1. Location-Centric Storage for Wireless Sensor Networks Kai Xingn 1, Xiuzhen Cheng 1, and Jiang Li 2 1 Department of Computer Science, The George Washington University 2 Department of Systems & Computer Science, Howard University The 2nd IEEE International Conference on Mobile Ad-hoc and Sensor Systems MASS 2005 Reporter: Shin-Wei Ho

2. 2 Outline Introduction Example Applications Location-Centric Storage Performance Analysis Simulation Conclusion

3. 3 Introduction Nevertheless, sensor networks pose many new challenges. One of the challenges is how to  Store data efficiently to facilitate user query.  On-demand warning across the entire sensor network.

4. 4 Introduction(cont ’ d) There exists three canonical data storage methods  Local Storage (LS)  External Storage (ES)  Data-Centric Storage (DCS) These studies indicate that no one outperforms the other two in all situations.

5. 5 Introduction(cont ’ d) In fact, none of these methods targets the application scenarios considered in our LCS design. For example, on-demand warning requires  Zero delay  High reliability

6. 6 Introduction(cont ’ d) Location-centric storage (LCS),  Efficiently disseminate aggregated data based on the intensity of the data.  On-demand Warring Applications

7. Example Applications Context-Dependent Information Dissemination for Pervasive Computing On-Demand Warning in Surveillance Sensor Networks Roadway Safety Warning

8. 8 Example Applications -- Context-Dependent Information Dissemination for Pervasive Computing “The Computer for the 21st Century”, 1991 Where is the most closest gas station? I would like to pay $X.

9. 9 Example Applications -- On-Demand Warning in Surveillance Sensor Networks Enemy Allied Force

10. 10 Example Application -- Roadway Safety Warning “Zero Fatality, Zero Delay”, the World Congress on ITS (Intelligent Transportation Systems and Services) Car crashes Where should I go ?

11. Location-Centric Storage

12. 12 Location-Centric Storage Assumption  Sensors can obtain their own geometric coordinates (Sx, Sy) using GPS or other techniques.  A robust broadcasting protocol is in place such that event records can be properly disseminated.

13. 13 Location-Centric Storage When detecting an event, the home sensor creates a record with the following five fields:  The time indicating when the event occurs.  The location (i.e. the coordinates (Sx, Sy)) of the event. For simplicity, we assume an event collocates with its home sensor.  An integral intensity value (σ) that characterizes the event. Intensity values are application-specific. Ex: the time needed to clear the road in highway safety warning.  A Time-To-Live (TTL) as the expiration time (relative to the current moment) of the record.  The event type bearing other information of the event.

14. 14 Location-Centric Storage Event

15. 15 Location-Centric Storage Event 1 3

16. 16 Location-Centric Storage Event 1 3

17. 17 Location-Centric Storage Event 1 3 Query User

18. 18 Location-Centric Storage(cont ’ d)

19. 19 Performance Analysis

20. 20 Performance Analysis(cont ’ d) Store both data oddeven 11

21. 21 Performance Analysis(cont ’ d) 11 Store both data oddeven Contradicts 2

22. 22 Performance Analysis(cont ’ d) There are at most 4 different X coordinates. The same argument holds true for the Y coordinate. There fore there are at most 16 pairs of coordinates at which the nodes store both records.

23. 23 Performance Analysis(cont ’ d)

24. 24 Performance Analysis(cont ’ d) Remark: Theorem 5.1  No matter how big the intensity value is, there will be a fixed number of sensors that store the same records. (as long as the two event locations are not colinear in X and Y directions)

25. 25 Performance Analysis(cont ’ d)

26. 26 Performance Analysis(cont ’ d) Remark: Theorem 5.2  The average number of records stored in each node at any time is independent of the network size.  Therefore, the protocol is efficient Storage requirement Power consumption Highly Scalable

27. 27 Performance Analysis(cont ’ d)

28. 28 Performance Analysis(cont ’ d)

29. 29 Performance Analysis(cont ’ d) Remark: Theorem 5.2 & 5.3  LCS is fair to all nodes in storage space. Records are uniformly and independently generated  This is an intrinsic difference compared with DCS.

30. 30 Performance Analysis(cont ’ d)

31. 31 Performance Analysis(cont ’ d)

32. 32 Performance Analysis(cont ’ d)

33. 33 Performance Analysis(cont ’ d)

34. 34 Performance Analysis(cont ’ d)

35. 35 Performance Analysis(cont ’ d) Remark: Theorem 5.4  When the user resides in the broadcast region of an event, the query distance is no more than distance between the user and the home location of this event.  A user can only be notified of the events that occur within certain distance from the user.

36. Simulation LCS Performance Evaluation Comparative Study

37. 37 Simulation -- LCS Performance Evaluation Simulation setup  200 seconds  λ=2 i x 10 -3, where i is one of 0,…,8  The intensity σ is randomly chosen from [0, 6]  The TTL value is randomly chosen from [1, 100] in seconds.  The TTL value decreases by 1 every second.  A record is removed when it’s TTL value reaches zero.

38. 38 Simulation -- LCS Performance Evaluation Max-vs-averge storage ratio:

39. 39 Simulation -- LCS Performance Evaluation

40. 40 Simulation -- LCS Performance Evaluation

41. 41 Simulation -- LCS Performance Evaluation

42. 42 Simulation -- Comparative Study Comparison:  External Storage  Local Storage The authors did not compare LCS with DCS  Target different application scenarios  Employ a totally different set of input parameters.

43. 43 Simulation -- Comparative Study For example, the message overhead in DCS depends on  The number of event types  The hash function exploited But in LCS, events are stored and disseminated based on its home location and its characteristics  Seriousness  Price  Intention Therefore, the authors found that it is almost impossible to design a simulation study for fairly comparing LCS and DCS.

44. 44 Simulation -- Comparative Study The total number of messages generated vs. The network size The network size (N) = 40000

45. 45 Simulation -- Comparative Study The total number of messages generated vs. The number of quires The number of queries (Q) = 50

46. 46 Conclusion LCS: A novel distributed location-centric data storage protocol for sensor networks. The protocol has many nice features, as indicated by theoretical performance analysis and simulation study. Several simple application scenarios of LCS  Safety warning in highway sensor networks  On-demand warning in surveillance networks  context-dependent information mining in pervasive computing.

47. Thank you ! Question ?