Professional Documents
Culture Documents
8, AUGUST 2006
Abstract—We show that the problem of routing messages in a wireless sensor network so as to maximize network lifetime is NP-hard.
In our model, the online model, each message has to be routed without knowledge of future route requests. We also develop an online
heuristic to maximize network lifetime. Our heuristic, which performs two shortest path computations to route each message, is
superior to previously published heuristics for lifetime maximization—our heuristic results in greater lifetime and its performance is less
sensitive to the selection of heuristic parameters. Additionally, our heuristic is superior on the capacity metric.
1 INTRODUCTION
that routing requests r1 ; ; rj are successfully routed. An maximum node cost. Of the proposed five metrics, only the
online routing algorithm routes ri without knowledge of any first (minimum-energy path) and fourth (minimize node
rj , j > i. An offline routing algorithm, on the other hand, cost) have been implemented by Singh et al. [17]. They raise
determines the routes for each ri with full knowledge of all concerns about the difficulty of implementing the remain-
succeeding routing requests. Clearly, the lifetime obtainable ing three in a routing protocol.
by the best offline algorithm is at least as much as that Toh et al. [20] propose the MMBCR (min-max battery
obtainable by any online algorithm. The competitive ratio of cost routing) and CMMBCR (conditional MMBCR) online
an online routing algorithm is the maximum value of the algorithms to select a source-to-destination path. The
lifetime obtained using the best offline algorithm divided MMBCR algorithm selects a path P for which the minimum
by the lifetime obtained by the online algorithm; the of the residual energies (i.e., energy remaining following a
maximum is taken over all routing sequences R. route) of the sensors on P is maximum. Recognizing that to
maximize lifetime we need to achieve some balance
between the energy consumed by a route and the minimum
3 RELATED WORK residual energy at the nodes along the chosen route, Toh
Several authors have developed energy-efficient algorithms et al. [20] also propose a conditional MMBCR algorithm,
for point-to-point communication [4], [5], [6], [10], [11], [13], CMMBCR. In CMMBCR, we look for a minimum energy
[17], [18], [19], [20], [22]. The overall objective of these source-to-destination path in which no sensor has residual
algorithms is to either maximize the lifetime (time at which energy below a threshold . If there is no source-to-
a communication fails first) or the capacity of the network destination path with this property, then the MMBCR path
(number of successful communications over some fixed is used.
period of time in which the network continues to service In the MRPC (maximum residual packet capacity)
route requests even after a communication failure). lifetime-maximization heuristic of Misra and Banerjee [15],
Aslam et al. [3] have shown that there is no online the capacity, cðu; vÞ of edge ðu; vÞ is defined to be
routing algorithm with oðnÞ competitive ratio for the ceðuÞ=wðu; vÞ. Note that cðu; vÞ is the number of unit-length
lifetime maximization problem. Singh et al. [17] propose messages that may be transmitted along ðu; vÞ before node u
five metrics that may be used in the selection of the routing runs out of energy. The lifetime of path P , lifeðP Þ, is
path for energy efficient routing. The first of these is to use a defined to be the the minimum edge capacity on the path. In
minimum-energy path (i.e., a path in G for which the sum MRPC, routing is done along a path P with maximum
of the edge weights is minimum) from s to t. Such a path lifetime. Fig. 1 gives the MRPC algorithm. A decentralized
may be computed using Dijkstra’s shortest path algorithm implementation, as well as a conditional MRPC, CMRPC,
[16]. Although this metric tends to minimize the total (or are also described in [15]. CMRCP attempts to route on a
average) energy consumed over a sequence of routes, it minimum-energy path P with lifeðP Þ , where is a
doesn’t focus on the primary objective of maximizing specified threshold value. If there is no source-to-destina-
lifetime. This is because using a minimum-energy path for tion path P with lifeðP Þ , the MRPC path is used.
the current route request may prevent the successful Aslam et al. [3] propose the max-min zPmin -path
routing of future messages. algorithm to select routes that attempt to make this balance.
