Towards Trust Enhanced P2P Systems

1
Towards Trust Enhanced P2P Systems

• Shanghai Jiaotong University
• E-Learning Lab

2
Outline

• Overlay Networks Overview
• Trust Management
• Reputation based Trust Management

3
Variety of Service Overlays

• Storage overlay
• Caching
• Distributed storage systems
• Computation overlay
• Computation grid
• Adaptive objective-oriented compression
• Associative overlay
• Content searching
• Communication overlay
• Adaptive routing
• Multicast overlay

4
Service Overlays Examples

• Content Delivery Network (CDN)
  proxy based
• Akamai
• Peer-to-peer (P2P) Network
  end-system based
• Gnuteller
• KaZaA
• Freenet

5
Metrics for Overlay Construction

• Delay
• Delay or equivalent such as number of hops,
  geographic location
• The metric mostly considered in previous overlay
  construction protocols
• Bandwidth
• Mainly considered as a factor for node degree
  determination but not peer selection
• Loss rate
• Lifespan
• Node capacity (CPU capacity, buffer size)
• Trust

6
Combined Metrics for Overlay Construction

• Why?
• Overlay construction itself is complicated
  enough, is it worth to make it more complex?
• Delay as the single metric for overlay
  construction is simply not good enough1
• The best peer to contact is not always the
  closest one and it changes with the application
  needs
• Delay bandwidth loss rate are mostly orthogonal
  which makes their separate use insufficient
• The influence of these metrics to p2p application
  performance such as media streaming remains
  unexplored

1Source Mohammad Malli, Chadi Barakat, Walid
Dabbous Application lever versus network level
proximity Asian Internet Engineering Conference
2005
7
Trust in Real Life?
8
Trust

• A definition
• Trust is the belief in the competence of an
  entity to act dependably, securely, and reliably
  within a specified context".
• Example
• Alice trusts Bob moderately to forward data.

trust
context (forward data) trust value (moderately)
Alice
Bob
9
Trust Management
Collects information for trust
Monitors and re-evaluates the trust
Evaluates and establishes trust relationships
By content assurance and peer reliability, trust
management systems help secure decentralized
networks
10
Trust Management

• Policy-based trust management
• Trust is immutable.
• Reputation-base trust management (reputation
  systems)
• Suitable for P2P dynamics
• Predict the trust that can be invested in one
  peer from the history of its past behavior!

11
The Role of Reputation

• Real world transactions personal and corporate
  reputations
• Reputation is an assumption that past behavior is
  indicative of future behavior
• Higher the online reputation gt the more
  trustworthy the entity
• eBays Feedback Forum1

1Source eBay. ebay home page, http//www.ebay.com
, 2005.
12
A Reputation System

• Helps establish mutual trust (distrust) by
  assigning a reputation to each peer
• How? Aggregate, process and disseminate
  transaction-based feedback

• Challenges1 of a reputation system
• Provide information that allows peers to
  distinguish between trustworthy and
  non-trustworthy peers
• Encourage peers to be trustworthy
• Discourage participation from those who are not

1Source P. Resnick, K. Kuwabara, R. Zeckhauser,
and E. Friedman. Reputation systems.
Communications of the ACM, 2000.
13
Reputation Management Framework
Modeling
Generation
Storage
Communication
Security
Reputation System
Decentralized System
14
Reputation Generation

• A reputation is generated by participants
  undertaking a transaction.
• Who generates reputation
• Context of reputation file, peer, etc.
• Reputation on what
• Participants and roles

Reputation seeker
Reputation holder
Reputation evaluator
Service requester
Service provider
15
Reputation Generation

• Participants and behavior
• Honest
• 100 honest in all transactions
• Dishonest
• 100 dishonest in all transactions
• Dynamic personality
• Honest at some times and dishonest at others

16
Reputation Generation

• Feedback
• Rating scales 0-10, Good, Ok, Bad
• Context of reputation
• Peer-based PeerTrust1
• Resource-based
• Peer and resource-based XRep2

