1. Auction-based Model of BitTorrent Seminar in Distributed Computing 2010 ETH Zurich Yu Li

2. What is BitTorrent

3. How to download using BitTorrent STEP 1: Select Torrent File STEP 2: Download Torrent

4. How to download using BitTorrent STEP 3: Configuration STEP 4: Download File

5. Downloading procedure Individuals View –Seed –Leecher

6. Downloading procedure System View –I upload and download at the same time … Me

7. Auction-based Model [D. Levin, K. LaCurts, 2008] "BitTorrent is an Auction: Analyzing and Improving BitTorrent’s Incentives" (Q1) Do I upload to everyone? … Me … Me

8. Auction-based Model (Q2) Do I treat every peer equally? … Me (6Mbps) … Me (6Mbps) 2 1 426 322 4 2 6 2

9. Auction-based Model (Q3) Do I ignore others? … Me (6Mbps) … Me (6Mbps) 2 2 426 22 1.5 4 6 0.1 0.2 0.1 0.2 1.5

10. Auction-based Model Bids are uploads of round n-1 Auction clearing Me (Bandwidth W) 1812321 Round n-1 Round n 18 Me (Bandwidth W) 12 321 W/S Highest S-1 1 random peer W/S Select S peers (S=4) W/S

11. Sybil Attack Normal Bidder 3 1218 12 18Mbps Auctioneer (Me) provides 72Mbps bandwidth in total “Red” bidder gets 18Mbps n-1 round n round

12. Sybil Attack Selfish Bidder 34 18 1244 12 “Red” bidder sybil gets 2×18Mbps n-1 round n round 18Mbps Auctioneer (Me) provides 72Mbps bandwidth in total

13. Review of Question 2 (Q2) Do I treat every peer equally? … Me (6Mbps) … Me (6Mbps) 2 1 426 322 4 2 6 2

14. Countermeasure: Proportional share auction Proportional share 3 1218 12 36Mbps24Mbps 6 42 n-1 round n round “Red” bidder gets 24Mbps

15. Countermeasure: Proportional share auction Sybil attack resilient 36Mbps 8Mbps 6 42 n-1 round n round 34 18 1244 12 8Mbps “Red” bidder sybil gets 3×8=24Mbps

16. Comparison of Auction-based &Propshare Normal client Selfish client (using sybil) Standard Bittorrent 18 Mbps2×18 Mbps Propshare24 Mbps Propshare is Sybil proof

17. Which one is better for me? Under-reporting Me A X Y B X Y X Y A XY B XYXY A X Y B X Y X Y A X Y B X Y X Y A X Y B X Y X Y X X Under- reporting

18. Under-reporting algorithm –Run algr. when peer is about to lose interest. –Determine the most common piece among available X Y Z X YZ X Y Z XY Z Tom Peer 1 Peer 2 Peer 3 X YZ Alice X Y Z X Y Z X Y Z Peer 1 Peer 2 Peer 3 X Y Z Tom X YZ Alice

19. Under-reporting

20. Countermeasure: Gossiping Me X Y Z Z X Y Z Costs overhead B A

21. Countermeasure: TrInc [D.Levin, J.R. Douceur 2009] "TrInc: Small Trusted Hardware for Large Distributed Systems" TrInc (Trusted Incrementer) –Prevents lying –Decreases overhead TPM (Trusted Platform Module) by Intel

22. Countermeasure: TrInc System equipped with TrInc

23. Countermeasure: TrInc Individual equipped with TrInc

24. Countermeasure: TrInc How does TrInc work? Old Counter Value New Counter Value Encryption BitField BitField + New Counter Value + Old Counter Value

25. Countermeasure: TrInc Input –New Counter Value –BitField Output –Attestation Sample Output A B C … DYZ A B C … DYZ  A B C DEF G 3434

26. Countermeasure Me X Y Z Z X Y Z B A

27. Countermeasure: TrInc X 2  3 YZT X 1  2 YZT X 0  1 YZT X Y Z X YZT X 1  2 YZT X 2  3 YZT Bitfield Request Piece Providing Nonce X 2  3 YZT 3  3 Nonce 3  3 Nonce Me A B C D

28. Countermeasure: TrInc Individual equipped with TrInc

29. Auction-based Model TrInc Q & A