Académique Documents
Professionnel Documents
Culture Documents
Long Cheng , Sajal K. Das , Jiannong Cao , Canfeng Chen and Jian Ma
State Key Lab of Networking & Switching Tech., Beijing Univ. of Posts and Telecomm., China
Center for Research in Wireless Mobility and Networking (CReWMaN), University of Texas at Arlington, USA
Department of Computing, Hong Kong Polytechnic University, Hong Kong
Nokia Research Center, Beijing, China
{longc, das}@uta.edu, csjcao@comp.polyu.edu.hk, {canfeng-david.chen, jian.j.ma}@nokia.com
189
has the same transmission range . A packet is received with minimum edge cost. In fact, such minimum cost
successfully if the received signal power is greater than multicast tree is well-known as the Steiner tree problem.
the receiving power threshold. As most other work in this However, it is not the best solution for WSNs, the multicast
area, the network is modeled by an undirected graph = routing energy efficiency can be further improved. In this
(, ), where is a set of sensor nodes and represents example, the multicast tree involves 9 nodes, 8 edges, and
the set of communication links. For any two nodes 1 and requires 7 transmissions. Fig. 1(c) illustrates the minimum
2 (1 , 2 ), if node 2 is in the transmission power transmission multicast tree, which takes advantage of the
range of node 1 , then there is an undirected arc (1 , 2 ) broadcast nature of wireless communication. We can see
. that it only needs 4 transmissions. The low transmission
Suppose (, ) is a given multicast request, overhead makes this kind of multicast tree suitable for the
where is the source and specifies a set of energy constrained WSNs.
multicast receivers , i.e., all multicast receivers have the
same . Given a graph = (, ), a source R R R R R R
190
C. Multicast tree construction realizes that it is on the path to the source, and it marks
itself as a forwarder (by setting a flag ) for
MTMRP is similar to ODMRP [10] in that it also the current multicast session. Then, the forwarding node
applies the on-demand procedures to dynamically build updates the JoinReply message, filling its cached upstream
routes and maintain multicast group membership to avoid NodeID of this multicast session as the NexthopID. In this
channel overhead and improve scalability. When a multi- way, the JoinReply is propagated by each forwarder until it
cast source has data to send, it floods a multicast request reaches the source via the reverse path of the corresponding
(, ). To keep consistency with previous work, JoinQuery. Finally, this process constructs the routes from
we also refer to this multicast request as the JoinQuery the source to all multicast receivers.
message. 3) Biased backoff scheme: Specifically, we need two
1) JoinQuery propagation: When a multicast source definitions which are used in MTMRP.
has data to send, it will flood a JoinQuery. A Join- Definition 1. (RelayProfit) We define RelayProfit of a
Query message is at least composed of the following forwarding node as the number of multicast receivers that
elements: MessageType, NodeID, SourceID, GroupID, Se- this forwarding node covers1 among its neighbors, but not
quenceNumber, HopCount and PathProfit (defined in Sec- including the multicast receivers that have already been
tion 4.3.3). specifies that it is a JoinQuery covered by other forwarding nodes.
message. is maintained by the source
node and increased each time the source floods a new The RelayProfit is dynamically calculated according to
JoinQuery. together with the and the overheard JoinReply nearby. If a node overhears that
identifies a JoinQuery to prevent loops a multicast receiver sends out a JoinReply, it will mark
and to discard stale routes. HopCount is the hop distance this neighbor as a covered receiver. Given the one hop
from the source thus far. NodeID, HopCount and PathProfit neighbor table and a JoinQuery message, the forwarding
will be updated at each hop. node knows the number of uncovered multicast receivers
When a node receives a non-duplicate JoinQuery, it in the same multicast group among its neighbors.
stores the upstream (for reverse path learning), Definition 2. (PathProfit) The PathProfit of a route from
, and into the the source to current forwarding node is the total num-
routing table and will forward this JoinQuery. Instead of ber of multicast receivers that this path covers, but not
rebroadcasting the JoinQuery immediately, like DODMRP including the RelayProfit of current forwarding node.
