+1443 776-2705 panelessays@gmail.com

Topic: Dynamic Manycasting in Optical Split-Incapable WDM Networks for Supporting High-Bandwidth Applications.

Review Guidelines:

1. One page, single space, margin 1 inch (top, bottom, left, right), font (times new roman, size 10)

Supporting document are attached.

Dynamic Manycasting in Optical Split-Incapable WDM Networks
for Supporting High-Bandwidth Applications

Jeremy Plante, Arush Gadkar, and Vinod Vokkarane
Department of Computer and Information Science, University of Massachusetts, Dartmouth, MA

Email: {jplante, agadkar, vvokkarane}@umassd.edu

Abstract—With the advent of bandwidth intensive applications,
the demand for manycast networking capabilities has become an
essential component of wavelength division multiplexed (WDM)
optical networks. Traditionally, the manycast functionality is
accomplished by splitting a signal all-optically, thereby creating
a light-tree, which originates from the source node and reaches
a subset of the destination nodes. To support the manycasting
functionality in an optical network that is Split-Incapable, i.e., the
optical cross connects are incapable of switching an incoming
optical signal to more than one output interface, one must
implement a logical overlay to the underlying optical layer. A
naı̈ve approach to accomplish this is by creating a set of unicast
lightpaths that originate at the source node and terminate at a
subset of the destination nodes of the manycast request. However,
for a large number of requests this approach leads to a poor
utilization of network resources. To overcome this problem, we
propose two overlay approaches: Manycasting with Drop at
Member Node (MA-DAMN) and Manycasting with Drop at Any
Node (MA-DAAN). In these solutions, we achieve manycasting
by creating a set of lightpath routes (possibly multiple hops)
in the overlay layer. We consider a dynamic traffic model and
propose efficient heuristics to solve the MA-DAMN and MA-
DAAN problems with a goal of minimizing the total number
of wavelengths required to satisfy the requests. Moreover, we
present a simple heuristic to approximate the unicast approach
of the naı̈ve method. Our results demonstrate that both the
overlay approaches reduce the wavelength consumption by ap-
proximately 33 − 45% over the naı̈ve manycasting via WDM
unicast approach for real-world large-scale networks.1

Index Terms—Manycasting, WDM, Overlay, Lightpath Routes,
Split-Incapable, O-E-O.

I. INTRODUCTION

Optical wavelength division multiplexed (WDM) networks
have proven to be an important technology to provide high-
bandwidth and services to the next-generation Internet. Many-
casting has caught the attention of several researchers during
the recent past, due to the emergence of many distributed
applications [1], [2]. Distributed applications, such as video
conferencing, distributed interactive simulations (DIS), grid
computing, storage area network (SAN), and distributed con-
tent distribution network (CDN) applications require large
amounts of bandwidth and an effective communication be-
tween a single source and a set of destinations. For example,
the data generated by the Large Hadron Collider is frequently
sent to multiple Department of Energy (DOE) laboratories
across the United States over the DOE’s Energy Sciences
network (ESnet)2. This data may be stored at multiple research

1This work was supported by the Department of Energy (DOE) COMMON
project under grant DE-SC0004909.

2www.es.net : Energy Sciences Networks of the Department of Energy
(DOE).

laboratories for distributed backup.
Manycasting [3], is a powerful communication framework

that can be used to support such applications. Manycasting is
a generic version of Anycasting3 that supports communication
from a sender to any k out of m (k ≤ m) candidate des-
tinations where the candidate destination set, |Dc| = m, is a
subset of nodes in the network. By changing the parameters of
a manycast request, one can also perform unicast (k = m = 1),
multicast (k = m > 1), and anycast (k = 1 < m). The key
difference between manycast and multicast is that in multicast,
all destinations are specified a priori, whereas in manycast
only a subset of candidate destinations must be chosen. We
can use manycasting to choose some subset of these locations,
while ensuring that the underlying network is used efficiently.

