Light-trail based Hierarchy
The Optimal Multicast Route in WDM Networks Without Splitters and Converters
Dinh Danh Le and Massinissa Merabet
LIRMM, University Montpellier 2, 161 rue Ada, 34095 Montpellier Cedex 5, France
Keywords:
WDM Network, Multicast Routing, TaC Cross-connect, Light-path, Light-trail, Light-hierarchy, Light-trail
based hierarchy, Wavelength Minimization, Integer Linear Programming (ILP).
Abstract:
Multicasting in WDM core networks is known as an efficient way of communications in high-speed multime-
dia applications. However, costly and complicated fabrication prevents multicast capable switches (splitters)
from deploying in the proposed architectures. Besides, in practical routing cases, the state of the network
is given by a directed graph. Accordingly, this paper investigates the multicast routing without splitters in
directed asymmetric topologies. The objective is to minimize the number of wavelengths used and then find
the best cost solution among those requiring the same number of wavelengths. In the case of no splitters, a
set of light-paths starting from the multicast source covering all the destinations is known as the traditional
solution. In this paper, we introduce two new concepts namely light-trail based hierarchy and light-path based
hierarchy, and develop two ILP formulations for them. Theoretical analysis and simulation results show that
the optimal solution is a set of light-trail based hierarchies. Particularly, our light-trail based solution achieves
fewer wavelengths required (up to 21.95% saved) while keeping slightly lower cost (up to 3.79% saved) com-
pared to light-path based solution.
1 INTRODUCTION
Being capable of supporting heavy load communica-
tions, all-optical networks are promising to be seri-
ous candidates for high-speed backbone networks. In
pure optical routing, the messages are transmitted us-
ing Wavelength Division Multiplexing (WDM) tech-
nology without electronic processing in the condition
that the computed routes should satisfy optical con-
straints (Zhang et al., 2000).
1.1 Multicast Routing Problems under
Optical Constraints
Multicast is known as the efficient way of communi-
cations to perform data transmission from a source
to several destinations. In traditional IP electronic
networks, the solutions are known as spanning trees
computed in the topology graph. In optical networks,
however, the multicast routes do not necessarily cor-
respond to trees but some structures complying sev-
eral (optical) constraints. Among the constraints, the
availability of light splitters in the switches are of-
ten the most hard ones. Light splitters (or multicast
capable switches) are special nodes capable of split-
ting an incoming signal from a predecessor to sev-
eral successors. Nevertheless, splitters are expensive
and complicated in fabrication. Besides, splitting in-
duces considerable power loss (inverselyproportional
with the number of outgoing ports (Ali and Deogun,
2000b)). Therefore, we assume to study the multi-
cast routing on the WDM networks without splitters.
In addition, since wavelength converters (the devices
that can shift a passing signal from one wavelength to
the other (Mukherjee, 2006)) are costly and immature
enough, we also exclude them from this study.
Regarding the objectives of multicast problems,
the requirements of economizing the networks re-
sources (e.g., the wavelengths) are first thing to con-
cern. Besides, among the possible routes, the least-
cost one is preferred. The total cost of the routes is
defined as the summation of the costs of the individ-
ual links of the routes and the cost of each link could
be any types of metrics including distance, monetary
cost, etc., depending on the network entity that we are
trying to minimize. However, it is usually hard to si-
multaneously minimize both metrics. Therefore the
trade-off solution is more interesting.
Another important aspect to consider is the kind
of examined networks. Most of the studies in optical
66
Le D. and Merabet M..
Light-trail based Hierarchy - The Optimal Multicast Route in WDM Networks Without Splitters and Converters.
DOI: 10.5220/0004705300660073
In Proceedings of 2nd International Conference on Photonics, Optics and Laser Technology (PHOTOPTICS-2014), pages 66-73
ISBN: 978-989-758-008-6
Copyright
c
2014 SCITEPRESS (Science and Technology Publications, Lda.)
multicast routing problems are carried out on symmet-
ric networks under the assumption that there are two
opposite directed fibers between every pair of con-
nected switches. In this case the networks are well
modeled by undirected (Zhang et al., 2000), (Din,
2009) or bidirected graphs (Ali and Deogun, 2000a),
(Zhou et al., 2010). However, it is more practical
and general to investigate the routing problems on
asymmetric networks that can be modeled by an ar-
bitrary directed graphs (or digraphs) where each arc
represents a directed fiber between a pair of nodes.
