Private Reputation Schemes for P2P Systems
Roslan Ismail
1
, Colin Boyd
1
, Audun Josang
2
, and Selwyn Russell
1
1
Information Security Research Centre, Queensland University of Technology,
Brisbane, QLD 4001 Australia
2
Distributed Systems Technology Centre, University of Queensland, Brisbane, QLD
4072 Australia
Abstract. The risks from participating in P2P transactions are rela-
tively high. To mitigate such risk a reputation scheme could be applied.
Reputation schemes have emerged as a promising means for enabling
electronic transactions with strangers. In order to gain optimal results
from the reputation scheme, the privacy of feedback provider should be
correctly addressed. The feedback provider should be allowed to leave a
feedback without fear of retaliation. Unlike in centralized schemes, pri-
vacy seems impractical for P2P systems especially when accountability
of feedback is also required. This paper considers how privacy can still
b e provided within the accountability requirement.
1 Introduction
Peer-to-peer (P2P) systems are becoming a popular medium for e-commerce.
Intuitively, these systems offer several advantages compared to centralized sys-
tems such as being cheaper, more convenient, faster, and also allowing expanded
scalability of the systems. Nevertheless, the risks from participating in P2P trans-
actions are relatively high. It is easy to create a phantom transaction and based
on such a transaction feedback is given and calculated to produce the reputation
rating. Typically this happens because there is no trusted authority to monitor
a transaction conducted between a peer and its counterpart.
In contrast, in a centralized system for example, eBay
3
, each transaction is
monitored by an authority. This measure ensures that the submitted feedback is
based on completed transactions. P2P systems commonly have different require-
ments compared to centralized systems. In the context of reputation schemes,
P2P systems require peers themselves to calculate and manage their reputation
value on their own. P2P systems may be roughly divided into two categories.
The first category is pure P2P systems while the second is mediated P2P systems.
The former operates without involvement of an authority while in the latter the
authority participates in certain tasks. In terms of practicality the mediated sys-
tems are preferred as it is easy to cheat in the pure P2P systems. In addition,
use of unaudited information makes pure P2P unsuitable for formal e-commerce.
It is vital to monitor each transaction that takes place in the P2P systems
to prevent false feedback. Recently, Fahrenholtz and Lamersdorf [5] proposed a
3
http:www.ebay.com
Ismail R., Boyd C., Josang A. and Russell S. (2004).
Private Reputation Schemes for P2P Systems.
In Proceedings of the 2nd International Workshop on Security in Information Systems, pages 196-206
DOI: 10.5220/0002674201960206
Copyright
c
SciTePress
hybrid solution (known as the FL scheme hereafter) which combines centralized
and distributed methods to monitor transaction activities. There is an authority
known as a portal to monitor the feedback process conducted between a peer and
its counterparts. Unlike the centralized system, the reputation scheme in the FL
scheme is managed by a peer itself. Although the FL scheme seems promising
against a false feedback it lacks privacy. No privacy is provided for the feedback
provider while submitting feedback. As a result the link between the feedback
provider and the submitted feedback is available to the recipient.
Privacy is a vital topic in many electronic systems such as e-voting, e-cash
and e-auction. This is equally true for reputation systems. In fact, privacy can
help to solve the problem of collecting sufficient negative feedbacks. The feed-
back providers are usually reluctant to leave negative feedback, even when it is
appropriate, for fear of retaliation. Unlike the centralized systems where privacy
may not be difficult to implement privacy seems hard for P2P systems.
Another vital property of reputation scheme for P2P systems is account-
ability. Accountability here means that each feedback should be legitimate. To
achieve this property each feedback needs to be signed by the feedback provider.
However, by signing the feedback the identity of the feedback provider can easily
be traced. Therefore, privacy and accountability seem to contradict one another.
This conflict has motivated us to explore a novel way of providing sufficient pri-
vacy while at the same time ensuring that accountability of the feedback is not
compromised.
In this paper a reputation scheme for P2P systems is applied in which peers
calculate and store their reputation on their own without any involvement from
an authority. The authority functions as an entity to monitor the process of
delivering feedback to appropriate participants. This is vital to ensure privacy is
preserved and at the same ensuring feedback is based on a legitimate transaction.
