1

Multicast Routing in Wireless Sensor Networks

First M.Simek
1
, Second D.Komosny
1
, Third R.Burget
1
and Fourth J.S.Silva
2
, Member, IEEE

1
Department of Telecommunications, Brno University of Technology, Czech Republic
2
Department of Informatics Engineering, University of Coimbra, Portugal

A paper deals with the state of the nowadays research of the multicast routing protocols for wireless sensor networks. We describe
how to establish efficient multicast tree for wireless sensor network environment and introduce the two base groups of the multicast
routing protocols. For these groups, blind flooding and geographic protocols, we introduce the main protocols representatives.


Index Terms network, wireless, sensor, multicast, routing

I. INTRODUCTION
In present, a set of protocols were proposed for the
multicast routing in the wireless Ad-hoc networks. But, most
of them are not suitable for the implementation in the sensor
networks environment. Above all, due to the modest
resources of the sensor nodes such as memory and bandwidth
limitations, the low computational power, limited battery etc.
For instance, the most Ad-hoc routing protocols assume the
relatively high links bandwidth ~ 2Mbps, but the sensor motes
with 802.15-4 radio base operate at a nominal transmission
rate of 250kbps.
The main characteristics for multicast routing protocols
evaluation are:

i) Packet Delivery Ratio: The ratio of the number
packets actually delivered to the destinations versus
the number of data packet supposed to be received.
ii) Number of packets transmitted per packets delivered:
Count of every individual transmission of packets by
each node to count of packet delivered.
iii) Routing Overhead (Number of control packets
transmitted per packets delivered): It express the
effectiveness of controls packets for data delivering.
iv) Number of packets and control packets per packets
delivered: The total packet overhead.
II. MULTICAST TREE CONSTRUCTION
A. Construction
The construction of multicast tree in the wireless ad-hoc
environment is quite different from the multicast tree
construction in the fixed IP networks. In such fixed networks,
there are minimal restrictions in terms of data processing
overhead and minimizing the cost of the established multicast
tree. These protocols build the multicast tree routed from the
source with the minimum number of hops, so-called source
(shortest) path tree SPT. The main goal in the research of
wireless ad-hoc multicast protocols is to find the efficient
solution with the regarding to minimizing the cost of multicast
tree with the consideration of the mobile scenario. In the ad-
hoc network, the mobility of end nodes is the common and
expected behavior whilst in the nowadays wireless sensor
networks, the mobility could be presented by the fast changing
topology as the sensor nodes become sleep or awake. The
cost of the multicast tree needs to be considered in such
wireless networks mainly because of the bandwidth limitations
and power-energy restrictions. The issue of the computing the
multicast tree with the minimal cost is known as the Steiner-
tree problem and it is also proved that this problem is NP-
complete [1]. Crucial condition is to determinate the particular
design metrics for each type of networks. Apparently, the
shortest past is the explicit criterion for the multicast tree
creation in fixed IP networks. In the wireless ad-hoc and
sensor networks the metric definition is dependent on the
resource conditions and the behavior (static / mobile) of such
network. We have to take into consideration two possible
metrics: i) number of hops relaying the multicast packet, ii)
number of transmission induced by the multicast
communication. The challenge arises with the attempt to find
the network conditions for which the i) or ii) is suitable metric
cost. Indeed, the network scenario for which the both metrics
are crucial can exist.
Fig. 1 shows the tree construction scenarios with the terms
of the different metric costs. Multicast tree in Fig. 1 a) is
constructed in terms of the finding the shortest path to all
receivers. In this case, SPT uses 8 nodes (included source) and
induces 5 transmissions. Steiner Tree algorithm establishes the
multicast tree with the minimum hop counts. In same scenario,
it uses two less nodes against SPT and induces also less
number of transmissions (see Fig. 1b)), but it is dependent
upon the network topology. Further, the multicast process is
different in fixed and wireless networks, since the designated
router in fixed network needs to make as many copies as the
number of receivers exist in his downstream leaf. In wireless
ad-hoc networks, the packet relaying is based on the single
message broadcasting to all neighbors. Thus, it is necessary in
such networks to reformulate the design metric for the
multicast tree construction. Rather than be considering in the
minimum hop count upon delivering multicast data from
source to multicast group, the more crucial metric becomes the
number of transmission that is incurred by the multicast
communication. The multicast tree construction in terms of the
transmission number minimization is shown in Fig. 1 c). The
number of nodes used is higher than with Steiner Tree, but the
overall low number of transmissions (also message overhead)
2


