Reputation Based Trust

1
Reputation Based Trust

• The using of reputation to accomplish trust
  between users on the Internet

M.Voloin, R.Gore, Ibe2roc. PF UPJ, Koice,
Slovakia
2
Contents

• What is Reputation Based Trust
• Various systems for Reputation Based Trust on
  Internet
• Human driven environments
• Auctions systems
• Ebay
• Amazon Auctions
• Yahoo Auctions (retired from June 2007)
• Expert Sites like www.Allexperts.com
• Reviews sites like www.Epinions.com
• Other
• Automated environments
• P2P networks
• Computing markets and grids
• (One such system we describe here in detail)

3
What is Reputation Based Trust?

• Anonymity on the Internet
• Viruses
• Why we trust our friends
• Name recorded past activities
• available and visible for everybody
• Reputation Based Trust

4
Ebay

• How to sell goods somebody from Australia? How to
  make it secure?
• Feedback forum on Ebay
• Ebay is Auction system
• Bussiness on Ebay

5
(No Transcript)
6
(No Transcript)
7
(No Transcript)
8
Reputation Based Trust in Automated Environments

• P2P networks
• Distributed computation markets

9
Distributed Computation Markets

• The Servers provides service complicated
  time-consuming computations And the Clients pay
  for this.
• How to trust a server that it has least-costly,
  fastest service, how it advertised?
• How to ensure that Server performed all work and
  didn't give fake results?
• Solution Proof by computation ringers and
  threshold witnessing mechanism

10

• Step 1 Witness selection
• Step 2 B sends service request
• Step 3 Server selection process
• Step 4 Executing of computation
• Step 5 Returning the execution proofs and
  computation result
• Step 6 Signing new rating by witnesses

11
Securing reputation mechanism in computing markets
12
Witnessing details
13
Building blocks

• Rating Store management
• Every participant stores the most recent data for
  every other participant
• Reputation value can be created or changed only
  if at least c1 participants agree.
• Witness Selection
• B select 2c1 witnesses randomly
• creates a multicast channel for the witnesses
• sends the (signed) job description f, the set of
  input valuesx1...xa, the maximum time B is
  willing to wait for job completion, the maximum
  amount B is willing to pay for the computation

14

• Server Selection
• The 2c1 witnesses selects the most suitable
  service provider (A)
• A is added to the witness multicast group.
• One of the witnesses multicasts the job
  description received from B.
• Threshold Witnessing
• Ringer Generation
• Each witness (Wj) selects one random value xz
  from the input set specified by B in the job
  description and computes a ringer rj H(f(xz))
• Wj sends SWj(H(Id(Wj) sid rj)), its identifier,
  sid, the ringer, together with the signed digest
  and Wj's public key certifcate to A
• A waits to receive 2c1 valid messages
• A sends a multicast message to all the witnesses
  with concatenation of all the signed ringers
  received
• The witnesse, inquire the remaining witnesses for
  their ringers.

15

• Revealing the Ringers
• A performs the computation and reveals the input
  values xz hidden in the 2c1 ringers
• A creates a single message containing Swj
  (H(Id(Wj) sid rj)) and SA(H(Id(A) sid z)),
  for j 1...2c1. There also are the results of
  the computation, f(x1),...,f(xA), along with its
  signed digest.
• Each witness Wj verifies the correctness of only
  its own ringer
• If any witness Wj discovers that rj ltgt H(f(xz)),
  Wj sends a multicast message to all the other
  witnesses revealing this fact
• The witnesses are able to verify the claim by
  computing the correct answer to Wj's ringer and
  compare it with the answer sent back by Alice

16

• Signature Generation
• each witness Wj is able to compute A's new rating
  
• Each Wj then generates a verifiable signature
  share of A's new reputation
• Wj sends this value, its certified verification
  key VKj and A's new rating in clear, to all the
  other witnesses, using the group's multicast
  channel
• Each witness waits to receive c correct signature
  shares for the same new reputation of A as the
  one generated by itself.
• each witness is able to generate the signed new
  rating of A locally
• Reputation Distribution
• The results of the computation are returned to B
  and the new reputation of A is distributed (by
  the first witness is in charge on the broadcast
  channel to all the participants in the system)
• Note that a witness cannot simply send an
  incorrect reputation since it will be easily
  detected
• Punishing Malicious Witnesses

17
Securing reputation mechanism in computing markets
18
Witnessing details
19
Possible attacks?

• Bad-mouthing (incorrect negative feedback)
• Ballot-stuffing (un-earned positive feedback)
• Lazy behavior
• Improvement adding fake ringers
• Sybil Attacks
• Mobile Virus Attacks

20
Conclusions

• With reputation based trust we can make Internet
  more secure and usable in more areas of our
  activities.

Thank you.