Supporting Multipoint-to-Point
Communications in All-Optical WDM Networks
Fen Zhou
IRISA / INSA Rennes
Campus de Beaulieu
Rennes, France, 35042
Email: fen.zhou@irisa.fr
Mohand Yazid Saidi
IRISA / University of Rennes 1
Campus de Beaulieu
Rennes, France, 35042
Email: msaidi@irisa.fr
Mikl
´
os Moln
´
ar
IRISA / INSA Rennes
Campus de Beaulieu
Rennes, France, 35042
Email: molnar@irisa.fr
Bernard Cousin
IRISA / University of Rennes 1
Campus de Beaulieu
Rennes, France, 35042
Email: bernard.cousin@irisa.fr
Abstract—The routing and wavelength assignment (RWA)
problem for multipoint-to-point communications in all-optical
wavelength division multiplexing (WDM) networks is investigated
in this paper. Two efficient algorithms, namely Reverse Shortest
Path Tree routing (RSPT) and k-Bounded Edge Disjoint Path
routing (EDPR), are proposed. The problem of minimizing the
total cost while establishing a multipoint-to-point session can
be solved by RSPT algorithm in polynomial time. Nevertheless,
EDPR algorithm produces a significant reduction in the max-
imum number of wavelengths required per link (i.e., the link
stress) by a multipoint-to-point session. Simulations demonstrate
the efficiencies of these two algorithms in supporting multipoint-
to-point communications in WDM networks.
Index Terms—All-optical WDM Networks, Multipoint-to-point
Communications, Routing and Wavelength Assignment (RWA),
Light-startree
I. INTRODUCTION
All-optical Wavelength Division Multiplexing (WDM) net-
works, which are built on the concept of wavelength routing,
have been deployed in the Internet infrastructure for decades to
satisfy increasing demand for bandwidth [1]. As optical cross-
connects (OXCs) become mature and commercially available,
it enables a light beam to pass an intermediate optical node
without the optical-electrical-optical (OEO) conversion. By
widely employing optical OXCs, all-optical WDM networks
overcome the electro-optic bottleneck and thus are gifted at
providing huge bandwidth, low signal attenuation and dis-
tortion as well as favorable transmission delay [2]. Relying
on their inherent advantages, WDM networks are capable
of supporting bandwidth-driven multimedia services in the
Internet, such as VoIP, video conference, online community-
based communications (e.g., blog, voting), etc. It broadly
consists of four traffic patterns: point-to-point communications
(or unicast), point-to-multipoint communications (multicast),
multipoint-to-point communications (MP2P) and multipoint-
to-multipoint (MP2MP) communications.
To support unicast communications in all-optical WDM
networks, lightpaths are introduced to reduce the number of
hops that a packet has to traverse [3]. A lightpath is a set of
consecutive wavelength channels, which connects a transmitter
at a source node to a receiver at a destination node.
To satisfy the multicast services in all-optical WDM net-
works, the lightpath is extended to a light-tree [3] to minimize
the number of transceivers and the average hop distance for
given traffic demands. A light-tree can be viewed as a point-to-
multipoint generalization of a lightpath. But, the network node
should be equipped with optical power splitters to construct
the multicast light-trees. In the case of sparse light splitting,
where only a small fraction of network nodes are capable of
splitting, the light-forest [4] concept is introduced.
For multipoint-to-multipoint case, the core-based tree
(CBT, [5]) structure could be applied with some proper
modifications. In a core-based tree structure, a core node is
employed to collect messages from all the source nodes and
then forward the collected messages to the destination nodes.
Supporting multipoint-to-point communications is also im-
portant and indispensable for wavelength-routed WDM net-
works. To the best of our knowledge, no literature has ad-
dressed this problem before. Therefore, in this paper, we
investigate the routing and wavelength assignment (RWA)
problem for multipoint-to-point communications in all-optical
WDM networks. The following assumptions are made in our
study (i) the number of wavelengths supported per fiber link
is sufficient to establish a multipoint-to-point communication,
(ii) wavelength converters are not available due to the high
cost and complex architecture, (iii) the optical traffic grooming
switches [6], which enable to aggregate low speed traffic
streams into one high capacity wavelength, are also unavail-
able because of the lack of intelligent network control and
automatic provisioning functionality [7].
