Scholarly article on topic 'Agent Based Predictive Data Collection in Opportunistic Wireless Sensor Network'

Agent Based Predictive Data Collection in Opportunistic Wireless Sensor Network Academic research paper on "Computer and information sciences"

CC BY-NC-ND
0
0
Share paper
Academic journal
Procedia Computer Science
OECD Field of science
Keywords
{"Opportunistic sensor networks" / "mobile agents" / "wireless sensor networks" / "data collection" / "agent based data gathering"}

Abstract of research paper on Computer and information sciences, author of scientific article — Rupali Rohankar

Abstract Opportunistic wireless sensor networks have significant applications like remote monitoring where node density is usually low than average. The traditional deployment of sensor networks has larger density, confirmed communication links and multiple routing paths. This does not apply to opportunistic networks. Due to sensing nodes constraints and sparse connectivity, the data collection in OWSN is a difficult task. Mobile agents can be injected into the network for performing network exploration to collect metadata of nodes. Ant-AODV routing protocol provides least overhead in mobile agent routing. Based on the metadata a predictive model can be built to find the probable contact time and location of the mobile nodes and sink in OWSN. This paper simulates an agent based predictive data collection algorithm (PLA). The simulation results are then compared with other opportunistic algorithms and the data collection performance improves almost by 33.2%.

Academic research paper on topic "Agent Based Predictive Data Collection in Opportunistic Wireless Sensor Network"

CrossMark

Available online at www.sciencedirect.com

ScienceDirect

Procedia Computer Science 57 (2015) 33 - 40

3 International Conference on Recent Trends in Computing 2015 (ICRTC-2015)

Agent Based Predictive Data Collection in Opportunistic Wireless Sensor Network

Rupali Rohankara*

_aAssitant Professor, Amity University, Noida, India, rrohankar@amity.edu_

Abstract

Opportunistic wireless sensor networks have significant applications like remote monitoring where node density is usually low than average. The traditional deployment of sensor networks has larger density, confirmed communication links and multiple routing paths. This does not apply to opportunistic networks. Due to sensing nodes constraints and sparse connectivity, the data collection in OWSN is a difficult task. Mobile agents can be injected into the network for performing network exploration to collect metadata of nodes. Ant-AODV routing protocol provides least overhead in mobile agent routing. Based on the metadata a predictive model can be built to find the probable contact time and location of the mobile nodes and sink in OWSN. This paper simulates an agent based predictive data collection algorithm (PLA). The simulation results are then compared with other opportunistic algorithms and the data collection performance improves almost by 33.2%.

© 2015The Authors.PublishedbyElsevierB.V. Thisis an open access article under the CC BY-NC-ND license (http://creativecommons.Org/licenses/by-nc-nd/4.0/).

Peer-reviewunderresponsibilityof organizing committee of the 3rd International Conference on Recent Trends in Computing 2015 (ICRTC-2015)

Keywords— Opportunistic sensor networks, mobile agents, wireless sensor networks, data collection, agent based data gathering

1. Introduction

In Wireless Sensor Networks (WSN) data collection is an important task as stated by [1]. Data collection is either done periodically or whenever an event is triggered. Since the sensor nodes are usually randomly deployed in large geographical area data collection has large volumes stated [25]. This data is then forwarded to a sink node for further processing. Since, the area covered is large and due to limited transmission range, not all the nodes can directly communicate with the sink node. Therefore, the data gathered from surrounding nodes is communicated through intermediate relay nodes using multi-hop communication [23]. On the other hand, some of the applications where the monitoring area involve thick forests, volcano sites and so on, the deployment of sensor nodes is a tedious task and they are dropped from air. The sensor nodes get unevenly spread in the monitoring site and may not be in the transmission range of other relay nodes. The data collection

1877-0509 © 2015 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.Org/licenses/by-nc-nd/4.0/).

Peer-review under responsibility of organizing committee of the 3rd International Conference on Recent Trends in Computing 2015 (ICRTC-2015) doi:10.1016/j.procs.2015.07.361

then is done using mobile nodes as the full coverage is required and full connectivity is impractical [23]. They are known as Opportunistic Wireless Sensor Network (OWSN) [2].

