THE EIGENTRUST ALGORITHM FOR REPUTATION MANAGEMENT IN P2P NETWORKS

1
THE EIGENTRUST ALGORITHM FOR REPUTATION
MANAGEMENT IN P2P NETWORKS

• Sepandar D. Kamvar Mario T. Schlosser Hector
  Garcia-Molina
• Stanford University
• International World Wide Web Conference
• Budapest, Hungary 2003

2
PAPER CONTRIBUTION

• Eigentrust decreases the number of downloads of
  inauthentic files
• Isolates malicious peers from the network
• Basic Idea Each peer i is assigned a global
  trust value this reflects the experiences of
  all the peers in the network with peer i

3
PAPER FLOW

• Basic Eigentrust
• Distributed Eigentrust
• Secure Eigentrust
• Experiments and Results

4
PAPER FLOW

• Basic Eigentrust
• Distributed Eigentrust
• Secure Eigentrust
• Experiments and Results

5
EIGENTRUST BASIC ALGORITHM

• Compute local trust value
• Normalize local trust value
• Aggregate local trust value (Distributed)

6
EIGENTRUST BASIC ALGORITHM

• Compute local trust value
• Normalize local trust value
• Aggregate local trust value

7
COMPUTE LOCAL TRUST VALUE

• Step1 Peer i downloads a file from peer j
• Positive transaction gt tr(i,j) 1
• Negative transaction gt tr(i,j) -1
• Step2 Local trust value sij sum of ratings of
  individual transactions that peer i has
  downloaded from peer j
• Sij S tr(i,j)

8
EIGENTRUST BASIC ALGORITHM

• Compute local trust value
• Normalize local trust value
• Aggregate local trust value

9
NORMALIZE LOCAL TRUST VALUE

• Why normalize?
• To avoid maliciousness

This ensures that all values will be between 0
and 1
10
EIGENTRUST BASIC ALGORITHM

• Compute local trust value
• Normalize local trust value
• Aggregate local trust value

11
THE NOTION OF TRANSITIVE TRUST

• Context of Trust Good Service Vs. Good Referrals
• This Paper
• A peer will have a high opinion of those who have
  provided it authentic files
• The global reputation of each peer i is given by
  the local trust values assigned to a peer i by
  other peers, weighted by the global reputations
  of the assigning peers

12
AGGREGATE LOCAL TRUST VALUE

• A peer asks his friends/acquaintances about
  their opinions
• of other peers

• where tik represents the trust that peer i places
  in peer k
• based on asking his friends.
• Probabilistically, if an agent were searching
  for reputable
• peers, it would crawl the network using the
  following rule at
• each peer i, it will crawl to peer j with the
  probability of cij

13
FORMAL NOTATION

• C matrix cij where C is the normalized local
  trust matrix
• ti vector containing the values tik
• ti CT ci where ci is the local trust
  vector

Now, peer i gains a view of the network wider
than his own experience t (CT)2 ci
(asking friends friends) t (CT)n ci
(complete view of the network) If n is large, ti
will converge to the same vector for every peer
i. At convergence, t is the eigenvector of C.
14
EIGENTRUST BASIC ALGORITHM

• Compute local trust value
• Normalize local trust value
• Aggregate local trust value

15
EIGENTRUST PRACTICAL ISSUES

• A Priori Notion of Trust
• Inactive Peers
• Malicious Collectives

16
A PRIORI NOTION OF TRUST

• Assuming that some peers in the network are
  trustworthy
• Define a distribution p over pre-trusted peers P,
  such that pi 1 / P
• In the presence of malicious peers, t (CT)n p
  will converge faster so p is used as the start
  vector

17
INACTIVE PEERS

• If peer i does not download from anybody else,
  or assigns a 0 score to all peers, cij will be
  undefined

• Redefine cij so that if a peer i does not know
  anyone, or does not trust anyone, he will choose
  to trust the pre-trusted peers