The remaining four metrics proposed in [17] are This algorithm selects a path that uses at most z Pmin
maximize time to network partition, minimize variance in energy, where z is a parameter of the algorithm and Pmin is
node energy levels, minimize the node cost of each the energy required by the minimum-energy path. The
transmission (the cost of a node is some function of the selected path maximizes the minimum residual energy
amount of energy used so far by that node), and minimize fraction (energy remaining after route/initial energy) for
Authorized licensed use limited to: IEEE Xplore. Downloaded on November 25, 2008 at 23:33 from IEEE Xplore. Restrictions apply.
1050 IEEE TRANSACTIONS ON COMPUTERS, VOL. 55, NO. 8, AUGUST 2006
nodes on the route path. Notice that the possible values for routing request is rejected (admission control); otherwise,
the residual energy fraction of node u may be obtained by the route is done using path P . Fig. 3 gives the algorithm.
computing ðceðuÞ wðu; vÞÞ=ieðuÞ, where ceðuÞ is the (cur- The CMAX algorithm of Fig. 3 has a complexity
rent) energy at node u just before the route. This advantage over the max-min zPmin algorithm of Fig. 2.
computation is done for all vertices v adjacent to u. Hence, The former does only one shortest-path computation per
the union, L, of these values taken over all u gives the routing request, while the latter does Oðlog nÞ, where n is
possible values for the minimum residual-energy fraction the number of sensor nodes. Although admission control is
along any path. necessary to establish the logarithmic competitive-ratio
Fig. 2 gives the max-min zPmin algorithm. bound for CMAX, we may use CMAX without admission
Several adaptations to the basic max-min zPmin algo- control (i.e., route every request that is feasible) by setting
rithm, including a distributed version, are described in [3]. ¼ 1. Experimental results reported in [12] suggest that
Kar et al. [12] develop an online capacity-competitive (the CMAX with no admission control outperforms max-min
capacity is the number of messages routed over some time zPmin on both the lifetime and capacity metrics.
period) algorithm, CMAX (capacity maximization), with For the performance evaluation of CMAX, Kar et al. [12]
logarithmic competitive ratio. On the surface, this would introduce another route selection algorithm, which they call
appear to violate the ðnÞ bound of [3]. However, to achieve max min. We refer to this algorithm as maxRE here. In
this logarithmic competitive ratio, the algorithm CMAX maxRE, the routing path is selected so as to maximize the
does admission control. That is, it rejects some routes that minimum residual energy fraction of sensors on the path.
are possible. The bound of [3] applies only for online This algorithm is equivalent to max-min zPmin with z ¼ 1.
algorithms that perform every route that is possible. Chang and Tassiulas [4], [5] develop a linear-program-
Let ðuÞ ¼ 1 ceðuÞ=ieðuÞ be the fraction of u’s initial ming formulation for lifetime maximization. This formula-
energy that has been used so far. Let and be two tion requires knowledge of the rate at which each node
constants. In the CMAX algorithm, the weight of every edge generates messages. Wu et al. [21] propose routing based on
ðu; vÞ is changed from wðu; vÞ to wðu; vÞ ððuÞ 1Þ. The connected dominating sets to maximize network lifetime.
shortest source-to-destination path P in the resulting graph Stojmenovic and Lin [19] and Melodia et al. [14] develop
is determined. If the length of this path is more than , the localized algorithms to maximize lifetime and Heinzelman
Authorized licensed use limited to: IEEE Xplore. Downloaded on November 25, 2008 at 23:33 from IEEE Xplore. Restrictions apply.
PARK AND SAHNI: AN ONLINE HEURISTIC FOR MAXIMUM LIFETIME ROUTING IN WIRELESS SENSOR NETWORKS 1051
Authorized licensed use limited to: IEEE Xplore. Downloaded on November 25, 2008 at 23:33 from IEEE Xplore. Restrictions apply.