Traditionally, one can support manycasting at the optical
layer using a technique called Light-Trees [4], which em-
ploys light-splitters in the optical crossconnects (OXCs). A
fundamental problem in supporting manycast over networks
like the ESnet is that the underlying optical-layer [5] is
Split-Incapable i.e., the OXCs are incapable of switching an
incoming optical signal to more than one output interface. To
overcome this problem, one can implement a simple unicast
approach, wherein we establish an independent point-to-point
communication channel (lightpath) between the source and
each selected candidate destination of the manycast request
set. This naı̈ve Manycasting via WDM Unicast (MA-VWU)
approach to the manycast overlay problem results in an
inefficient consumption of bandwidth. As an alternative, we
present two multi-hop solutions which will provide a set
of lightpath-routes from source to every selected manycast
destination using Steiner tree [6] routing: 1) Manycasting with
Drop At Member Node (MA-DAMN), which limits lightpath
termination to only members (source or destination nodes) of
the manycast request, and 2) Manycasting with Drop At Any
Node (MA-DAAN), which relaxes this constraint and allows
a lightpath to be potentially terminated at any node in the
network.

In this paper, we consider a dynamic traffic model, wherein
the manycast requests arrive to the network according to
some stochastic process4. We present heuristics to solve the
manycast overlay problems for the MA-DAMN and MA-
DAAN schemes, with the goal of minimizing the number of
wavelengths required to satisfy all the manycast requests. We

3Anycasting is defined as the communication paradigm in which the user
has the ability to choose a single destination from a group of candidate
destinations, unlike deciding it a priori as in unicast.

4We consider the static version of the manycast overlay problem in [7].

compare the performance of MA-DAMN and MA-DAAN with
a simple heuristic to solve the MA-VWU problem. The re-
mainder of this paper is structured as follows: In Section 2 we
define the MA-VWU, MA-DAMN, and MA-DAAN problems
formally and present the corresponding heuristics in Section 3.
Simulation results are presented in Section 4, and Section 5
concludes the paper.

II. PROBLEM DEFINITIONS

Given a network topology graph G(V, E), with V nodes
and E links, a manycast request can be defined as R =
(sr, Dr, K

′), where sr is the source node of the request
(sr ∈ V ) and Dr is the set of candidate destination members
(nodes) (Dr ⊆ V − {sr}). For a manycast request with K
destination members, we represent the destination set as Dr =
{d1, d2, . . . , dK}. K′ ≤ K is the number of nodes necessary
to reach in order to successfully service the manycast request.
In the MA-VWU problem, we establish lightpaths from the
source node of the manycast request to each of the K′ selected
destination nodes. In doing so, we create K′ unicast routes
(single logical-hop) in the overlay network. In the MA-DAMN
model, we find a set of lightpath routes that start at the source
node and reach the K′ destination members possibly via
multiple logical-hops. While creating these lightpath routes,
we take into account that we can terminate (i.e., drop) a
lightpath only at nodes that belong to the destination set.
In what follows, we first outline the differences between the
MA-VWU and the MA-DAMN problems with the help of an
example and then formally state the problem definitions.

Illustrative Example

We consider a simple six-node network with bi-directional
links. For the sake of simplicity let us assume that the initial
network state consists of just a single manycast request R1:{1,
(2,5,6), 2}. Fig. 1 and 2 demonstrate how the MA-VWU and
MA-DAMN models service this request.

1 2 3

654

(a) Physical topology routing.

1 2 3

654

(b) Lightpaths in the logical overlay.

Fig. 1. MA-VWU network state after satisfying request R1.

Fig. 1(a) and 1(b) depict the state of the network upon
servicing R1 using the MA-VWU model. More specifically,
Fig. 1(a) shows the routing of lightpaths on the physical
topology. For simplicity, we select destinations based on their
relative proximity (in physical hops) to the source. For R1,
the selected destinations are nodes 2 and 5. Note that MA-
VWU establishes a unique lightpath route to each selected
destination. Fig. 1(b) shows the logical view of the lightpaths

1 2 3

654

(a) Physical topology routing.

1 2 3

654

(b) Lightpaths in the logical overlay.

Fig. 2. MA-DAMN network state after satisfying request R1.

1 2 3

654

(a) Physical topology routing.

1 2 3

654

(b) Lightpaths in the logical overlay.

