FriendFinder: A Privacy Aware Grid Based Location Service

Transcription

1 FriendFinder: A Privacy Aware Grid Based Location Service Ove Andersen Laurynas Siksnys Jeppe R. Thomsen Aalborg University Department of Computer Science Selma Lagerløfs Vej 300, DK-9220 Aalborg Ø Denmark December 8, 2008 Jeppe, Laurynas, Ove FriendFinder 1/24

2 Motivation Related Work Project about a friend finding location based service (LBS). Notify a mobile user when any of his friends are within a certain distance. Dealing with location privacy issue. The server is said to be untrustworthy. Users do not want the LBS to know anything about their positions. Issuing continuous queries on private data. Jeppe, Laurynas, Ove FriendFinder 2/24

3 Motivation Related Work Motivation How can location privacy be obtained if the LBS is untrusted? Can an anonymizer, which will be a single point of attack, be excessed? Is it possible to both retain sufficient privacy requirements and keep communication costs at a minimum? Not much work done on querying private data - most related work only considers private queries on public data. Jeppe, Laurynas, Ove FriendFinder 3/24

4 Motivation Related Work Anonymizers are not Necessary Private Queries in Location Based Services: Anonymizers are not Necessary. Proposes a framework to support private location-dependent queries. Does not require an anonymizer, since it s a single point of attack. Uses cryptographic techniques to ensure privacy. Communication and CPU intense, because of encryption. Jeppe, Laurynas, Ove FriendFinder 4/24

5 Motivation Related Work Buddy Tracking Buddy tracking - efficient proximity detection among mobile friends. Proposes a centralized server and a peer-to-peer method for tracking friends. Using the strips on the peer-to-peer algorithm - ensures privacy and reduces communication cost. Using a quadtree algorithm for the centralized algorithm. Still lot of communication using the peer-to-peer algorithm. Centralized server algorithm.has a lot of overhead and is outperformed when lots of users has joined the service. Privacy is not discussed in this article. Jeppe, Laurynas, Ove FriendFinder 5/24

6 The contribution Solutions for Problems Agreed on the problem and its setting Develop a user location secure proximity detection approach Designed a Friend-finder service Develop a prototype Performed a prototype testing Jeppe, Laurynas, Ove FriendFinder 6/24

7 Solutions for Problems We want a solution for friend-finder LBS Solution must be strong against attempts to intercept exact locations of any user The solution must be practical Jeppe, Laurynas, Ove FriendFinder 7/24

8 Solutions for Problems u 1 u 2 Untrusted LBS server Friends relationship Wireless communication Jeppe, Laurynas, Ove FriendFinder 8/24

9 Example of the Proximity Detection Solutions for Problems LBS server Jeppe, Laurynas, Ove FriendFinder 9/24

10 Example of the Proximity Detection Solutions for Problems LBS server Jeppe, Laurynas, Ove FriendFinder 9/24

11 Example of the Proximity Detection Solutions for Problems LBS server Jeppe, Laurynas, Ove FriendFinder 9/24

12 Example of the Proximity Detection Solutions for Problems LBS server Jeppe, Laurynas, Ove FriendFinder 9/24

13 Example of the Proximity Detection Solutions for Problems LBS server Jeppe, Laurynas, Ove FriendFinder 9/24

14 Privacy-aware Proximity Detection Idea Solutions for Problems 5,b5 d d 4,b4 Col Col+1 3,b3 u 2 a3 a4 a2 a3 2,b2 u 1 Row Row+1 1,b1 b3 b4 b2 b3 0,b0 0,a0 1,a1 2,a2 3,a3 4,a4 5,a5 In Proximity Jeppe, Laurynas, Ove FriendFinder 10/24

15 Privacy-aware Proximity Detection Idea Solutions for Problems 5,b5 d d 4,b4 Col Col+1 3,b3 u 1 u 2 a3 a4 a2 a3 2,b2 Row Row+1 1,b1 b3 b4 b3 b4 0,b0 0,a0 1,a1 2,a2 3,a3 4,a4 5,a5 In Proximity Jeppe, Laurynas, Ove FriendFinder 10/24

