DESIGN AND PERFORMANCE EVALUATION OF MAXIMUM REMAINING ENERGY
CONSTRAINED DIRECTED DIFFUSION ROUTING ALGORITHM FOR WIRELESS
SENSOR NETWORKS
Jae Yong Lee*, An Kyu Hwang, Byung Chul Kim
Data Communications Laboratory, Department of Information Communications Engineering,
Chungnam National University, 220 Gung-dong Yusung-ku Daejeon, 305-764 Korea
Email: {jyl, akhwang, byckim}@cnu.ac.kr
Tel: +82-42-821-6865, Fax: +82-42-823-5586
Abstract: Since the sensor network nodes have a small size and limited battery power, there have been many
studies for reducing their energy consumption. Each sensor node can show different energy usage according
to the frequency of event sensing and data transmission, thus they have different lifetime. So, some nodes
may run out of energy that causes disconnection of paths and reduction of network lifetime. In this paper, we
propose a new energy-efficient routing algorithm for sensor networks called Maximum Remaining Energy
constrained Directed Diffusion (MRE-DD) routing, that selects a least energy-consuming path among the
paths formed by nodes with highest remaining energy grade, and provides long network lifetime and
somewhat uniform energy consumption by nodes. Simulation results show that our algorithm extends the
network lifetime and enhances the network reliability by maintaining distribution of remaining energy
relatively uniform among sensor nodes.
Keywords: Wireless sensor networks, sensor network routing, network lifetime, directed diffusion
1. INTRODUCTION
Due to the development of semiconductor technology, wireless communication and sensor technology, it is
possible to produce abundantly low-cost and tiny sensor nodes that have sensing, processing and wireless
communication functions. We can build many kinds of sensor networks for various applications that require
sensing and communicating useful environmental information by using many low-cost sensor nodes. Sensor
network is a kind of ad-hoc networks that find routing paths between nodes without network infrastructure.
One of the important advantages of sensor networks is that we can get useful information from many places
dangerous or difficult to access by distributing small sensor nodes (Akyildiz et al., 2002).
In order to maintain proper operation of sensor networks, it is important to use the network resources such as
node energy efficiently, because it is difficult to replace the randomly installed sensor nodes that exhaust
their resources. Since almost every sensor node is operated by battery power and one of the largest energy
consuming functions is routing process, it is very important to adopt an energy-efficient routing algorithm in
sensor networks.
There have been several studies for the energy-effective routing algorithms for sensor networks. Directed
diffusion (DD) routing algorithm (Intanagonwiwat et al., 2003) tries to find a routing path that consumes the
least energy for information transfer among many paths between a sensor node and an information collecting
gateway. Its energy consumption could be minimal, but the same path is used repeatedly until one of nodes
in the path spends all the energy it has and that path is disconnected. Thus, the lifetime of sensor networks is
reduced. The distribution of remaining energy of nodes becomes uneven, which make the network
unreliable. Another important study for energy-efficient routing for sensor networks is the Energy Aware
Routing (EAR) (Shah et al., 2002). When sensing information is transferred, it does not select energyshortest path, but it chooses next node probabilistically by considering energy waste, remaining energy and
etc. Since it does not use the same path repeatedly, lifetime of a path can be extended compared to DD.
35th International Conference on Computers and Industrial Engineering
1299
However, probabilistic path selection in the EAR algorithm does not enforce the use of energy-shortest path,
which may have the packet traverse more nodes and consume more energy. Thus, the EAR cannot extend the
lifetime of sensor networks, except for the time of the first node failure.
In this paper, we propose a new energy-efficient routing algorithm for wireless sensor networks, called
“maximum remaining energy constrained directed diffusion (MRE-DD) routing”. The MRE-DD tries to find
an energy-shortest path as the DD does. In addition, it selects a path that has more remaining energy in the
traversing nodes. First, it assigns some grades to the sensor nodes according to the remaining energy level,
for example, 20% interval of remaining energy level. When it chooses the next hop for information transfer,
it selects and reinforces a shortest-energy path only among the highest energy-grade node group. Thus, it
does not persist to use the same shortest energy path as in the DD, because the remaining energy level of the
shortest-energy path would be eventually lower, then the MRE-DD will choose another shortest-energy path
among the highest energy-grade node group.