Fig. 1 Comparison of a) SPT, b) Steiner Tree, c) Min.Tx Tree multicast tree
construction.

makes this type of the construction suitable for the deployment
to the bandwidth-constrained networks such the wireless
sensor networks are. On the other hand, the complexity of
computing minimal tree with the minimizing of path cost still
leads the research community to deal with multicast route
discovery process using the broadcast flooding solution also
known as Broadcast-Storm Problem.
B. Definition of Multicast Tree Problem
In the mathematical expression, the wireless ad-hoc
network is mostly modeled by the undirected graph G(V,E)
where V is the set of vertices and E is the set of edges. Lets
have r as a transmission range of each node v V. The edge
between the v
1
and v
2
is established if dist (v
1,
v
2
) < r.
For the exact problem definitions, some expressions need to
be formulated. The problem is to find the set of nodes F V
relaying the multicast data from source to the destination (end
nodes joined to the particular multicast group). From the
formulation (1) describing the problem of finding the set of
relaying nodes in unicast communication, we are able to
define the formulation for multicast communication as well
(2):

V s.t. sFD is connected (1)

V s.t. sFR is connected (2)

, where the source s V, the destination D V and the set of
receivers R V [1]. As the literature [2] describes, the
number of transmissions referred as C
t
(t) required for
delivering single message from the source s to the set of
receivers R via the set of relaying nodes F creating multicast
tree t T can be evaluated as:


Fig. 2 Mobility solution of VLM
2



C
t
(t)=1+|F
t
| (3)


As was mentioned above, in wireless ad-hoc scenario, each
relay node rebroadcast just only one packet, thus number of
transmission is equal to number of relay nodes |F
t
| whereas the
source contributes also with the one initiated packet. Hence,
for the minimizing the message overhead, we have to also find
the minimal number of relaying nodes |F
t
|. In the next section,
the two optimization algorithms proposed in [2] are briefly
described.
III. BLIND FLOODING PROTOCOLS
The main shortcomings of the broadcast flooding described
in [2] also called as broadcast storm problem used for
multicast tree discovery process are:

i) induces the great data overhead,
ii) the intermediate nodes need to maintain the heavy
multicast table,
iii) the energy wasting,
iv) the maintaining of the fresh multicast tree in the mobile
scenario incurs the often message
A. Very Lightweight Mobile Multicast VLM
2

VLM
2
is link layer protocol providing the unicast,
broadcast, multicast routing for the forward sink-to-nodes
direction and unicast, broadcast for backward nodes-to-sink
direction. Downstream unicast routing is performed by the
multicast group establishment with only one concrete member.
This protocol also offers the feasible solution for the WSNs
with the hierarchical (least one powerful basestation) and the
mobile scenario; whereas the symmetric lines are expected to
be guaranteed (S1 sensor is able to communicate with S2 and
contrariwise). Due to 8-bit nodes and groups identifiers used,
the number of nodes and multicast groups is restricted to 255,
and 248 respectively. However, this numbers are for WSNs
environment more than sufficient.
The main idea of mobility solution consists in the DtB
(Distance to Base) timer maintained by all nodes. The timer
indicates the last time when the DtB value was updated. If the
timer expires, the nodes increment DtB value and generate the
DtB_Update message to discover possible changes in the own
position (for more detailed information see Fig.2. In the
simulation described in the efficiency of the upstream and the
3