It is more practical because even if the network is
designed to be bidirected, when some demands hold
some of the network resources, the resultant topology
graph is then no longer bidirected but (arbitrarily) di-
rected, therefore the routing for subsequent demands
will be calculated on a digraph. It is also more gen-
eral because bidirected graphs are special cases of di-
graphs where every arc has its reverse one.
1.2 Related Works
Due to its interest, WDM multicast routing has been
investigated intensively in the literature and several
propositions exist to adapt multicast routing algo-
rithms to the optical constraints (cf. (Zhang et al.,
2000) for some basic algorithms and (Zhou and Poo,
2005) for a survey).
The problem of minimizing the number of used
wavelengths was investigated at first in (Li et al.,
2000). The considered network is assumed to be
equipped with splitters and wavelength converters,
and it is considered as a set of wavelength graphs
where the arcs representing the wavelengths avail-
able in the corresponding fibers. The objective is
to construct a light-tree satisfying optical constraints
such that the number of required wavelengths is min-
imized. The NP-hardness of the problem is proved,
and an approximation algorithm has been proposed.
The case of switching without splitters in sym-
metric networks has been discussed in (Ali and De-
ogun, 2000a). The problem is to find a Multiple-
Destination Minimum Cost Trail that starts from a
source and spans all the destinations with minimizing
the total cost of the links traversed. To ensure a fea-
sible solution, a low-cost cross-connect architecture
called Tap-and-Continue (TaC) has been proposed to
replace splitters. TaC cross-connects can tap a sig-
nal with small power at the local station and forward
it to one of its output ports. The problem is proved
to be NP-hard and then a heuristic (namely MDT) is
proposed. The advantage of MDT heuristic is that
only one wavelength (and one transmitter) is suffi-
cient for each multicast request (i.e, the wavelength
is minimized). However, due to multitude of round-
trip traversing, a large number of links is required in
both directions, and the total cost of the light-trail is
always very high.
The multicast routing problems without splitters
in asymmetric networks has been studied in (Le et al.,
2013). The problem is proved to be NP-hard, and
two heuristics namely Farthest First and Nearest First
are proposed. These heuristics based on light-trails.
Thanks to the interesting properties of light-trails, the
number of wavelengths can be considerably saved. In
comparison to the heuristics proposed in (Din, 2009),
they provide better solutions with fewer wavelengths
required and lower total cost. However there are no
exact solutions given to calculate their approximation
ratios.
In this paper, we study the multicast routing in
asymmetric WDM networks without splitters and
converters. Our objective is to minimize the number
of used wavelengths and then try to minimize the total
cost. To solve the problem we introduce a new con-
cept called light-trail based hierarchies and develop
two ILP formulations to search for the exact solutions.
We theoretically and experimentally show that the op-
timal solution is a set of light-trail based hierarchies.
The structure of the paper is the following. Section 2
presents the problem modeling and performance met-
rics. The concept of light-trail based hierarchy and
its benefits are given in Section 3. Then the ILP for-
mulation for it is presented in Section 4, followed by
the experimental results on their performancesin Sec-
tion 5. We conclude our paper in Section 6.
2 PROBLEM MODELING AND
METRICS
The considered network is modeled by the topology
graph G = (V, A), a simple digraph in which each
arc represents the availability of a directed fiber be-
tween a pair of nodes (we suppose that there are at
most two opposite directed fibers between any pair of
nodes). As mentioned in Section 1, we deal the multi-
cast problem in the networks which are not equipped
with any splitters but TaC cross-connects. We sup-
pose that each fiber has the same set W of available
wavelengths and each arc a A is associated with a
positive value cost(a). Given the multicast request
r = (s, D), in which s V is the source node and
D V \ {s} is the set of destinations, the routing
problem is to compute the light-structures (e.g., light-
trees) from s covering all the destinations simultane-
ously. These light-structures must comply the follow-
ing constraints (Zhou and Poo, 2005):
Light-trailbasedHierarchy-TheOptimalMulticastRouteinWDMNetworksWithoutSplittersandConverters
67
Wavelength Continuity Constraint: In the absence
of wavelength converters, the same wavelength
should be used continuously on all the links along
a light-structure.