Fig. 3. MA-VWU network state after satisfying request R2.

reserved in Fig. 1(a). Similarly, Fig. 2 demonstrates the
servicing of R1 according to the MA-DAMN model at both
the (a) physical layer, and (b) logical layer. Note that in this
case, the manycast request is serviced using multiple hops in
the logical (overlay) layer. It is also important to note that the
lightpaths are allowed to originate/terminate only at destination
nodes of the manycast request. From Fig. 1 and 2, it is easy
to verify that both models require only a single wavelength to
service R1.

Given the current network states for MA-VWU and MA-
DAMN, consider the arrival of the second request R2: {4,
(2,3,5), 2}. Fig. 3 and 4 show how this request can be satisfied
using MA-VWU and MA-DAMN respectively. Once again,
Fig. 3(a) shows MA-VWU’s physical routing, while Fig. 3(b)
shows its logical routing. Note that MA-VWU will route from
the source node 4 to selected destinations 2 and 5 using
independent lightpaths to each destination. Since R1 already
routed to nodes 2 and 5 along wavelength 1 (shown with a
broken line), R2 must be serviced using a second wavelength
(shown with a solid line) according to the wavelength conti-

1 2 3

654

(a) Physical topology routing.

1 2 3

654

(b) Lightpaths in the logical overlay.

Fig. 4. MA-DAMN network state after satisfying request R2.

nuity constraint (WCC)5 and the wavelength clash constraint6.
MA-DAMN however, can service R2 without the need for a
second wavelength as shown in Fig. 4. MA-DAMN therefore
requires only a single wavelength to service both manycast
requests, compared to the required two wavelengths under the
MA-VWU approach. However, this decrease in the number of
wavelengths comes at the expense of an increase in the average
number of logical hops (2 for MA-DAMN, 1 for MA-VWU),
which results in a greater end-to-end delay.

For this specific example, MA-DAAN too requires just
a single wavelength to service both the requests, however,
in some cases MA-DAAN tends to outperform MA-DAMN.
Note that we aim to minimize the network wide wavelength
count, not the maximum number of wavelengths on any single
link. In MA-DAMN and MA-DAAN we service a given
multicast request by setting up lightpath routes in the overlay
network, wherein we potentially terminate the lightpath at
some intermediate node. This incurs an O-E-O conversion
at that node, and it is thus fair to allow a free wavelength
conversion at that node. For a given lightpath however, we do
adhere to the WCC.

The problem statements can now be formally stated follows:
Definitions: Given a network G = (V, E), and a dynamic

arrival process for the manycast requests, the MA-VWU
model must assign, for each request, a lightpath and a wave-
length from the source node of the request to K′ destination
nodes, in such a manner that the number of wavelengths
required is minimized while satisfying the WCC. In the MA-
DAMN model we must assign, for each request, a set of
lightpath-routes (possibly multiple-hop) in the overlay layer, to
reach K′ destination nodes, in such a way that the number of
wavelengths required is minimized while satisfying the WCC.
The solution must also take into consideration that for any
given manycast request, no lightpath terminates/originates at
a non-member node (i.e., not a candidate destination node)
of the request. The MA-DAAN problem is similar to MA-
DAMN, with the exception that we now provide the flexibility
to terminate/originate a lightpath at/from any non-member of
the request set.

III. MANYCAST OVERLAY HEURISTICS

In this section we propose a heuristic for providing many-
cast capability over split-incapable optical WDM networks.
The Manycast – Shortest Path Overlay (MA-SPO) heuristic
described in Algorithm 1, can be applied to solve both the
MA-DAMN and MA-DAAN problems with subtle differences
as outlined below.

The input to MA-SPO is a set of predetermined routes,
routeList, which contains the set of all shortest-path (SP)
routes from one manycast request member to every other
member in the case of MA-DAMN, or the set of all unicast SP
routes for every network node pair in the case of MA-DAAN.
The routes in this list are sorted on Line 3 of Algorithm 1, with

5A lightpath must occupy the same wavelength across every physical link
of the network it traverses.

6No two lightpaths may occupy the same wavelength simultaneously.

