See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/4345062

On energy aware routing in wireless networks

Conference Paper · October 2007

DOI: 10.1109/BROADNETS.2007.4550500 · Source: IEEE Xplore

CITATIONS READS
16 77

3 authors, including:

Sridhar Radhakrishnan Venkatesh Sarangan

University of Oklahoma Tata Consultancy Services Limited
157 PUBLICATIONS   1,548 CITATIONS    92 PUBLICATIONS   1,138 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Book Project View project

All content following this page was uploaded by Sridhar Radhakrishnan on 07 January 2015.

The user has requested enhancement of the downloaded file.

 On Energy Aware Routing in Wireless Networks
Aravind B. Mohanoor, Sridhar Radhakrishnan* Venkatesh Sarangan
School of Computer Science Computer Science Department
University of Oklahoma, Norman, OK 73019 Oklahoma State University
(aravindmc, Sridhar)@ou.edu Stillwater, OK 74078
* Corresponding Author. [email protected]

Abstract—Online energy aware routing in wireless networks energy required by node u (resp. v) to communicate to node v
is the problem of finding energy efficient routes that (resp. u). The residual energy of a path is defined as the
maximize the network lifetime without the knowledge of minimum energy level of any node in the path (metric 1).
future message flows. To maximize network lifetime, the The max-min routing paradigm suggested in the literature [1,
paths for message flows are chosen in such a way that the 4, 7] proposes to find the path where the residual energy is the
total energy consumed along the path is minimized while maximum and forwards packets through this path termed as
avoiding energy depleted nodes. Finding paths which the maximum residual energy path. The energy consumed
consume minimum energy and finding paths which do not along a path (or simply the energy of a path) is the sum of the
use energy depleted nodes lead to conflicting objectives. In weights on the edges along the path (metric 2).
this paper, we propose a two-phased energy-aware routing Notable routing strategies which utilize the concept of the
strategy that balances these two conflicting objectives by residual energy (either directly or indirectly) proposed so far
transforming the routing problem into a multi-metric include MMBCR [7], MRPC [1] and max-min zPmin [4]. These
widest path problem. We find that the proposed approach research works also caution that merely using the residual
outperforms the best known algorithm in literature. We energy strategy may lead to higher energy consumption in the
also demonstrate a simple but insightful relationship network, since the energy consumed along the data forwarding
between the total energy required along a path and the path is not taken into consideration. They suggest that a good
minimum remaining energy of a node along the path. energy-aware routing technique should balance two different
goals: choosing a path with maximal residual energy and
Keywords - Wireless networks, Energy aware routing, choosing a path with minimal energy consumption. We note
Combinatorial algorithm, QoS metrics. that the residual energy along a path is a concave metric1,
whereas the energy consumed along a path is an additive
I. INTRODUCTION metric.
Energy management in wireless networks is of paramount The approach in max-min zPmin [4] attempts to balance
importance due to the limited energy availability in the metric 1 and metric 2 by calculating a path based on the
wireless devices. Since wireless communication consumes a residual energy levels, but then rejecting any path whose total
significant amount of energy, it is important to minimize the energy is more than a factor z times the minimum energy path.
energy costs for communication as much as possible by We note that the quality of its solution depends on an
practicing energy aware routing strategies. Such routing empirically generated parameter z, and does not always
strategies can increase the network lifetime. In this paper, we provide an optimal solution. The MRPC algorithm, which is a
focus on developing routing strategies for a multiple hop generalization of the MMBCR algorithm, uses the residual
wireless network which has significant energy constraints, like ‘packet capacity’ instead of the residual energy for
a multihop mesh network where all the nodes are powered by optimization. As we will illustrate later, even the MRPC
battery or other external power sources such as solar energy. algorithm can fail to maximize a network’s lifetime. Chang
One way to quantify network lifetime is through the number of and Tassiulas [2] combine metrics 1 and 2 into a single metric
packets that can be transferred in the network before the and run Dijkstra’s on this new metric. While it is a good
source and destination get disconnected from each other. A heuristic, this method does not actually optimize either metric.