Distinct Wavelength Constraint: Two light-
structures should be assigned with different wave-
lengths unless there are edges (or arcs) disjoint.
Degree Constraint: In the absence of light-
splitters, all the nodes (except the source) in every
light-structure should have the degree that do not
exceed two.
Without loss of generality, let LS be the set of
light-structures LS
i
, i = 1, .. . , k computed for the
given request r. Since each light-structure consumes
a distinct wavelength, the number of wavelengths
needed to performthe multicast request r is equal to k:
No. Wavelengths(r) = k. The total cost of the light-
structures is the summation of cost all the arcs in all
light-structures LS
i
:
TotalCost(r) =
k
i=1
cost(LS
i
) =
k
i=1
aLS
i
cost(a).
In our study, we first minimize the number of used
wavelengths, then try to minimize the total cost
among the solutions with the same minimal wave-
lengths.
Traditionally, the solutions correspond to light-
trees in general cases or light-paths in the case of
no splitters and no converters (as it is considering in
this paper). However, the nodes can be traversed sev-
eral times with the same wavelength as long as there
are different incoming and outgoing ports for each
passing (Zhou et al., 2010). Consequently, the solu-
tions are not necessarily sets of light-paths but sets of
light-trails. In Section 3, we introduce a new light-
structure based on light-trails call light-path based hi-
erarchies. We will prove that the problem with light-
path based solutions is NP-hard. We then compare it
with the light-path based solution to find a better solu-
tion for the considering problem. Its ILP formulation
is given in Section 4.
3 LIGHT-TRAIL BASED
HIERARCHIES
Before defining the new concept light-trail based
hierarchy, let us first introduce the concept light-
hierarchy proposed in (Molnar, 2011).
Based on the fact that the multicast routes are not
necessarily sub-graphs but any types of structures that
retain the connectivity and spanning properties, a hi-
erarchy is proposed to replace the traditional solutions
(e.g., paths, trees, etc.). It is a graph related structure
obtained by a homomorphism of a tree in a graph. Re-
call that in graphs, a homomorphism can be defined
as follows. Let Q = (W, F) and G = (V, E) be two
(both undirected or directed) graphs. Q is called the
base graph, and G is the target graph. An applica-
tion h : W V maps a vertex in W to each vertex
in V is a homomorphism if the mapping preserves the
adjacency: (u, v) F (h(u), h(v)) E. If Q is a
connected graph without cycle (a tree) then the triple
(Q, h, G) defines a hierarchy in G. If both graphs Q
and F are directed, the triple (Q, h, G) defines a di-
rected hierarchy
1
in G (Molnar, 2011). In term of
optical routing, light-hierarchy is defined as a hierar-
chy using a single wavelength. Equivalently, a light-
hierarchy is a hierarchy that has no duplicated arc but
is free of repetition of nodes (Zhou et al., 2010).
According to the definition of light-hierarchy,
when the base graph Q is a rootedtree without branch-
ing vertices (except the root correspondingto the mul-
ticast source
2
), i.e, Q is a star, the triple (Q, h, G) de-
fines a special light-hierarchy. It corresponds to a set
of rooted arc-disjoint trails in the target graph G, so a
single wavelength is needed to serve it. For this rea-
son, we call it light-trail based hierarchy (LTH).
Especially, if the mapping h is injective (i.e., each
vertexinW associates with only one vertex inV), then
the hierarchy has no duplicated vertices (and so no
duplicated arcs well as), and it corresponds to a set of
rooted elementary trails (trails without repetition of
vertices) or paths, in G. This has been considered as
the traditional solution for the problem we are exam-
ining (Din, 2009). We will call it a light-path based
hierarchy (LPH) in order to distinguish with a general
light-trail based hierarchy.
Figure 1 shows an example of a light-trail based
hierarchy. Each vertex of the star Q is associated with
an unique vertex of the graph G. In the reverse di-
rection, some vertices of G are mapped from several
vertices in Q (nodes a and f). A vertex in Q can be
labelled by the vertex in G which it is associated. To
distinguish the occurrences related to the same vertex
v in G, we will use the labels v
1
, v
2
, ..., v
k
in Q (and in
the hierarchy H as well). Notice that the degree of a
vertex occurrence v
i
in the hierarchy H is defined as
the degree of the corresponding vertex occurrence v
i
in the base graph Q (Molnar, 2011). It is important to
verify the degree constraint stated in Section 2.
