Supporting Reputationbased Trust for Peertopeer Electronic Communities

1
Supporting Reputation-based Trust for
Peer-to-peer Electronic Communities

• Li Xiong, Mudhakar Srivatsa, Ling Liu
• Distributed Data Intensive System Lab
• College of Computing

2
Outline

• Introduction
• Trust Model
• Implementation Issues
• Experiments and Results
• Conclusions Future Work

3
P2P Electronic Communities
4
Problem Statement

• Risks and threats in P2P
• Gnutella Example
• No trusted third parties
• Main security techniques without trusted third
  parties
• Micropayments
• Reputation based trust systems - building trust
  through social control

5
Trust Definitions

• McKnight et al.
• Trusting belief is the extent to which a peer
  believes that another peer is trustworthy in this
  situation.
• Trustworthy means one is willing and able to act
  in the other entitys best interest.
• Consistency, Willingness, Competency, and Honesty

6
Reputation Systems - Challenges

• Effective trust model
• Accurately and Effectively capture the
  trustworthiness of peers
• Ability to cope with malicious behaviors of peers
• Ability to adapt to different communities and
  situations
• Implementation
• Decentralized implementation
• Secure implementation
• Experimental evaluation

7
Outline Trust Model

• Introduction
• Trust Model
• Common problems with current reputation systems
• Proposed trust parameters
• Proposed trust metrics
• Implementation Issues
• Experiments and Results
• Conclusions Future Work

8
Trust Model Issues

• Dishonest feedback
• Differentiate between honest and non-credible
  feedback
• Various Contexts
• Incentive to Rate
• Malicious/Strategic behavior of peers
• Alter node behavior strategically and dynamically

9
PeerTrust Parameters

• Feedbacks in terms of amount of satisfaction
• Feedback scope number of transactions
• Feedback credibility
• Trust value based
• Similarity based
• Adaptive transaction context factor
• transaction size
• transaction category
• Adaptive community context factor
• provide incentives for rating others
• Utilized pre-trusted peers or trust authorities

10
General and Basic Trust Metric

• General Metric
• Basic Metric

11
Handling Dishonest Feedback

• Conventional reputation metric
• Average based
• PeerTrust model
• Feedback credibility to differentiate credible
  and non-credible feedback
• Credibility measures
• Trust Value Based
• Feedback Similarity Based

12
Trust value based credibility measure (TVM)

• CRTVM(u) TV(u)
• Assumptions
• Untrustworthy nodes are likely to submit false
  feedbacks
• Trustworthy nodes are likely to be more honest
  questionable
• Problems
• large population of malicious nodes
• collusions

13
Personalized Similarity based credibility Measure
(PSM)

• CRwPSM(u) Sim(u, w)
• Intuition
• Peers who file similar ratings
• Similarity measures
• Vector based cosine measure
• Root mean square based distance measure
• Benefit
• Personalized similar raters are given more
  weight
• Handles large fractions of malicious nodes
• Handles malicious cliques very well

14
Strategic dynamic behavior

• Issues
• Misbehave after earning high reputation
• Alternate between good and bad behavior at
  regular or arbitrary frequencies
• Desired properties
• Reflect the dynamic behavior of peers quickly
• Hard to build, easy to drop differentiate
  improvement and decrease of behavior
• Reflect consistent behavior of peers
• Tolerate occasional unintentional errors

15
Handling strategic dynamic behavior - PID Model

• Rn(t) Reputation-based trust value of node n at
  time t computed using feedback ratings
• TVn(t) a Rn(t)
• ß ?t0t Rn(x) dx
• ? d/dx (Rn(x)) xt

16
Incorporating History

• Assume trust value of node n is available for the
  last maxH intervals
• Hni ?k1maxH Rni-k wk / ?k1maxH wk
• Optimistic Vs pessimistic weights
• wk ?k-1 (exponentially weighted sum)
• wk 1/Rni-k (inverse trust value weighted sum)

17
Reflecting Fluctuations

• Dni Rni Hni
• Vni a Rni ß Hni
• ?(Dni) Dni
• ?(x) ?1 if x 0, ?2 otherwise
• ?1 lt ß lt ?2
• TVni can now handle steady and sudden
  behavioral changes

18
Outline - PeerTrust Implementation

• Introduction
• Trust Model
• Trust Implementation Strategies
• System Architecture and Trust Data Location
• Scalable and efficient trust data lookup
• Secure Trust Data Communication
• Experiments and Results
• Conclusions Future Work

19
PeerTrust System Architecture
20
Trust Data Location
21
Secure Trust Data Dissemination

• PKI based scheme
• Confidentiality
• Integrity
• Replication
• Data loss
• Peer dynamics

22
Outline - Experimental Results

• Introduction
• Trust Model
• Trust Implementation Issues
• Experiments and Results
• General Simulation Setting
• Experiment 1 Robustness against Dishonest
  feedback
• Experiment 2 - Benefit of PeerTrust peer
  selection to P2P systems
• Experiment 3 Robustness against strategic
  dynamic behavior
• Conclusions Future Work

23
General Simulation Setting

• P2P System Model
• Fixed number of peers
• Percentage of untrustworthy peers
• An untrustworthy peer acts malicious with certain
  rate
• Threat Model
• Dishonest feedback
• Non-collusive and collusive setting
• Strategic dynamic behaviors
• Different models

24
Experiment 1 Effect of Dishonest feedback

• Goal
• Understand the effect of malicious behavior of
  peers in providing dishonest feedback
• Simulation Design
• Non-collusive setting
• Collusive setting
• Evaluation Metric
• Trust computation error the root-mean-square of
  the computed trust value and the real assigned
  trust value

25
Experiment 1 Effect of of malicious peers
providing dishonest feedback
collusive
Non-collusive

• Average based metric deteriorates when of
  malicious peers increases
• TVM breaks down with malicious peers gt 50 in
  non-collusive setting and with very small
  malicious peers in collusive setting
• PSM stays effective

26
Experiment 1 Effect of frequency that malicious
peers provide dishonest feedback
collusive
Non-collusive

• Average based metric deteriorates when malicious
  rate of malicious peers increases
• Malicious peers are able to confuse the system by
  acting trustworthy sometimes
• TVM breaks down with collusion

27
Experiment 2 Benefit of Reputation Based trust
mechanism

• Goal
• Understand the benefit of PeerTrust peer
  selection to P2P systems
• Simulation Design
• 3 Systems
• Peer Selection
• Evaluation Metric
• Transaction success rate the ratio of number of
  successful transactions over total number of
  interactions.

28
Experiment 2 - Transaction Success Rate
Non-collusive
collusive

• Reputation helps peers avoiding malicious peers
• Different trust mechanisms have different
  performances
• Collusion can render the whole system based TVM
  useless

29
Experiment 3 Dynamic malicious behaviors
30
Experiment 3 Effect of Dynamic changing
behaviors

• Compare dominant a, ß and ? parameters
• a follow actual behavior, disregard history and
  fluctuations
• ß change trust value slowly and steadily
• ? amplify sudden changes in the behavior of a
  node
• Non-adaptive incapable of adjusting quickly

31
Discussion

• Common attacks and threats in P2P
• PeerTrust alleviates or resolves some security
  concerns in P2P
• New vulnerabilities are introduced by reputation
  based systems
• PeerTrust tries to minimize the security
  weaknesses

32
Summary and Ongoing Work

• Summary
• PeerTrust model
• Implementation issues
• Experimental validation
• Ongoing Work
• Techniques to resist attacks
• Secure Implementation
• Integration into P2P applications

33
Thank you!