Algorithm 1: Manycast Shortest Path Overlay (MA-SPO)
input : Manycast Request: R = (sr, Dr, K′),

Dr = {d1, d2, . . . , dK}
: List of appropriate SP routes: routeList

output: Manycast overlay tree yielding the fewest
number of additional wavelengths to the network

1 AltTrees[K] = NULL
2 routedDestinations = 0
3 sort(routeList)
4 for each di ∈ Dr do
5 Treek = NULL
6 Treek.add(SP(s, di))
7 update routedDestinations
8 for each routei ∈ routeList do
9 if routedDestinations ≥ K′ then

10 break

11 if (routei.source ∈ Treek) & (routei.dst ̸∈
Treek) & (routei.dst ∈ Ds) then

12 Treek.add(routei)
13 update routedDestinations

14 AltTrees.add(Treek)

15 return min(AltTrees)

the shortest routes (in terms of physical hop count) first. MA-
SPO will generate K-alternate logical manycast trees, where
K = |Dr|. These alternate trees are stored in an array of
size K named AltTrees in Line 1. Each of these alternate
trees is determined within the loop starting at Line 4. For
each manycast tree corresponding to a particular candidate
destination di, the first logical hop is mandated to be the SP
from the manycast source sr to di, and this hop is added
to the tree at Line 6. It is possible that this first hop to
di may actually route through one or more other dj ∈ Dr,
and this appropriate number of successfully reached manycast
destinations is updated on Line 7. MA-SPO will then inspect
each route in the now sorted routeList to determine its
suitability for inclusion in the manycast tree. In order for a
given route to be included in the tree, its source node must
already be in the tree, and its destination must not. However,
the route’s destination must be a member of Dr. The first
route in routeList which satisfies these criteria will be added
to the tree as in Line 12. This continues until K′ destinations
have been routed to along a single tree, at which point the tree
will be added to the collection of alternate trees as in Line 14.
Once K-alternate trees (each containing at least K′ destination
members, if the tree is not empty) have been identified, the
least-cost tree is selected as the manycast overlay tree which
will satisfy the request in Line 15. The least-cost tree is the
one which will result in the smallest number of additional
wavelengths required to be added to the current network in
order to satisfy the given request. If more than one tree shares
this minimum wavelength count, the tie is broken by selecting
the tree which incurs the fewest logical hops from sr to all
K′ destinations. In the case of MA-DAMN, all routes in
routeList will originate at a member node and terminate at

another member node, thus a signal may be dropped only at
a member of the manycast request. In the MA-DAAN case
however, any node already included in the tree may be selected
as the Steiner point for the next lightpath to be sourced from,
thereby relaxing the MA-DAMN constraint that lightpaths may
only originate at either the source or destination members, and
increasing the routing flexibility of MA-SPO.

In Fig. 5, we present an example to illustrate how the
MA-SPO generates K-alternate trees for a given manycast
request. Shortest paths between nodes are ordered according to
a simple bakery algorithm. For example, if node 1 is to route
to node 5, it will route along node 2 rather than node 4 since
node 2 has a lower index. Similarly assume that the routeList
is ordered, such that if a route from sr to a destination di has
the same weight as a route from dj to di, the route from sr
is chosen over its competitor. For the manycast request R =
2:{4,5,6}:2, K = 3 alternate trees will be generated by the
MA-SPO algorithm, but only K′ = 2 destinations must be
reached in each tree. The first tree requires that its first hop
be from node 2 to node 4. The corresponding route is via
node 1. The next shortest route is directly from node 2 to
node 5. These two paths complete the first tree. The second
tree will first route from node 2 to node 5. This now means
that the shortest paths to both nodes 4 and 6 are sourced
at the closest member, in this case node 5. Relying on the
previously mentioned bakery algorithm structure, node 4 is
the one selected as the second manycast destination. The first
hope of the final tree is mandated to route from node 2 to
node 6. The appropriate path is via the intermediate node 3.
The second destination in this tree is node 5 which is one hop
from both nodes 2 and 6. According to the order of routes
in routeList, the source for this route should be the source
node 2. These three distinct alternate trees are shown in Fig. 5.
The current state of the network largely affects which of these
trees is chosen for satisfying the manycast request. We select
the tree that yields the fewest additional wavelengths to the
network. Let us assume that trees 1 and 2 require no new
wavelengths, and tree 3 requires 1 new wavelength. Tree 3 is
no longer in contention, and the tie is broken by the fact that
tree 2 contains 1 logical hop to destination node 5, but two
hops to node 4 for a total of 3 logical hops, as opposed to tree
1 which only contains 2 logical hops. For the given request,
tree 1 is selected and added to the network.