With LTH solutions, the considering problem is
1
In this paper we just consider directed hierarchies, but
sometimes the word directed is omitted for the sake of sim-
plicity.
2
Because the source can be equipped with multiple
transmitters, so it can inject the same wavelength to differ-
ent successors.
PHOTOPTICS2014-InternationalConferenceonPhotonics,OpticsandLaserTechnology
68
a
d
b
c
e
g
f
h
s
a
1
d
1
b
1
c
1
e
1
g
1
f
1
h
1
s
1
a
2
f
2
Q
h
Figure 1: Mapping of vertices from a star for a light-trail
based hierarchy.
NP-hard. It can be deduced from the following theo-
rem.
Theorem 3.1. Let k be any fixed positive integer. If
P 6= NP, then there is no polynomial time algorithm
to check whether W
k, where W
is the minimum
number of wavelengths needed for the given multicast
request.
Proof. We reduce the well-known directed Hamilto-
nian Path problem to our problem. It is known that to
decide whether a given graph is Hamiltonian is NP-
complete (Garey and Johnson, 1979).
Let G = (V, A) be a given directed graph. We build
a graph G
by replacing each vertex v V by two
new vertices v
1
and v
2
and linking v
1
to v
2
by the arc
(v
1
, v
2
). Each predecessor of v becomes a predecessor
of v
1
and each successor of v becomes a successor of
v
2
.
We build a graph H by making k copies of G
, G
1
=
(V
1
, A
1
), G
2
= (V
2
, A
2
), ..., G
k
= (V
k
, A
k
), and adding
two new vertices s, z connected by the arc (s, z) (s
is considered as the source). Then we make z adja-
cent (predecessor) to all v
1
-vertices of each copy of
G
(Figure 2). We suppose that D = V
1
V
2
... V
k
.
It is easy to check that H admits a solution W
k for
the light-trail based problem if and only if G admits a
directed Hamiltonian path.
Lemma 3.1. For the problem of minimizing the num-
ber of wavelengths, the path-based solution is not op-
timal.
Proof. Let us consider the topology given in Figure 1.
We suppose that the source node is s and D = V {s}
a
b
c
d
e
f
g
(a) Input graph G
(b) Modified graph H
Figure 2: Illustration of the proof (polynomial transforma-
tion of G). For simplicity we suppose that k = 1.
is the set of destinations. The two solutions: light-
trail based hierarchy (LTH) and the light-path based
hierarchy (LPH) are shown in Figure 3. As shown
in Figure 3a), only one LTH is sufficient to span all
the destinations satisfying the aforementioned degree
constraint. On the other side, as shown in Figure 3b),
two LPHs (each using one wavelength) are required,
i.e., two different wavelengths needed to span all the
destinations. Hence, in this case, the light-path based
solution can not be optimal.
a
d
b
c
e
g
f
h
s
b) LPH solutions
LTH
a) LTH solution
LPH
1
LPH
2
a
d
b
c
e
g
f
h
s
Figure 3: LTH and LPH solutions for the same multicast
request.
Theorem3.2. The optimal solution for the problem of
minimizing the number of wavelengths in non-splitter
WDM networks is a set of light-trail based hierar-
chies. The number of required wavelengths is at least
Light-trailbasedHierarchy-TheOptimalMulticastRouteinWDMNetworksWithoutSplittersandConverters
69
one (in the best case) and limited by that needed for
the optimal light-path based hierarchies.
Proof. Assume that the problem always has feasible
solutions, i.e., there is at least one directed path from
the multicast source to each destination and there are
enough wavelengths to route.
Let us first recall the definition of light-trail based
hierarchy. The base graph Q is a star, so it satisfies the
aforementioned degree constraint. According to the
definition of vertex degree in the corresponding hier-
archy, the hierarchy H also satisfies the degree con-
straint.