An optical multipoint-to-point communication generally in-
volves a simultaneous light signal transmission from several
source nodes to the same sink node. In all-optical WDM net-
works, two basic physical constraints are always encountered.
The first one, namely wavelength continuity constraint, implies
the same wavelength should be retained along a lightpath
without wavelength conversion. And the other one, called the
distinct wavelength constraint, does not permit to assign the
same wavelength for two lightpaths sharing a common fiber
link. For multipoint-to-point communications in all-optical
networks, the distinct wavelength constraint is even more
strict. The light signals emitted by two source nodes can
be carried on the same wavelength only if the lightpaths
from these two source nodes to the same sink node are edge
disjoint. Thereby, we introduce a light-startree structure to
Fig. 1. An example of light-startrees
route a multipoint-to-point communication. A light-startree
is consisted of several edge disjoint lightpaths terminated
at the same sink node but assigned only one wavelength.
An example of light-startrees is given in Fig. 1. In the left
dash-dotted rectangle, a multipoint-to-point communication
demande mp2p
¡
(s
1
, s
2
, s
3
), d
¢
arrives. Two light-startrees in
the right dash-dot rectangle (i.e., wavelength w
1
is assigned
to LST
1
, and w
2
to LST
2
) can be used to route this session.
Based on the light-startree structure, two efficient routing
and wavelength assignment (RWA) algorithms are presented
in our paper. The proposed Reverse Shortest Path Tree routing
algorithm is able to minimize the wavelength channel cost
(i.e., the total cost). To reduce the maximum number of
wavelengths required per link (i.e., link stress) by a multipoint-
to-point communication request, a k-Bounded Edge Disjoint
Path routing algorithm is introduced. Extensive simulations are
done to assess the performance of these two algorithms.
The rest of this paper is organized as follows. Section II
formulates the RWA problem for multipoint-to-point com-
munications in WDM networks. Two RWA algorithms are
introduced in Section III. Numerical results are obtained in
Section IV. Finally, a summary of this paper is made in
Section V.
II. MULTIPOINT-TO-POINT
COMMUNICATIONS IN ALL-OPTICAL WDM NETWORKS
In this article, the routing and wavelength assignment
(RWA) problem for multipoint-to-point communications in all-
optical WDM networks is studied. A WDM network can be
modeled by an undirected graph G(V, E, c, W ). V denotes the
set of nodes in the network. Each edge e ∈ E is associated
with a weight function c(e), which signifies the cost of fiber
link channel e. c(e) is additive along a lightpath LP (u, v),
where u and v are two end point nodes of the lightpath. W
denotes the number of wavelengths that could be supported
in a fiber link. We consider a multipoint-to-point session
mp2p(S, d), which requires to set up simultaneous lightpaths
from each source s
i
∈ S to the same destination d. Due to the
distinct wavelength constraint in WDM networks, two light-
paths could be assigned the same wavelength only if they are
edge disjoint. Hence, the multipoint-to-point communication
can be realized either by an individual unicast path to the
destination for each source or a set of light-startrees. A light-
startree is a set of edge-disjoint lightpaths, which share the
same receiver end point and occupy the same wavelength.
One light-startree may not be sufficient to span all the source
nodes of a multipoint-to-point session, and thus several light-
startrees may be used with each on a different wavelength.