We find and compare the network lifetime performance of the three algorithms, the DD, the EAR, and the
MRE-DD by simulation. We also measure the fairness index of the remaining energies for sensor nodes.
Simulation results show that the MRE-DD algorithm extends the lifetime of sensor networks and increases
the network reliability by keeping the remaining energy of nodes relatively uniform.
This paper is organized as follows. In section 2, we briefly review the related works for energy-efficient
sensor network routing algorithms. We propose the MRE-DD routing algorithm for wireless sensor networks
in section 3. The performance evaluation of the proposed algorithm is shown by simulation in section 4. We
present summary and conclusion in section 5.
2. RELATED WORKS
Since every node in sensor networks cannot always reach and communicate each other, sensor nodes should
have multi-hop routing functions in order to transmit the received data to the direction of the destination.
Since one of the largest energy consuming functions in sensor nodes is wireless communication and routing
process, routing algorithms for sensor networks should be designed considering its energy efficiency. In
other words, it should select an energy-efficient path that consumes less energy for data transmission, and it
should distribute energy consumption rather uniformly among sensor nodes to increase the network
survivability and to prevent network disconnection by energy depletion in some localized area. In this
section, we review two routing algorithms the Directed Diffusion (DD) (Intanagonwiwat et al., 2003) and
Energy Aware Routing (EAR) (Shah et al., 2002) for sensor networks that have been designed considering
energy efficiency.
2.1 Directed diffusion
The directed diffusion (DD) algorithm (Intanagonwiwat et al., 2003) uses three steps in sensing data
transmission. The first step is that the sensing tasks (interests) are disseminated throughout the sensor
network as shown in Fig. 1 (a). The interests can be broadcasted by using some flooding algorithm to the
entire network or to some selected direction. The second step is that the intermediate nodes relaying interests
create and save the information of the previous hop as their routing information called gradients as shown in
Fig. 1 (b). A gradient is utilized as a routing state toward the information collecting (sink)
(a) Interest propagation
(b) Initial gradient setup
(c) Best path reinforcement
Fig. 1 Operations of the directed diffusion routing algorithm.
1300
35th International Conference on Computers and Industrial Engineering
node when sensing data is relayed. The gradient direction is set to the neighboring node from which the
interest is received. Through this process, interests are propagated to the sensing nodes and sensing data
starts flowing toward the originators of interests by using multiple gradient paths. As the third step, the
originator selects and reinforces one or a small number of these paths that are better for sensing data transfer
as shown in Fig. 1 (c). After these three steps, the DD uses the least energy-consuming path for data transfer
among multiple paths.
The DD algorithm described above accomplishes energy-efficient routing by using an energy-shortest path
from sensing nodes to the information collecting nodes. Although it is one of the most energy saving routing
method for the whole sensor network, however, the same path and same nodes are used repeatedly until one
of nodes in the path spends all of its energy and that path is disconnected. Thus, the DD algorithm has a
shortcoming that the lifetime of sensor networks is reduced. Also, the distribution of remaining energy of
nodes becomes uneven, which make the network unreliable.
2.2 Energy aware routing
Another important study for the energy-efficient routing for sensor networks is the Energy Aware Routing
(EAR) (Shah et al., 2002). When sensing information is transferred, it does not select energy-shortest path,
but it chooses next hop node probabilistically by using the information such as energy consumption,
remaining energy level of the nodes, etc. It tries to use neighbor nodes somewhat uniformly, thus it does not
exhaust node energy by not using the same path and nodes repeatedly as in the DD algorithm. Although it
may not utilize the energy-shortest path, it can extend the lifetime of a path compared to the DD. The
simulation results reported in (Shah et al., 2002) showed that the network lifetime until the first node failure
is extended 40% compared to the DD. In the EAR, the probability Pji that a specific node j selects node i as a
next hop is given by
Pji =
1/ C ji
∑ 1/ C
k ∈FT j
(1)
jk
where C ji is the cost between node j and node i and FTj is the forwarding table of node j. We can calculate
the C ji as
C ji = eαji Riβ
(2)
In (2), e ji is the amount of consumed energy for data transfer between node j and node i, Ri is the remaining
energy for node i. We can select constants α and β appropriately. Fig. 2 show the probabilistic selection of
neighboring nodes from node j in the EAR algorithm. The probability for node k becomes higher as C jk
increases.
