Wireless Sensor Networks, energy efficiency and path recovery

Transcription

1 Wireless Sensor Networks, energy efficiency and path recovery PhD dissertation Anne-Lena Kampen Trondheim 18 th of May 2017

2 Outline Introduction to Wireless Sensor Networks WSN Challenges investigated Energy usage in WSN - consumption model Energy reduction during forwarding Connectivity in WSN Path recovery algorithms Probability of path disconnections Balancing of energy consumption Conclusion 2

3 Wireless Sensor Networks WSN Sensor nodes, or nodes for short Sink Data collection Radio and antennae for wireless communication Forward messages on behalf of other nodes Typically low cost devices with limited energy, memory, and microcontroller capacity Source/ Sensor node: Nodes that generate data Sensor data is gathered at the sink Does not have the capacity limits Application areas for WSN Surveillance and tracking Industrial processes, Military solutions, Geriatric care, Environment monitoring * Sink 3

4 Data forwarding Since we assumed multihop WSNs, a routing protocol is needed to create forwarding paths Used the IETF routing protocol standard developed for WSN: Routing Over Low power and Lossy networks (RPL), RFC 6550 The next slide gives a brief presentation of the characteristics of RPL that is essential to our work 4

5 Routing Over Low power and Lossy networks RPL [RFC 6550] Creates a Destination Oriented Directed Acyclic Graph(DODAG) S The DODAG is formed as the sink initiate broadcast of Destination Information Option messages (DIO) containing: The DODAGVersionNumber that defines the DODAG version Managed by the sink, renew the DODAG The rank that define the nodes distance to the sink Parent list A preferred parent: the next-hop/ successor node Trickle timer Control the nodes DIO emission Sink start DIO transmission S Select a preferred parent and re-broadcast DIO S Parent group of child 5

6 CHALLENGES INVESTIGATED ENERGY USAGE AND NETWORK CONNECTIVITY Increase lifetime Energy usage in WSN Energy consumption balance Disconnected nodes degrades the data accumulated Recovery mechanisms Probability of path disconnections 6

7 Energy usage in WSN The main energy consumer is the radio The radio states switches between transmission, reception, idle and sleep The power consumed in during transmission energy consumed during receiving. Sleep state consumes the least power Idle state is controlled by the MAC protocol. We assume homogeneous nodes, hence idle power consumption does not affect the relative consumption among the nodes 7

8 Energy consumption model k 1 : represent the constant level of energy that is consumed when transmitting a bit. k 2 : represent the energy that is proportional to the radiated power when a bit is transmitted. k 3 :energy consumed per received bit. d: transmission range of the nodes. D: distance between the source and the sink. D/d: the minimum number of times the packet has to be relayed to reach the sink. b: number of data bits λ: Node density Energy consumption for one hop transmission: E TX = (k 1 +k 2 d 2 ) b Receiving energy consumption: E RX = k 3 πd 2 λ b Thus, total energy consumption is E TOT = b D d (k 1+k 2 d 2 + k 3 πd 2 λ ) d D 8

9 Energy optimal transmission rang By differentiating E TOT, the energy optimum transmission range is: d opt = 1 k 2 k 1 + k 3 k 1 π λ k 1 =2.7*10-5 k 2 =1.7*10-11 k 3 =2.9*10-5 Overhearing is the main energy consumer The energy optimal transmission range is the shortest range needed to keep the network connected Number of covered nodes when using energy optimal transmission rage: 0.62: AT86RF : CC

10 Receiver in the border area of the sender s transmission range Short transmission rage -> high number of hop # retransmissions increases when receiver is located in the border area of the sender s transmission range Reason to investigate the impact of retransmissions 10

11 Receiver in the border area of the sender s transmission range Packet delivery ratio (PDR) changes at the border of the transmission range Fermi-Dirac function is a suitable model f ( x) 1 1 e x x x 1 0 Packet delivery ratio (PDR) as a function of distance between sender and receiver. The red dots are measured data adopted from Y-D. Lee et.al., *Red dots: Y.-D. Lee, D.-U. Jeong, H.-J. Lee, Performance analysis of wireless link quality in wireless sensor networks, in Proceedings of the 5 th International Conference on Computer Sciences and Convergence Information Technology (ICCIT), 2010,pp