Besides, to guarantee the distinct wavelength con-
straint, the mapping h associates a vertex of Q to a
vertex of G such that no two arcs in Q are associ-
ated with the same arc in G, i.e., no duplicated arcs in
H. The worst case happen when the mapping func-
tion (h) is injective, i.e., there are no duplicated ver-
tices (hence no duplicated arcs) in the resultant hierar-
chies. These light-trail based hierarchies correspond
to sets of light-paths (as we call light-path based hier-
archies). Thus, even if the duplication of nodes is not
possible to diminish the number of wavelengths, this
number of wavelengths needed for light-trail based
solution is equal to the number of light-path based hi-
erarchies in the worst case.
However, in general cases, the mapping can gener-
ate several duplicated vertices in the resultant hierar-
chies. As shown in the above example, these vertices
can help to visit more destinations in one trail. As the
result, these duplicated vertices reduce the number of
wavelengths required to cover all the destinations. In
the best case, a set of trails which can be colored by
just one wavelength is sufficient (Figure 3a)).
Lemma 3.2. The optimal solution for the problem of
minimizing the number of wavelengths does not nec-
essarily minimize the total cost of the solution in non-
splitter WDM networks.
Proof. Consider an example in Figure 4 where
there is a trail that spans all destinations nodes
(s, 0, 1, 0, 2, 0, 3, 0, ..., k 2, 0, k 1, 0, k).
Just only one wavelength is sufficient for this
trail. Therefore, this light-trail based hierarchy is
the optimal solution in term of number of wave-
lengths. In contrast, the light-path based solu-
tion can be found corresponding to the set of paths
{(s, 0, 1), (s, 0, 2), (s, 0, 3), ..., (s, 0, k
1), (s, 0, k)}. All these paths share the arc (s, 0).
So the number of wavelengths needed to perform the
multicast is equal to k.
Now we suppose that the cost of arc (s, 0) is equal
to 1, all the others have costs of 10. Accordingly,
the light-trail based solution consumes cost(LTH) =
Figure 4: Graph G = (V,A), a source node s and a set D =
{1 , 2 , 3 , ... , k 1 , k }.
1+2(k 1)10+10) = 20(k 1)+11. Whereas,
the light-path based solution consumes cost(LPH) =
k 1 + k 10 = 11 k. Obviously, cost(LTH) >
cost(LPH), k > 1. Hence, the lemma follows.
4 ILP FORMULATION FOR
LIGHT-TRAIL BASED
HIERARCHIES
In this section, we formulate the considering problem
with the solution corresponding to a set of light-trail
based hierarchies. Let us recall that each LTH can
be composed by a set of rooted arc-disjoint trails (and
thus, each requires a distinct wavelength). The fact
that one wavelength can may not sufficient to cover
all the destinations, several LTHs (i.e, several wave-
lengths) may be needed.
Notations and Network Parameters:
G = (V, A): The directed graph with a set V of
nodes and a set A of arcs
W: The set of wavelengths available on each arc
λ: A wavelength λ W
: An big enough integer such that >
aA
cost(a)
In(m): The set of nodes which have incoming arcs
to node m in G
Out(m): The set of nodes which have outgoing
arcs from node m V
(s, D): A multicast request
Indeg(m): The in degree of node m
Outdeg(m): The out degree of node m
a(m, n): The arc from node m to node n
c
m,n
: The cost of the arc a(m, n)
ILP Variables:
L
λ
m,n
: Binary variable. Equal to 1 if wavelength λ
is used on arc a(m, n) on wavelength λ; equal to 0
otherwise.
PHOTOPTICS2014-InternationalConferenceonPhotonics,OpticsandLaserTechnology
70
F
λ
m,n
: Commodity flow, integer variable. De-
notes the number of destinations served by the arc
a(m, n) on wavelength λ.
w(λ): Binary variable. Equal to 1 if wavelength λ
is used by the light-trails, equal to 0 otherwise.
ILP Formulation:
The primary objective is to minimize the number of
wavelengths required. Secondly, among the wave-
length optimal solutions, the one with the lowest cost
will be chosen. To achieve this, a big enough integer
is introduced which is superior to the summation
of costs of all the arcs in the graph, i.e., >
aA
c
m,n
.
Accordingly, the general objectivefunction can be ex-
pressed as follows.
Minimize : ·
λW
w(λ) +
λW
nV
mOut(n)
c
m,n
· L
λ
m,n
(1)
This objective function is subject to a set of con-
straints which are listed below.