Fig. 3 ADMR route discovery process
downstream transmissions were evaluated. It is shown that
VLM
2
contends with the high mobility environment that
induces the heavy packet loss with the increasing speed of
nodes. This packet loss is incurring because of the increasing
frequency of the Subscription message generation by the
multicast members nodes as topology quickly changes.
Likewise, the greater control message (DtB_update) overhead
appears as the nodes become more mobile, since the
increasing number of links fault need to be reestablished. As
reference [3] describes some disadvantages, the mentioned
system has no capability to automatically choose the set of
nodes and subscribe them to the specific multicast group. The
present solution assumes that the nodes know own multicast
address that is preprogrammed in the operation system.
B. Adaptive Demand-driven Multicast Routing ADMR
ADMR protocol [4], [5] uses also the common broadcast
flooding process for the multicast route discovery process.
Initially, sender broadcast the Route_Discovery message to
determinate the shortest path to the members of the multicast
group. The intermediate nodes compare the hops number of
the discovery message with the own maintained number of
hops to source. If the message hops count is less than nodes
own hops, the node updates the routing information. In this
downstream direction, intermediate nodes only merely
rebroadcast the discovery message to neighbors. While the
receivers response with the Receiver-Join message as
describes Fig. 3, the intermediate nodes receiving this message
identify itself as a forwarder for the multicast pair (S,G). Each
node pertaining to delivery multicast tree maintains the tree
routing tables: Node, Membership and Sender Table that are
individually indexed by the (S,G) pair. The content of the
particular tables is shown in Fig.4..

Index Node Table Membership Table Sender Table
(S,G)
previous hop
forwarder/sender

sender for G1
sender for G5
.
path cost
sequence number

Fig. 4 The routing tables of all nodes belonging to the forwarder multicast tree

To afford the sufficient memory space mainly for the Node
tables becomes the critical issue with the increasing number of
multicast groups emerged in the sensor network.
One of the ADMR weaknesses (as the origin protocol for
ad-hoc network) is the routing metric using the minimum hop
count principle for the route discovery process. However, the
path selection with the minimal number of hops may not be
the best solution, since may exist the else longer links but with
the much better parameters of transmission quality.
Nevertheless, as describes [4], the several alternate selection
metrics overcoming this issue were proposed yet. It is
necessary also point out, that ADMR was originally proposed
for the 802.11 wireless ad-hoc network with the 1Mbps of
average bitrate. And thus, the regular sender flood process for
maintaining of the fresh multicast delivery tree appears as the
efficient and suitable solution in this 802.11 wireless
environment. However, in the sensor 802.15.4 wireless
interfaces with the five times less bitrate about 200kbps, this
sender flooding becomes the overhead originator No.1. At last,
we can summarize the main aspects restraining the ADMR
deployment into the sensor networks. These are: the source
regular flooding, the insufficient memory space for routing
tables for the great-scale network and also the volume of the
ADMR upon conversing to nesC code.
IV. GEOGRAPHIC MULTICAST ROUTING PROTOCOLS
In particular, the main advantage that geographic routing
brings is the data routing without any knowledge of the
network topology. That is, the nodes need not to maintain the
heavy multicast routing tables and refresh it in periodic
interval such as in the flooding-based solutions mentioned
above in chapter 3. The geographic routing protocols take
advantage of the position information of all nodes in the WSN.
The intermediate nodes belonging to the multicast tree
forwarding the multicast data in accordance with the list of
nodes included in the each packet header that is to be
forwarded. From the mentioned, the principal limitation of the
geographic-based routing ensues. It is the increasing size of
the packet header as the network scale grows, mainly as the
number of nodes belonging to the multicast tree is increasing.
In the next chapters, the known geographic multicast routing
protocols are briefly described. Some of them deal with the
energy-efficiency, others with the cost of the tree in terms of
number of the forwarding nodes.
A. Grid Multicast Protocol GMP
The key advantage of the Grid Multicast as the energy-
efficient multicast protocols consists in the data routing from
the sink node to the multicast destination via the energy-
shortest geographic distance. It relies at the theory that the
longer hop between the pair of nodes consumes more energy
than the smaller one. Therefore, it uses the rectilinear hop by
hop communication in the sensor network formed in the grid
shape.
The grid shape of the sensor network is crucial condition for
this algorithm, but as literature [7] describes, the precise grip
deployment is not necessary. In the random deployment
scenario, the virtual grid is constructed with the grid unit size
4
of the sensor radio range divided by 5. Thus, in the each grid
unit, there is just one awake sensor, while the rest of sensors in
given unit are in the sleeping mode. Due to the grid unit size,
each node is able to communicate with the adjacent nodes
from the other units in the rectilinear way and thus to save an
energy. The Grid Multicast introduces the Steiner Point
node (see node classified as SP in Fig.5) as the contributed
nodes for the reducing of the distance from sink to the nodes
belonging to the multicast group.