[6], we introduce a backoff time at the JoinQuery forward-
ing node intentionally, based on a new metric which we
will give the definition below. The aim of this operation = ( ) (1)
is to amplify the differences of a JoinQuerys traversing =1
delay along different paths. This enables the JoinQuery
to travel faster along the selected path according to our where {, 1 , 2 , . . . , , } is a path connecting the
defined metric than other paths. source and current forwarding node . ( )
When the backoff timer expires, the JoinQuery is re- denotes the RelayProfit of node . We know that Path-
broadcasted. When it reaches a multicast receiver, this Profit is non-decreasing as the JoinQuery spreads over the
receiver marks itself as a covered multicast receiver, then network.
generates and broadcasts a JoinReply message. This Join- The heuristic of MTMRP lies in the biased backoff
Reply will return to the multicast source along the reverse scheme. As we learned from the motivation, giving the
path of the JoinQuery. Note that in case of receiving the node with larger RelayProfit and PathProfit higher priority
same JoinQuery several times, the receiver shall only reply to take the forwarding task may possibly reduce the
the first received JoinQuery and neglect others. number of forwarding nodes and increase the number of
2) JoinReply propagation: A JoinReply message is leaf nodes. That is to say, both RelayProfit and PathProfit
at least composed of the following information: Mes- influence the backoff time calculation. Thus, we introduce
sageType, NodeID (refers to the lasthop NodeID), Nex- two kinds of backoff delays: 1) relay backoff delay ;
thopID, ReceiverID, SourceID, GroupID and Sequen- and 2) path backoff delay , which are defined as Eq.
ceNumber. NexthopID specifies the selected next hop node. 2 and Eq. 3, respectively.
ReceiverID specifies the node who originates the JoinRe-
ply. = 2( ) (2)
When a node receives a JoinReply, no matter whether 1 In this work, covered refers to connected, denoting that there is
it is a multicast receiver or not, it checks if it is the a multihop route connecting the source to a multicast receiver via some
selected next hop of the JoinReply. If that is the case, it forwarding nodes.
191
RP = 1 RP = 1
[4,5]
PP = 0 PP = 0
N=3 A D G
A C RP = 1
A C RP = 1 RP = 1
PP = 1 PP = 1 PP = 2
S S HC = 2
192
to the multicast receiver who originates this JoinReply. It 1 Procedure: void RecvJoinQuery (Packet *p)
only marks itself as a forwarder for the current multicast
session. 2 if Non-duplicate JoinQuery then
Similarly, when a multicast receiver receives a Join- 3 if I am a multicast receiver then
Query message, before replying the JoinReply, it first 4 if There is a forwarder among my
checks whether there is any forwarder in its neighbor table. neighbors then
If there is, it will remain silent. If not, it then sends out a 5 remain silent;
JoinReply. Besides, considering the uncertainty of wireless 6 else
7 send out JoinReply;
8 end
G I 9 mark myself as a covered multicast
receiver;
path handover
H
10 end
C
F
11 calculate and call
A Backoff( , );
S /* It will schedule a backoff timer
E whose value is , then call
B D JoinQuery forwardJoinQuery(p) when the timer
JoinReply expires */
12 else
Fig. 4. An example illustrating the path han- 13 Drop();
dover scheme (PHS) in MTMRP 14 end
Algorithm 1: Algorithm in MTMRP when a node
channel, e.g., contention or collision, it is still possible that receives a JoinQuery message
a node receives a JoinReply, but it has marked itself as
a forwarder. In this case, it will discard this JoinReply.
When a multicast receiver receives a JoinReply, if it is packet duplications, a forwarder shall only forward data
the selected next hop of the JoinReply and it has been the first time it receives.
covered, it will mark itself as a forwarder and then discard It is possible that the discovered routes between source
this JoinReply. and multicast receivers break, e.g., a forwarder runs out of
PHS can not only reduce the transmission cost in energy. In this situation, the route recovery mechanism can
multicast tree construction phase, but also prune redundant be adopted. For example, if a multicast receiver detects a
routes for multicast routing. An example is shown in Fig. missing forwarder through periodical HELLO messages,
4. When node receives a JoinReply originated from , it can broadcast a route error message to repair the failed
since node has overheard a JoinReply from before, link locally [22] or even trigger the source to initiate a new
it knows that there is a forwarder among its neighbors. multicast routing construction process [23].
