Resisting Sybil Attacks on Peertopeer Marketplaces

1
Resisting Sybil Attacks on Peer-to-peer
Marketplaces

• Jonathan Traupman
• Advisors Prof. Doug Tygar
• Prof. Robert Wilensky

2
Introduction

• Reputation systems are a key component of many
  peer-to-peer systems, especially marketplaces
  (e.g. eBay, Amazon Marketplace)
• In order to work, reputations must have value
• Value also encourages cheating
• Sybil attacks use fluid online identities to
  unfairly increase/decrease a targets reputation
• Very little defenses in deployed systems

3
PageRank-like algorithms

• Demonstrably scalable (Google)
• Less complex than max-flow
• Claim to be sybilproof
• Use graph structure to find sybils
• EigenTrust Kamvar et al. 2003
• PageRank applied to reputation graphs
• Some modifications for negative FB
• Not directly applicable to markets
• No decision procedure

4
EigenTrust not Sybilproof!

• Uniformly distributed startset (Google) clearly
  not sybilproof Clausen 2004
• Use of a single, fixed startset (EigenTrust)
  opens up the possibility of corruption
• Even with a non-uniform startset, sybils can
  capture the random surfer and increase an
  attackers rank

5
Relative Rank

• Transformation of PageRank
• Easy to discriminate between good/bad users
  regardless of experience

EigenTrust
Relative Rank
6
Relative Ranks and Sybils
Effect on master node
Reputation of sybils
7
RAW Random Acyclic Walks

• Changes the random walk process
• Random walk not allowed to revisit nodes on the
  same walk
• Destroys Markov property
• No closed form or iterative algorithm like
  PageRank
• Uses a Monte Carlo simulation (e.g. Fogaras
  Rácz, 2004)
• Fully personalized
• Initial simulation is expensive
• Queries are very fast

8
RAW Results
Effect on master node
Reputation of sybils
9
Summary Conclusion

• Attacking reputation systems
• Can severely compromise effectiveness
• RAW
• Highly sybil-resistant reputation system
• Adapts PageRank-like systems to P2P markets
• Effective robustness
• Cant eliminate attacks
• Can mitigate their effects

10
(No Transcript)