1052 IEEE TRANSACTIONS ON COMPUTERS, VOL. 55, NO. 8, AUGUST 2006
small relative to minRE are assigned a high weight because current energy). There is also a sensor or node responsible
of the term. Thus, the weighting function discourages the for global routing among sensor hosts. This global routing
use of edges whose use on a routing path is likely to result sensor treats each zone as a single network vertex and uses
in the failure of a future route. an estimate of the current energy in the zone. Kar et al. [12]
Fig. 5 gives our OML algorithm to select a path for have pointed out the same zone-based strategy may be
request ri . This algorithm may be used as a heuristic for employed to derive a zone-based CMAX algorithm.
lifetime maximization. The resulting heuristic is the OML Aslam et al. [3] also describe a strategy to arrive at a
heuristic. Note that our OML algorithm performs two distributed version of their min-max zPmin algorithm. The
shortest-path computations for each route request. By strategy employed in [3] may also be employed to arrive at
contrast, the CMAX algorithm of [12] does only one a distributed version of OML. Kar et al. [12] use a limited
shortest-path computation per route request, the max-min flooding approach to develop a distributed version of their
zPmin algorithm of [3] does Oðlog nÞ shortest-path computa- CMAX algorithm. In this limited flooding approach, each
tions per route request, and the MRPC algorithm of [15] node computes the shortest path (using edge weights as
does Oðlog nÞ depth or breadth-first searches per route described for CMAX) to the destination and forwards the
request. Note that a shortest path can be computed in message to the next hop on this shortest path. With some
Oðn log n þ eÞ time, where n and e, respectively, are the periodicity, each sensor broadcasts its current energy level
number of vertices and edges in the graph. Also, a depth or to sensors within some distant r from it. Hence, when
breadth-first search requires Oðn þ eÞ time. So, the complex- computing a shortest path to the destination, a sensor has
ity of OML and CMAX is Oðn log n þ eÞ per route request, the global topology, a reasonably recent measure of the
while that of max-min zPmin is Oðn log2 n þ e log nÞ and that energy of nearby sensors and the initial (or much less
of MRPC is Oðn log n þ e log nÞ. recent) energy of distant sensors. The distributed algorithm
Note that, although both the CMAX and OML algorithms is augmented with loop avoidance defenses. Clearly, this
have a ðuÞ 1 term in the edge weighting function, the two limited flooding strategy may be employed to arrive at a
algorithms use a different ðuÞ function. In the case of CMAX, distributed OML algorithm as well.
ðuÞ ¼ 1 ceðuÞ=ieðuÞ is the fraction of u’s initial energy that
has been used so far. So, CMAX discourages the use as relays
7 EVALUATION
of sensors that have depleted a large fraction of their initial
energy (even though such sensors may have a large amount of Our experimental setup is similar to that employed by Kar
energy remaining). In OML, ðuÞ ¼ minRE=ceðuÞ. Hence, et al. [12] and Aslam et al. [3]. We randomly populate either
OML discourages the use as relays of sensors whose current a 10 10 or a 25 25 grid with sensors. The sensors are
energy is much less than minRE. deployed at randomly selected grid points. The energy
required by a single-hop transmission between two sensors
is 0:001 d3 , where d is the Euclidean distance between the
6 DISTRIBUTED OML two sensors.
OML, as presented in Section 5, is a centralized algorithm Simulations reported in the literature already have
that requires a computational node with complete informa- established the superiority of CMAX over max-min zPmin ,
tion regarding the topology of the sensor network and ME, and maxRE [12], as well as the superiority of MRPC
current energy levels of all sensors. In real sensor networks, and CMRCP over MMBCR, CMMBCR, and ME [15]. Misra
these requirements often are impractical. We may develop a and Banerjee [15] report that, for a sensor transmission
hierarchical zone-based version of OML using the same radius rT (i.e., the maximum distance that a sensor can
strategy as used in [3] to arrive at a zone-based max-min transmit in a single hop) of 2.9 units (a unit being the
zPmin algorithm. Such a zone-based algorithm divides the distance between two adjacent grid points) and larger,
sensors into zones, each zone containing a set of geogra- MRPC performs slightly better than does CMRPC, whereas,
phically close sensors. Each zone has a host sensor that does for a smaller transmission radius such as 1.5 units, CMRPC
local routing (the host sensor in a zone may change with has a better performance. When sensors are deployed in
time and may, for example, be the sensor with maximum hostile and otherwise hard to reach environments, it is
Authorized licensed use limited to: IEEE Xplore. Downloaded on November 25, 2008 at 23:33 from IEEE Xplore. Restrictions apply.