Fig.5 Multicast tree construction, a) traditional model of SPT, b) Grid
Multicast model using rectilinear communication

The only aim of Steiner Point as a tree member is the
packet forwarding from the sink node. The advantage of the
Grid Multicast is data routing by means of Location-Guided
Grid Routing that has no use for the maintaining of the
multicast routing tables in the intermediate nodes. The
position of the intermediate nodes belonging to the created
multicast tree is placed in to the multicast packet header, thus
upon the receiving, the intermediate nodes have position
information of the downstream multicast nodes. But since the
nodes have the narrow memory space, with the increasing
number of the intermediate nodes whose addresses are
included in the packet header, this solution has the restricted
capabilities. Hence, the authors of [7] try to reduce the size of
packet header by using of the grid coordinate information,
instead of IP address or geographic position information as the
identifier of each node belonging to the multicast tree. In the
comparison with the similar geographic-based protocols such
as the LGT [8], the Grid Multicast achieves the better results
in terms of the energy-efficiency even if the nodes are not
deployed in the accurate grid. In terms of the number of nodes
involved in the multicast tree transmission, this multicast
protocol induces the bigger bandwidth consumption. It is
obvious, since it aims to use more nodes (as the energy-
efficient solution) consuming the larger bandwidth resources.
B. Geographic Multicast Routing GMR
GMR protocol is geographic localized protocol using only
position information for the multicast routing in the sensor
networks, which avoids the undesirable broadcast flooding. It
uses the new heuristic neighbor selection algorithm [6] that
reduces the cost of the constructed multicast tree in terms of
the number of nodes selected for the multicast data
forwarding, it means in terms of the number of nodes creating
the multicast tree. The key of GMR is mostly based on the
idea described in the chapter 2, the idea of the most important
metric in the ad-hoc network is not being the number of nodes
involved in the multicast session but the number of
transmission induced by this session. The GMR uses the new
evaluative parameter (lets call it nv = neighbor value) for
the selection of the best candidate sets of nodes for the data
forwarding to the multicast destination. Statement for the nv
parameter is following:

(4)

The n
n
stands for number of neighbor of given set, D
w
stands for the distance to the multicast destination from the
current node without neighbor selection (direct
communication) and finally the D
n
is the new distance from
selected neighbor node to the multicast destination. The
algorithm subsequently selects the set of forwarding neighbor
nodes with the minimal nv whereby ensures that the minimal
number of nodes will used for the multicast transmission. For
the clarification, take into consideration the scenario in Fig. 6.


Fig. 6 GMR Scenario of the neighbor selections


The node N is selecting the adjacent nodes as the forwarders
for the data being transmitted. In the Fig. 6a) the node N
selects the 3 neighbor nodes F1,F2,F3 for which computes the
nv parameter, separately for each of sets being leafs of
concrete F node. For instance, the nv parameter for the F1
node is computed as the reciprocal value of the D1 (origin
distance) and D2 (new distance) distinction. We can call this
distinction as the d
12.

1 |1| |2| |3|

2 |11| 12| |13|
(5)


And thus, the overall nv parameter for the scenario in Fig. 6a)
is:

(6)
In the statement (6), the x stands for the number of selected
forwarding nodes. The key of GMR consists in the selection of
the set of forwarding nodes with the smallest nv parameter, it
means that the set of forwarding nodes from Fig. 6b) will
selected as the forwarders if:

5

(7)
nv
ab
=nv parameter for scenario in Fig. 6a) and Fig. 6b)
respectively. The authors of GMR [6] made an efficiency
comparison with the similar based protocol PBM [9]. The
authors of PBM attempt to find the optimal relation between
the number of selected forwarding nodes and efficiency of the
created path. For the evaluation of the possible subsets FS,
(each selected subset fs FS) of forwarding nodes use the
function described in the statement (8).