suitable energy-aware routing strategy for wireless networks is Park and Sahni [5] present the Online Maximum Lifetime
to use those wireless nodes with high energy levels and avoid (OML) heuristic, which is an enhancement of the CMAX
those with low energy levels. algorithm presented by Kar et al [3]. OML uses a two-step
Wireless networks for energy-aware routing techniques approach where they remove those edges with low energy
are modeled as graphs wherein, a vertex is a wireless device from the graph, and then run Dijkstra’s on a graph where the
and an edge between two vertices indicate that they are in edge weights have been modified in such a way that the paths
direct communication range of each other. The weight on a found usually use nodes with high energy levels and edges
vertex indicates the energy level available at that sensor node
and the weight on an edge (u, v) represents the amount of 1
For definitions of concave and additive metrics, see Wang and
Crowcroft[8].
This work was supported by the US FHWA through grants SAFTEA-LU
1934 and SAFETEA-LU 1702.
with low energy costs. They report the best performance in the cost of reception explicitly to avoid clutter in our discussions.
current literature. Such a cost can be easily incorporated in our proposed work.
In this paper, we present a polynomial time combinatorial Given a source s, a destination t, and a single packet to be
technique which can provide a good balance between metrics routed, we can define two problems formally:
1 and 2. The proposed technique first maximizes the concave a. Minimum energy path problem: Find a path P(s, t) with
metric (the residual energy of a path) and then minimizes the minimum e(P(s, t)).
additive metric (energy consumed along a path). We b. Maximum residual energy path problem: Find a path P(s,
qualitatively justify why this order of optimization – concave t) with maximum r(P(s, t)).
first, additive second – is better than the other possible order –
additive first, concave second. On an illustrative topology, we Let G0 be set to the initial network G. Assume that P0(s,
show that the proposed two-phased routing prolongs the t) is a path in G0. Now after routing a single packet along the
network lifetime better than other routing techniques. Our path P0(s, t) and following the decrease operation we obtain a
simulation studies also show that the performance of the new network G1. In the network G1 the edge weights are the
proposed technique is superior than that of the best known same as in G0 but the nodes energy levels are different. If a
routing approach proposed in the literature (Park and Sahni node u’s energy level becomes 0 after the decrease operation
[5]). An additional incentive for using the proposed approach the node u and the edges (u, v) ∈ E are removed from the
is that it allows us to address (i.e. it can be easily combined network. For the second packet we can again find a path P1(s,
with) other QoS metrics such as delay, which can be beneficial t) in G1 and the process continues until there exists no path
for resource constrained networks. between s and t in some network Gk. That is, we can send at
The paper is organized as follows. Section II provides the most k packets from s to t before the network is disconnected.
definition of the network lifetime problem and the solution The goal of the network lifetime problem with respect to a
proposed. Section III discusses the performance of our source s and destination t is to find paths P0(s, t), P1(s, t),…,
approach and other approaches on a benchmark topology. In Pk-1(s, t), such that the value of k is maximized.
Section IV, we deduce a relationship between the total energy Our solution to the network lifetime problem as follows.
consumed along a path and the residual energy of edges on the We modify the graph G into an energy graph EG = (V, E΄) as
graph, which provides additional insight into the nature of the follows. We leave the vertices intact but replace each single
problem. In Section V we discuss how other QoS metrics such undirected edge in G with two directed edges. The weight of a
as delay and error rate can be used in the second phase of our directional edge in EG is made equal to the difference between
solution. We discuss the rationale behind the chosen order of the originating node’s energy level and the transmission cost
optimizing the two metrics (in our problem the energy along the edge. This is also the residual energy of a node as
consumed along a path and the residual energy at the nodes) in defined in Li et al [4]. In Figure 1 (a) we have shown an
Section VI. Section VII discusses our simulation setup and example wireless network and in Figure 1 (b), the
results. Section VIII concludes the discussion. corresponding energy graph.
II. PROBLEM DEFINITION AND SOLUTION