In OWSN data collection is usually done using mobile sink and needs one-hop communication. The sensing nodes therefore temporarily store the data being sensed and wait till the sink node arrives in the communication range [21]. They make use of store-carry-forward mechanism for disseminating the sensed data. Since, the communication is opportunistic a definite routing path cannot be defined in OWSN and new routing methods are needed. In [4], data collection is done in one-hop or multi-hop by mules (mobile data collector) and then they forward the data collected to the sink node. Mobility graph based on contact time prediction between mobile buses and sensing nodes is simulated in [5]. In [6] mobile sink node is controlled and routed on a pre-decided path. A predictive data collection algorithm (PLA) is implemented in this paper. The graph is modeled based on the metadata collected by mobile agents. The sink is then programmed based on the calculations and graph to collect data from the sensing node [16].

A mobile agent migrates from one node to another and accesses the information or performs a task on a node [8]. Mobile agents have been efficiently deployed in WSN to perform data dissemination and fusion thus saving node energy and bandwidth. They best serve the distributed operations in an opportunistic network and well support disconnected, asynchronous operations [9-10]. In [12], data forward probability is calculated and only neighbors with high probability are forwarded the data. In [13], a friendship based routing is used to know the friendly relations based on their behavior between nodes (which are people in social networking group) and the messages are forwarded accordingly. [14] describes the fixed link structure for forwarding data in adhoc networks. [15-16] describe the applications of mobile agents in monitoring wireless sensor networks. The PLA algorithm, utilizes mobile agents to explore network. Network is explored by collecting metadata like residual node energy, statistics of node contact with mobile sink. The number of agents that are sent on the network is determined based on the task for which they will be created and adapt to changing network topology.

Section II discusses predictive location pattern and agent based network exploration. An example predictive graph is created in Section III and mobile agent migration is described. Section IV illustrates simulation results followed by conclusion and future scope in section V.

2. Predictive Pattern & Agent Based Network Exploration

2.1. Network model

Consider a homogeneous network with a sink node and N sensing nodes with unique id. The values of the parameters sensed from the environment periodically are store in a record on the node with a unique id. The nodes are constrained with energy and computation space and can identify their own locations. The nodes movement is cyclic with some kind of pattern. The sink is actuated accordingly for data communication. When two nodes are in transmission range of each other they temporarily establish the communication link to perform opportunistic data dissemination.

i) Communication opportunity: Two nodes m and have communication opportunity if they are in contact with each other for more than a predetermined threshold.

ii) Path: for a group of nodes if (m1, m2, ... mk-1) if there is an opportunity for communication between m1 -m2, m2 - m3, .. mk-1 - mk, then a path is said to exists between m1 and mk.

Fig.l. .Agent teas e d. Data Gathering in OWSN ~

iii) Cost of data collection: For N sensing nodes, if average k-bits are exchanges during each contact and t is transmission cost then the cost of data collected across mj - mk path is given as in equation (1) below:

Cost = £ (k * t) (1)

2.2. Predictive Location Based Algorithm

This section describes the agent based predictive location algorithm (PLA). When the mobile nodes move around, the contact nodes change defining a new path to the sink. If there is a pattern in the node movement, we can predict the next possible contact between the nodes and sink. PLA has following phases:

• Sink collects network metadata like node movement, contact duration, through mobile agents.

• Sink creates a graph to predict contact duration and location with a node based on metadata.

• Sink is actuated based on the predictive model for collecting sensed data from nodes.

• Data collection points are updated whenever necessary.

2.3. Network Exploration using mobile agents

The algorithm proposed PLA makes use of predictive model for data collection which is built from the information collected by mobile agents. Specifically the mobile agents collect the metadata of each node such as connection history with other nodes or sink node in the network, the contact duration, node residual energy value, and location information of the sensing nodes. Ant-AODV routing protocol for network exploration as the routing protocol [8]. The information gathered is further processed to compute the contact time and location of nodes and sink. Mobile agents are created and destroyed at sink node and they operate autonomously following a unique path. OWSN networks are sparsely populated, and therefore density of spatial nodes is negligible. The network is said to be completely explored if all the N nodes are visited by the mobile agent. The processing time needed by agent at a node is predictable [18]. Mobile agents are adaptable to the surrounding conditions such as latency requirements and energy available at each node. A mobile agent server is established at sink node and it exhibits various functions of agents like inter-agent communication, security features and migration of agents to different nodes and back to sink. Besides, each node has a static agent that maintains routing table on the host node, or finds alternate routes to sink node and stores node statistics. Mobile agent migrates to different nodes and collects the required information from static agent residing on the node. The routing table on mobile nodes is simultaneously updated with the information gathered from mobile agent by the static agent [19].