16 Privacy-aware Proximity Detection Idea Solutions for Problems 5,b5 d d 4,b4 Col Col+1 3,b3 u 1 a3 a4 a2 a3 2,b2 u 2 Row Row+1 1,b1 b2 b3 b3 b4 0,b0 0,a0 1,a1 2,a2 3,a3 4,a4 5,a5 In Proximity Jeppe, Laurynas, Ove FriendFinder 10/24

17 Privacy-aware Proximity Detection Idea Solutions for Problems 5,b5 d d 4,b4 Col Col+1 3,b3 u 1 a3 a4 a2 a3 2,b2 Row Row+1 1,b1 u 2 b1 b2 b3 b4 0,b0 0,a0 1,a1 2,a2 3,a3 4,a4 5,a5 In Separation Jeppe, Laurynas, Ove FriendFinder 10/24

18 Limitation and solutions of this approach Solutions for Problems Limitations of this approach: 1 It is inefficient for big number of users 2 A proximity detection distance is fixed (2d 2) Solutions for problems: 1 Grouping of friends 2 Dynamic grid approach Jeppe, Laurynas, Ove FriendFinder 11/24

19 Grouping of friends Solutions for Problems The solution: Users are grouped into friend-groups A single grid is assigned for every friend-group Consequences: When user location changes, less secret values must be delivered to LBS server Jeppe, Laurynas, Ove FriendFinder 12/24

20 Grouping of friends Solutions for Problems Friends relationship Friend-groups Jeppe, Laurynas, Ove FriendFinder 12/24

21 Grouping of friends Solutions for Problems Jeppe, Laurynas, Ove FriendFinder 12/24

22 Dynamic Grid Approach Solutions for Problems The solution: Assign a stack of grids with decreasing cell size for every friend-group Extend a basic proximity detection approach to support stack of grids Consequences: Any pair of friends in friend-group will be able to choose a prefered proximity distance from a list Jeppe, Laurynas, Ove FriendFinder 13/24

23 Dynamic Grid Approach Solutions for Problems L prox (u 1,u 2 )=2 u 1 u 2 Level 0 Level 1 Level 2 Level 3 Jeppe, Laurynas, Ove FriendFinder 13/24

24 Dynamic Grid Approach Solutions for Problems L prox (u 1,u 2 )=2 u Switch to L1 Switch to L1 1 u 2 In proximity! Level 0 Level 1 Level 2 Level 3 Jeppe, Laurynas, Ove FriendFinder 13/24

25 Dynamic Grid Approach Solutions for Problems L prox (u 1,u 2 )=2 u 1 u 2 In proximity! Level 0 In separation! Level 1 Level 2 Level 3 Jeppe, Laurynas, Ove FriendFinder 13/24

26 Dynamic Grid Approach Solutions for Problems L prox (u 1,u 2 )=2 u 1 u 2 In proximity! Level 0 Level 1 Level 2 Level 3 Jeppe, Laurynas, Ove FriendFinder 13/24

27 Dynamic Grid Approach Solutions for Problems L prox (u 1,u 2 )=2 u Switch to L1 1 u 2 In proximity! Level 0 Level 1 Level 2 Level 3 Jeppe, Laurynas, Ove FriendFinder 13/24

28 Dynamic Grid Approach Solutions for Problems L prox (u 1,u 2 )=2 u Switch to L2 Switch to L2 1 u 2 In proximity! Level 0 In proximity! Level 1 Level 2 Level 3 Jeppe, Laurynas, Ove FriendFinder 13/24

29 Dynamic Grid Approach Solutions for Problems L prox (u 1,u 2 )=2 u In proximity with u In proximity with u u 2 In proximity! Level 0 In proximity! Level 1 In proximity! Level 2 Level 3 Jeppe, Laurynas, Ove FriendFinder 13/24

30 Friend-finder service designs Designs of the Friend-finder service Testing Methodology Tests on Group Size Test Comparison Two Friend-finder service designs: Basic Design Uses basic proximity detection approach Dynamic Grid Design Uses dynamic grid approach Features: Employs users grouping into friend-groups Assumes an existence of trusted 3rd party server Jeppe, Laurynas, Ove FriendFinder 14/24