Let G = (V, E) represent a wireless network with nodes V and

edges E. Let w(u) (also referred to as the residual energy of a
node), u ∈ V represent the available energy at node u. Let c
(u, v), (u, v) ∈ E be the energy required to transmit a packet
from node u to node v. We assume that c (u, v) = c (v, u), for
all (u, v) ∈ E.
Let P(v0, vk) = v0, v1, …, vk, be a path in G. The energy of
the path P(v0, vk) denoted e(P(v0, vk)) is Figure 1: A graph showing energy levels at nodes and energy required to
k −1
transmit at each edge. Figure 1 (b) shows the corresponding energy graph.
e(P(v0, vk)) = ∑ c (v
i=0
i , v i +1 )
Given a source node s and a destination node t, we then
The residual energy of a path P(v0, vk) denoted r(P(v0, vk)) run our two-phased routing algorithm on this energy graph EG
is to find a suitable path between s and t. In Phase I, we apply a
r(P(v0, vk)) = mini(vi), 0 ≤ i < k. variant of the Dijkstra’s algorithm shown in Figure 2 to find a
path with the maximum residual energy. Phase I of our
When a packet is sent along P(v0, vk), we need to perform solution will return a path whose residual energy will be the
the following energy decrease operation on each node along maximum in the network. Let the path returned by phase I
the path except on the node vk: w(vi) = w(vi) – c(vi, vi+1), 0 ≤ i < have a residual energy of B. It is to be noted that there could
k. That is, after the packet is sent by a node, the energy level be many paths in the network between s and t with a residual
of the node is decremented by the amount of energy required energy of B.
to send the packet. In our model, we have not included the
 In Phase II, we choose from the set of all paths with a before the network gets disconnected [4], where n is the
residual energy of B, a path which has the lowest energy number of nodes in the network. The first message will take
consumption. If E” is the set of edges whose residual energy is the path (1, n-2, n). The second message will take the path (1,
less than B, this can be accomplished by first pruning those n-3, n), and so on until the (n-4)th message will take the path
edges from EG and using Dijkstra’s algorithm to find the least (1, 3, n).
energy cost path on EG\E”. If there are many such paths, we
arbitrarily choose one among them. It can be noted that this
algorithm can also handle the energy cost of reception if such
information were available. We would need to modify step 1 5 n-4
of the RELAX procedure to add the energy cost of reception. 4 n-3
It must be noted that the pruning is temporary, in other words, n-2
the edges are restored before the next route computation. 3
Each time a path is computed, we will invoke Dijkstra’s 2 n-1
algorithm twice in sequence. Hence it can be seen that our
algorithm has a complexity equal to k times the complexity of 1 n
Dijkstra’s algorithm, where k is the number of packets
transmitted. Figure 3: How do we apply max-min on this graph?
Banerjee and Misra[1] define the residual packet capacity
MaxCapacity(EG)
as the number of packets which can be transmitted by a node
// s – source node at its current energy level. Directly applying MRPC on this
// weight(u,v) = capacity of edge (u, v) in graph is identical to max-min, and may end up sending at most
graph EG 20 packets in the worst case. But they also provide another
// width(u) = weight function for a node in algorithm called the CMRPC, where they define a parameter γ
graph EG which represents the threshold energy level of the critical
1 width[s] = 0 nodes. When nodes reach this energy level, they shift from
width[v] = weight(s, v) if v є Adj[s] minimum energy routing to maximum residual capacity
width[v] = 0 for all other nodes routing.
2 S
s Suppose the parameter γ is set at 0.5 (representing 50% of
3 Q
V[EG] - s node’s energy). The authors of the MRPC algorithm do not
4 while Q ≠ Φ mention how they make the choice of minimum energy paths
5 do u
EXTRACT-BEST(Q) when there is more than one. Suppose we use the sequence
6 S
S U {u}
{(1, n-2, n), (1, n-3, n),..., (1, 4, n), (1, 3, n)}, we can send 5(n-
7 for each vertex v є Adj[u]
8 if v does not belong to S 4) messages using the straight line edges after which each
9 do RELAX(u, v, weight(u, v)) forwarding node (except 2 and n-1) will be left with energy 10
+ є. Since we reach the threshold value, we start using the
EXTRACT-BEST
maximum residual capacity paths. Using the maximum
1 return u є Q where u has maximum width residual capacity paths, only 10 more messages could be sent,
RELAX (u, v, weight(u,v)) for a total of 10 + 5(n-4) messages. In fact, we may not be able
1 if width[v] < min(width[u], weight(u, v)) to send more than 10γ(n-4) + 20(1- γ) messages. The
2 width[v] = min(width[u], weight(u,v)) maximum value of this quantity happens when γ = 1 (which
3 Pred(v)
u means packet capacity is never used), in which case we can
still send only 10(n-4) messages.
Figure 2: Dijkstra’s shortest path algorithm modified to compute the
maximum residual energy path. On the other hand, if one were to use the two-phased
approach we have proposed, the following paths will be used
for routing. We will repeat the sequence {(1, n-2, n-1, n), (1,
III. PERFORMANCE OF THE PROPOSED APPROACH n-2, n-1, n), (1, n-3, n), (1, n-4, n), (1, n-5, n)…, (1, 4, n), (1,
We compare the performance of the proposed two-phased 3, n)} before source and destination get disconnected.
approach with other existing approaches on an illustrative Consequently, it is easy to see that a total of 10(n-3) messages
topology shown in Figure 3 (from Li et al [4]). can be sent before the nodes run out of energy.
In the network shown in Figure 3, each node (other than IV. RELATIONSHIP BETWEEN THE TOTAL ENERGY AND THE
source 1 and destination n) has energy 20 + є. The weight of RESIDUAL ENERGY OF PATHS
each edge (along the semi-circle) is set to 1, but the weight of
each straight edge is set to 2. The energy of the source is While the maximum residual energy path computation
infinite. Li et al. [4] state that using a single-pass max-min identifies the bottleneck edge and allows us to discover the
approach, it is possible that only twenty messages can be sent maximum residual energy subgraph, we can also define what
before the network gets disconnected. The authors then state are called residual energy constrained subgraphs.
that using the straight line edges 10(n-4) messages can be sent
Definition: Let EG(w) represent the subgraph constructed those edges in EG which have residual energies less than w.
from the original residual energy graph EG, by pruning all
Destination t Destination t Destination t
2 40 40 40

