You are on page 1of 55

# .

.

Course “Algorithmic Foundations of Sensor Networks” Lecture 4: Energy Balance Algorithms
Sotiris Nikoletseas
Department of Computer Engineering and Informatics University of Patras, Greece

March 22, 2013

.

.

.

.

.

.

1 / 55

A “canonical” problem: Local Event Detection and . Data Propagation

A single sensor, p, senses a local event E . The general propagation problem is the following: “How can sensor p, via cooperation with the rest of the sensors in the network, propagate information about event E to the control center?”

.

.

.

.

.

.

2 / 55

. Representative data propagation protocols

• Directed Diffusion (DD): a tree-structure protocol (suitable

for low dynamics)
• LEACH: clustering (suitable for small area networks) • Local Target Protocol (LTP): greedy, single-path

optimization (best for dense networks)
• Probabilistic Forwarding Protocol (PFR): redundant

optimized transmissions (good efﬁciency / fault-tolerance trade-offs, best for sparse networks)
• Energy Balance Protocol (EBP): guaranteeing same per

sensor energy (prolong network life-time)

.

.

.

.

.

.

3 / 55

. Low Energy Adaptive Clustering Hierarchy (LEACH)

• The network is partitioned in clusters. • Each cluster has one cluster-head. • Each non cluster-head sends data to the head of the

cluster it belongs to.
• Cluster-heads gather the sent data, compress it and send

it to the sink directly.

.

.

.

.

.

.

4 / 55

. The basic idea of PFR
PFR probabilistically favors redundant transmissions toward the sink within a thin zone of particles around the optimal line connecting the particle sensing the event E and the sink.

Thin Zone of particles
. . . . . .

5 / 55

. The Forwarding Probability Data is propagated with a suitably chosen probability p. .. To favor near-optimal transmissions the following probability is used: Pfwd = E p2 p1 1 2 ϕ π S . 6 / 55 . . while it is not propagated with probability 1 − p. . .

. . • We deﬁne the success ratio: I Ps = k K • We deﬁne the average energy per particle Eavg as: ∑n Eavg = i Ei n where n is the number of particles deployed. .. . Let Ei be the available (over time) energy for particle i . . Particle i is active–alive if Ei > 0. • Let hA be the number of “active” particles. Let Is be the injection rate of events. 7 / 55 . LEACH Let K be the total number of events and k the number of events successfully reported to the sink. . Comparison of PFR.

9 PFR 0.7.8 for Is = 0.05 0.8 0. .05. • For H-TEEN I Ps → 0. .4 0. • For large networks.1 0 0 200 400 600 800 1000 1200 1400 1600 Network Size • For small networks.85 ≤ I Ps ≤ 1.8. .7 H-TEEN 0. and I Ps → 0. Success Rate over Network Size 1 PFR 0.8 Success Rate 0..8 0.6 0. (HTEEN: Hierarchical extension of LEACH) .05 H-TEEN 0.2 0.5 0. . . 0.3 0.3 for Is = 0. 8 / 55 . success rate drops: • For PFR I Ps → 0.

6 0.3 0.2 0.8 0.6 0. .7 H-TEEN 500x500 H-TEEN 1500x1500 Success Rate 0.5 0.7 0.9 Event Rate Is ∈ [0.8 0.4 0.5 0. Success Rate over Injection Rate 1 PFR 500x500 0.2 0. .1 0 0 0. .3 0. . • PFR seems to be less affected by changes in Is .1 0. . .05.9 PFR 1500x1500 0.4 0..8] • For large networks the success rate of H-TEEN drops signiﬁcantly when Is increases. 9 / 55 . 0.

Average Energy over Time 1 PFR 500x500 Average Energy per Particle (J) PFR 1000x1000 0.9 H-TEEN 500x500 H-TEEN 1000x1000 0. . . . 10 / 55 ..8 0.6 0.7 0. H-TEEN is more energy efﬁcient. the energy consumption of H-TEEN increases dramatically. • For large network size. .5 0 1000 2000 3000 4000 5000 6000 7000 8000 Simulation Rounds • For small network size. • The energy consumption of PFR is almost identical in both cases. . .

. • For large network size H-TEEN “drains” particles rapidly. . . .. 11 / 55 . . . Number of Alive Particles over Time 2600 PFR 500x500 2350 PFR 1000x1000 H-TEEN 500x500 Alive Particles 2100 H-TEEN 1000x1000 1850 1600 1350 1100 0 3000 6000 9000 12000 15000 Simulation Rounds • For small network size PFR “drains” particles rapidly.