To carry out a multipoint-to-point session, a RWA algorithm
should try to save as many wavelength channel cost as possible
while overcoming the optical constraints. The total cost of a
multipoint-to-point communication can be calculated by the
sum of each fiber link channel cost in the set of light-startrees
(LST ) computed out. i.e.,
c[mp2p(S, d)] =
X
LST
i
∈mp2p(S,d)
X
e∈LST
i
c(e) (1)
Besides this, the maximum number of wavelengths required
per link for a multipoint-to-point session (i.e., link stress) is
also important for WDM networks. This is because, physically,
the number of wavelengths supported on a fiber link is limited
to a fixed number. If a multipoint-to-point session spends
too many wavelengths in one fiber link, then the number
of wavelengths left for other sessions will be limited, which
thus results in the blocking of some sessions. Hence, balanc-
ing the multipoint-to-point traffic and avoiding the blocking
should also be taken into account by an efficient routing
algorithm. Generally, the set of light-startrees computed out
for a multipoint-to-point session are not edge disjoint and thus
the link stress equals the number of light-startrees and at most
the number of source nodes in a multipoint-to-point session.
III. PROPOSED SOLUTIONS
In this section, two RWA algorithms are proposed for
multipoint-to-point communications in all-optical WDM net-
works. The first one, namely Reverse Shortest Path Tree rout-
ing algorithm, tries to optimize the total cost for a session. The
second one, namely k-Bounded Edge Disjoint Path Routing
algorithm (EDPR), is primarily designated to minimize the
link stress for a multipoint-to-point session. This algorithm
balances the traffic load on the network fiber links. Apart from
the primary objective, algorithm EDPR also tries to optimize
the total cost as the second objective.
A. Reverse Shortest Path Tree Routing Algorithm (RSPT)
Algorithm RSPT solves the routing and wavelength as-
signment problem for a multipoint-to-point session separately.
The shortest path tree SPT rooted at destination d is first
constructed. Then, the edges in the shortest path tree SPT are
reversed in order to transmit light signals from each source s ∈
S to destination d. After the search of routes, the wavelength
assignment is implemented to satisfy the distinct wavelength
constraint. Different wavelengths have to be employed for any
pair of source nodes if their shortest paths to d share at least
a common link. In the absence of optical traffic grooming
switches [7], algorithm RSPT is able to provide the optimal
total cost for a MP2P session. In addition, it can be solved in
the time O(|V |log|V | + |E|) by Dijkstra’s algorithm.
B. k-Bounded Edge Disjoint Path Routing Algorithm (EDPR)
Although algorithm RSPT optimizes the total cost, it is
not competitive for minimizing the link stress. This can
Algorithm 1 k-Bounded EDPR Algorithm
Input: Graph G(V, E, c) and mp2p(S, d).
Output: A set of light-startrees LST
i
each on a different
wavelength w
i
for mp2p(S, d).
1: i ← 1 {the serial number of a light-startree}
2: j ← 0 {the serial number of a renewed graph}
3: S SET ← S, G
j
← G(V, E, c)
4: while (S SET 6= φ) do
5: LST
i
← {d}
6: while (d can be reached from S SET in G
j
within
bound k) do
7: Compute the shortest path tree rooted at desti-
nation d in the renewed graph G
j
8: for (s ∈ S SET ) do
9: Find the nearest source
˜
s to destination d
such that
SP
G
j
(
˜
s,d)
SP (
˜
s,d)
≤ k
10: end for
11: LST
i
← LST
i
∪ SP
G
j
(
˜
s, d)
12: G
j+1
← G
j
\ {e|e ∈ SP
G
j
(
˜
s, d)}
13: S SET ← S SET \ {
˜
s}
14: j ← j + 1
15: end while
16: Assign wavelength w
i
to LST
i
, i ← i + 1
17: end while
be explained as follows. Algorithm RSPT only routes the
multipoint-to-point traffic over the shortest paths, but it does
not consider the relevant relation between the paths from
the sources to the destination. Especially, some fiber links
around the destination and leading to source nodes will be
frequently used. Since different light signals are transmitted
over each shortest path, the fiber links commonly shared
by these shortest paths are highly loaded in the absence of
costly optical traffic grooming switches [7]. The more traffic
is concentrated on a fiber link, the more wavelengths will be
required. Although sometimes we can not avoid the frequent
usage of the links adjacent to the destination, we can at least
balance the traffic among these fiber links. As the traffic load
is balanced and evenly distributed, the maximum number of
wavelengths needed per fiber link (i.e., link stress) is reduced.