4
4

70 70 70
90

2
90 90
75 4 45 75 4 45 45
5 5 5 75 5

5
5

5
5 5
100 100 100 100
100 100

3
5

3
5
3
5 5 5 5
80 4 110 5
80 4 110 5 150 80 4 110 5
150 50 50 150 50

5
5

Source s 2 Source s 2 Source s 2

200 100 60 200 100 60 200 100 60

(a) (b) (c)

Figure 4 Total energy of minimum energy path vs the constraint residual energy (width). (a) Original graph – this is also EG(38), as 38 is the lowest residual
energy edge in the residual energy graph. The minimum energy path has been marked with arrows and has energy = 10 units. (b) EG(66) and minimum energy
path = 16 units (c) EG(85) and minimum energy path = 20 units.

For example, for the original graph EG shown in Figure 4(a), in the figure has an initial residual energy of 19, and a total
we can observe that the graphs in Figures 4(b) and 4(c) show energy of (n-1) units. The path (1, n-2, n-1, n) also has a
examples of residual energy-constrained subgraphs EG(66) residual energy of 19 but consumes only 4 units of energy for
and EG(85). The bidirectional edges have not been shown for transmission. When we perform such route computations
the sake of clarity. The edges which are removed from the repeatedly, it can lead to substantial savings in energy.
original graphs are such that their residual energies fall below
the threshold limit along both directions. Using this
construction, we can see that the total energy required for the
minimum energy path on a subgraph increases as the residual
energy constraint of the subgraph increases. If we were to
repeatedly compute the minimum energy path for all the
possible residual energy values of a graph (a graph can have at
most O(m) such discrete values, where m is the number of
edges) we would obtain the non-decreasing graph similar to
the one shown in Figure 5. This can be easily proved.

Lemma 1:

Let Emin(w) represent the minimum energy of a path in EG(w).