LTH Structure Constraints:
Source Constraint:
λW
mIn(s)
L
λ
m,s
= 0 (2)
1
λW
nOut(s)
L
λ
s,n
|D| (3)
Constraints (2) and (3) ensure that the source s
must not have any incoming arcs in a LTH, but must
have at least one outgoing arc on some wavelength
and the total number of outgoing arcs from s should
not exceed the number of destinations, i.e., |D|.
Destination Constraint:
1
λW
mIn(d)
L
λ
m,d
|D| 1, d D (4)
Constraint (4) guarantees that each destination
should be spanned in at least one LTH but at most
|D| 1 LTHs.
Non-source Node Constraint:
nOut(m)
L
λ
m,n
nIn(m)
L
λ
n,m
, λ W, m V \ {s}
(5)
Since all the nodes are MI nodes that are equipped
with TaC option, they can be transited several times.
However, constraint (5) ensures that the number of
outgoingarcs should not exceed the number of incom-
ing ones for every LTH.
Non-member Nodes Constraint:
nOut(m)
L
λ
m,n
=
nIn(m)
L
λ
n,m
, λ W, m V \ (s D)
(6)
Constraint (6) makes sure that non-member nodes
can be either not used or served only as the intermedi-
ate nodes. In this case, the number of outgoing arcs is
equal to the number of incoming ones in every LTH.
Constraints (5) and (6) also imply that only destina-
tions can be leaf nodes.
Relationship between L
λ
m,n
and w(λ):
w(λ) L
λ
m,n
, m, n V, λ W (7)
w(λ)
mV
nV
L
λ
m,n
, λ W (8)
Constraints (7) and (8) indicate that wavelength λ
is used in a LTH if and only if at least one arc uses it.
However, the above set of constraints is not
enough to guarantee the connectivity of the LTHs as
shown in (Zhou et al., 2010). To solve this problem,
we use the community method that is proposed
in (Yu and Cao, 2005). We introduce an other
variable, commodity flow F
λ
m,n
, as the support of the
variable L
λ
m,n
in order to make sure the continuity and
connectivity of the resultant LTHs.
Connectivity Constraints:
Source Constraint:
λW
nOut(s)
F
λ
s,n
= |D| (9)
Constraint (9) indicates that the sum of flows emit-
ted by the source is equal to the number of destina-
tions in the given multicast session.
Destinations Constraints:
λW
nIn(d)
F
λ
n,d
=
λW
nOut(d)
F
λ
d,n
+ 1, d D (10)
nIn(d)
F
λ
n,d
1
nOut(d)
F
λ
d,n
nIn(d)
F
λ
n,d
, λ W, d D
(11)
Constraints (10) and (11) ensure that each destina-
tion must be consumed totally one and only one flow
in all the LTHs. These constraints also guarantee that
each destination is reachable from the source.
Non-member Nodes:
nIn(m)
F
λ
n,m
=
nOut(m)
F
λ
m,n
, λ W, m V \ (s D)
(12)
Equation (12) ensures that non-member nodes are
only served as intermediate nodes without consuming
any flows.
Light-trailbasedHierarchy-TheOptimalMulticastRouteinWDMNetworksWithoutSplittersandConverters
71
Relationship between L
λ
m,n
and F
λ
m,n
:
F
λ
m,n
L
λ
m,n
, m, n V, λ W (13)
F
λ
m,n
|D| · L
λ
m,n
, m, n V, λ W (14)
Equations (13) and (14) indicate that an arc should
carry a positive number of flows if it is used in a LTH,
and this number should not exceed the total flows
emitted by the source.
It is worth noting that with the supplementarycon-
nectivity constraints, the constraints (3) and (4) are
now relaxed.
5 EXPERIMENTAL RESULTS
In this section we present the experimental results of
the LTH solution for the concerned problem in com-
parison to the traditional LPH solution. In order to
make the comparison, we also develop an ILP formu-
lation for the LPH solution.
5.1 LPH Structure Constraint
Like the difference between light-path and light-trail
structures, LTH allows cycles whereas LPH does not.