PARK AND SAHNI: AN ONLINE HEURISTIC FOR MAXIMUM LIFETIME ROUTING IN WIRELESS SENSOR NETWORKS 1053
Fig. 6. Average lifetime for different values. Fig. 7. Lifetime statistics for OML and CMAX.
reasonable to assume that the density (i.e., number of network; 10 networks), the OML algorithm resulted in a
sensors divided by the number of grid points in the region) lifetime greater than or equal to that obtained by CMAX!
of operational sensors is small. For example, in our As can be seen from Fig. 6, the performance of CMAX is
experiments (as well as those reported in [12], [3], between quite sensitive to the chosen value of , whereas OML has a
8 percent and 20 percent of the eligible sensor sites have a relatively stable performance for between 104 and 1011 . In
sensor). In these environments, an rT 2:9 will almost further experiments, we used ¼ 1011 for both OML and
always mean that we start with a disconnected sensor CMAX.
network. Therefore, in our experiments, we use rT > 5. A similar experiment was conducted to determine a
Misra and Banerjee [15] focus on wireless nodes. The suitable value for c in the definition of . This experiment
only data set described by them has a wireless node at every revealed that, for our data sets, lifetime was relatively
eligible location, ensuring network connectivity even when insensitive to the choice of c. However, lifetime was
rT ¼ 1. This data set seems inappropriate for the applica- (marginally) best when c ¼ 0:001 r3T (recall that, in our
tions considered in this paper. As a result of the relative single-hop energy model, 0:001 d3 units of energy are
evaluations reported in the literature and our choice of needed to transmit a distance d).
rT > 5, we focus our simulations on the relative perfor- 7.2 Effect of Transmission Radius
mance of OML, CMAX, and MRPC. To determine the effect of transmission radius on the
7.1 Selecting OML and CMAX Parameters performance of OML, CMAX, and MRPC, we used a 25
25 grid. Ten networks, each obtained by placing 50 sensors
To determine suitable values for and , we experimented
at randomly chosen grid points, were considered. For each
with 10 10 grids into which 20 sensors were randomly
network, 10 route request sequences were generated. The
placed. A total of 10 random placements of 20 sensors were
initial energy at each sensor was set to 30 and the
considered. In other words, we used 10 random sensor
transmission radius rT was varied from 7 to 30. Fig. 8
networks defined over a 10 10 grid. Each network had
shows the average lifetime and average energy consumed
20 sensors. The transmission radius and initial energy for
per session (i.e., source-to-destination route). The lifetime
each sensor were set to 1 and 30, respectively. For our first
obtained using either OML or CMAX is significantly larger
experiment, we set c ¼ 0:001r3T in the definition of .
than that obtained using MRPC. For small transmission
Network lifetime was determined for ¼ 10i , 1 i 12.
radii, CMAX and OML have similar performance. As the
For each network and combination, the lifetime over
transmission radius is increased from a low of 7, the lifetime
10 random route request sequences was measured. So, for
obtained by both CMAX and OML increases. However,
each value, 100 lifetime measurements (10 networks with
while the lifetime obtained by OML is fairly stable for
10 sequences each) were made. The average lifetime for rT 10, the lifetime obtained by CMAX declines rapidly as
these 100 measurements is reported in Fig. 6. Fig. 7 gives the rT increases beyond 10. Similarly, the energy consumed per
average lifetime for 9 of the 10 networks used in the study session is much larger when MRPC is used than when
of Fig. 6. However, this time, the average is over the either OML or CMAX is used. For rT < 10, CMAX used
120 lifetimes for the 10 sequences times 12 values used for slightly less energy per session than did OML. However, for
each network. The standard deviation in the lifetime over larger values of rT , CMAX was more energy frugal.
these 120 tests also is reported. The rows labeled Avg Imp
give the average increase in lifetime obtained by OML over 7.3 Effect of Sensor Density
that obtained by CMAX and the rows labeled Std Dev Imp Our next study compared the effect of sensor density on
give the standard deviation in this lifetime increase. The performance. For this study, we placed n sensors,
results for the 10th network used in our study are similar to n 2 f40; 50; 60; 70; 80; 90; 100g, at random locations in a 25
those reported in Fig. 7. In all 1,200 lifetime tests (120 per 25 grid. For each choice of n, we experimented with
Authorized licensed use limited to: IEEE Xplore. Downloaded on November 25, 2008 at 23:33 from IEEE Xplore. Restrictions apply.