18
MALICIOUS COLLECTIVES

• Break collectives by having each peer place at
  least some trust in the peers P that are not a
  part of the collective
• How? Probabilistically, an agent crawling the
  network is less likely to get stuck crawling a
  malicious collective
• Challenge To make sure that no pre-trusted peer
  is a member of a malicious collective (that would
  compromise the algorithm)

19
PAPER FLOW

• Basic Eigentrust
• Distributed Eigentrust
• Secure Eigentrust

20
DISTRIBUTED EIGENTRUST

• Each peer stores its local trust vector ci and
  global trust vector ti
• Computation, Storage, Message Overheads are
  minimal
• In P2P networks, each peer has limited
  interactions with other peers. So most of the
  trust values will be 0.
• In the case of a network with heavily active
  peers, limit the number of local trust values cij
  that each peer can report

21
DISTRIBUTED EIGENTRUST

• Observations
• Algorithm converges fast around 100 query cycles
  for a network of 1000 peers
• Computed global trust values do not change
  significantly any more after a low number of
  iterations

22
PAPER FLOW

• Basic Eigentrust
• Distributed Eigentrust
• Secure Eigentrust

23
SECURE EIGENTRUST

• Earlier Each peer computes and reports its own
  trust value ti
• Drawback Malicious peers can easily report false
  trust values
• Secure Eigentrust
• Trust value of a peer must not be computed by and
  reside at the peer itself
• Trust value of one peer will be computed by more
  than one other peer (use of majority vote to
  avoid malicious results)

24
SECURE EIGENTRUST

• M peers (score managers) compute trust value of
  peer i
• Use Distributed Hash Table (DHT) to get M score
  managers

• Hash unique ID of peer (IP address, TCP port)
  using hash functions h1, h2, h3 into points in
  the logical coordinate space
• The peers corresponding to the points become the
  score managers.

25
SECURE EIGENTRUST PROPERTIES

• Anonymity
• A peer at a specific coordinate cannot find out
  for whom it computes the trust value
• Randomization
• Peers that enter the system cannot select at
  which coordinates in the hash space they want to
  be located
• Redundancy
• Several score managers compute the trust value
  for one peer

26
PAPER FLOW

• Basic Eigentrust
• Distributed Eigentrust
• Secure Eigentrust

27
USING GLOBAL TRUST VALUES IMPLICATIONS

• Isolating Malicious Peers
• Highly trusted peers gt overloading a few peers
• Select peers probabilistically based on trust
  value
• Limits number of unsatisfactory downloads
• Balances network load
• Allows newcomers to build up their reputation
• Peers may also bias their choice of download
• Avoid download from a peer that has given it bad
  service, even if gives rest of the network good
  service

28
USING GLOBAL TRUST VALUES

• Incentivize freeriders to share
• Reward reputable peers (Increased connectivity,
  greater BW)
• Incentive to share files
• Gives non-malicious peers an incentive to delete
  inauthentic files

29
PAPER FLOW

• Basic Eigentrust
• Distributed Eigentrust
• Secure Eigentrust
• Experiments and Results

30
SIMULATION SETTINGS
31
EXPERIMENT LOAD DISTRIBUTION MODEL
32
EXPERIMENT THREAT MODELS
33
EXPERIMENT THREAT MODELS A,B
34
EXPERIMENT THREAT MODEL C
35
EXPERIMENT THREAT MODEL D
36
THANK YOU!
37
PROBABILISTIC INTERPRETATION

• Markov Chain Collection of random values (t1,
  t2,,tn)
• whose probabilities at a time interval depends on
  the value
• of the number at a previous time.
• t2 CT t1
• .
• .
• tn CTtn-1 (convergence at tn )
• tn1 CTtn (here, tn tn1)
• tn1 CTtn1 (eigenvector definition, Av cv)
• At convergence, t is the eigenvector of C.