Complexity: For a network with V nodes, and E links,
without considering the complexity of the underlying uni-
cast lightpath reservations for an overlay manycast tree, the
complexity of identifying a single tree can be represented by
the product of the number of routes traversed in routeList,
O(V 2), and the maximum number of nodes to traverse in
a single tree, O(V ). Considering MA-SPO establishes K-
alternate trees, the worst-case complexity of Algorithm 1 can
be expressed as O(KV 3).

In the following section, we compare the performance of the
MA-SPO heuristic to a simple Manycast-Shortest Path Unicast
(MA-SPU) heuristic, which provides a sub-optimal solution
to the MA-VWU problem by establishing individual unicast

R = 2:{4,5,6}:2

1 2

4 5

3

6

2

5

Tree 1

1

4

2

5

Tree 2

4

3

6

Tree 3

2

5

Fig. 5. Illustrative example of MA-SPO for a specific manycast request.

lightpaths from the source node to the nearest K′ candidate
destination nodes. Due to space restrictions, we did not include
a full description of the MA-SPU.

IV. SIMULATION RESULTS

We present simulation results of the MA-SPO and MA-SPU
heuristics described in the previous section for dynamically
arriving manycast connection requests on the National Science
Foundation Network (NSFnet) and an augmented Energy
Sciences Network (ESnet) shown in Fig. 6.

(a) 14-node NSFnet.

(b) Augmented Energy Sciences Network (ESnet).

Fig. 6. Networks used for heuristic evaluation.

Manycast requests arrive according to a Poisson process
with average arrival rate λ and exponentially distributed hold-
ing times with an average service rate µ. Each request set
consists of 105 manycast requests. The results presented in
this section represent the average of 30 unique request sets.
The source node of each request is uniformly distributed
over all nodes in the network. For each manycast request,
the size of the candidate destination set (K) is uniformly
distributed across the interval [3, Dmax] (where Dmax is a
parameter which represents the maximum number of manycast

10 20 30 40 50 60 70 80 90 100
5

10

15

20

25

30

35

40

45

50

Network Load

A
v
e

ra
g

e
n

u
m

b
e

r
o

f
w

a
v
e

le
n

g
th

s

MA−VWU

MA−DAMN

MA−DAAN

Fig. 7. NSFnet: Average number of wavelengths (Dmax = 10).

destinations). Therefore, for a particular manycast request
r, |Dr| = K. For each request, the destination nodes are
also uniformly distributed across the network. The minimum
number of destination nodes (K′) to reach is set to


|Dr|
2


.

We simulate each heuristic with different values for Dmax,
and record the average number of wavelengths. We also
plot the 95% confidence intervals for each heuristic. Note
that each fiber of the network is assumed to have sufficient
number wavelengths to eliminate blocking. The network load
in Erlangs is calculated as the ratio of the average arrival rate
to the average service rate (λ/µ). Note that we will hereafter
refer to the MA-SPO implementations of MA-DAMN and
MA-DAAN as simply MA-DAMN and MA-DAAN. Similarly,
we will refer to the MA-SPU heuristic as MA-VWU.

In Fig. 7 and Fig. 8 we show the average number of
wavelengths required by the three heuristics for the NSFnet
and ESnet for Dmax= 10. For the NSFnet it is observed that
the MA-DAMN achieves approximately 38% improvement (in
terms of the number of wavelengths) over MA-VWU. Simi-
larly in Fig. 8, for the ESnet it is observed that MA-DAMN
achieves approximately 47% improvement over MA-VWU.

10 20 30 40 50 60 70 80 90 100
10

20

30

40

50

60

70

80

Network Load

A
v
e

ra
g

e
n

u
m

b
e

r
o

f
w

a
v
e

le
n

g
th

s

MA−VWU

MA−DAMN

MA−DAAN

Fig. 8. ESnet: Average number of wavelengths (Dmax = 10).