Therefore, it marks itself as a forwarder and then discard
the JoinReply received from . We can see that the origi- V. Simulation
nal path is handovered from [ ]
to [ ] by introducing PHS. This oper- In this section, we present simulation results to evaluate
ation saves 2 transmissions and involves 1 less extra node. the performance of MTMRP using ns-2 [24] simulator
Regarding how a node handles a received JoinQuery by comparing it with DODMRP [6], ODMRP [10] and
and JoinReply in MTMRP, we illustrate the pseudo codes MTMRP w/o PHS (MTMRP without adopting the path
of RecvJoinQuery() and RecvJoinReply() in Algorithm 1 handover scheme). All the results have been averaged over
and Algorithm 2, respectively. 100 Monte Carlo runs.
Once the multicast routing tree is constructed, data for- To fully evaluate the MTMRP, we carry out both grid
warding is very simple. Like ODMRP [10] and DODMRP topology and random topology tests in a 200 200
[6], when the multicast source sends out data packets, square area. In the grid topology, 10 10 sensor nodes are
only those nodes whose is active (marked as uniformly placed in the field, forming a two dimensional
forwarders) will take the forwarding task. To avoid the grid. The random topology with 200 sensor nodes is
193
loss exponent. We set = = 1, = = 1.5,
1 Procedure: void RecvJoinReply (Packet *p)
= 1, and = 4 in our simulation.
2 if Non-duplicate JoinReply then We select three evaluation metrics:
3 if My NodeID == JoinReply.NexthopID then
Normalized transmission overhead: This metric
4 if There is a forwarder among my
measures the communication overhead and energy
neighbors then
efficiency. We define the normalized transmission
5 mark myself as a forwarder;
overhead as the number of transmissions required to
6 Drop();
deliver a data packet from the source to all multicast
7 else
receivers.
8 if I am a forwarder then Number of extra nodes: This metric is defined as
9 Drop(); the number of extra nodes involved in a multicast
10 else if I am a multicast receiver and tree. Under the premise of reducing the transmission
has been covered then overhead, MTMRP tends to involve less extra nodes.
11 mark myself as a forwarder; Average relay profit: This metric is defined as the
12 Drop(); average of each relays RelayProfit in a multicast tree.
13 else In some sense, it also reflects the energy efficiency of
14 get JoinReplys nexthop; a multicast routing protocol.
15 call forwardJoinReply(p);
16 end B. Simulation results
17 end
18 else 1) Grid topology test: We start our evaluation with a
19 if JoinReply.ReceiverID!=JoinReply.NodeID grid topology, to first measure the performance gains of
then MTMRP before moving onto more general topologies. 20
20 mark this neighbor as a forwarder; multicast receivers are randomly selected at each round.
21 else Fig. 5 shows the results when we vary the multicast group
22 mark this neighbor as a covered size, i.e., the number of the multicast receivers.
receiver; Fig. 5(a) reports the changes of the normalized trans-
23 end mission overhead at increasing the multicast group size.
24 Drop(); The figure clearly shows that MTMRP outperforms
25 end DODMRP and ODMRP in terms of energy efficiency.
26 else
When the multicast group size is small, the gap between
27 Drop(); MTMRP and DODMRP is not obvious. The reason is
28 end
that the multicast receivers are very sparsely distributed in
Algorithm 2: Algorithm in MTMRP when a node the network when multicast group size is small. However,
receives a JoinReply message as the multicast group size increases, we observe that
MTMRP shows obvious superiority to other two protocols.