Contribution.
The contributions of the paper are twofold: analyzing the secu-
rity of the Fahrenholtz and Lamersdorf scheme and introducing privacy to the
scheme.
Organisation of the paper.
The remainder of the paper is structured as fol-
lows. Section 2 lists related work. Section 3 reviews the Fahrenholtz and Lamers-
dorf scheme. Section 4 presents our proposal. Finally section 5 discusses several
issues and then concludes the paper.
2 Related Work
To understand the implementation of privacy in the reputation scheme, consider
a scenario of evaluating the performance of a lecturer in a University. At the end
of a semester registered students will be given an evaluation form so that they
can leave feedback on the performance of lecturers. To protect the privacy of
the students, they are not required to write their names on the evaluation form.
By doing so the link between the feedback and students is untraceable. Usually
197
a trusted party ensures that only enrolled students can give feedback and each
student can only complete one evaluation form. In the context of e-commerce,
the feedback providers are the registered students, the feedback targets are the
lecturers and finally the authority refers to the university authority.
Recently, a number of reputation schemes for P2P systems have been pro-
posed for various purposes [3–7]. Gupta et al. [6], for example, proposed a scheme
to calculate peers’ reputation with the help of an authority known as a central-
ized reputation calculation agent (CRCA). CRCA is used to maintain consis-
tency in calculating the submitted feedback to produce reputation for peers. In
addition, it also can prevent manipulation of the reputation. Since the scheme is
aimed at P2P systems the reputation is returned to peers to manage. The scheme
is also concerned with protecting the submitted feedbacks from being manipu-
lated by requiring the feedback provider to encrypt and sign the feedback before
sending it to CRCA.
Cornelli et al. [3] and Damiani et al. [4] have proposed two schemes (the basic
and advanced schemes) to seek a reputable ‘servent’ (a combination of client and
server) for downloading files in P2P systems. Both the schemes employ voting
mechanisms to evaluate recommendations collected from others in searching for
a reliable servent. An entity receiving the higher vote is a reputable entity. The
basic scheme provides partial privacy as it hides the identity of the feedback
provider but IP address is in clear form. The IP address is required to verify the
vote’s origin. The advanced scheme, on the other hand, discloses the identity of
voter so that the voter credibility can be assessed. Integrity and non-repudiation
of the feedback are assured via encryption and signature mechanisms.
Liau et al. [7] proposed a reputation scheme for P2P systems based on cer-
tificate mechanisms. This scheme is a pure P2P reputation scheme; a peer itself
is in charge on the management of reputation. This improves the storage and
integrity of the reputation rating. The reputation certificate is propagated and
evaluated b efore it can be accepted as a reputation reference. The checking of
the reputation certificate is conducted by contacting the recent preceding rater.
In a case where the preceding rater is not available the next predecessor rater
should be contacted and so on until an available preceding rater is found. To
preserve the integrity of the reputation certificate the rater signs the updated
certificate.
In all the schemes above the privacy of the feedback provider is not given a
fair treatment. Rather the schemes are focussed on how to provide integrity of
the feedback. However, we argue that to collect a sufficient amount of feedbacks,
especially negative feedback, privacy should be seriously taken into account.
Otherwise the problem of eliciting negative feedback remains unsolved. Privacy
in fact empowers participants to leave negative feedback when appropriate with-
out fear of possible retaliation. Table 1 presents properties hold by the reviewed
schemes. The denotes a full feature is available while ? denotes a partial feature
is provided. Out of three schemes reviewed, we choose to improve the scheme
due to Fahrenholtz and Lamersdorf because it has two suitable features for P2P
systems. The first feature is a monitor mechanism to check that transactions
198
Table 1. Summary of properties in several reputation schemes
Integrity
Non-Repudiation
Privacy
Monitoring
Liau et al. scheme
Cornelli et al. scheme
?
Damiani et al. scheme
?
Gupta et al. scheme
Fahrenholtz and Lamersdorf scheme
have been conducted by the peers. By doing so a feedback is assured to be based
on a legitimate transaction. The second feature is a mechanism which prevents
peers from discarding unfavorable feedback collected by them.