Based on this fact, a k-Bounded Edge Disjoint Path Routing
algorithm is proposed.
Different from the RSPT algorithm, the k-Bounded Edge
Disjoint Path Routing algorithm considers the relationship
between the lightpaths from all sources to the destination while
searching the routing paths. It aims to find the maximum edge
disjoint paths for the source nodes. To achieve this, an edge
removal action is introduced during the lightpath computation.
The edge removal action takes place whenever the lightpath
from a source node to destination d is found. At the j
th
step,
the source
˜
s nearest to destination d in the renewed graph
G
j
is added to light-startree via the shortest path SP
G
j
(
˜
s, d)
such that it holds
SP
G
j
(
˜
s,d)
SP (
˜
s,d)
≤ k. To avoid using the same link
twice, all the links in the lightpath specified for a source will
be removed from the graph (refer to step-12). Then, the nearest
source is computed in the renewed graph and the shortest
path in the renewed graph is specified as the lightpath for the
nearest destination. It is true that, most of the time, one light-
startree is not sufficient to span all the sources. In case that
the destination is not reachable from the unspanned sources
in the renewed graph, another light-startree assigned a new
wavelength will be computed over the initial topology using
the same technique.
IV. NUMERICAL RESULTS
Numerical results are obtained by the simulations. The USA
Longhaul topology, which is viewed as a good candidate
for the future backbone network of USA, is selected as the
WDM network testbed. Destination and source nodes are
assumed to be distributed uniformly and independently in the
topology for each multipoint-to-point session. An equal cost of
1 unit hopcount cost is assumed on all the links to simplify
the simulation. Two metrics are considered to evaluate the
quality of the computed light-startrees for mp2p(S, d):
• Total Cost: wavelength channels cost in the light-startrees
computed for mp2p(S, d). It is calculated by equation (1).
• Link Stress: the maximum number of wavelengths
required per link to span all the source nodes in
mp2p(S, d).
A. Impact of Bound k in EDPR Algorithm
The bound k should be chosen carefully, since it has a great
impact on the performance of the algorithm. Here, we study
the quality of light-startrees when the bound k varies, which
could be a good reference for the selection of bound k.
In Fig. 2, the cost approximation ratio of EDPR algorithm is
plotted when the session members ratio corresponds to 10%,
50% and 100% respectively. Since the cost of RSPT is optimal,
the cost approximation ratio is computed by
c(EDP R)
c(RSP T )
. As
indicated, the approximation ratio becomes worse as the the
bound k grows. We can also see that, the approximation ratio
stays stable when k increases beyond a certain value. Because
the diameter of a network is a fixed value (e.g., USA Longhaul
Network has a diameter of 8), the value of
SP
G
j
(
˜
s,d)
SP (
˜
s,d)
is limited.
Once k increases big enough, EDPR algorithm can get the
same total cost as that when k is infinite.
The tendency of link stress versus the bound k is shown in
Fig. 3. The link stress of EDPR algorithm diminishes as the
bound k grows. The EDPR algorithm can reach the best link
stress when k = 1.8. After that value, the link stress keeps
steady. This can be interpreted as follows. As k increases, the
condition of the possible candidate lightpaths from a source
to destination d is relaxed. Then more possible lightpaths can
be found for a source. However, similar to the total cost, there
will be no longer any constraint on the length of a possible
lightpath if k is large enough. Consequently, link stress will
be stable.