12 / 55 . . • PFR tends to strain the particles which lie close to the sink more. while it is “expensive” in large network areas and high injection rate. . . . . . • LEACH/H-TEEN behaves very well on small network size and low injection rate settings. Conclusions • PFR behaves quite well for all network sizes and injection rates..

. 13 / 55 . . . • In a direct transmission scheme the distant nodes tend to be overused. “How can we achieve equal energy dissipation per node in order to prolong the network lifetime by avoiding early network disconnection?” . The Energy Balance Protocol (EBP) All protocols tend to “strain” some speciﬁc nodes in the network. . • In a hop-by-hop scheme the nodes closer to the sink tend to be overused.. .

.. . . . Solution: “Each node chooses randomly whether to propagate the event one-hop closer to the sink or to transmit it directly to the sink”. . A Probabilistic Solution • Direct transmission cost is much larger than that of one-hop transmission and exhausts distant nodes. . The goal is to “balance” the two types of transmissions and achieve equal average energy dissipation per sensor. 14 / 55 . • One-hop transmissions are cheap but tend to overuse nodes that lie closer to the sink.

The EBP Protocol (I) • Network Partition: • Partition the network into n sectors. . . . . 15 / 55 . “slices". .. of width R (the transmission range). .

. The EBP Protocol (II) • Data Propagation: Each node in sector i propagates its messages according to the following rule: . . 16 / 55 . ..Propagate the message directly to the sink with probability 1 − pi . The choice of pi is made such as the average per sensor energy dissipation is the same for all sensors in the network. . .Propagate the message to sector i − 1 with probability pi . . .

An Instance of the Execution There is a message on node i . 17 / 55 . . .. . . . .

. . n} . j ∈ {1. • Ei : the total energy dissipated in sector i . . We want: E [Ej ] E [Ei ] = Si Sj ∀i . . . . . • Sensor nodes are spread random uniformly in the network area.. . Balanced Average per Sensor Energy Dissipation • Si : the area size of sector i . 18 / 55 .

. . . Let ϵij a random variable that measures the energy that dissipates the sector i so as to handle message j . The Model & Assumptions The events occur in random uniform positions in the network. 19 / 55 . E [ϵij ] = [i 2 − pi (i 2 − 1)]cR 2 cR 2 c (iR )2 with probability pi with probability 1 − pi . .. { ϵij = Clearly. .

. 20 / 55 . . . Computation of pi (message handling) • Let hi the number of messages that handles sector i . Clearly: hi = fi + gi By linearity of expectation: E [hi ] = E [fi ] + E [gi ] .. . . • Let fi the number of messages that were forwarded to sector i . • Let gi the number of messages that were generated in sector i .

. 21 / 55 . . . . . . n} . .. . Clearly: pi +1 = E [fi ] E [fi +1 ] + E [gi +1 ] E [fi ] = pi +1 E [hi +1 ] = pi +1 · (E [fi +1 ] + E [gi +1 ]) The Energy Balance Property: ] [ ∑h ] [∑ hi j ϵ ϵ k =1 ik k =1 jk =E E Si Sj ∀i . Lemma . Computation of pi . j ∈ {1. The following relationship holds: .

. . . 22 / 55 . .. . A recurrence relation for pi a(i + 1)E [fi +1 ] − (d (i ) + a(i ))E [fi ] + d (i − 1)E [fi −1 ] = a(i )E [gi ] − a(i + 1)E [gi +1 ] where a(i ) = Initial conditions: i2 2i − 1 E [fn ] = 0 d (i ) = (i + 1)2 − 1 2i + 1 E [f0 ] = n .

n. we compute the exact values of probabilities pi . • First we transform the recurrence into a simpler recurrence with only two successive terms (whose coefﬁcient is 1). we solve the initial one. 23 / 55 . . . . • Having computed the values of E [fi ] for i = 1. .. . . Computation of pi Structure of the rest of solution evaluation. . . • Having solved the latter recurrency. . .

The exact solution . 24 / 55 . . . .The solution: E [fi ] = − n−i ∑ k =1 ∏n−i +1 j =k ∏n−i j =k a(n − j ) d (n − j )  n −k ∑ · (a(j )E [gj ] − a(j + 1)E [gj +1 ]) + a(1) · E [f1 ] j =1  where i −1 ∏ i a(i ) = 1 ..Thus we can calculate pi ’s: pi + 1 = E [ fi ] E [fi +1 ] + E [gi +1 ] .Easily computed in a repetitive manner . . .