3. Predictive Graph for Actuating Sink

Mobile agent is first authenticated by the target node and then it processes and fuses the information on the node before carrying it along. Network metadata on sink is regularly updated at certain intervals. The sensing data and the node path history is collected by the agent and passed to the sink. This movement data of nodes is stored at sink in a table as record including parameters as shown in the table 1. The movement table is then frequently updated.

Table 1: Nodes movement table

Time Stamp Nodes Link Contact Duration of Contact

10 m ^ n 5

12 n ^ o 10

19 o ^ p 2

27 p ^ q 15

m ^ r, m ^ o p ^ s_

From the information collected in the table 1, the temporal graph is created to predict contact locations for nodes and sink. For distant node contact the sink can be actuated to move to the appropriate location according to the graph. This results in saving unwanted transmission energy and effectively maximizing network lifetime. 3.1. Creating Temporal Graph

The contact time T probability is calculates as in equation (2):

Tc = ^1/2T - K

Where, T is the current epoch and at k epoch nodes m and n meet. If two nodes do not meet each other then k will become 0. The length of the contact time TL is given as duration for which the nodes m and n meet:

Tl = 1/(T -1)£ p * k, T > 1

p denotes the duration for which nodes m and n are in contact with each other, k denotes the contact time. Consider T =4, and nodes (m, n) encounter at 3rd, 4th epoch then the probability of contact between them is CT = 1/ 8 - 3 + 1/ 8 - 4 = 0.45. if CT meets some threshold (h), then a link is added to the contact graph between m and n. Consider h > 10, from nodes movement table 1. The table 2 shows contact probability CT values at different instances and the corresponding contact graph as shown in Fig. 2.

Table 2: Contact Duration Probability Ct_

Time Stamp

7 10 15

23 32 39

Nodes Link

Contact

n3 n4 n5

Node Location

(17,25) (20,30) (25,35) (40,55) (45,50) (33,40)

(x 10-1)

12 4 22 9 17 15

Fig. 2: Nodes Contact Graph

4. Simulation

4.1 Simulation Environment

Extensive simulations have been performed for comparing the proposed algorithm PLA with algorithm in [7], hereafter known as ADA, and algorithm in [8], known as MAR. The implementation of PLA is done in Matlab. Consider N = 20 sparsely scattered in an area of 160m x 160m. The simulation is performed 30 times to find the average values for each result. A packet size is 512B and 1 packet per 5seconds is generated by a node. Since the communication cost is more important we consider only communication cost.

4.2 Simulation Results

The comparison of algorithms is made on the basis of communication cost for variable network size, radio

m 40 50 Number saf. Nodes

Fig. 3: Messages Transmitted Vs. Network Size Fig. 4: Data Collected (packets) Vs. Network Size

range and moving speed. Finally the impact of network exploration using mobile agents is discussed based on the communication cost and latency [19-20].

A. Network Size

As seen from fig. 3 the increase in network size leads to significant increase in the number of transmission messages. PLA has the smallest number of transmission messages required. While MAR has the highest messages transmitted. PLA reduces the transmission messages by 33% as compared to MAR and is almost similar to ADA when the network size grows to its maximum. As in fig. 4, the data collected by PLA is about 94% on average and is almost 10% higher than those of ADA and MAR. This is because PLA moves along the pattern learnt from mobile agents Further, PLA also buffers less number of messages saving storage space. Also, the sink in PLA is actuated according to the data gathering points in contact graph which reduces energy consumption due to unnecessary transmissions.

