1. 1 Credentials Revocation in Vehicular Networks: Design & Evaluation Ghita Mezzour Panos Papadimitratos

2. 2 Overview Introduction Regional CRL CRL broadcast at low rate Results Conclusion

3. 3 System model – General CA R Root CA CA A Region A CA B Region B CA C Region C

4. 4 System model – Regional CA Certification authority (CA) Road Side Units (RSUs)  Wired communication with the CA  Wireless communication with vehicles Each vehicle has  A unique identity V  A pair of private and public keys {k V, K V }  A certificate Cert {V, K V, L f, attr} CA Each message  Signed  Accompagnied by the sender’s cert  Accepted only within the region of the responsible CA

5. 5 Problem statement Vehicles can ‘misbehave’  Attackers : tampered software and hardware  Mulfunctioning devices  Stolen vehicles  Administrative reasons Once detected, it is necessary to revoke their credentials

6. 6 Challenges & Constraints Scalability  Large number of revoked vehicles  Large number of equipped vehicles that need the revocation information Communication between RSUs and vehicles  Non-pervasive  Short contact times  Bandwidth constrained

7. 7 Classical credential management schemes (1/2) Certificates revocation lists (CRLs)  Long lived certificates e.g. 1 year  CRL contains not yet expired certificates that were revoked  CA periodically issues a CRL  CRL can become very large

8. 8 Classical credential management schemes (2/2) CRL and  -CRL  CRL issued e.g every month   -CRL issued e.g every day or week  Problem if some revocation piece is not received Short lived certificates  Short cert lifetime e.g. 1 day or 1 week  Get a new certificate when certificate expires  Overhead of issuing new Certs

9. 9 Related work [RPAJH JSAC 2007] propose two revocation schemes  Revocation of the Trusted Component (RTC) Reduces the number of Cert in the CRL  Requires to geographically localize vehicles  Revocation using Compressed Certificate Revocation Lists (RC 2 RL) CRLs are lossly compressed using Bloom Filters Scalable  Some legitimate nodes may get revoked as well

10. 10 CRL based approach Widely used and tested in many systems Robust No false positive  Scalability issues

11. 11 Agenda Introduction Regional CRL CRL broadcast at low rate Results Conclusion

12. 12 CRL size Expected CRL size E(N CRL ) = N v * p * r * (L f /2) N v Total number of vehicles p Percentage of equipped vehicles r Percentage of revoked vehicles per day L f Certificate lifetime France N v = 5.10 6, 3.10 5 stolen vehicles per year => 100 – 200 KBytes

13. 13 Foreigner Cert (1/2) { a, K a, fr} B {K B } Root {a, K a } A B A { a, K a, fr} B {a, K a } A Regular Cert of vehicle a by CA A {a, K a, fr} B Foreigner Cert of vehicle a by CA B

14. 14 Foreigner Cert (2/2) Delivery protocol Characteristics  CAs have global revocation information  Need to present a valid regular Cert  Short lifetime  Only valid inside B a B {a, current time} ka, {a, Ka} A {a, K a, fr} B, {B, K B } Root {a, ACK, current time} ka If a CRL A

15. 15 Revocation – Misbehavior in the home region A a Insert {a} in CRL A B {a, K a } A a in CRL A

16. 16 Misbehavior of a Revocation – Mibehavior in a host region B Insert {a,fr} in CRL B A Insert {a} in CRL A C {a, K a } A a in CRL A {a, K a } A {a, K a, fr} B a not in CRL A

17. 17 Foreigner Cert lifetime Short lifetime  Journeys in host regions are typically short One week or one month lifetime  Small overhead of issuing foreigner Certs  Foreigner Certs in CRLs  Periodical check of regular Certs that were issued a foreigner Cert One day lifetime  Overhead of issuing new foreigner Certs if long journey  Implicit revocation: no foreigner Certs in CRLs

18. 18 Summary CAs need global revocation information Vehicles needs regional revocation information CRL of a region A contains  Certs of region A  Foreigner Certs of foreign vehicles that misbehaved while in A Small number Short lifetime => Short CRLs

19. 19 Agenda Introduction Regional CRL CRL broadcast at low rate Results Conclusion

20. 20 CA - vehicles communication Satellites  Wide coverage  Satellite receivers may not be compulsory  Low and expensive bandwidth  Satellite usage loyalties Cell phones  Expensive WLAN, buses  City infrastructure  Present in remote areas RSUs  Non-pervasive  Short contact times  Bandwidth constrained  VANET infrastructure

21. 21 Background - Erasure codes Erasure codes for data transmission  The data is cut into M pieces  The blocks are encoded into N >> M encoding pieces  Reception of any slightly larger subset of pieces is enough to recover the original data

22. 22 Background – Fountain codes Fountain codes e.g. Raptor code for data transmission  The data is cut into M pieces  The blocks are encoded into a potentially limitless encoded symboly  Reception of any (1 +  )M subset of pieces is enough to recover the data

23. 23 How it works (1/2) CRL is encoded using an Erasure code / fountain code RSUs broadcast the encoded CRL pieces Vehicles collect CRL pieces as they encounter RSUs Vehicles recover the entire CRL when they receive enough pieces

24. 24 How it works (2/2) Erasure code: RSUs  Shuffles the N pieces pseudorandomly  Broadcasts them  When the N pieces are over, it starts the broadcast again Fountain code: RSUs  Broadcast the encoded pieces

25. 25 Summary Broadcast based on Erasure/fountain codes  No collaboration between RSUs  No synchronized Broadcast schedule Requirements  Vehicles complete the CRL reception fast  Small overhead to the system

26. 26 Agenda Introduction Regional CRL CRL broadcast at low rate Results Conclusion

27. 27 Number of pieces to receive Number of pieces to be received to complete the reception of the CRL (99.99% confidence)  Erasure codes M Number of uncoded CRL pieces N Number of encoded CRL pieces  Raptor code M Number of CRL pieces  Code parameter affects the compltexity

28. 28 CRL bcst Bandwidth B RSU CA v R D R Time to complete the CRL Total time to complete the CRL P tot Number of pieces to be received sz Size of a CRL piece + overhead v Speed of the vehicle B Bandwidth of the CRL broadcast R Range of RSUs D Distance between encountering RSUs

29. 29 Coding schemes comparison Total number of pieces to be received to complete the reception of the CRL (99.99% cofidence) vs. Number of pieces in the CRL

30. 30 Broadcast bandwidth – RSU range Time duration to complete the reception of the CRL vs. CRL broadcast bandwidth 200 KB CRL, D = 500m, v = 60 km/h

31. 31 Vehicle speed – Distance between RSUs Time duration to complete the CRL vs. vehicle speed 200KB CRL, B = 3KBytes/s, R = 300m

32. 32 City vs. Highway scenario City scenario V = 40 km/h, dense RSUs Highway scenario V = 120 km/h, less dense RSUs 200 KB CRL

33. 33 References M. Raya, P. Papadimitratos, I. Aad, D. Jungels, and J. –P. Hubaux, Eviction of Misbehaving and Faulty Nodes in Vehicular Networks, IEEE Journal on Selected Areas in Communications (JSAC), Special Issue on Vehicular Network, 4th Quarter, 2007 Ronald L. Rivest. Can we eliminate certificate revocation lists? In Rafael Hirschfeld, editor, Financial Cryptography, volume 1465, page 178-183, anguilla, British West Indies, February 1998. Springer

34. 34 Conclusion Revocation is crucial for VANET Challenging due to special environmental constraints CRL approach can be adapted  Regional CRL (Foreigner Certs)  Low rate bandwidth (Erasure/fountain codes)