1 , 0 1 (8)
Here N is the rate between the number of selected
forwarding nodes to the total number of possible forwarding
(adjacent) nodes and S stands for the rate between the sum of
distances from nodes in FS toward the destinations to the sum
of distances from actual node (selecting forwarders) toward
the destinations. As the authors present, the main issue is to
find the optimal value of parameter that defines the
efficiency of the selected solution. As the results from [6]
show, the GMR notably outperform the PBM in the terms of
the number of induced transmissions. As the density of
adjacent neighbors of evaluated node varies from 5 to 10, the
GMR achieves the 2-35% (density of 5) and 30-95% (density
of 10) better results. The range of percentage results depends
on the increasing value of parameter for PBM from lower
value 0 to the maximum value 1. In terms of the number of
transmission required for the message delivering to for
instance 25 destinations with the density of 7, the GMR
induces about 700 transmissions whilst the PBM about 900
transmissions. From these results, it is possible to highlight the
two fundamental advantages of GMR. It is the suitable
efficiency in terms of the number of transmission inducing and
the independency on the other parameters. But the question
arises, what results this protocol achieves upon the real
implementation to the sensor motes with the constrained
memory space and also how will battle with the bigger
number of destinations, since the performed simulations
consider just only small number of multicast receivers.
V. CONCLUSION
We have described a main principles, advantages and
shortcomings of the multicast routing protocols for wireless
sensor networks. As one can see, from the mentioned
shortcomings of these protocols, the proposing of the new and
energy-effective multicast protocol for the wireless sensor
network is still the open challenge.
ACKNOWLEDGMENT
This work was supported by the FRVS grant agency,
project no. 1514/2008
REFERENCES
[1] Lim H., Kim C., Multicast Tree Construction and Flooding in Wireless
Ad Hoc Networks, Proc. MSWiM00, Boston, MA. August, 2000, pp.
6168.
[2] Ruiz, P.M., Gomez-Skarmeta A.F., Approximating Optimal Multicast
Trees in Wireless Multihop Networks, in Proc. 10th IEEE Symposium on
Computers and Communications, ISCC 2005, La Manga,Spain, June
2005, pp. 686691.
[3] Sheth A., Shucker B., Han R.: VLM2 : A Very Lightweight Mobile
Multicast System For Wireless Sensor Networks, In Proceeding of IEEE
WCNC 2003.
[4] Chen B., Muniswamy-Reddy KK., Welsh M.: Ad-hoc Multicast Routing
on Resource-Limited Sensors Nodes, In Proceeding of REALMAN 06,
May 26, 2006, Italy.
[5] Jetcheva J. G., Johnson D. B.: Adaptive Demand-Driven Multicast
Routing in Multi-Hop Wireless Ad Hoc Networks. In 2001 ACM
International Symposium on Mobile Ad Hoc Networking and
Computing (MobiHoc2001), October 2001.
[6] Sanchez J.A., Ruiz P.M., Stojmenovic I. GMR: Geographic Multicast
Routing for Wireless Sensor Network, In Proceeding in Sensor, Mesh
and Ad-hoc Communication and Network, IEEE, Secon 06,
2006,Virginia, USA.
[7] Zeng G., Wang Ch., Xiao L., Grid Multicast: an Energy-Efficient
Multicast Algorithm for Wireless Sensor Networks
[8] Chen K., Nahrstedt K.,: Effective location-guided overlay multicast in
mobile ad hoc networks, International Journal of Wireless and Mobile
Computing(IJWMC), vol. 3, 2005.
[9] Mauve M., Fuler H., Widmer J., Lang T.: Position-Based Multicast
Routing for Mobile Ad-Hoc Networks, TR-03-004, Department of
Computer Science, University of Mannheim, March, 2003.
[10] Chen K., Nahrstedt K.: Efficient location-guided tree construction
algorithms for small group multicast in Manet. In INFOCOM, pages
11801189, 2002.