Given the residual energies of the graph EG in increasing
order as (w1, w2, …, wm), i.e. w1 ≤ w2 ≤ … ≤ wm, then Emin(w1)
≤ Emin(w2) ≤ … ≤ Emin(wk).
Figure 5 Total energy of path Vs residual energy
Proof:
Let G1 = EG(w1) and G2 = EG(w2) where w1 ≤ w2. Any edge
V. USING OTHER METRICS FOR THE SECOND PHASE
in G2 also exists in G1, by definition. Thus the minimum
energy path in G1 cannot have higher energy than the Using a similar argument as in Section IV, we can generalize
minimum energy path in G2. In other words, Emin(w1) ≤ to say that any path-based metric (such as delay, error rate,
Emin(w2). By induction, we get the result. jitter) will have a similar graph. This has two implications for
QoS metrics on wireless networks.
We gain some useful insight from the relationship between the If our priority is to perform routing according to QoS
residual energy along a path and the minimum energy path constraints, then the problem of also ensuring that we choose
possible for such a residual energy. For one, knowing that we energy aware paths is easily handled by using a maximum
have a lower energy path, we could avoid routing the packet residual energy (widest) path approach. Since the maximum
along a path with identical residual energy but which residual energy path computation yields a subgraph whose
consumes a much higher energy for the entire path. This is edges already have high residual energy, the other QoS
precisely the reason why a maximum residual energy path still constraints could be computed on the resulting subgraph. Thus
turns out to be a poor choice in the benchmark topology an existing polynomial solution for the QoS constraint
considered in Figure 3. We can note that the path along the arc problem will not be worsened by adding energy awareness.
Contrast this with the issue of choosing the absolute minimum VII. EXPERIMENTAL RESULTS
energy path which would make the original problem (e.g.
minimum delay) NP-Complete, since the problem of A. Simulation Settings
computing paths which simultaneously optimize two additive In our experimental study, we compare the performance of the
metrics is known to be NP-Complete [8]. proposed shortest-widest path (SWP) and on-line maximum
A second benefit comes from the use of the property life-time (OML) heuristic proposed by Park and Sahni [5].
mentioned in the earlier section. Figure 5 provides a clue as to Though several works have been proposed in the literature,
the maximum possible residual energy we could use for reference [5] has established the superiority of OML over
finding a path with a given constraint value for the second other existing works. Therefore, we compare SWP with OML
stage metric. For example, we can plot the minimum possible alone.
value of the additive constraint (e.g. delay) on the y-axis of
Figure 5 as a function of the residual energy. This means that Topologies Used: We use a topology which is identical to
given an upper bound on the additive constraint, we can find that used for OML. We randomly populate a 25×25 grid with
the highest residual energy for which it is feasible. Repeatedly 50 nodes. We add edges to the network if the nodes are within
utilizing this idea might actually ensure that the paths used are each others’ transmission range, which is decided by the
always having high residual energy while also satisfying the transmission radius rT. The energy cost of transmitting a single
additive constraint. In general this should give us better energy packet is calculated as 0.001×d3 where d is the Euclidean
awareness. In contrast, other QoS metrics cannot be combined distance between the nodes. These settings are identical to the
in the Online Maximum Lifetime heuristic (which has the best ones used for OML.
performance in current literature) since it modifies the graph
in such a way that the original information about the graph Session failures tolerated: In the literature, network life-time
weights are lost. is traditionally measured using the number of packets that can
be transferred in the network until a session failure occurs. In
VI. DISCUSSIONS ON THE ORDER OF OPTIMIZATION other words, the life-time is measured until the time when two
We could also consider the other possible order of nodes get disconnected. In practice however, certain classes of
optimization for this problem, namely the additive metric first networks such as sensor networks, prove to be useful and
(minimum energy) and the concave metric second (widest). continue with their functioning even though a single node pair