1 1.5 2 2.5 3 3.3
1
1.05
1.1
1.15
1.2
Bound k
Approximation Ratio of Total Cost
groupsize=28
groupsize=14
groupsize=3
Fig. 2. Approximation of Total Cost versus Bound k
1 1.5 2 2.5 3 3.3
0
2
4
6
8
10
12
14
16
Bound k
Link Stress
groupsize=28
groupsize=14
groupsize=3
Fig. 3. Link Stress versus Bound k
2 6 10 14 18 22 26 28
10
20
30
40
50
60
70
80
90
100
110
Group Size
Total Cost
RSPT
EDPR−k=∞
EDPR−k=1.5
Fig. 4. Total Cost against Multipoint-to-point Session Group Size
2 6 10 14 18 22 26 28
0
2
4
6
8
10
12
14
16
18
Group Size
Link Stress
RSPT
EDPR−k=∞
EDPR−k=1.5
Lower Bound
Fig. 5. Link Stress against Multipoint-to-point Session Group Size
B. Comparison of RSPT and k-Bounded EDPR
In Figs. 4 and 5, the quality of light-startrees computed
by RSPT and EDPR algorithms are compared. The lower
bound of link stress plotted in Fig. 5 is calculated by
max
©
d
|S|
Deg(d)
e, 1
ª
. Based on the simulation results, both the
total cost and the link stress race up rapidly when the MP2P
session group size increases. The RSPT algorithm is able to
get the best total cost, while EDPR algorithm with k = ∞
results in the best link stress which is very close to the lower
bound. When k = 1.5, the total cost of EDPR algorithm is
better than that of k = ∞, while its link stress is less favorable
than that of k = ∞. So, a tradeoff should be found between
the total cost and the link stress to satisfy different types of
multimedia multipoint-to-point traffics.
V. CONCLUSION
Routing and wavelength assignment is a very important
issue for supporting multipoint-to-point communications in
all-optical WDM networks. Distinct wavelengths should be
employed by source nodes to send light signals to the same
destination d, if their lightpaths leading to d are not edge
disjoint. The light-startree structure is introduced to route
multipoint-to-point communications, which consists of several
edge disjoint lightpaths assigned the same wavelength and
terminated at the same destination. To establish a multipoint-
to-point session, a set of light-startrees, each with a different
wavelength, may be required. As the number of wavelengths
supported in one fiber link is limited, not only the total cost
but also the link stress are important parameters to measure
the quality of light-startrees computed. For this reason, the
Reverse Shortest Path Tree routing algorithm is proposed,
which is able to to minimize the total cost in polynomial
time. Meanwhile, the k-Bounded Edge Disjoint Path Routing
algorithm is presented to route a multipoint-to-point session
with as fewer wavelengths as possible. It results in significant
reduction of the link stress. Numerical results obtained demon-
strate the efficiency of the proposed algorithms in supporting
all-optical multipoint-to-point communications.
REFERENCES
[1] George N. Rouskas. Optical layer multicast: rationale, building blocks,
and challenges. IEEE Network, 17(1): 60-65, 2003.
[2] Biswanath Mukherjee. WDM Optical communication networks: progress
and challenges. IEEE Journal on Selected Aeras in Communications,
18(10): 1810-1824, 2000.
[3] Laxman H. Sahasrabuddhe, Biswanath Mukherjee. Light-trees: optical
multicasting for improved performance in wavelength-routed networks.
IEEE Communications Magazine, 37(2): 67-73, 1999.
[4] Xijun Zhang, John Wei, Chunming Qiao. Constrained multicast routing
in WDM networks with sparse light splitting. IEEE/OSA Journal of
Lightware Technology, 18(12): 1917-1927, 2000.
[5] Jianping Wang, Xiangtong Qi, Mei Yang. Routing and wavelength assign-
ment for core-based tree in WDM networks. Computer Communications,
29: 1896-1904, 2006.
[6] Osama Awwad, Ala I. Al-Fuqaha, Ammar Rayes. Traffic grooming,
routing, and wavelength assignment in WDM transport networks with
sparse grooming resources. Computer Communications, 30: 3508-3524,
2007.
[7] Keyao Zhu, Hui Zang, Biswanath Mukherjee. A comprehensive study
on next-generation optical grooming switches. IEEE Journal on Selected
Areas in Communications, 21(7): 1173-1186, 2003.
