Kai Li Division of Computer Science University of Central Florida

1. Mobile Data Collection Networks for Wireless Sensor Networks* Kai Li Division of Computer Science University of Central Florida

2. Traditional Wireless Sensor Networks Wireless Sensor Networks are composed of a large number of small devices (irreplaceable in many applications), called wireless sensors, which are normally distributed in an ad hoc manner. • Wireless sensors gather information, such as pressure, humidity, temperature, speed etc. • Wireless sensors typically share some common characteristics, such as small size, low power, low cost etc.

3. Applications • Military • Battle damage assessment, nuclear, biological and chemical attack detection, etc. • Environmental • Forest fire detection, habitat monitoring, etc. • Health applications • Patients monitoring, drug administration, etc. • Industry • Production temperature, humidity, pressure control

4. Wireless Sensor Networks . . . Internet Sink/Base Station Wireless Sensor

5. Data Collection in Traditional WSNs • Data Collection is of paramount importance • sensing data needs to be routed to the sink or base-station (sometimes further transmitted over the Internet through them) for further analysis or applications. • Data transmission over the ad hoc network formed by resource-constrained sensors • data generated at each sensor can only reach their neighbors within communication range • data go through multiple sensors on their way to the sink

6. Data Collection Issues in traditional WSNs Communication is a major energy consumer for those energy-constrained sensors • Multi-hop communication → sensors assume dual roles: data source and data forwarder • Fixed routing path (towards static sink) → sensors in the neighborhood of the sink will deplete their energy faster (“energy hole problem”), which render the WSN dysfunctional prematurely

7. WSNs with Mobile Elements Adding Mobile Elements is considered to be a promising solution to the aforementioned problem. Existing approach includes • Mobile Sink Approach • Mobile Messenger Approach

8. Mobile Sink Approach • Sinks moves to different locations in the network field during WSN lifetime • Sojourn for a time interval at each location • When sojourning at each site, routing path of sensors are updated and traffic is redirected towards the current sink site Mobile Sink

9. Mobile Sink Approach • Advantages • The “neighborhood” of the sink does not remain unchanged any more, thus distributing the burden of those sensors over the whole network, preventing premature cessation of network operation. • Limitations • Sensors still do multi-hop communication • Mobile sinks are not feasible for some applications (e.g. it’s not possible for them to have access to Internet in harsh environments) • Not scalable for large scale WSNs

10. Mobile Messenger Approach • Sinks are static • Mobile messengers start out from sink site, following a path, to visit each sensor • Sensors upload their data to the messenger (in a single hop ) when they approach. • Mobile messengers go back to he sink to deliver the collected data Sink Mobile Messenger

11. Mobile Messenger Approach • Advantages • sensors transmit data in a single hop, and do not forward data for other sensors • low communication overhead w.r.t. routing • energy consumption at each sensor is greatly reduced. • Limitations • every sensor has to wait a long time for the messenger to approach, thus resulting in long or even unpredictable latency • long wait may result in sensor buffer overflows, thus reducing data delivery ratio

12. Our Approach—the MDCNet • We propose a new data collection paradigm—the Mobile Data Collection Network (MDCNet) for WSNs that features • Energy efficiency (single-hop communication model) • Short latency (compared with mobile messenger approach) • High data delivery ratio

13. MDCNet - A new data collection paradigm • MDCNet is a self-deployedmesh network • formed by Mobile Relay Nodes (MRNs) (e.g. mini robot, autonomous vehicles), each serving a certain number of sensors • with partial and intermittent connection among MRNs (MRN only communicate with other neighboring MRNs when it needs to transmit data) • through which data could be uploaded by sensors in a single hop and electronically transmitted towards the sink or base-station

14. MDCNet - A new data collection paradigm A Conceptual View of MDCNet Sensor Mobile Relay Node

15. MDCNet - A new data collection paradigm • The MDCNet is designed with the following three major considerations • the number of sensors each MRN serves should be balanced to reduce sensor contention • sensor’s data should be collected in a timely manner to avoid data loss caused by sensor buffer overflows • data relay among MRNs should conform to a reliable protocol to guarantee safe arrival at the sink • Each of the above three requirements is satisfied by corresponding techniques

16. MDCNet - A new data collection paradigm • Load-balanced Area Partitioning • Deterministic Area Partitioning (DAP) • Adaptive Search and Conquer (ASC) • Local Data Collection Protocol • Data Relay Protocol