Since the EAR chooses next hop probabilistically, it does not dry up the remaining energy for a specific
node, thus it can extend the lifetime of the first failed node. However, since it requires heavy computation for
probabilistic calculation in each node and it may transmit data through non-optimal path that may cause
some routing loop, it is possible to consume more energy than that of the DD. Therefore, the EAR
Fig. 2 Probabilistic selection of next hop in the EAR algorithm
could spend more energy, fail to extend the lifetime of sensor networks, except for the time of the first node
failure.
35th International Conference on Computers and Industrial Engineering
1301
3. MAXIMUM REMAINING ENERGY CONSTRAINED DIRECTED DIFFUSION ROUTING
In this section, we propose an energy-efficient sensor network routing algorithm called Maximum
Remaining Energy Constrained Directed Diffusion (MRE-DD) that extends the network lifetime until
network partition due to energy depletion as well as the time for the first node failure. In the MRE-DD, when
it reinforces a path between sensor and gateway, it tries to select a path with high remaining energy level
among multiple paths. Its objectives are to choose an energy-shortest path with high remaining energy, so
that it may use less energy for data transfer and have the remaining energy level relatively uniform for the
entire network, which makes the network more reliable.
In Fig. 3, we show the procedure for the MRE-DD algorithm. The first step is for “interest propagation”, and
the second step is for “gradient setup in the intermediate nodes” as in the DD. In the third step, it selects one
appropriate path with highest remaining energy to be reinforced. Although the gateway node in the DD
reinforces the path that receives sensing data earlier, in the MRE-DD try to select and reinforce a shortestenergy path among paths with highest remaining energy grade. In order to do this, each sensing data is
transferred to the gateway node with an updated remaining energy level of the path. The gateway node
selects first the highest energy path group, and then reinforces one shortest path among them. The sensing
data is transferred along the reinforced path until the remaining energy level of some nodes in the path
becomes lower. At that time, the algorithm tries to reinforce another path with highest remaining energy
level for data transfer, on the contrary the DD persists to use the reinforced path until energy depletion. So,
the MRE-DD always tries to keep using the least energy-consuming path with highest energy, that results in
relatively uniform energy usage among network nodes. As a result, it often changes the routing path to
another one with better remaining energy level. Finally, when there is no path with remaining energy greater
than 0, it decides that the lifetime of sensor networks is exhausted.
Fig. 3 Operation procedure for the MRE-DD routing algorithm
3000 m
Sink
3000 m
Source
(a) Grid network model
Fig. 4 Sensor network models for simulation
1302
(b) Random network model
35th International Conference on Computers and Industrial Engineering
4. PERFORMANCE EVALUATION
In this section, we present some simulation results for the network lifetime and fairness index of the
remaining energy for the three sensor network routing algorithms, the DD, the EAR and our MRE-DD. We
implement simulation program by using sim++ (Cubert et al., 1995) package. As for the common simulation
parameters, we set the initial energies of the sensing nodes 107 mJ, those of the intermediate nodes 105 mJ,
energy consumption for one data transmission 15 mJ, and that for data reception 5 mJ. We let α = 1 and β =
-1 for the EAR simulation. We take two network models for the simulation. One is a grid network model
shown in Fig. 4(a), in which network nodes are uniformly located. The other is a random network model
shown in Fig. 4 (b), in which we locate 500 nodes randomly in a square plane with size 3000m × 3000m. We
don’t include energy consumption for signaling message transfer and internal computation for routing.