25 / 55 . . . A closed approximate form for pi If E [fi ] ≃ E [fi −1 ] then pi = 1 − 3x (i + 1)(i − 1) 1 2 for 3 ≤ i ≤ n p2 = x and can be set to E [fi ] ≃ E [fi −1 ] is realistic because: • Si ≃ Si −1 • pi ≃ pi −1 . . . .. .

.e. when far away from the sink it is better to move hop-by-hop. . .. . . i. 26 / 55 . when we approach the sink it is better to transmit directly in order to bypass the critical region (and since energy consumption is small). Comments on pi . . . to avoid spending too much energy.when i becomes small.e. pi is large. pi is small.when i is large. i.

Pierre Leone. . José Rolim in DCOSS 2006 . .An Optimal Data Propagation Algorithm for Maximizing the Lifespan of Sensor Networks Aubin Jarry. . 27 / 55 . . . Olivier Powell.

Generalized data propagation algorithm. . . allows to jump over bottle-neck nodes .. . Lifespan Maximization . 28 / 55 . . . Note: do hops to “intermediate” slices help? .Slicing of the network . Our algorithm balances energy consumption: .

mixed propagation: only single hops and direct to sink transmissions . Main Theoretical Result . Lifespan is maximized by a mixed data propagation algorithm . . . Application . lifespan) .mixed strategies beat every other possible strategy (wrt. . . 29 / 55 . We use this fact to propose a distributed optimal data propagation algorithm .. . . . Lifespan Maximization .

. Energy cost: Sending a message from slice i to j costs (i − j )2 E/msg fi . Model 1 . . sink 1 2 3 4 . 30 / 55 . . .. .j is the message rate from slice i slice to slice j msg/t sink 0 1 4 9 16 1 1 0 1 4 9 2 4 1 0 1 4 3 4 9 16 4 9 1 4 0 1 1 0 2 .

1 f3.j is the message rate from slice i slice to slice j msg/t sink f0.1 f4. 31 / 55 . .0 f1.4 f4.2 3 f0.2 f2.2 f1.4 f1. Energy cost: Sending a message from slice i to j costs (i − j )2 E/msg fi .3 f1.3 f4. .0 f3..1 f2.4 2 .3 f2. sink 1 2 3 4 . .0 f2.2 f3. .1 f1.4 f3.4 f2.2 f4.3 f3.1 2 f0.0 1 f0.3 4 f0. .0 f4. Model 1 .

gi : event generation rates (i. . . LP to maximize the generalized-ﬂow: . . .i = Tgi Detection rates ∑N 2 j =0 fi . data injected at i ) .j = j =0 fj . Generalized-ﬂow maximization . Given a WSN: Maximize the generalized network ﬂow • Given the detection rates f0.i Flow equations . .e. f0. 3 .j (i − j ) ≤ bi Energy constraint ∑N ∑N j =0 fi .. 32 / 55 .i := gi • Given the available energy: bi . Maximize T in: 1 . Problem .maximize time T ⇒ lifespan maximization . 2 . . .

33 / 55 . Problem .j = 0 if j ̸∈ {0. . .i := gi • Given the available energy: bi .i = Tgi Detection rates ∑N 2 j =0 fi . . LP to maximize the mixed-ﬂow: . . . . Given a WSN: Maximize the mixed network ﬂow • Given the detection rates f0. i − 1} mixed ﬂow constraint . Mixed-ﬂow maximization . 3 . . 2 ..i Flow equations fi .j (i − j ) ≤ bi Energy constraint ∑N ∑N j =0 fi . Maximize T in: 1 . f0. 4 .j = j =0 fj .

i := gi • Given the available energy: bi . 4 . 2 . Maximize T in: 1 . f0.j (i − j ) ≤ bi Energy constraint ∑N ∑N j =0 fi .i Flow equations fi . Given a WSN: Maximize the mixed network ﬂow • Given the detection rates f0.. .j = 0 if j ̸∈ {0. . 34 / 55 . (4) guarantess no hops to “intermediate” slices (3) guarantees ﬂow preservation . LP to maximize the mixed-ﬂow: . Problem .i = Tgi Detection rates ∑N 2 j =0 fi . . . 3 . Mixed-ﬂow maximization . .j = j =0 fj . i − 1} mixed ﬂow constraint . .