17. Load-balanced Area Partitioning: DAP approach • Assumption • sensor locations are known a priori • Centralized administration of MRN deployment is possible • Simple partition • Evenly divide the region into several parts • Associate each partition with a mobile relay node • MRN moves back and forth in a snake-like pattern to collect data from sensors

18. Load-balanced Area Partitioning: ASC approach The Adaptive Search and Conquer (ASC) approach has the following characteristics • It assumes no knowledge of sensor locations • No centralized deployment (i.e. decentralized self-deployment) is required • MRNs cooperatively and incrementally search and conquer different regions until the whole WSN has been covered

19. Load-balanced Area Partitioning: ASC approach • These Target Areas will be explored by other MRNs, upon their receipt of the NOTICE message from the MRN that claimed the bottom left region as its Service Area 1 2 • 4th Expansion by 2R 2nd Expansion by 2R • After conquer the area as its service area, the MRN will move in a snake-like pattern the service area to collect data from sensors 3 • MRNs set a random timer in the beginning, the MRN whose timer expires first will be the first one to start out and at the same time send out a TIMEOUT message. Others will cancel their timer upon receipt of this message. L+2R 1st Expansion by 2R 3rd Expansion by 2R L+2R

20. Local Data Collection Protocol • The sensor is not served within a predefined time frame . (This is to make sure that a sensor does not get repetitive service when MRN is within its communication range ) Mobile Relay Node Sensor 1. HELLO message 3. START message Move& broadcast If satisfy service requirement, then stop and set a wait timer time Cancel timer &receive data . . . 4. Start sending data packet 2. ACK message 5. FINISH message timeout End Session

21. Data Relay Protocol Data relay hierarchy of DAP Sink 7 8 9 1 4 5 6 4 5 2 2 1 3 6 3 6 7 8 O(Sink Location)

22. Data Relay Protocol In the ASC approach, the data relay hierarchy is automatically established as MRNs cooperatively search the sensor field. Sink 1 2 4 3 7 8 6 5 9 6 9 2 8 7 1 3 5 4 O(Sink Location)

23. Data Relay Protocol MRN (child) MRN (parent) Serving sensors HELP message start sending data packet FINISHmessage stop serving & seek help stop and set a wait timer Cancel timer &receive data transmit data . . . timeout Ready message Resume serving sensors Session end time

24. Simulation Environment • Simulation Environment: NS2 • Network Topology: 100m by 100m • Sensor Nodes • data generation rate:10bit very 0.1 seconds • buffer capacity: 10KB • communication radius: 7m • random distribution • Mobile Relay Nodes: • moving speed: 2m/s • communication radius: 40m

25. Performance Metrics • Data delivery ratio ratio of the data packets delivered to the sink and the data packets generated by the sensors • Latency sensors’ average service interval by mobile relay nodes • Deployment time (for ASC approach) average searching time of mobile relay nodes

26. Effect of Sensor Density When there are less than 300 sensors, MRN has not reach full load. Thus, in our setting 300 is the full load point • DAP cannot dynamically adjust the size of its service area with the increase of number of sensors (i.e. its load keep increasing) • ASC tries to keep given workload (number of sensors to serve), and adjust its service area size accordingly When sensors are very sparse, the 1st MRN takes a long time to conquer a service area, which dominates the deployment time As sensors density increases, more MRNs are dispatched, which contributes to the gradual increase in deployment time

27. Effect of Load factor (ASC approach) performance does not degrade anymore, because partition number has reached minimum • When load factor exceeds 60, the number of partitions can not decrease any more (i.e. at least 4 parts) • Taking all factors (latency, data delivery ratio, cost in terms of number of MRNs needed) into account, the optimal load factor in our setting should be between 40 to 50. Too many sensors per MRN. They are overloaded Latency shows rapid increase around 50, as a result of overload

28. Conclusion and Future Work • A Mobile Data Collection Network has many advantages: • Single-hop communication saves sensors energy to the largest extent • Electronic transmission of sensing data over MDCNet contributes to shorter latency • Self-deployment and distributed cooperation of MRNs is fit for large scale WSNs • Future Work • Further reduce assumptions such as location awareness (i.e. GPSs) • Generalization to irregular-shaped service areas • Consideration of obstacles and/or constrained path

29. kienhua@cs.ucf.edu