3 Fahrenholtz and Lamersdorf Scheme
Fahrenholtz and Lamersdorf [5] proposed a distributed reputation scheme for
P2P networks. In the FL scheme an entity called a portal is used to monitor
transactions conducted between peers. The portal also records the number of
transactions conducted, as well as the number of feedbacks obtained by each
peer. This is achieved via the use of ticket and nonce. The ticket typically con-
tains identification of a peer and its counterpart, and the nonce. The nonce is
extracted from the transaction ticket and it needs to be submitted alone with
the questionnaire form. These mechanisms can prevent peers from discarding
unfavorable feedbacks collected. To achieve integrity and non-repudiation of the
feedback each one will be encrypted and signed before sending it one to another.
3.1 Outline of FL Scheme
Figure 1 shows the entities in the scheme and their interactions. There are two
types of entities in the scheme; a trusted third party (TTP) known as a portal
and the peers. The peers are required to register with the portal before com-
mencing with a transaction. Each peer is required to create a key pair (private,
public) when completing the registration. The scheme contains five phases; au-
thentication of reputation management system subjects, service location for a
context-specific transaction partner, selection of transaction partner, domain de-
pendent transaction and rating partner. For simplicity we only review phase 2
and 5 (for further details consult [5]).
System Setup. Let U
1
and U
2
be two peers who want to transact one to
another, and A be an authority. The identities of U
1
and U
2
are denoted by
199
U
1
U
i
Portal
Request Nonce & Send Nonce
Send Nonce
Transact & Rate One to Another
Mediator
Mediation
Mediation
Fig. 1. Abstract View of FL scheme
id
U
1
and id
U
2
, respectively. S
X
(m) denotes that m is signed by X, E
Y,Z
(m)
denotes m is encrypted using a shared key between Y and Z, and trans-
action tickets for U
1
and U
2
are denoted by d
A,U
1
=(id
U
1
, id
U
2
, r
A,U
2
) and
d
A,U
2
=(id
U
1
, id
U
2
, r
A,U
1
), respectively. The notations r
A,U
1
and r
A,U
2
repre-
sent the nonce issued by an authority A for user U
1
and U
2
, respectively.
Transaction Partner Selection. Identification of a suitable partner must
take place before a transaction begins and it is assumed that this process has
already taken place beforehand. Rather we continue to the next step where
U
1
requests the authority A to issue transaction tickets to itself and its
counterpart. Upon receiving this request, two tickets d
A,U
1
, d
A,U
2
are issued
and signed by A where d
A,U
1
is for the transaction ticket for U
1
and d
A,U
2
is for U
2
. To ensure confidentiality and integrity of these tickets, they are
encrypted with the key shared between A and U
1
, and A and U
2
, respectively.
The following are the protocol messages sent by A.
1. A U
1
: E
A,U
1
(d
A,U
1
, S
A
(d
A,U
1
))
2.
A U
2
: E
A,U
2
(d
A,U
2
, S
A
(d
A,U
2
))
Rating of Partner. Upon completing the transaction, U
1
and U
2
can start
to rate the performance one to another. U
1
sends the Q
0
naire
Cxt,U
1
along
with the nonce to U
2
, and vice versa. Upon receiving the Q
0
naire
Cxt,A
, U
1
and U
2
can start to fill it with a feedback before sending it to one another.
To protect the integrity of the Q
0
naire, it is encrypted with the key shared
between U
1
and U
2
. To confirm that the completed Q
0
naire has already been
submitted to one another U
1
and U
2
send the nonce to A. The nonce can be
extracted from the transaction ticket. Up on receiving the nonces A sends an
acknowledgement to both U
1
and U
2
to indicate the status of the sending
nonce: either it is fine or an error is reported. The nonce functions as a means
to prevent U
1
and U
2
from discarding unfavorable feedbacks. The protocol
executed is described as follows.
1.
U
1
U
2
: E
U
1
,U
2
(Q
0
naire
Cxt
, S
U
1
(Q
0
naire
Cxt,U
2
), r
A,U
2
)
200
2. U
2
U
1
: E
U
1
,U
2
(Q
0
naire
Cxt,U
1
, S
U
2
(Q
0
naire
Cxt,U
1
), r
A,U
1
)
3.