1054 IEEE TRANSACTIONS ON COMPUTERS, VOL. 55, NO. 8, AUGUST 2006
Fig. 8. Average lifetime and energy consumption as a function of transmission radius. (a) Average lifetime. (b) Average energy per session.
Fig. 9. Lifetime and energy consumed as a function of n, rT ¼ 10. (a) Average lifetime. (b) Average energy per session.
Fig. 10. Lifetime and energy consumed as a function of n, rT ¼ 20. (a) Average lifetime. (b) Average energy per session.
rT 2 f10; 20; 30g. The initial energy at each sensor was set to the lifetime using OML was at least as large as that obtained
30. For each n and rT combination, we generated 10 random using CMAX. MRPC consistently underperformed both
placements and, for each placement, we generated 10 random OML and CMAX. As can be seen from the figures, for all
request sequences. Fig. 9, Fig. 10, and Fig. 11 graph, as a three choices of rT , the percentage increase in lifetime
function of n, the average lifetime and average energy obtained by OML versus CMAX and MRPC increases as the
consumed per session for each rT . For all 700 test instances, sensor density increases. When n ¼ 100 and rT ¼ 30, the
Authorized licensed use limited to: IEEE Xplore. Downloaded on November 25, 2008 at 23:33 from IEEE Xplore. Restrictions apply.
PARK AND SAHNI: AN ONLINE HEURISTIC FOR MAXIMUM LIFETIME ROUTING IN WIRELESS SENSOR NETWORKS 1055
Fig. 11. Lifetime and energy consumed as a function of n, rT ¼ 30. (a) Average lifetime. (b) Average energy consumed per session.
average lifetime using OML is 31.9 percent larger than sensor was set to 30 units and an rT value of 20 was used.
when CMAX is used and is 55 percent larger than when The remaining parameters were set to the same values used
MRPC is used. These percentages at rT ¼ 20 are 16.5 percent in our earlier experiments.
and 53.9 percent and, at rT ¼ 10, they are 4.3 percent and Fig. 12 shows the average lifetime over 10 different
53 percent. Although at rT ¼ 10, the energy consumed per randomly generated request sequences for each of the
session by OML and CMAX is virtually the same, OML and 20 random networks. The results are consistent with those
CMAX consume, on average, 33.2 percent less energy than reported in [12]: CMAX is superior to max-min zPmin . The
consumed by MRPC. When rT ¼ 20 OML consumes, on new information is that min-max zPmin is superior to MRPC.
average, 9.1 percent less energy per session than does
CMAX and 35.5 percent less energy than consumed by 7.5 Network Capacity
MRPC; at rT ¼ 30, this reduction in energy consumed per As mentioned in Section 3, the CMAX algorithm was
session is 17.8 percent and 35.5 percent. developed originally to maximize network capacity—the
number of successful routes in a given time period. Online
7.4 Comparison with max-min zPmin routing algorithms for capacity maximization may employ
The CMAX algorithm was compared experimentally to the an admission control policy to determine which route
max-min zPmin algorithm of [3] in [12]. From this requests they will service. Algorithms developed for
comparison and our experimental results reported so far, lifetime maximization are required to service every route
we expect that OML and CMAX will result in better request for which there is a feasible path. We may use a
lifetimes than min-max zPmin . However, the standing of lifetime maximization algorithm in a capacity maximization
min-max zPmin relative to MRPC is not known. To resolve application by using the admission control policy “service
this, we ran a lifetime experiment using 20 randomly the current route request if there is a feasible path for this
generated networks, each obtained by randomly placing request.” We compared the performance of OML, CMAX,
50 sensors on a 25 25 grid. The initial energy at each and MRPC using the capacity metric. For our experiment,
Fig. 12. Lifetime under various algorithms. Fig. 13. Capacity under various algorithms.
Authorized licensed use limited to: IEEE Xplore. Downloaded on November 25, 2008 at 23:33 from IEEE Xplore. Restrictions apply.