B. Varying radio range of nodes

From Fig. 5 & 6, it can be seen that number of message transmitted and number of data packets collected increases with the increasing radio range. PLA is better as compared to ADA and MAR in transmission message as well as data collection with varying range. Radio range when grows maximal, the transmission and

data collection grows to 90%, which is the best performance. Practically, achieving this performance depends on hardware and will be very expensive for each transmitted data unit.

C. Speed of nodes

The nodes in opportunistic networks are mobile and faster they move more nodes they can encounter and

15 2№ 25 g m

Radio Range Radle flan3e

Fig. 5: Radio Range Vs. Messages Transmitted F'£- Radio Range Vs. Data packets Collected

exchange data packets. As seen in Fig. 7 & 8 data transmission and data packets collected increase with the speed of node. When the speed is low, the transmission messages needed by PLA are reduced by almost 33.2% when compared with ADA and MAR. While, the data collection increases by almost 7-10% by PLA. The reduction in message transmission and increased data packets is due to the fact that PLA initially builds a movement table and actuates the sink accordingly. The routing table is built by other algorithms only when nodes come in contact with each other. This further reduces the delay caused in data collection by PLA. Hence, PLA results in considerable energy saving thus maximizing network lifetime.

201--1-1-1-1-1-1-1- -1-1-1-1-1-1-1-

3 3 4 J S 1 8 9 10 g 5 8 7 8 9 10

^'peedtif no'des'{mis] Speed of.nodes.'{»i5i

Fig. 7: Speed Vs. Messages transmitted Fig. 8: Speed Vs. Data packets collected

4.3 Discussion

It can be concluded from the simulation results that by using ant like mobile agents and implementing the ant-AODV on demand routing protocol the route discovery latency and connectivity between nodes or sink is increased. The initial prediction based on metadata simplifies the sink movement to a certain location at a particular instance of time. Since, the routing table is frequently updated the unnecessary transmission of control messages is reduced thus saving energy of the node. Further, as discussed earlier if a node is low on energy the mobile agent does not carry any status packet for transmitting an event. Any node failure can be subsequently detected. Ant-AODV being hybrid and combines goodness of both the algorithms the routing overhead is the least and similar to AODV.

5. Conclusion & Future Scope

In this paper, a mobile agent based predictive algorithm (PLA) is implemented for data collection in opportunistic sensor networks. The PLA is an efficient algorithm and is based on metadata of node movements. This metadata obtained from network exploration is done by mobile agents using ant-AODV routing protocol. The mobile agents send the status packet only when a new event is detected. A contact graph based on contact prediction CT is prepared. If the connection time TL is more than a certain threshold (h) then the communication link is established between the nodes for data exchange. As seen from the experimental results the number of messages transmitted between the nodes is reduced by 33.2% saving considerable energy of the nodes and increasing data collection rate by 10 to 12% when compared with other algorithms like ADA and MAR. In future, PLA implementation considering energy efficiency with certain additional features like data redundancy in OWSN and contact failure probability can be worked on.

References

1. W. Dargie, Christian P., (2010): Fundamentals Of Wireless Sensor Networks: Theory And Practice, Wiley .Series on Wireless Communications and Mobile Computing, pages. 7-10.

2. D. Culler, D. Estrin, M. Srivastava, (2004) "Overview Of Sensor Networks" IEEE Computer Society, August 2004. 41-49.

3. M. Chen, T. Kwon, Y. Yuan, C.M. Leung, (2006) "Mobile Agent Based Wireless Sensor Networks", Journal Of Computers, Vol. 1, No.1.

4. R. C. Shah, S. Roy, S. Jain, and W. Brunette, (2003) "Data MULEs: modeling and analysis of a three-tier architecture for sparse sensor networks," Ad Hoc Networks, vol. 1, no. 2-3, pp. 215- 233.

5. J. Luo, X. Xu, Q. Zhang, (2010) "Delay tolerant event collection in sensor networks with mobile sink," in Proceedings of the 29th INFOCOM '10, pp. 2471-2479.

6. A. Sabharwal, A. Chakrabarti, B. Aazhang, (2003) "Using predictable observer mobility for power efficient design of sensor networks," in Proceedings 2ndIntn'l Conf. on Info. Proc. in Sen. Net., pp. 129-145, Springer.