It can save about 3 transmissions on average2 . We also see
generated by the setdest tool in ns-2. The multicast source that DODMRP approaches to ODMRP as the multicast
node is positioned at bottom left (0, 0), and all the group size increases.
multicast receivers are randomly generated at each round. Fig. 5(b) illustrates the number of extra nodes involved
The sensor node transmission range is taken 40. in a multicast routing as the multicast group size changes.
and are set 0.001 and 4, respectively. The MAC layer It is seen that both DODMRP and MTMRP show ob-
protocol is IEEE 802.11 MAC. TwoRayGround channel vious improvements compared with ODMRP. Although
model is adopted as the signal propagation model, where DODMRP involves least number of extra nodes in this
the shadowing fading factor is not considered. Therefore, test, MTMRP is comparable with DODMRP.
for a unique distance, given the transmission power , the Fig. 5(c) plots the average relay profit under differ-
receiving power is a deterministic value as expressed ent multicast group sizes. As the multicast group size
in Eq. 5,
2 The multicast routing performance depends much on the distribution
2 2
() = (5) of multicast receivers in a network. In our simulation, all the multicast
receivers are randomly generated in each round. For some randomly
generated distributions, e.g., when multicast receivers are very scattered
where is the distance between the sender and receiver; or crowded, MTMRP wont show obvious advantage over other protocols.
and are the antenna gains; and are the Therefore, we believe 3 transmissions saving on average under the
antenna heights; is the loss factor, and is the path- randomly generated distributions is not a minute reduction.
194
Normalized transmission overhead
30 22 3.5
3
20
25 2.5
18 2
20 MTMRP 16 1.5
MTMRP w/o PHS 1
14
15 DODMRP
0.5
ODMRP 12
0
5 10 15 20 25 30 35 40 45 50 55 60 5 10 15 20 25 30 35 40 45 50 55 60 5 10 15 20 25 30 35 40 45 50 55 60
Multicast group size Multicast group size Multicast group size
(a) Normalized transmission overhead (b) Number of extra nodes (c) Average relay profit
30 7 MTMRP
35
28 6.5 MTMRP w/o PHS
Number of extra nodes
6 DODMRP
increases, the average relay profit for different protocols average relay profit.
increases as well. MTMRP always provides highest aver-
From Fig. 5 and Fig. 6, we also see that MTMRP
age relay profit. Although the differences seem very small,
behaves a little better than MTMRP w/o PHS in all metrics,
it still reflects the energy efficiency of a multicast routing
which benefits from the cases where a longer path is
protocol. This is because that all the results have been
handovered to a shorter path in PHS.
averaged over all forwarders.
2) Random topology test: In this test, 15 multicast 3) Tuning system parameters: In this experiment we
receivers are randomly selected at each round. The simu- adjust the system parameters and used in MTMRP.
lation results are shown in Fig. 6. Fig. 7 shows the effect under different and in grid
Fig. 6(a) depicts the performance comparison on the topology, where 20 multicast receivers are randomly se-
normalized transmission overhead. When the value of lected at each round. We see that MTMRP is more affected
multicast group size is within 10 to 40, the reduction of by the system parameters. This is due to the biased backoff
transmission overhead for MTMRP is obvious. Although scheme, the larger and , the more amplified difference
the performance is more influenced by the uncertainty in of routing latency at each hop. For example, when = 3
random topology, MTMRP shows more or less advantages and = 0.001, there is no significant difference between
over other two protocols averagely. DODMRP and MTMRP. As and increase, MTMRP
Fig. 6(b) shows the changes of the number of involved achieves a distinct advantage over other protocols. Note
extra nodes under different multicast group sizes. When the that the price paying for the reduced transmission cost
multicast group size is small, MTMRP is comparable with for DODMRP and MTMRP is the introduced backoff
DODMRP. However, as the multicast group size increases, delay at each hop during the multicast tree construction
less extra nodes are involved in DODMRP. phase. However, during the data forwarding phase, the
Fig. 6(c) illustrates the results of the average relay transmission overhead can be reduced significantly. When
profit. Similar as the grid topology test, the average relay tuning and in random topology, where 15 multicast
profit for different protocols increases at increasing the receivers are randomly selected at each round, Fig. 8 shows
multicast group size and MTMRP always provides highest the same results as in Fig. 7.