12 Energy reduction during forwarding Main conclusion was that overhearing is the main energy consumer Thus we proposed a method for reducing overhearing 12

13 Energy consumption [µj] Solution to reduce overhearing energy consumption: Preventing nodes form receiving packets not addressed to them Sender Sleeping nodes Energy consumption for transmitting with different node densities CSMA Transmission range Receiver- Next hop node Omnet++ is used for all our simulations unless otherwise stated Output power [dbm] Our suggested solution 13

14 Path recovery algorithms Observed unpredictable energy consumption Nodes will deplete their energy at different epochs Path breaks may occur The reconnection strategies defined in the RFC of RPL have weaknesses Suggest recovery algorithms Main focus: discover all available loop-free paths 14

15 RPL: node enters poisoned state Node D depletes In order to reconnect, node P must have a neighbor at equal or better rank as the former parent node Routing paths before node depletes 2hop 1hop D S Depleting node Neighbors of P P Node entering poisoned state Routing path before node depletes 15

16 Method1: Sequence Number Introduces locally sequence numbers to verify freshness of discovered paths S Broadcasting of request Poisoned node S Reply unicasted toward requesting node 16

17 Method 2: ACK Unreliable poisoning May create routing loops Proposed solution: Reliable poisoning of sub-dag Routing paths before node depletes Depleting node S To achieve reliability: Include parent node information in ordinary DIO Poison message ACK message B DIO D 17

18 Number of management packets Number of unconnected nodes Comparing recovery methods JGuo SeqNum ACK Number of neighborhood nodes Management packets overhead Number of management packets vs. number of nodes in the network Standard RPL JGuo SeqNum ACK Classic Number of neighborhood nodes Number of unconnected nodes Number of unconnected nodes vs. number of nodes in the network. [1] J. Guo, C. Han, P. Orlik, J. Zhang, and K. Ishibashi, "Loop-Free Routing in Low-Power and Lossy Networks," in SENSORCOMM 2012, The Sixth International Conference on Sensor Technologies and Applications, 2012, pp

19 Alternative method: On-demand global recovery Motivation for the method: Recovery method may not succeed although they introduce overhead Nodes that enters poisoning state broadcast an increase-sequencenumber request. When the sink received the request it initiate the global recovery algorithm. No need for periodic DODAG refreshment; long term energy consumption is reduced 19

20 What is the probability that at node enters poisoning state? Do we need recovery methods? An analytical evaluation of the probability that depleted nodes would create path breaks. Simulations were performed to support the analytical results. 20

21 The probability that node N does not enter poisoning state The probability that node N node does not enter Extreme point 1 poisoning state when its parent deplete is: E[P(not poisoned)] = p more than one in Ai at least one in Ai f y dy Extreme point 2 f y : probability density function for the position, y, of node N 21

22 The probability for at least one recovery node p more than one in A at least one in A Extreme point 1 P 2 or more nodes in area A = P 1 or more nodes in area A Extreme point 2 1 P 1 node in area A P 0 nodes in area A 1 P 0 nodes in area A = 1 A(y) e A(y) 1 22

23 Probability density function The cumulative distribution function, F(y) Probability density function, f(y) Extreme point 1 F y = π y 2 π r 2 = y2 r 2 π 2r 2 π r 2 3r 2 f y = 2y 3r 2 Extreme point 2 Extreme point 1 Extreme point 2 F y = π h r +y 2 +π h r 2 π h r+r 2 π h r 2 = y(2hr+y) (1+2h)r 2 f y = 2(hr+y) (1+2h)r 2 23

24 Simulations Probability that a dedicated recovery method is not needed to mend disconnections Java simulation performed for path breaks at: Extreme point 1) Extreme point 2) Counted the number of nodes in the shaded area #simulation with two or more nodes in the area # simulations with at least one node inthe area 1000 simulation runs for each node density 24