10 20 30 40 50 60 70 80 90 100

1

1.1

1.2

1.3

1.4

1.5

Network Load

A
v
e

ra
g

e
n

u
m

b
e

r
o

f
lo

g
ic

a
l
h

o
p

s

MA−VWU

MA−DAMN

MA−DAAN

Fig. 9. ESnet: Comparison of logical hops (Dmax = 10).

For both the networks it is seen that MA-DAAN achieves
marginal performance improvement over MA-DAMN. This is
at the expense of a a slightly higher average number of logical
hops, as shown in Fig. 9 and Fig. 10 respectively. Note that
the MA-VWU heuristic establishes end-to-end to lightpaths
from the source node of a request to each selected candidate
destination node. Hence the average number of logical hops
is always 1.

In Fig. 11 we plot the average runtimes (for successfully
servicing all 105 manycast requests) for the three heuristics
on the ESnet. It can be verified that the run times for MA-
VWU are in the order of few tens of seconds. The MA-SPO
heuristics not only create logical lightpath trees in the overlay
network, but generate K-alternate trees for every request, and
then select the best tree for use in establishing lightpaths
on the underlying layer. This increases the computational
complexity and it can be observed that the MA-DAMN and
MA-DAAN implementations of MA-SPO take close to two
hundred seconds to complete. Due to space restrictions, we do
not show the execution times for the NSFnet, but the results
follow a similar trend to those of the ESnet.

In Tables I – IV, we show the results obtained for the

10 20 30 40 50 60 70 80 90 100

1

1.1

1.2

1.3

1.4

Network Load

A
v
e

ra
g

e
n

u
m

b
e

r
o

f
lo

g
ic

l
h

o
p

s

MA−VWU

MA−DAMN

MA−DAAN

Fig. 10. NSFnet: Comparison of logical hops (Dmax = 10).

10 20 30 40 50 60 70 80 90 100
40

60

80

100

120

140

160

180

200

220

240

Network Load

A
v
e

ra
g

e
r

u
n

ti
m

e
i
n

s
e

c
o

n
d

s

MA−VWU

MA−DAMN

MA−DAAN

Fig. 11. ESnet: Comparison of run times (Dmax = 10).

TABLE I
NSFNET: AVERAGE NUMBER OF WAVELENGTHS (Dmax = 6).

Load MA-VWU MA-DAMN MA-DAAN %MA-DAMN %MA-DAAN

10 13.13 9.33 8.57 28.93 34.77
20 16.80 12.20 11.33 27.38 32.54
30 20.07 14.50 13.73 27.74 31.56
40 22.87 16.80 15.63 26.53 31.63
50 25.27 19.10 17.70 24.41 29.95
60 28.27 21.10 19.43 25.35 31.25
70 30.67 22.97 21.37 25.11 30.33
80 33.20 24.73 23.07 25.50 30.52
90 35.90 26.77 24.90 25.44 30.64

100 37.97 28.67 26.53 24.50 30.11

NSFnet and ESnet for Dmax = 6, 8 respectively. We show
the percentage improvement (in terms of wavelength usage) of
MA-DAMN (%MA-DAMN) and MA-DAAN (%MA-DAAN)
as compared to MA-VWU.

For both networks, it can be observed that MA-DAMN
clearly out-performs MA-VWU, and achieves a 30 − 45%
improvement. The MA-DAAN achieves a further 2 − 6%
improvement over the MA-DAMN. As the number of many-
cast destinations increase, the performance of both MA-SPO
implementations improve relative to MA-VWU, while the im-
provement of MA-DAAN over MA-DAMN remains relatively
consistent across network loads and request set sizes.

TABLE II
ESNET: AVERAGE NUMBER OF WAVELENGTHS (Dmax = 6).