1056 IEEE TRANSACTIONS ON COMPUTERS, VOL. 55, NO. 8, AUGUST 2006
we generated 20 networks by randomly placing 20 sensors [12] K. Kar, M. Kodialam, T. Lakshman, and L. Tassiulas, “Routing for
Network Capacity Maximization in Energy-Constrained Ad-Hoc
in a 10 10 grid. Each sensor started with 30 units of energy Networks,” Proc. IEEE INFOCOM 2003.
and rT was set to 1. For each of the 20 networks, a random [13] M. Maleki, K. Dantu, and M. Pedram, “Power-Aware Source
sequence of 10,000 message routing requests was generated. Routing Protocol for Mobile Ad Hoc Networks,” Proc. Int’l Symp.
Low Power Electronics and Design (ISLPED ’02), pp. 72-75, 2002.
The remaining parameters were as used in our earlier [14] T. Melodia, D. Pompili, and I. Akyildiz, “Optimal Local Topology
experiments. Knowledge for Energy Efficient Geographical Routing in Sensor
Fig. 13 shows the average number (for each of the Networks,” Proc. IEEE INFOCOM, 2004.
[15] A. Misra and S. Banerjee, “MRPC: Maximizing Network Lifetime
20 random networks, 10 request sequences of size 10,000 for Reliable Routing in Wireless Environments,” Proc. IEEE
each were tried) of routing requests successfully completed Wireless Comm. and Networking Conf. (WCNC), 2002.
by each of the three test algorithms. Network capacity using [16] S. Sahni, Data Structures, Algorithms, and Applications in Java,
second ed. Silicon Press, 2005.
OML is approximately 6.7 percent higher, on average, than
[17] S. Singh, M. Woo, and C. Raghavendra, “Power-Aware Routing in
when CMAX is used; the capacity is about 58.1 percent Mobile Ad Hoc Networks,” Proc. ACM/IEEE MOBICOM, 1998.
higher using OML rather than MRPC. [18] A. Spyropoulos and C. Raghavendra, “Energy Efficient Commu-
nications in Ad Hoc Networks Using Directional Antenna,” Proc.
IEEE INFOCOM, 2002.
8 CONCLUSION [19] I. Stojmenovic and X. Lin, “Power-Aware Localized Routing in
Wireless Networks,” IEEE Trans. Parallel and Distributed Systems,
We have shown that the lifetime maximization problem is vol. 12, no. 11, pp. 1122-1133, Nov. 2001.
NP-hard. We have also proposed a new online heuristic- [20] C. Toh, H. Cobb, and D. Scott, “Performance Evaluation of
Battery-Life-Aware Routing Optimization Schemes for Wireless
OML-for lifetime maximization. Extensive simulations Ad Hoc Networks,” Proc. IEEE Int’l Computing Conf. (ICC ’01),
show that our new heuristic is superior to previously 2001.
published heuristics for lifetime maximization both in terms [21] J. Wu, M. Gao, and I. Stojmenovic, “On Calculating Power-Aware
of providing a larger lifetime and in terms of sensitivity to Connected Dominating Sets for Efficient Routing in Ad Hoc
Wireless Networks,” J. Comm. Networks, vol. 4, no. 1, pp. 59-70,
algorithm parameters. Additionally, our proposed heuristic 2002.
provides larger network capacity than provided by compet- [22] G. Zussman and A. Segall, “Energy Efficient Routing in Ad Hoc
ing heuristics. The superior performance of OML is due to Disaster Recovery Networks,” Proc. IEEE INFOCOM, 2003.
its making a conscious effort to balance the desire to
Joongseok Park received the BS degree in
minimize total energy consumption as well as the desire to computer information science from Ohio State
prevent the depletion of a sensor’s energy. University in 1994, the MS degree in computer
science from Oklahoma State University in 1999,
and the PhD degree in computer and information
ACKNOWLEDGMENTS science and engineering from the University of
Florida in 2005. Currently, he is a senior engineer
This research was supported, in part, by the US National at Samsung Electronics in Korea. His research
Science Foundation under grant ITR-0326155. interests are wireless sensor networks, routing
algorithm, and network security.