In other words, for LPH structures, there is at most
one incoming arc to every node (except the source)
for every given wavelength. Thus, to make ILP for-
mulation for LPH we just add one more constraint to
the constraints of the ILP formulation for LTH pre-
sented in Section 4. This constraint can be formulated
as follows:
nIn(m)
L
λ
n,m
1, λ W, m V \ {s} (15)
5.2 Simulation Settings
The two ILP formulations are implemented in C++
using GLPK v4.45 package (Makhorin, 2010). We
have carried out series of simulations with ran-
dom graphs generated using LEDA v.6.3 library
(Mehlhorn and Naeher, 2010). All the considered
graphs are directed. Due to the fact that ILP pro-
grams do not scale well, we just test with relative
small graphs in which the number of nodes N =
{20, 30, 40, 50}. The density value (the ratio between
the number of arcs and the number of nodes) is fixed
to 2. Graphs with this density are considered as sparse
graphs. We suppose that sparse graphs well reflect the
common core optical networks.
The costs of arcs are randomly selected from the
set of integer {1, 2, .., 20}, and the set of destina-
tions D are also randomly selected with different size
|D| = {10%, 20%, .., 50%} of the number of nodes.
To ensure that there is a feasible solution for all in-
stances, the selected graph must be connected and
have at least one directed path from the source to each
destination. Moreover, in order to guarantee an ac-
ceptable confidence interval, for a certain graph and
for each size |D|, we run 100 simulations with differ-
ent sources and destination sets. For each simulation,
the number of wavelengths used and the total cost of
the routes (hierarchies) are computed as the resultant
performance metrics to evaluate the two ILP formula-
tions.
Besides, to accelerate the ILP computation speed,
we first employ the Farthest First heuristic proposed
in (Le et al., 2013) to the light-trail based hierarchy
ILP and the Farthest Greedy heuristic proposed in
(Din, 2009) to the light-path based hierarchy ILP to
get the upper bound for the number of wavelengths
used. These heuristics are known to be good ones for
the same concerned problem applying the two consid-
ering approaches respectively. With these heuristics,
we benefited much of the time saved to accomplish
the simulations.
5.3 Simulation Results
The overall simulation results are presented in Table
1. As it is expected, light-trail based hierarchy solu-
tion (marked as LTH in the table) outperforms light-
path based hierarchy counterpart (marked as LPH) in
both number of wavelengths used and the total cost.
For the number of wavelengths used, the LTH solu-
tion always consumes fewer wavelengths than LPH
one. In particular, the ratio of saved wavelengths of
LTH solution is up to 19.84% (N = 20), 21.47% (N =
30), 16.09% (N = 40) and 21.95% (N = 50). On aver-
age, this ratio is 11.43%, 10.75%, 9.82% and 12.71%
respectively. This is obvious, because in general, al-
lowing the repetition of nodes in a LTH, more des-
tinations can be covered by a LTH than by a LPH.
Therefore, in term of minimizing the number of wave-
lengths, LTHs requires fewer or at most equal to the
number of wavelengths needed by LPH solution. This
is compatible with the Theorem 3.2.
For the total cost, there are few instances in which
the total costs are better with LPH solution, shown
as some negative reduced ratios in the table. This is
compatible with the Lemma 3.2. However, in general
the LTH solution results in lower costs with the saved
ratio up to 3.79% (N=50), and on average this ratio
is 1.31% (N=20), 0.46% (N=30), 0.90% (N=40), and
2.49% (N=50). In short, even though the total cost is
the second optimized objective, it is also better with
LTH based solution.
PHOTOPTICS2014-InternationalConferenceonPhotonics,OpticsandLaserTechnology
72
Table 1: Performance comparison between light-trail based
hierarchy (LTH) and light-path based hierarchy (LPH) so-
lutions.
Size Wavelengths Total cost
N=20
|D| LPH LTH ց LPH LTH ց
2 100 100 0% 5758 5630 2.22%
4 138 113 18.12% 9762 9543 2.24%
6 123 121 1.63% 10591 10565 0.25%
8 126 101 19.84% 8997 8884 1.26%
10 199 164 17.59% 14004 13922 0.59%
AVG 11.43% 1.31%
N=30
|D| LPH LTH ց LPH LTH ց
3 102 100 1.96% 6416 6281 2.10%
6 114 109 4.39% 11474 11253 1.93%
9 163 128 21.47% 14701 14377 2.20%
12 161 151 6.21% 20598 20277 1.56%
15 203 163 19.70% 22069 23281 -5.49%
AVG 10.75% 0.46%
N=40
|D| LPH LTH ց LPH LTH ց
4 120 109 9.17% 9198 9132 0.72%
8 170 154 9.41% 18436 18349 0.47%
12 174 146 16.09% 22640 22158 2.13%
16 196 171 12.76% 26584 26255 1.24%
20 302 297 1.66% 33441 33467 -0.08%
AVG 9.82% 0.90%
N=50
|D| LPH LTH ց LPH LTH ց
5 114 105 7.89% 14417 13981 3.02%
10 151 126 16.56% 19859 19165 3.49%
15 246 192 21.95% 28187 28310 -0.44%
20 204 179 12.25% 33614 32750 2.57%
25 266 253 4.89% 54694 52620 3.79%
AVG 12.71% 2.49%
6 CONCLUSIONS
In this paper we address the multicasting problem in
all-optical networks without splitters and converters.