In the grid sensor network model consisting of 25 × 25 nodes in Fig. 4(a), each node has maximum 8
neighboring nodes. We apply up to 4 flows with different sensing nodes and information collecting
(gateway) nodes. For the MRE-DD, we divide the remaining energy level into 5 classes with 20 % interval.
In Fig. 5(a), we show the time to the first dead node (FDN) and the network lifetime until no path is
remained for the three routing algorithms. For the FDN time, the MRE-DD algorithm shows the best
(longest) performance among them. For the network lifetime, the EAR shows the worst performance, and the
MRE-DD has the longest lifetime, but very similar result to the DD. Actually, the EAR cannot use the
50000
1
0.95
40000
Fairness Index
Lifetime(Cycle)
0.9
30000
20000
10000
0.85
0.8
DD
EAR
MRE-DD
0.75
0
0.7
1
2
3
4
5000
Flows
DD_FDN
EAR_FDN
MRE-DD_FDN
10000
15000
20000
25000
Time(Cycle)
DD_lifetime
EAR_lifetime
MRE-DD_lifetime
(a) Network lifetime
(b) Fairness index of the remaining energy
Fig. 5 Simulation results for the grid network model
1.00
60000
0.95
50000
Fairness Index
Lifetime(Cycle)
0.90
40000
30000
20000
0.85
DD
EAR
MRE-DD
0.80
0.75
10000
1
2
Flows
DD_FDN
EAR_FDN
MRE-DD_FDN
3
4
0.70
5000
10000
DD_lifetime
EAR_lifetime
MRE-DD_lifetime
15000
20000
25000
Time(Cycle)
(a) Network lifetime
(b) Fairness index of the remaining energy
Fig. 6 Simulation results for the random network model
35th International Conference on Computers and Industrial Engineering
1303
optimal energy path and thus wastes some energy in data transfer because of its probabilistic path selection,
which may result in some routing loops and short network lifetime.
We often use the fairness index to show the property of even distribution of a number of values. It is defined
as follows when there are n values R1, R2,…, Rn (Jain et al., 1984),
 n

fairness index =  ∑ Rk 
 k =1 
2
 n 2
 n∑ Rk 
 k =1 
(3)
We show in Fig. 5(b) the fairness index of the remaining energy for each node. The Figure indicates that the
MRE-DD algorithm keeps the remaining energy of sensor nodes relatively more uniform than the other two
routing algorithms, which makes sensor networks more reliable. Fig. 6 shows the simulation results of the
network lifetime and the fairness index for the random network model. We see from the Figure that the
performance advantage of the MRE-DD increases more for the random network case.
5. CONCLUSION
In this paper, we have proposed a new energy-efficient routing algorithm for sensor networks called
Maximum Remaining Energy constrained Directed Diffusion routing, which selects a least energyconsuming path among multiple paths formed by nodes with highest remaining energy. Simulation results
show that our algorithm extends the network lifetime and enhances the network reliability by maintaining
remaining energy distribution relatively uniform among sensor nodes. For the large scale wireless sensor
networks, the MRE-DD algorithm will become one of the promising energy-efficient routing methods.
REFERENCES
1. Akyildiz, I.F., Su, W., Sankarasubramaniam, Y., & Cayirci, E. (2002). A survey on sensor networks.
IEEE Communications Magazine, 40(8), 102-114.
2. Cubert, R., & Fishwick, P. (1995). Sim++ manual. URL:
http://www.cise.ufl.edu/~fishwick/simpack/simpack.html
3. Intanagonwiwat, C., Govindan, R. & Estrin, D. (2003). Directed diffusion for wireless sensor networks.
IEEE/ACM Trans. Networking, 11(2), 2-16.
4. Jain, R., Chiu, D., & Hawe, W. (1984). A quantitative measure of fairness and discrimination for
resource allocation in shared computer systems. DEC Research Report TR-301.
5. Shah, R.C., & Rabaey, J.M. (2002) Energy aware routing for low energy ad hoc sensor networks. IEEE
WCNC 2002, 350-355.
1304
35th International Conference on Computers and Industrial Engineering