U
1
A : E
A,U
1
(r
A,U
1
)
4.
U
2
A : E
A,U
2
(r
A,U
2
)
3.2 Analysis of the FL Scheme
Notation.
Let N
U
be the number of transactions carried out by user, F
A
de-
notes the number of nonces received by the authority, F
U
denotes the number
of feedbacks obtained and recorded by the user.
The b ehavior of the relying party depends on the relative sizes of N
U
, F
A
and F
U
. We consider several different cases.
1. N
U
= F
A
= F
U
.
This outcome means that the number of transactions made by a user U is
equal to the number of feedbacks for the user recorded by the authority and
the user. This is an ideal case where all the participants follow the protocol
honestly. Typically in this case the reputation rating of the user is accepted.
2. F
U
= F
A
and F
U
< N
U
.
This outcome means that the number of feedbacks recorded by user U is
equal to the number of feedback recorded by authority but it is less than the
number of transactions recorded by the authority. There are occasions when
some feedback providers may not return their feedback after completing the
transactions. This could be quite common especially when there is no reward
for submitting a feedback. In addition, fear of the consequences due to the
given feedback is another factor which causes lack of interest to leave a
feedback. In this case the reputation of the peer is commonly accepted.
3. F
U
> F
A
and F
A
= N
U
This outcome means that the number of feedbacks recorded by user U is
greater than the number of feedbacks and transactions made and recorded
by the authority. The user may create some phantom feedbacks in order to
boost his reputation. Thus, in this case the reputation of the user is not
accepted.
4.
F
U
< F
A
and F
A
= N
U
This outcome means that the number of feedbacks recorded by a user U
is less than the number of feedbacks and transactions made recorded by
the authority. The user may have discarded some of unfavorable feedbacks
submitted for him. The reputation of the peer is not accepted.
5. F
U
> F
A
and F
U
< N
U
This outcome means that the number of feedbacks recorded by a user U is
greater than the number of feedbacks recorded by the authority but less than
the number of transactions made. Some feedback providers may choose not
to return their nonces to A. As a result A’s record on the number of nonce
received will be less than the number of feedback record by the user. The
reputation of the user could be accepted depending on a proof provided by
the user.
201
Although the FL scheme provides sufficient protection from the outsider at-
tacks it is exposed to attacks by the internal players. For simplicity the commu-
nication line is assumed reliable. The nonce is always reached to the authority.
In the FL scheme the portal is a trusted party. However, this may not be valid
in some cases. We would like to point out some important facts regarding the
above outcomes. For example, in case 2 an attack could be launched by the dis-
honest portal. The portal may learn the number of transactions conducted by
U
1
, as well as the number of feedbacks collected by U
1
from the returned nonce.
With this knowledge a phantom nonce can be added to U ’s account. As a result
U
1
’s reputation could be rejected as there is discrepancy in the record of the
authority and U
1
.
A major restriction the FL scheme p ossesses is that the feedback target may
learn the link between the feedback provider and the feedback because it is in
clear form. As a result the feedback providers may be reluctant to leave honest
feedback especially in the case of negative feedback due to the consequences they
may suffer later. In a real e-commerce environment negative feedback is needed
to counterbalance positive feedback so that the produced reputation can reflect
true behavior of users.
4 Improved Scheme
We propose an improvement of the FL scheme by introducing privacy as the
major concern. The players and processes are similar to the FL scheme. To ensure
that privacy can be achieved the number of a peer’s counterparts should be more
than one, and preferably a large number. This requirement is essential to allow
unlinkability. The importance of unlinkability to maintain privacy is discussed by
Maitland et al. [8]. Without sufficient number of players privacy seems impossible
to implement. Another vital consideration is the timing of delivering of the
feedback. If the feedback is sent immediately after the transaction is completed
then the link between the feedback and the feedback provider can easily be
formed. To avoid such undesirable outcome delay of the delivery of the feedback
to a certain time later could be undertaken. This could either be a randomised
delay, or delivery of feedbacks could be batched after a threshold number of
feedbacks has been received.