The problem is to find the multicast routes which min-
imize the number of required wavelengths with a low
cost. The problem is proved to be NP-hard, and two
exact solutions are presented in the forms of ILP for-
mulations (one for light-path based hierarchy, and the
other for light-trail based hierarchy). The theoretical
analysis pointed out that the optimal solution for the
problem in term of wavelength minimization corre-
sponding to a set of light-trail based hierarchies. The
simulations are carried out to verify it. Once again,
the experimental results showed that not a set of light-
path based hierarchies but a set of light-trail based
hierarchies are the optimal solution. Moreover, al-
though it does not necessarily deduce the cost optimal
solution, the light-trail based solution also appears to
be a good solution when consuming a lower cost com-
pared to the traditional light-path based solution in
general.
REFERENCES
Ali, M. and Deogun, J. S. (2000a). Cost-effective imple-
mentation of multicasting in wavelength-routed net-
works. IEEE/OSA Journal of Lightwave Technology,
18:1628–1638.
Ali, M. and Deogun, J. S. (2000b). Power-efficient design of
multicast wavelength routed networks. IEEE Journal
on Selected Areas in Communications, 18:1852–1862.
Din, D.-R. (2009). Heuristic Algorithms for Finding Light-
Forest of Multicast Routing on WDM Network. In-
formation Science and Engineering, 25:83–103.
Garey, M. R. and Johnson, D. S. (1979). Computers
and Intractability: A Guide to the Theory of NP-
Completeness. W. H. Freeman & Co, New York, NY,
USA.
Le, D. D., Molnar, M., and Palaysi, J. (2013). Multicast
Routing in WDM without Splitters. Infocommunica-
tions Journal, 5(2):1–10.
Li, D., Du, X., Hu, X., Ruan, L., and Jia, X. (2000). Min-
imizing number of wavelengths in multicast routing
trees in WDM networks. Networks, 35(4):260–265.
Makhorin, A. O. (2010). GNU Linear Programming Kit
(GLPK).
Mehlhorn, K. and Naeher, S. (2010). LEDA-a Library of
Efficient Data Types and Algorithms.
Molnar, M. (2011). Hierarchies to Solve Constrained Con-
nected Spanning Problems. Technical Report 11029,
LIRMM.
Mukherjee, B. (2006). Optical WDM Networks. Springer.
Yu, O. and Cao, Y. (2005). Mathematical formula-
tion of optical multicast with loss-balanced light-
forest. In Global Telecommunications Conference,
2005. GLOBECOM ’05. IEEE, volume 4, pages 5 pp.–
1972.
Zhang, X., Wei, J., and Qiao, C. (2000). Constrained multi-
cast routing in WDM networks with sparse light split-
ting. IEEE/OSA Journal of Lightwave Technology,
18:1917–1927.
Zhou, F., Molnar, M., and Cousin, B. (2010). Light-
Hierarchy: The Optimal Structure for Multicast Rout-
ing in WDMMesh Networks. In Computers and Com-
munications (ISCC), 2010 IEEE Symposium on The
15th IEEE Symposium on Computers and Communi-
cations (ISCC2010), 2010, pages 611 – 616, Riccione
Italie.
Zhou, Y. and Poo, G.-S. (2005). Optical multicast over
wavelength-routed WDM networks: A survey. Op-
tical Switching and Networking, 2(3):176 – 197.
Light-trailbasedHierarchy-TheOptimalMulticastRouteinWDMNetworksWithoutSplittersandConverters
73