31 Testing Methodology Designs of the Friend-finder service Testing Methodology Tests on Group Size Test Comparison System Test Settings 200 users. 20 Position Points per user. Each user only in one group. Proximity level: 6 Four data sets 5 Users per Group (40 Groups). 10 Users per Group (20 Groups). 20 Users per Group (10 Groups). 50 Users per Group (4 Groups). Jeppe, Laurynas, Ove FriendFinder 15/24

32 5 Users in each Group / 40 Groups Designs of the Friend-finder service Testing Methodology Tests on Group Size Test Comparison Jeppe, Laurynas, Ove FriendFinder 16/24

33 10 Users in each Group / 20 Groups Designs of the Friend-finder service Testing Methodology Tests on Group Size Test Comparison Jeppe, Laurynas, Ove FriendFinder 17/24

34 20 Users in each Group / 10 Groups Designs of the Friend-finder service Testing Methodology Tests on Group Size Test Comparison Jeppe, Laurynas, Ove FriendFinder 18/24

35 50 Users in each Group / 4 Groups Designs of the Friend-finder service Testing Methodology Tests on Group Size Test Comparison Jeppe, Laurynas, Ove FriendFinder 19/24

36 Designs of the Friend-finder service Testing Methodology Tests on Group Size Test Comparison Total Messages, Percent-wise distribution Jeppe, Laurynas, Ove FriendFinder 20/24

37 Total Messages Designs of the Friend-finder service Testing Methodology Tests on Group Size Test Comparison Jeppe, Laurynas, Ove FriendFinder 21/24

38 Future Work Questions Novel approach. System has a very low running cost in terms of messages Very resilient to attacks. Jeppe, Laurynas, Ove FriendFinder 22/24

39 Future Work Questions Future Work Possible extensions 1 Different grid cell shapes. Jeppe, Laurynas, Ove FriendFinder 23/24

40 Future Work Questions Future Work Possible extensions 1 Different grid cell shapes. 2 Notification of other users getting in proximity. Jeppe, Laurynas, Ove FriendFinder 23/24

41 Future Work Questions Future Work Possible extensions 1 Different grid cell shapes. 2 Notification of other users getting in proximity. 3 Notification of a user getting in proximity of a static object. Jeppe, Laurynas, Ove FriendFinder 23/24

42 Future Work Questions Future Work Possible extensions 1 Different grid cell shapes. 2 Notification of other users getting in proximity. 3 Notification of a user getting in proximity of a static object. 4 Notification of large groups of friends in proximity. Jeppe, Laurynas, Ove FriendFinder 23/24

43 Future Work Questions Future Work Possible extensions 1 Different grid cell shapes. 2 Notification of other users getting in proximity. 3 Notification of a user getting in proximity of a static object. 4 Notification of large groups of friends in proximity. Other Directions 1 Analyze user-levels on Server to find congested areas. Jeppe, Laurynas, Ove FriendFinder 23/24

44 Future Work Questions Future Work Possible extensions 1 Different grid cell shapes. 2 Notification of other users getting in proximity. 3 Notification of a user getting in proximity of a static object. 4 Notification of large groups of friends in proximity. Other Directions 1 Analyze user-levels on Server to find congested areas. 2 History analysis of user grid patterns, e.g to find carpooling opportunities. Jeppe, Laurynas, Ove FriendFinder 23/24

45 Future Work Questions Future Work Possible extensions 1 Different grid cell shapes. 2 Notification of other users getting in proximity. 3 Notification of a user getting in proximity of a static object. 4 Notification of large groups of friends in proximity. Other Directions 1 Analyze user-levels on Server to find congested areas. 2 History analysis of user grid patterns, e.g to find carpooling opportunities. 3 Make the idea fully or partially peer-to-peer Jeppe, Laurynas, Ove FriendFinder 23/24

46 Future Work Questions Questions Interesting future work? Would you use such a system? How can a group of friends, through an untrusted server, exchange private information (e.g. an encryption key) Jeppe, Laurynas, Ove FriendFinder 24/24