195
MTMRP MTMRP
MTMRP w/o PHS MTMRP w/o PHS
DODMRP DODMRP
30 24
28 22
26 20
24 0.03 18 0.03
0.025 0.025
22 3 0.02 16 3 0.02
0.015 0.015
4
5
0.01 4
5
0.01
0.005 0.005
N 6 N 6
Fig. 7. Tuning system parameters and Fig. 8. Tuning system parameters and
in grid topology in random topology
4) A closer observation: To have an intuitive under- (PHS), which can prune redundant routes in multicast rout-
standing of the above tests, we give a closer observation. ing. We compared MTMRP with existing protocols using
We show snapshots of multicast routing paths taken by ns-2 simulator. Simulation results demonstrated that our
three different protocols in Fig. 9 and Fig. 10. The hollow proposed MTMRP could effectively improve the energy
blue circles represent sensor nodes deployed in the field; efficiency for multicast routing.
the red crosses represent the multicast receivers; and the
filled black circles represent the forwarders in multicast
routing.
Acknowledgment
As shown in Fig. 9, MTMRP needs 26 transmissions
and involves 21 extra nodes. 32 transmissions and 20 extra The authors would like to thank the anonymous re-
nodes are required in DODMRP. While for ODMRP, it viewers for the helpful comments and suggestions. The
needs 33 transmissions and involves 29 extra nodes to com- work of Long Cheng was done while he was a vis-
plete the multicast routing task. The snapshots of routing iting scholar at CReWMaN Lab at the University of
paths for different protocols in random topology are shown Texas at Arlington. This work is partially supported by
in Fig. 10. In this example, the required transmissions National Natural Science Foundation of China (NSFC)
are 16, 21, 24 for MTMRP, DODMRP and ODMRP, under grant No.60873241; National 863 High Technol-
respectively. ogy Program of China under grant No.2008AA01Z217
From the above simulation results, we know that only and No.2009AA01Z210. NSF Grant IIS-0326505, CNS-
reducing the number of extra nodes [6] does not neces- 0721951 and CNS-0916221 and AFOSR Grant A9550-08-
sarily reduce the transmission overhead. To improve the 1-0260. The work of S. K. Das is also supported by (while
energy efficiency of multicast routing, it can be achieved serving at) the National Science Foundation. Any opinion,
by reducing the number of forwarding nodes and at the findings, and conclusions or recommendations expressed in
same time increasing the number of leaf nodes. Con- this material are those of the authors and do not necessarily
clusively, compared with existing distributed protocols, reflect the views of the National Science Foundation.
MTMRP reduces the energy cost for multicast routing.
References
VI. Conclusion
[1] K. Romer and F. Mattern, The design space of wireless sensor
networks, IEEE Wireless Communications, vol. 11, no. 6, pp. 54
In this work, we proposed MTMRP, a heuristic dis- 61, 2004.
tributed multicast routing protocol for WSNs. The design [2] L. Wang and Y. Xiao, A survey of energy-efficient scheduling
objective of MTMRP is to reduce the transmission cost mechanisms in sensor networks, Mobile Networks and Applica-
tions, vol. 11, no. 5, pp. 723740, 2006.
for multicast routing in WSNs. MTMRP introduces the [3] X. Jia, D. Li, and F. Hung, Multicast routing with minimum energy
biased backoff scheme and basically uses only one-hop cost in ad hoc wireless networks, in IEEE GLOBECOM 04, Nov
neighborhood information to construct a multicast tree. It 2004, pp. 28972901.
[4] P. Ruiz and A. Gomez-Skarmeta, Approximating optimal multicast
also takes advantage of the broadcast nature of wireless trees in wireless multihop networks, in IEEE Symposium on
communication and introduces the path handover scheme Computers and Communications, June 2005, pp. 686691.
196
200 200 200
180 180 180
160 160 160
140 140 140
Y-axis (m)
Y-axis (m)
Y-axis (m)
120 120 120
100 100 100
80 80 80
60 60 60
40 40 40
20 20 20
0 0 0
0 20 40 60 80 100 120 140 160 180 200 0 20 40 60 80 100 120 140 160 180 200 0 20 40 60 80 100 120 140 160 180 200
X-axis (m) X-axis (m) X-axis (m)
(a) MTMRP, 26 transmissions, 21 extra nodes (b) DODMRP, 32 transmissions, 20 extra nodes (c) ODMRP, 33 transmissions, 29 extra nodes
Fig. 9. Snapshots of routing paths for different protocols in grid topology, 20 multicast receivers are
randomly selected
Y-axis (m)
Y-axis (m)
Fig. 10. Snapshots of routing paths for different protocols in random topology, 15 multicast receivers
are randomly selected
[5] A. Shaikh and K. Shin, Destination-driven routing for low-cost hoc multicast routing protocol, Mobile Networks and Applications,
multicast, IEEE Journal on Selected Areas in Communications, vol. 7, no. 6, pp. 429439, 2002.