Load MA-VWU MA-DAMN MA-DAAN %MA-DAMN %MA-DAAN

10 17.10 10.80 10.37 36.84 39.38
20 22.33 14.77 14.23 33.88 36.27
30 27.27 18.00 17.63 33.99 35.33
40 31.30 21.40 20.90 31.63 33.23
50 35.50 24.10 23.53 32.11 33.71
60 39.53 27.07 26.23 31.53 33.64
70 43.07 29.83 29.10 30.73 32.43
80 46.00 32.53 31.57 29.28 31.38
90 49.73 34.90 34.10 29.83 31.43

100 53.30 37.47 36.50 29.71 31.52

TABLE III
NSFNET: AVERAGE NUMBER OF WAVELENGTHS (Dmax = 8).

Load MA-VWU MA-DAMN MA-DAAN %MA-DAMN %MA-DAAN

10 15.48 9.57 8.97 38.21 42.09
20 19.93 12.63 12.27 36.61 38.45
30 23.41 15.27 14.73 34.80 37.07
40 27.55 17.60 16.93 36.12 38.54
50 30.66 19.57 19.10 36.17 37.69
60 33.59 21.90 20.87 34.79 37.87
70 36.72 23.63 22.80 35.65 37.92
80 39.03 25.77 24.67 33.99 36.81
90 42.14 27.50 26.53 34.74 37.03
100 44.66 29.70 28.13 33.49 37.00

TABLE IV
ESNET: AVERAGE NUMBER OF WAVELENGTHS (Dmax = 8).

Load MA-VWU MA-DAMN MA-DAAN %MA-DAMN %MA-DAAN

10 20.10 11.00 10.70 45.28 46.78
20 27.48 14.93 14.37 45.66 47.72
30 32.62 18.50 17.97 43.29 44.92
40 37.28 21.47 21.07 42.41 43.48
50 42.21 24.40 24.23 42.19 42.58
60 46.00 27.53 26.80 40.14 41.74
70 50.76 30.17 29.50 40.57 41.88
80 55.07 33.27 32.33 39.59 41.29
90 58.86 35.77 34.80 39.24 40.88
100 63.48 38.47 37.30 39.41 41.24

V. CONCLUSION

The steady emergence of bandwidth-intensive applications
effects the need for manycast communication essential in op-
tical WDM networks. Split-incapable networks do not support
all-optical manycast splitting, thus creating the problem of
establishing the same functionality as a virtual overlay to the
unicast-only optical layer. In this paper, we have presented
two heuristic solutions to the manycast overlay problem,
MA-DAMN and MA-DAAN. Through extensive simulations,
we have demonstrated their potential to substantially reduce
wavelength utilization (33−45%) across large-scale networks
over naı̈ve single-hop unicast routing (MA-VWU) to multiple
destinations, for realistic, dynamic traffic patterns.

REFERENCES
[1] R. Jain, “Internet 3.0: Ten problems with current Internet architecture and

solutions for the next generation,” in Proc., IEEE MILCOMM, Oct. 2006.
[2] M. D. Leenheer, F. Farahmand, K. Lu, T. Zhang, P. Thysebaert, B. Vol-

ckaert, F. D. Turck, B. Dhoedt, P. Demeester, and J. P. Jue, “Anycast
algorithms supporting optical burst switched grid networks,” in Proc.
IEEE International Conference on Networking and Service, (ICNS-’06),
Silicon Valley, CA, USA, Jul. 2006, pp. 63–69.

[3] X. H. Q. She, V. M. Vokkarane, and J. P. Jue, “Manycasting over optical
burst-switched networks,” in Proceedings, IEEE ICC, Jun. 2007.

[4] L. H. Sahasrabuddhe and B. Mukherjee, “Light trees: optical multicasting
for improved performance in wavelength routed networks,” vol. 37, no. 2,
pp. 67–73, Feb 1999.

[5] “Advanced networking for distributed petascale science: R&D challenges
and opportunities,” 2008. [Online]. Available: http://www.er.doe.gov/
ascr/ProgramDocuments/ProgDocs.html

[6] H.Takahashi and A.Matsuyama, “An approximate solution for the steiner
problem in graphs,” Math. Japonica, 1980.

[7] A. Gadkar, J. Plante, and V. M. Vokkarane, “Manycasting: Energy-
efficient multicasting in wdm optical unicast networks,” in Proc. Globe-
com, Dec. 2011.