25 E(P) Findings The expected value of the probability that a disconnected node does not enter poisoning state. Next to leaf-node 1 Lesson learned: Disconnections are likely to occur! 0,8 0,6 0,4 0,2 2-hop nodes Node next to the leaf node: Analytical Java 2 hop node: Analytical Java Node density 25

26 Balancing of energy consumption Depletion of Single Point of Failure Node (SPOF) may disconnect the network Suggest three new methods for energy balancing Evaluated and compared arrange of energy balancing algorithms 26

27 Balancing algorithms Round-robin among all the nodes in the parent list Weighted round-robin based on energy information in DIO messages prediction parents energy consumption energy information conveyed in ACK packets Use the highest residual-energy parent Weighted round-robin while avoid lowest-energy parent 27

28 Methods suggested Randomize parent selection Randomly select a preferred-parent among the nodes in the parent list Weighting round-robin based on SPOF-parent energy level SPOF node: Nodes that are parent lists containing only one item. Child node advertise: Min[ my energy, SPOF parent energy] Weighting round-robin based on eavesdropping Nodes read the source and destination address information in the overheard traffic. 28

29 Residua energy, % of fully charged battery Findings Energy balance Residual-energy in the nodes after each node has generated 100 data packets Residual energy in the nodes Simulations: The nodes are randomly distributed Area 800m times 800m area. Transmission range is 141m. Node density: 20 nodes inside tx-range Use highest residualenergy parent --- SPOF ---- Round robin --- Random parent --- Classic RPL --- Rank 29

30 Conclusion - Energy usage in WSN The energy-optimal transmission range is shorter than the range needed to keep the network connected The optimal transmission range is, therefore, the shortest range needed to avoid network partition. Reducing number of overhearing nodes a significant impact on the energy consumption The border area of the disk model should be replaced by a more accurate model The Fermi-Dirac function may be used model the relationship between packet delivery ratio (PDR) and the distance between the nodes. 30

31 Conclusion - Energy consumption balance Energy imbalance occurs in network running the classic RPL routing protocol. Randomly selecting a preferred parent among the node in the parent list has a balancing effect. In high density networks the residual-energy of the most depleted node is increased more than 10% compared to classic RPL The highest balancing effect is obtained when nodes use the parent with the current highest amount of residual-energy. Increases the residual-energy of the most depleted nodes by 25% in high density networks. 31 3

32 Conclusion Network connectivity Dedicated recovery methods are needed to mend 25% to 60% of the path breaks. Include recovery methods or adjusted the periodic global update frequency according to the networks recovery delay requirements. Successful path recovery approaches require participation from a high number of nodes =>Message overhead Our suggested on-demand method may reduce the long-term energy consumption, especially in stable networks 32

33 Our work in the field Energy Reduction in Wireless Sensor Networks by Switching Nodes to Sleep During Packet Forwarding Anne-Lena Kampen, Knut Øvsthus, Lars Landmark and Øivind Kure. Proceedings of the 6th International Conference on Sensor Technologies and Applications (SENSORCOMM 2012), pp , 2012, ISBN: Reconnection strategies in WSN running RPL Anne-Lena Kampen, Knut Øvsthus and Øivind Kure 39th Annual IEEE Conference on Local Computer Networks Workshops (LCN2014), pp , 2014, Electronic ISBN: , DOI: /LCNW An Analysis of the Need for Dedicated Recovery Methods and Their Applicability in Wireless Sensor Networks Running the Routing Protocol for Low-Power and Lossy Networks An Analysis of the Need for Dedicated Recovery Methods and Their Applicability in Wireless Sensor Networks Running the Routing Protocol for Low-Power and Lossy Network Energy balancing algorithms in Wireless Sensor Networks Anne-Lena Kampen, Knut Øvsthus and Øivind Kure Proceedings of the 2015 Federated Conference on Computer Science and Information Systems (FedCSIS), Volume 5, pp , 2015, Electronic ISBN: , DOI: /2015F67 Modelling the Optimal Link Length in Wireless Sensor Networks for Two Different Media Access Protocols Knut Øvsthus, Espen Nilsen, Anne-Lena Kampen and Øivind Kure Sensors & Transducers Volume 185. Issue 2, pp , 2015, ISSN:

34 Thank you for listening! 34