This is also called the widest shortest path. However, since the gets disconnected. Therefore, we generalize the definition of
width is a concave metric, different paths with the same width network life-time as “the number of packets that can be
can be constructed by simply removing the bottleneck nodes. transferred in the network until s session failures are occur”,
However, the energy cost of a path is an additive metric and where s is a parameter to be set by the network manager. The
depends on all the edges in the path. If we do remove some value for s can be set based on the number of session failures
edge(s), we face a much more difficult constraint in choosing that can be tolerated by the application. We study the
other paths with the same energy cost. This means there can be performance of the routing schemes under different values of
several widest paths in a network (that may include almost all s.
the nodes in the network) among which we could choose the
one with the least energy cost. Over time, this would mean Session Length: Another parameter that we generalize in our
that the energy burden of forwarding packets is shared among simulation study is session length. Earlier works assume that
many more packets, which is a desirable situation. However, a single packet is transmitted in a session between a given
there may be only a few minimum energy paths, and hence node-pair. However, in reality, it is highly likely that multiple
choosing the widest (maximum residual energy) among them packets will be exchanged in a session between two nodes.
may not utilize all the nodes in the network very effectively. Therefore, in our experiments, we assume that k packets are
The order in which we wish to optimize the metrics plays transmitted in a session between a given node-pair. We vary
an important role in determining the existence of a single-pass the value of k and observe the performance of the different
algorithm for our problem. Sobrinho [6] shows that there routing schemes. As in other works in the literature, we
exists a single pass algorithm for determining a multi-metric calculate the route afresh for each packet transmission.
path only when the metrics governing the path are isotonic.
Sobrinho also proves that the shortest-widest path is non- Traffic pattern used: We conduct our experiments assuming
isotonic. Since minimal energy path is the ‘shortest’ path in a any-to-any communication model, i.e. source-destination
terms of energy consumption, and the maximum residual pairs are selected at random and packets are transmitted
energy path is the ‘widest’, a single-pass algorithm for between them.
discovering the minimum energy-maximum residual energy
path cannot be constructed. Unless otherwise mentioned, we use the following default
values: there are 50 nodes placed randomly on a 25×25 grid,
the transmission radius is set to 8, the session length is set to 1
(single packet), the number of session failures tolerated is set
to 1 (first failure to find a route), the initial energy level for
each node is set to 30 and any-to-any communication pattern that when we allow the number of session failures to be
is assumed. greater than one, we benefit from an improved lifetime.
Figures 8 and 9 show the performance of SWP and OML
B. Effect of transmission radius
as the number of session failures tolerated is increased. From
We vary the transmission radius from 7 to 30 and evaluate the the figures we again observe that, SWP outperforms OML
performance of SWP and OML algorithms (Figures 6 and 7). with respect to both network life-time and the average residual
We can notice that SWP equals or betters the lifetime of OML energy in the nodes. We use an initial energy level of 50, to
for all values of transmission radii. However, it must be noted better highlight the performance difference between the two
that a higher value of transmission radius comes with an algorithms. The lifetime is measured by calculating the
associated energy cost for the nodes, which will then lose average over ten different runs at the specific session failure
energy much more rapidly when not transmitting data. Figure point. In other words, we find the number of packets
7 shows the remaining energy as a fraction of the total initial transmitted before the sth session failure over multiple runs and
energy in the network. We notice that SWP can actually send use their average.
more packets using lesser energy.