35 / 55 . . it maximizes the mixed-ﬂow it maximizes the generalized-ﬂow .. . . . . We can maximize the generalized-ﬂow problem using: • a mixed-ﬂow • a local property (energy-balance) . Mixed-ﬂows are optimal . 2 . If there exists an NRG-balanced mixed ﬂow 1 . Result . . . Application .

• Find m: the lowest potential neighbour • If potential (m) < potential (self ) then send data to m • Else send data directly to the sink . 3 . . . Algorithm Propagate Data . . Each node has a potential potential (n) ≃ EnergySpent (n) 2 . . Each node knows its list of neighbours Vn Each node knows the potential of its neighbours . Inputs . .A distributed algorithm . . 1 . 36 / 55 . . Deﬁnition .

37 / 55 . .A distributed algorithm . . . . Illustration . .

. . • The algorithm balances energy . Remark . .A distributed algorithm . 38 / 55 . • The algorithm produces a mixed ﬂow . Illustration . . .

• Let λi be the probability that the event occurs in slice i . 3. Inputs . • {X (t )}t ≥0 is a Markov Chain • If λi > 0. We consider   XN (t ) − XN −1 (t ) XN −1 (t ) − XN −2 (t )   X (t ) =   . . . the Markov chain . 2. A Markov chain approach . .Stability of the Algorithm .   . . . X2 (t ) − X1 (t )  t . . • Xi (t ) is the energy spent by slice i at time t = 1. 0 0 .   We would like to have X (t ) →   39 / 55 . . Remarks . . . 0 is irreducible      . . .

... Theorem . we give sufﬁcient (martingale type) conditions for stability Stability: the MC can leave the origin but will always come back . . . i = 2. 40 / 55 . with { } i2 N −1 A= X ∈R : |Xi (t ) − Xi −1 (t )| ≤ . If i 2 λi > (i − 1)2 λi −1 .. N 2 . .. . the Markov chain X (t ) is stable around A. . . Stability of the algorithm Note: λi > 0 is necessary for irreducibility but not always realistic Thus.

. . Simulations • Scatter nodes randomly over a region • Events are generated randomly • Data is propagated according to the algorithm .. . . . 41 / 55 .

. . .. 42 / 55 . . First Simulation • 1000 sensors randomly dispersed over a 10m disc • Sink at the center of the disc • Potential function potential (n) = Energy (n) . .

computed by an LP) L: maximum possible ﬂow without direct transmissions . First Simulation .. . 43 / 55 . . Messages received by sink . max{nodes} (EnergySpent ) F : the ﬂow of the algorithm U : maximum possible ﬂow (ofﬂine. Network ﬂow . . . .

Remarks . First Simulation . . • Close to optimal .. • Distributed and on-line! . 44 / 55 . . . . .

. .. .the online algorithm performs very well . First Simulation . . .off-line ideal ﬂow U balances the energy load . 45 / 55 .

.. . . . . 46 / 55 . Second Simulation • 600 sensors randomly dispersed over 2 intersecting 30o sector graphs of 10m diameter with one sink at the narrow end of each sector • events can be reported to either sink .

. . . 47 / 55 . Second Simulation • 600 sensors randomly dispersed over 2 intersecting 30o sector graphs of 10m diameter with one sink at the narrow end of each sector • events can be reported to either sink .. . .

Second Simulation • 600 sensors randomly dispersed over 2 intersecting 30o sector graphs of 10m diameter with one sink at the narrow end of each sector • events can be reported to either sink . . .. . . 48 / 55 . .

. . 49 / 55 . . 3 . . . . Summary 1 . Prove that optimal solution belongs to a subset of realistic data propagation algorithms: a mixed strategy which is energy balanced Propose a distributed algorithm based on theoretical results Show that the algorithm is efﬁcient (Markov chain context and simulations) 2 . .

sector widths must be ﬁne-tuned • As expected. Adjusting transmission ranges to avoid energy holes Design Guidelines for Maximizing Lifetime and Avoiding Energy Holes in Sensor Networks with Uniform Distribution and Uniform Reporting S. .. 50 / 55 . leads to uneven energy depletion. the width of sectors in energy-balanced network decreases as we near the sink . . . each sending roughly the same number of reports to the sink • They prove that to minimize energy spent. . . Stojmenovic in INFOCOM 2006 • Uniformly distributed sensors. Towards energy balance. Olariu. all sectors must have the same width (which they evaluate) • This choice. I. however.