vol. 15, no. 3, pp. 373381, 1997. [16] L. Junhai, X. Liu, and Y. Danxia, Research on multicast routing
[6] K. Tian, B. Zhang, H. Mouftah, Z. Zhao, and J. Ma, Destination- protocols for mobile ad-hoc networks, Computer Networks, vol. 52,
driven on-demand multicast routing protocol for wireless ad hoc no. 5, pp. 988997, 2008.
networks, in IEEE ICC 09, June 2009, pp. 15. [17] T. Kunz, Multicasting in ad-hoc networks: Comparing maodv and
[7] G. H. A. D. de Morais Cordeiro, C., Multicast over wireless mobile odmrp, in Proceedings of the Workshop on Ad hoc Communica-
ad hoc networks: present and future directions, IEEE Network, tions, 2001, pp. 186190.
vol. 17, no. 1, pp. 5259, 2003. [18] S. Guo and O. W. W. Yang, Energy-aware multicasting in wireless
[8] C. E. Perkins and E. M. Royer, Multicast ad hoc on-demand ad hoc networks: A survey and discussion, Computer Communi-
distance vector (maodv) routing, IETF Intemet Draft: draft-ietf- cations, vol. 30, no. 9, pp. 21292148, 2007.
manet-maodv-00.txt, 2000. [19] C. Papageorgiou, P. Kokkinos, and E. Varvarigos, Energy-efficient
[9] S. Anmol, S. Brian, and H. Richard, VLM: a very lightweight multicasting in wireless networks with fixed node transmission
mobile multicast system for wireless sensor networks, in IEEE power, in ACM IWCMC 09, 2009, pp. 958962.
WCNC 03, Mar 2003, pp. 19361941. [20] A. Penttinen, Minimum cost multicast trees in ad hoc networks,
[10] S. J. Lee, W. Su, and M. Gerla, On-demand multicast routing in IEEE ICC 06, June 2006, pp. 36763681.
protocol in multihop wireless mobile networks, Mobile Networks [21] H. Frey, F. Ingelrest, and D. Simplot-Ryl, Localized minimum
and Applications, vol. 7, no. 6, pp. 441453, 2002. spanning tree based multicast routing with energy-efficient guar-
[11] J. Garcia-Luna-Aceves and E. Madruga, The core-assisted mesh anteed delivery in ad hoc and sensor networks, in WOWMOM 08,
protocol, IEEE Journal on Selected Areas in Communications, 2008, pp. 18.
vol. 17, no. 8, pp. 13801394, Aug 1999. [22] M. Naderan-Tahan, A. Darehshoorzadeh, and M. Dehghan,
[12] C.-C. Chiang, M. Gerla, and L. Zhang, Forwarding group multicast ODMRP-LR: ODMRP with link failure detection and local re-
protocol (fgmp) for multihop mobile wireless networks, Cluster covery mechanism, in ICIS 09, 2009, pp. 818823.
Computing, vol. 1, no. 2, pp. 187196, 1998. [23] R. S. Gray, D. Kotz, C. Newport, N. Dubrovsky, A. Fiske, J. Liu,
[13] M. Lusheng Ji, Corson, Differential destination multicast-a manet C. Masone, S. McGrath, and Y. Yuan, Outdoor experimental
multicast routing protocol for small groups, in IEEE INFOCOM comparison of four ad hoc routing algorithms, in ACM MSWiM
01, 2001, pp. 11921201. 04, 2004, pp. 220229.
[14] S. I. Sanchez J.A., Ruiz P.M., GMR: Geographic multicast routing [24] Network simulator. [Online]. Available:
for wireless sensor networks, in SECON 06, 2006, pp. 2029. http://www.isi.edu/nsnam/ns/
[15] J. Xie, R. R. Talpade, A. Mcauley, and M. Liu, AMRoute: Ad
197