1. Dissuasive Methods Against Cheaters in Distributed Systems Kévin Huguenin Ph.D. defense, December 10 th 2010 TexPoint fonts used in EMF. Read the TexPoint manual before you delete this box.: A AAA A A

2. Distributed systems and Models cable fiber losses wireless losses computer Van Neuman 2 upon receive(x) y= x + y send y upon receive(x) y= x + y send y crashes, bugs

3. Fault models and Approaches bad very bad Crashes, losses Byzantine faults Dishonest users hardware peoplerational conspros 3

4. 4 Approaches: example Masking more roads Preventing speed governing Dissuading speed traps & fines

5. Detection Punishment Dissuasive Approach: How To 5

6. Human nature ◦Collaborative dissemination Social nature ◦Computation in Social Networks Outline 6

7. Collaborative dissemination 7

8. Epidemics Collaborative Dissemination Principle 8

9. Collaborative Dissemination Attacks and Dissuasion 9 Propose 12, 22, 31 (free SMS) 4x Request 12, 22 (free SMS) 4x Send 12, 22 (MMS) period Receive 12, 22, 31 4 guests 3x Propose lessLess guestsSend lessBias selection

10. Verifications Decision Collaborative Dissemination Challenges and Solution A B C I contacted C Yes he didDid he? Did B send what I asked? 0 punished score log CABDEFGH ZYZ Ok  No ZYZ Propose lessLess guestsSend lessBias selection 10

11. Social networks 11

12. E.g., polling ◦“Should partners be invited?” Computation in Social Networks 12 Yes! But it sounds cheesy… No! What if my partner had to learn? No way! I have to prevent this from happening But what if people find out?

13. Computation in Social Networks A new model of entities ◦Reputation ◦Privacy Computation ◦Set of entities ◦Input values ◦Compute ? 13

14. The S 3 problem Definition Scalable and Secure distributed computations in Social networks 14

15. The S 3 problem Definition: Candidate S 3 candidate quadruple where is an arbitrary set, is a metric space and is a symmetric function 15

16. The S 3 problem Definition: Scalability -Scalability message, spatial and computational complexities are 16

17. The S 3 problem Definition: Accuracy -Accuracy where 17

18. The S 3 problem Definition: Privacy Probabilistic anonymity For any trace generated from a non-trivial configuration For any coalition of faulty nodes For any non-faulty node Exists another trace (generated from ) s.t. 18

19. The S 3 problem Definition: Privacy Privacy: probabilistic anonymity ◦Discard trivial input configurations ◦(strong): trivial = inputs can be inferred from output alone ◦(weak): trivial = all inputs are equal 19

20. The S 3 problem Definition: faults Model of faulty-nodes: ◦Deviate from the protocol BUT ◦Never behave in such a way that their misbehavior is detected with certainty 20

21. Solving (√,√,weak)-S 3 Architecture groups of size (ring) 21

22. +1 +1 +2 +1 +4 22 Solving (√,√,weak)-S 3 Demo: Polling

23. Solving (√,√,weak)-S 3 Theorem: ◦The protocol S 3 computes aggregation functions for 23

24. Solving (√,√,weak)-S 3 Proof: Scalability Messages Memory 24

25. +1 25 Solving (√,√,weak)-S 3 Proof: Accuracy Attack: Voting

26. +1 +5 +1 -1  +1 26 Solving (√,√,weak)-S 3 Proof: Accuracy Attack: Counting

27. 27 Attack: Token corruption Solving (√,√,weak)-S 3 Proof: Accuracy

28. Impact of one faulty-node: ◦Voting: ◦Counting: ◦Aggregation along the ring: none Relative error is 28 Solving (√,√,weak)-S 3 Proof: Accuracy

29. Exists (w.h.p.) two equivalent traces where inputs are swapped Group with no faulty node p p q q 29 Solving (√,√,weak)-S 3 Proof: Privacy Output unchanged

30. Generalization Can compute the multi-set of inputs Can compute any regular function with a fixed input set 30

31. Conclusion & perspectives User-centric models practical solutions Boundaries Massively multi-player online games 31