1 Source L. Xiong and L. Liu. Peertrust
Supporting reputation-based trust for P2P
electronic communities. IEEE Trans. On KDE,
2004. 2 E. Damiani, et al. A reputation-based
approach for choosing reliable resources in P2P
networks. In Proc. of the ACM CCS, 2002.
17
Reputation Modeling

• Feedback
• Quantify the feedback
• Number of transactions
• ebay model
• Credibility of feedback
• The trustworthiness of source of feedback itself
• Transaction context factor
• Size a 1000 transaction should be evaluated
  more important than a 1 one
• Time recent transaction vs. older transactions
• Incentive schemes
• Rewards peers for submitting feedback

18
Reputation Modeling

• Modeling direct and indirect observations A -gt B
  -gt C so A-gtC
• Correlated trust Managing Trust, EigenTrust
• Separate trust metric PeerTrust

19
Reputation Modeling

• Managing Trust1
• Complaints-based model
• Once a cheater, always a cheater
• Not robust to varying personalities of peers

1Source K. Aberer and Z. Despotovic. Managing
trust in a Peer-2-Peer information system. In
Proc. of the CIKM,2001
20
Reputation Modeling

• EigenTrust1
• A trusts B, B trusts C gt A trusts C
• Matrix of normalized local trust and global trust
  values
• define a local trust value sij as the sum of the
  ratings of the individual transactions that peer
  i has downloaded from peer j sij sum(trij.)
• each peer i is assigned a unique global trust
  value that reflects the experiences of all peers
  in the network with peer i.
• global trust values correspond to the left
  principal eigenvector of a matrix of normalized
  local trust values.

1 Source S. Kamvar, M. Schlosser, and H.
Garcia-Molina. The eigentrust algorithm for
reputation management in P2P networks. In Proc.
of the Intl. WWW conference,2003.
21
Reputation Modeling

• EigenTrust1
• A trusts B, B trusts C gt A trusts C
• Matrix of normalized local trust and global trust
  values
• PeerTrust2
• Three basic parameters
• Feedback, number of transactions, credibility of
  feedback
• Two adaptive parameters
• Transaction context size, category, and time
  stamp
• Community context reward peers that give
  feedback
• Separate trust metric defined for credibility of
  feedback
• Peer w will use a personalized similarity between
  itself and another peer v to weght the feedback
  by v on any other peers.

1 Source S. Kamvar, M. Schlosser, and H.
Garcia-Molina. The eigentrust algorithm for
reputation management in P2P networks, 2003. 2
L. Xiong and L. Liu. Peertrust Supporting
reputation-based trust for P2P electronic
communities. IEEE Trans. On KDE, 2004.
22
Reputation Modeling Challenges

• The problem of dynamic peer personalities
• Avoid arithmetic summation schemes eBay
• Reputation fading PeerTrust
• The problem of unfair ratings
• Credibility of feedback PeerTrust

23
Reputation Modeling Challenges

• The problem of collusion (bad mouthing and ballot
  stuffing)
• EigenTrust Use of pre-trusted peers
• PeerTrust Feedback similarity function

24
Reputation Modeling Challenges

• The problem of collusion (bad mouthing and ballot
  stuffing)
• Collusion in Google page rank (google reputation)

node
referential link
The walker
X
1/2
1/3
Z
Y

• As time goes on, the expected percentage of steps
  the walker is at each node v converges to the
  PageRank weight PR(v).

25
Reputation Modeling Challenges
Collusion in Page Rank Google Reputation

• To increase their PR weight, i.e., the stationary
  weight in the random walk, the colluding nodes
  will stall the random walk.

26
Proposed Reputation Model for P2P Network
Algorithms such as Page-rank could be used to
generate a global reputation value for each peer
based the metrics of the number of transactions
and transaction context.
27
Proposed Reputation Model for P2P Network
peer
X
D
2
B
2
3
C
A
Transaction made weighted by transaction context
and number of transactions
When x tries to make an transaction with an
unknown peer A, it will compute a local
reputation value based on the feedbacks from its
neighbor peers who had evaluations on A and their
credibility values as well as As global
reputation value. The result describes the
extent of trust X can invest on A for making a
transaction.
28
Thanks