7. H. Ochiai, H. Ishizuka, Y. Kawakami, and H. Esaki, (2011) "Agent based sensor data gathering for agricultural applications," IEEE Sensors Journal, vol. 11, no. 11, pp. 2861-2868.

8. S. Marwaha, C.K. Tham, D. Srinivasan, (2002) "Mobile Agents based Routing Protocol for Mobile Ad Hoc Networks", Global Telecommunications Conference, 2002. GLOBECOM '02. IEEE Vol.1, pp. 163 - 167.

9. Y. Xu, H. Qi, (2007) "Mobile Agent Migration Modeling and Design for Target Tracking in Wireless Sensor Network", Ad Hoc Networks J., Vol.6,N0.1,pp.1-66.

10. B. Krishnamachari, Y. Yu, V. Prasanna, (2004) "Energy-latency tradeoff for data gathering in wireless sensor networks", Proc. IEEE INF0C0M'04, pp. 244-255.

11. S.J. Lee , W. Su, M. Gerla, (2001) "Mobility prediction & routing in ad hoc wireless networks", International Journal of Network Management, 11, pp. 3-30.

12. A. Lindgren, Doria A., O. Scheln, (2004) "Probabilistic routing in the connected networks", Lect. Notes in Computer Science, pp. 239 -254, Vol. 3126.

13. E. Bulut, B. Szymanski, (2012) "Exploiting friendly relations in mobile social networks", IEEE Transactions on Parallel & Distributed Systems, pp. 2254 - 2265, Vol.23, issue 12.

14. S. Madden, M. Franklin, J. Hellerstein, W. Hong, (2002) "Tiny Aggregation (TAG) for Ad-hoc sensor networks", Procedia, ACM Symposium on Operating System Design and Impplementation .

15. A. Maiwaring, J. Polastre, R. Szewczyk, David Culler, J. Anderson, (2002) "Wireless Sensor Network for habitat monitoring" Procedia, ACM Workshop of Wireless Sensor Networks and their applications, pp. 88 -97.

16. N. Xu, S. Rangwala, S. Chintalapudi, D. Estring (2004) "Wireless Sensor Networks for structural monitoring", Proceedings, ACM Sensys, pp. 13-24.

17. Rupali Rohankar, A L N Rao, (2014) " Agent Based Data Gathering in Opportunistic Wireless Sensor Networks" Proceedings, Intl. Conf. On Computing, Informatics & Networks, pp. 52-54.

18. H. Jiang, (2011) "Prediction or not? An energy-efficient framework for clustering-based data collection in wireless sensor networks", IEEE Transactions on Parallel & Distributed Systems, Vol. 22, No. 6, pp. 1064-1071.

19. L. He, J. Pan, J. Xu, (2013) "A Progressive Approach to Reducing Data Collection Latency in Wireless Sensor Networks with Mobile Elements", IEEE Transactions On Mobile Computing, Vol. 12, No. 7, Pp. 1308-1320.

20. M. Radi, B. Dezfouli, K. A. Bakar, M. Lee (2012), "Multipath Routing in Wireless Sensor Networks Survey and Research Challenges", Sensors 2012, Open Access Article ID 12, 650-685; doi: 10.3390/s120100650.

21. A. Li, J. Xiao (2012): Efficient Data Gathering Algorithm in Wireless Sensor Networks with Optimal-path Mobile Sink, Journal of Computational Information Systems 8: 22 9269-9279

22. D. Francesco, M., Das, S. K., and Anastasi, G. (2011): Data collection in wireless sensor networks with mobile elements: A survey. ACM Trans. Sen. Netw. 8, Issue 1, Article ID 7, 31 pages.

23. G. Anastasi, E. Borgia, M. Conti, and E. Gregori, (2011): A hybrid adaptive protocol for reliable data delivery in WSNs with multiple mobile sinks, Computer Journal, vol. 54, no. 2, pp. 213-229.

24. G. Xing, T. Wang, Z. Xie, W. Jia (2008): "Rendezvous planning in wireless sensor networks with mobile elements", IEEE Trans. Mob. Comp. 7, 12, 1430-1443.

25. H. Ammari (2012): On the problem of k-coverage in mission-oriented mobile wireless sensor networks, Elsevier J. Computer Networks 56, 1935-1950.