. An ofﬂine approach On the Energy Hole Problem of Nonuniform Node Distribution in Wireless Sensor Networks X. Chen. Das in MASS 2006 • They ﬁnd that in a circular sensor network with a nonuniform node distribution and constant data reporting. the unbalanced energy depletion among the nodes in the whole network is unavoidable • A suboptimal energy efﬁciency among the inner parts of the network is possible if the number of nodes increases with geometric proportion from the outer parts to the inner ones • They also present a routing algorithm with this node distribution strategy . K. . S. G. . . . Wu. 51 / 55 ..

S. Boukerche. 52 / 55 . off-line optimum solution derived by a linear program which maximizes the network lifetime and show that it achieves near-optimal performance for uniform sensor deployments . . . . Raptopoulos in MSWiM 2011 • They investigate both uniform and heterogeneous sensor placement distributions • They propose a distributed. Ch. adaptive data propagation algorithm that exploits limited. local network density information for achieving energy-balance while at the same time minimizing energy dissipation • They compare the protocol’s performance with a centralized. Exploiting limited local network density information Close-to-optimal energy balanced data propagation via limited. Nikoletseas. local network density information A. . . D.. Efstathiou.

. I. Kotsogiannis. . S.. . . Nikoletseas in MOBIWAC 2011 • This work is the ﬁrst studying the energy balance property in wireless networks where the nodes are highly and dynamically mobile • They propose a new diverse mobility model which is easily parametrized • They also present a new protocol which tries to adaptively exploit the inherent node mobility in order to achieve energy balance in the network in an efﬁcient way . Efstathiou. . 53 / 55 . Energy balance under mobility Energy balanced data propagation in wireless sensor networks with diverse node mobility D.

in Wireless Networks (WINET) Journal. 2004. Parallel Distrib. . Powell. C.P. Efthymiou. “An Adaptive Blind Algorithm for Energy Balanced Data Propagation in Wireless Sensor Networks”. 35-48. Nikoletseas and J. Rolim. In Proc. in the IEEE International Conference on Distributed Computing in Sensor Networks (DCOSS).A. . (JPDC). “Energy Efﬁcient Protocols for Sensing Multiple Events in Smart Dust Networks”. I. Efthymiou. Powell. Antoniou. . 12(6): 691-707 (2006).24. Rolim: “Energy balanced data propagation in wireless sensor networks”. 2005. Springer Verlag. 67(3): 302-317 (2007). S.S. 37th Annual ACM/IEEE Simulation Symposium (ANSS’04). . “An Optimal Data Propagation Algorithm for Maximizing the Lifespan of Sensor Networks”. Nikoletseas. .O.. Rolim. Chatzigiannakis. IPDPS 2004. in the proceedings of DCOSS 2006. . S.C. Mobile. . IEEE Computer Society Press. A. Kinalis and G. pp. Mylonas. A. J. P. “Energy optimal data propagation in wireless sensor networks”. Rolim. J. Volume 3267. References (I) . . Jarry. in Proc. LNCS. Nikoletseas. . O. 4th International Workshop on Algorithms for Wireless. J. pp. Leone. Comput. Leone. 15 . 54 / 55 . Leone. Ad-Hoc and Sensor Networks (WMAN ’04). Also. J. . P.

INFOCOM 2006. 1-12. ACM. .A. S. Boukerche. I. Wu. S. "On the Energy Hole Problem of Nonuniform Node Distribution in Wireless Sensor Networks". 2006 IEEE International Conference. ACM. Stojmenovic. In Proceedings of the 9th ACM international symposium on Mobility management and wireless access (MobiWac ’11). analysis and simulation of wireless and mobile systems (MSWiM ’11). 55 / 55 . NY. S. 2006. Efstathiou.. "Close-to-optimal energy balanced data propagation via limited. USA. . . . References (II) . 2011. USA. Nikoletseas. C. 180-187. "Design Guidelines for Maximizing Lifetime and Avoiding Energy Holes in Sensor Networks with Uniform Distribution and Uniform Reporting". Proceedings. 85-92. Mobile Adhoc and Sensor Systems (MASS). local network density information". Chen. "Energy balanced data propagation in wireless sensor networks with diverse node mobility". Efstathiou. . . .D. 163-166. 2011.S. G. In Proceedings of the 14th ACM international conference on Modeling. Olariu. Kotsogiannis. Nikoletseas. K. NY. 25th IEEE International Conference on Computer Communications. 2006. D.X. I. . Raptopoulos. Das. .