Unlike the FL scheme, our proposal does not use shared keys. Instead the
peers use their counterpart’s public key to encrypt the feedback. However, we
follow the practice of FL scheme in managing the administrative task. The au-
thority is still responsible for issuing a nonce to each peer. The nonce must to
be obtained before a transaction can take place. Besides issuing the nonce, the
authority also maintains a record of the returned nonces.
There are six phases in our proposal; requesting nonce, preparing and send-
ing token, signing and sending legitimate token, submitting feedback, calculat-
ing feedback and showing reputation. In our improved scheme, registration of
participants is not required. Thus, each participant is assumed to have a valid
certificate issued by the certificate authority CA.
202
There are several options can be taken to construct our improved scheme.
One can construct a scheme based on ring signature schemes [9]. Ring signatures
allow the identity of the signer to be hidden from recipients while retaining the
important advantage of enabling accountability of feedback to b e achieved. Thus
the two characteristics we require for reputation schemes are provided. However,
this option does not seem very practical to be implemented as it requires vastly
more computation to verify the signatures. There are two types of computation
required. The first is to verify signature of the n members in a ring signature
for a feedback provider. The second is to verify each of the d feedback providers.
As a result there are d × n computations are required. A second option is to
implement a scheme based on a bilinear ring signature scheme. This scheme
reduces the inefficiency faced by the first option but like the first scheme it is
still too inefficient to be really practical. To overcome this inefficiency, a token
based scheme is proposed. Since our scheme follows the same process as the FL
scheme where a nonce is required to be obtained b efore starting a transaction
this phase is not considered in the following protocol.
4.1 Protocol of the Scheme
The following protocol uses several notations as follows: ID
F P
denotes an iden-
tification of the feedback provider, E
X
denotes encryption using X’s public key
and Sig
X
denotes signing using X’s private key.
Preparing and Sending Token. F P prepares the transaction particulars m.
The content of m can be the date of transaction, the feedback target’s identifi-
cation, the amount of transactions and the given feedback. To ensure integrity is
achieved a pair (ID
F P
, m) is encrypted using TTP’s public key before sending
it to TTP for signing. In a variant of this procedure the token could be created
using electronic cash technology [2]. A coin is issued by the TTP for a particular
transaction and when submitting feedback the coin payment protocol is used.
The advantage of this option is that the feedback value can be hidden from the
TTP, but there is an extra computational cost. We do not consider this option
further in this paper.
F P T T P : E
T T P
(ID
F P
, m)
Signing and Sending Legitimate Token. Upon receiving the pair (ID
F P
, m)
from F P , T T P decrypts it and then verifies the correctness of m against a
database of transactions maintained by T T P itself. If the verification is suc-
cessful m is signed by T T P and then encrypted using the F P ’s public key. To
complete the phase, T T P sends m to the feedback provider. The signed m is
considered a legitimate token. Only the legitimate token can be used for sub-
mitting a feedback. Without using the legitimate token the feedback will not be
counted for calculation of reputation. A nonce is also issued by T T P and then
submitted to F P .
T T P F P : E
F P
(Sig
T T P
(m), nonce)
203
Submitting Feedback. F P sends to F T the legitimate token m which consists
of the feedback. To protect integrity of the legitimate token, m is encrypted using
the feedback target’s public key. F P also send the nonce to F T .
F P F T : E
F T
(Sig
T T P
(m), nonce)
Calculating Feedback. Upon receiving m, F T decrypts and then verifies the
TTP’s signature on m. If the verification is successful the legitimate token is
accepted otherwise it is rejected. The accepted token is then used to calculate
the reputation of the feedback target. F T sends the nonce to T T P to confirm
the feedback is received from the feedback provider. However, F T does not who
is the feedback provider.
T T P sends a signed list n to the feedback target. The list n consists of number
of the legitimate tokens issued by T T P for the feedback target. n is important
to prevent the feedback target from discarding the submitted legitimate tokens.
In addition, n also acts a means to convince the relying party that the calculated
reputation is based on the submitted feedback.
A F T : E
F T
(Sig
T T P
(n))
Showing Reputation.