Figure 8 Lifetime vs session failures

Figure 6 Lifetime vs Transmission radius

Figure 9 Energy remaining vs session failures

Figure 7 Energy Remaining vs transmission radius D. Effect of session length

C. Effect of session failures tolerated When we need to transmit more than a single packet from a
source to a destination, the length of the session is (perhaps
The number of session failures tolerated can be a number
much) greater than 1. So we evaluate the performance of the
s(>1), as set by the network manager. We vary s from 1 to 10
two algorithms when the number of packets transmitted per
and analyze the performance of the two algorithms. We notice
session, i.e. the session length, is varied to be more than 1. We
compute fresh routes for each packet transmitted for both the
algorithms. We compare the performance of SWP and OML
for session lengths of 1, 10, 25 and 100 in Figures 10 and 11.
Again, we find that the performance of the SWP algorithm is
better than that of OML.

E. Effect of node density

Figures 12 and 13 show the relative performance of the two

algorithms when the number of nodes in the network is
increased. We vary the number of nodes from 40 to 100 (in
increments of 10) and find the lifetime which is achieved. We
notice the same trends – SWP consistently outperforms OML
as the node density increases.

Figure 12 Lifetime vs Node density

Figure 10 Lifetime vs session length

Figure 13 Energy remaining vs Node density

F. Remarks

The number of path calculations is the same as the number of

packets which need to be sent. When we need to send a lot of
packets, we could avoid this situation by constructing the
energy graph based on k messages, where k is some
predetermined constant.
Gathering information about the energy levels of all
nodes has an associated energy cost (although all the previous
energy aware algorithms suffer from the same limitation). The
effects of this problem can also be partially offset by
calculating the energy graph for k messages at a time.

Figure 11 Energy remaining vs Session length VIII. CONCLUSIONS

The online energy aware routing problem is
conducive to a solution using the widest path approach. We
have shown through simulations that the proposed approach
easily outperforms the best known solution in the literature.
An additional incentive for using the proposed approach is that
it allows us to address other QoS metrics such as delay, which
can be beneficial for resource constrained networks. This is
due to the combinatorial nature of the solution which makes it [4] Li, Q., Aslam, J., and Rus, D., “Online power-aware routing in wireless
Ad-hoc networks,” Proceedings of the 7th Annual International
possible to enumerate and discover all favorable route choices. Conference on Mobile Computing and Networking (MobiCom), pp.97-
107, 2001.
[5] Park, J., and Sahni, S., “An online heuristic for maximum lifetime
REFERENCES routing in wireless sensor networks,”, IEEE Transactions on Computers,
v.55, no.8, pp.1048-1056, 2006.
[1] Banerjee, S. and Misra, A., “Energy Efficient Reliable Communication [6] Sobrinho, J.L., “Algebra and algorithms for QoS path computation and
for Multi-hop Wireless Networks,” to appear in the Journal of Wireless hop-by-hop routing in the Internet,” IEEE Conference on Computer
Networks (WINET). Communications (INFOCOM), v.2, pp. 727-735, 2001.
[2] Chang, J. and Tassiulas, L., “Maximum lifetime routing in wireless [7] Toh, C.-K., Cobb, H. and Scott, D.A., “Performance evaluation of
sensor networks,” IEEE/ACM Transactions on Networking, v.12, no.4, battery-life-aware routing schemes for wireless ad hoc networks,” IEEE
pp.609-619, 2004. International Conference on Communications (ICC), v.9, pp.2824-2829,
2001.
[3] Kar, K., Kodialam, M., Lakshman, T.V., and Tassiulas, L., “Routing for
Network Capacity Maximization in Energy-Constrained Ad-Hoc [8] Wang, Z. and Crowcroft, J., “Quality-of-service routing for supporting
Networks,” International Conference on Computer Communications multimedia applications,” IEEE Journal on Selected Areas in
(INFOCOM), pp.673-681, 2003. Communications, v.14, no.7, pp. 1228-1234, 1996.

View publication stats