. Net-
working, 1(5):534–546.
Mukherjee, B. (1997). Optical communication networks. McGraw-Hill.
Mukherjee, B., Ramamurthy, S., Banerjee, D., and Mukherjee, A. (1994).
Some principles for designing a wide-area optical network. Proc. IEEE
INFOCOM, pages 110–119.
Mukherjee, B., Ramamurthy, S., Banerjee, D., and Mukherjee, A. (1996).
Some principles for designing a wide-area optical network. IEEE/ACM
Trans. Networking, 4(5):684–696.
Ramaswami, R. and Sasaki, G. (1997). Multiwavelength optical networks
with limited wavelength conversion. Proc. IEEE INFOCOM, pages
489–498.
Ramaswami, R. and Sivarajan, K. (1996). Design of logical topologies
for wavelength-routed optical networks. IEEE JSAC, 14(5):840–851.
Ramaswami, R. and Sivarajan, K. (1998). Optical networks: a practical
perspective. Morgan Kaufmann Publishers.
Sahasrabuddhe, L. and Mukherjee, B. (1999). Light-trees: Optical mul-
ticasting for improved performance in wavelength-routed networks.
IEEE Communications Magazine, pages 67–73.
Zhang, Z. and Acampora, A. (1995). A heuristic wavelength assignment
algorithm for multihop WDM networks with wavelength routing and
wavelength re-use. IEEE/ACM Trans. Networking, 3(3):281–288.

View publication stats