Before a transaction can commence, F T sends the cal-
culate reputation and n to the relying party so that the relying party can evaluate
the validity of reputation. Due to a possibility of having huge number of tokens
to be verified the relying party could batch them. The scheme proposed by Bel-
lare et al. [1] can be employed which save the computation of verification. The
relying party has to verify two signatures: the TTP’s signature on the tokens
and the TTP’s signature on the list n.
5 Analysis
Privacy.
The improved scheme achieves conditional privacy where the feedback
provider is hidden from public except T T P . However, it requires trust to be
placed on the trusted third party not to reveal the identity of the feedback
provider otherwise the privacy of the feedback provider is compromised. In other
words T T P is assumed honest in performing its task. However, in a case where
this assumption is difficult to implement, for example, in the presence of the
dishonest TTP then threshold schemes could be implemented. This means a
number of T T P is required in which each individual T T P shares a portion of
identity of the feedback provider. Without sufficient number of T T P to form
the identity of the feedback provider, the privacy of the feedback provider is
preserved.
6 Conclusion
An analysis conducted on the scheme of Fahrenholtz and Lamersdorf reveals a
few security concerns. We have proposed an improved reputation scheme which
is based on the FL scheme. The improved scheme provides privacy for feedback
204
providers while submitting a feedback. With this property the feedback providers
can leave negative feedback without fear of retaliation from the other parties.
The token based solution is suitable to provide a simple privacy protection and
furthermore it is efficient in terms of computation required to verify signatures.
References
1. Mihir Bellare, Juan A. Garay, and Tal Rabin. Fast batch verification for modular
exp onentiation and digital signatures. In K. Nyberg, editor, EUROCRYPT ’98 ,
volume LNCS 1403, pages 236–250. Springer-Verlag Heidelberg, 1998.
2.
Stefan Brands. Untraceable off-line cash in wallets with observers, In D. R. Stinson,
editor, Advances in Cryptology - Crypto ’93, pages 302–318, Springer-Verlag, 1993.
3. Fabrizio Cornelli, Ernesto Damiani, Sabrina De Capitani di Vimercati, Stefano Para-
b oschi, and Pierangela Samarati. Choosing Reputable Servents in a P2P Network.
In Proceedings of the eleventh international conference on World Wide Web, pages
376–386. ACM Press, 2002.
4.
Ernesto Damiani, Sabrine De Capitani di Vimercati, Stefano Paraboschi, Pierangela
Samarati, and Fabio Violante. A reputation-based approach for choosing reliable
resources in Peer-to-Peer networks. In Proceedings of the 9th ACM conference on
Computer and communications security, pages 207–216. ACM Press, 2002.
5. D. Fahrenholtz and W. Lamersdorf. Transactional security for a distributed repu-
tation management system. In K. Bauknecht, A. Min Tjoa, and G. Quirchmayr,
editors, Proceedings of the 3rd International Conference on Electronic Commerce
and Web Technologies, volume 2455 of Lecture Notes in Computer Science, pages
214–223. Springer-Verlag, 8 2002.
6.
Minaxi Gupta, Paul Judge, and Mostafa Ammar. A reputation system for Peer-to-
Peer networks. In ACM 13th International Workshop on Network and Operating
Systems Support for Digital Audio and Video (NOSSDAV 2003), pages 144–152.
ACM Press, June 1-3 2003.
7.
Chu Yee Liau, Xuan Zhou, Stephane Bressan, and Kian-Lee Tan. Efficient dis-
tributed reputation scheme for Peer-to-Peer systems. In The 2nd International
Human.Society@Internet Conference, volume LNCS 2713, pages 54–63. Springer-
Verlag, 2003.
8. Greg Maitland, Jason Reid, Ernest Foo, Colin Boyd, and Ed Dawson. Linkability in
Practical Electronic Cash Design. In Proceedings of Information Security Workshop
(ISW 2000) , volume LNCS 1975, pages 149–163. Springer-Verlag, 2000.
9.
Ronald L. Rivest, Adi Shamir, and Yael Tauman. How to leak a secret. In C. Boyd,
editor, ASIACRYPT 2001, volume LNCS 2248, pages 552–565. Springer-Verlag,
2001.
205