REFERENCES
[1] G. Li, J. Aslam, and D. Rus, “On-Line Power-Aware Routing in Sartaj Sahni received the BTech (electrical
Wireless Ad Hoc Networks,” Proc. Mobicom, pp. 97-107, July 2001. engineering) degree from the Indian Institute of
[2] K. Kar, M. Kodialam, T.V. Lakshman, and L. Tassiulas, “Routing Technology, Kanpur, and the MS and PhD
for Network Capacity Maximization in Energy-Constrained Ad- degrees in computer science from Cornell
Hoc Networks,” Proc. Infocom, pp. 673-681, 2003. University. He is a Distinguished Professor and
[3] J. Aslam, Q. Li, and R. Rus, “Three Power-Aware Routing Chair of Computer and Information Sciences
Algorithms for Sensor Network,” Wireless Comm. and Mobile and Engineering at the University of Florida. He
Computing, vol. 3, pp. 187-208, 2003. is also a member of the European Academy of
[4] J. Chang and L. Tassiulas, “Routing for Maximum System Sciences, a fellow of the IEEE, ACM, AAAS, and
Lifetime in Wireless Ad-Hoc Networks,” Proc. 37th Ann. Allerton Minnesota Supercomputer Institute, and a Dis-
Conf. Comm., Control, and Computing, Sept. 1999. tinguished Alumnus of the Indian Institute of Technology, Kanpur. In
[5] J. Chang and L. Tassiluas, “Energy Conserving Routing in 1997, he was awarded the IEEE Computer Society Taylor L. Booth
Wireless Ad-Hoc Networks,” Proc. IEEE INFOCOM, 2000. Education Award and, in 2003, he was awarded the IEEE Computer
[6] J. Chang and L. Tassiulas, “Fast Approximate Algorithms for Society W. Wallace McDowell Award. He was awarded the 2003 ACM
Maximum Lifetime Routing in Wireless Ad-Hoc Networks,” Proc. Karl Karlstrom Outstanding Educator Award. He has published more
IFIP-TC6 Networking 2000, pp. 702-713, 2000. than 300 research papers and written 15 texts. His research publications
[7] C. Florens and R. McEliece, “Scheduling Algorithms for Wireless are on the design and analysis of efficient algorithms, parallel
Ad-Hoc Sensor Networks,” Proc. IEEE GLOBECOM, pp. 6-10, computing, interconnection networks, design automation, and medical
2002. algorithms. Dr. Sahni is a co-editor-in-chief of the Journal of Parallel and
[8] M. Garey and D. Johnson, Computers and Intractability: A Guide to Distributed Computing, a managing editor of the International Journal of
the Theory of NP-Completeness. W.H. Freeman, 1979. the Foundations of Computer Science, and a member of the editorial
[9] W. Heinzelman, A. Chandrakasan, and H. Balakrishnan, “Energy- boards of Computer Systems: Science and Engineering, International
Efficient Communication Protocol for Wireless Microsensor Net- Journal of High Performance Computing and Networking, International
works,” Proc. IEEE Hawaii Int’l Conf. System Sciences (HICSS), 2000. Journal of Distributed Sensor Networks, and Parallel Processing Letters.
[10] R. Kannan, L. Ray, R. Kalidindi, and S. Iyengar, “Energy He has served as a program committee chair, general chair, and been a
Threshold Constrained Geographic Routing Protocol for Wireless keynote speaker at many conferences. Dr. Sahni has served on several
Sensor Networks,” Signal Processing Letters, vol. 1, no. 1, pp. 79-84, US National Science Foundation and US National Institutes of Health
2003. panels and he has been involved as an external evaluator of several
[11] R. Kannan and S. Iyengar, “Game Theoretic Models for Reliable computer science and engineering departments.
Path-Length and Energy Constrained Routing with Data Aggre-
gation in Wireless Sensor Networks,” IEEE J. Selected Areas in
Comm., vol. 22, no. 6, pp. 1141-1150, 2004.
Authorized licensed use limited to: IEEE Xplore. Downloaded on November 25, 2008 at 23:33 from IEEE Xplore. Restrictions apply.