Trust management and incentive mechanism for P2P networks: Survey to cope challenges

Trust management and Incentive mechanism for P2PNetworks: Survey to cope challenges

Asia SamreenDepartment of computer science

University of [email protected]

Shahid HussainDepartment of Computer Science and Engineering,

Bahria University, [email protected]

Figure 1: Example indirect trust ofpeer X to peer Z calculated via peer V'srecommendation

whereas section 3 focuses on the incentive mechanisms,section 4 describe some problems regarding trust and incentivemanagement in P2P environment. Finally, conclusions andfuture challenges are given in section 5.

II. TRUST AND REpUTATION

Trust is the expectation about a peer's future behaviorbased on information about the peer's past behavior.Continuous trust values instead of discrete trust values arewidely accepted in representing degree of trust. Trust is themetric, which gives the reputation of a peer, the globalreputation values are assigned on the basis of trust relationshipamong peers [7].

A. Trust calculation and management

The trust of an entity with other entity is not a fixed valuebut can change dynamically depending on the behavior of anentity and context in the environment. Trust should beestablished from the viewpoint of both the parties. Requestor'strust with service provider may be different from the serviceprovider trust with the requestor [22]. There are several typesof trust such as

Direct Trust: The trust that an entity holds on a serviceprovider without any intermediate service provider or entity isknown as direct trust.

The interpreted formula for direct trust with the context ofP2P network is given below.

Abstract-Peer-to-Peer networks are very popular these daysdue to their use in data and resource sharing. Trust andreputation of a peer not only make it easy to take a decision toask for resource from the trustworthy node but also provide away to punish evil peers for malicious act. Incentive mechanismprovokes a peer to share resource and discourage the well knownfree riding. In this paper we studied proposed schemes anddiscuss some open issues to cope as the future research work aswell as we have provided solution of some problems such as todetect malicious peers, false rating problem and sudden changein the behavior of peers that can be helpful in designing a newnovel and robust frame work for trust and reputation basedincentive mechanism.

I. INTRODUCTION

P2P networks enable direct end to end communication anddistribution of resources and services among peers in adecentralized manner. P2P Systems pose too much higheramount of uncertainty due to quick change in Peer's behavioras such systems are designed for enabling Peers to shareresources and services [9], [3]. Thus entities participating in thesystem resource sharing must be identified and evaluatedbefore they are authorized to access resources orfunctionalities. Moreover any Peer can join or leave thenetwork at any time and each peer itself is responsible formaking local autonomous decision based on informationreceived from other peers in the network. Therefore, for such adynamic and highly autonomous network where usuallyinteractions happen between stranger peers, there are varioussecurity problems such as sending false information from evilpeers, strategically altering behavior of malicious peers, groupmanagement and membership control for multi-agent system,etc. [7],[3],[5]. There is also a need to verify trustworthiness ofshared files in P2P systems to prevent data from being lost [2]

.All these issues approach towards a novel

framework of trust management system that can build onpeers reputation. To appreciate cooperation from both theservice providers and service requestors, an incentivemechanism can also be combined with the trust and reputationmechanism. Incentive mechanisms not only benefit the honestpeers but also helpful to detect dishonest or evil peers. Thispaper provides a survey of some incentive mechanisms basedon Trust and Reputation.

The remainder of this paper is organized as follows. Section2 is concerned with the trust and reputation based schemes

978-1-4244-2824-3/08/$25.00 ©2008 IEEE

0.7

Tij=1-an

0.7*0.27=0.19

(1)

301

Proceedings ofthe 12th IEEE International Multitopic Conference, December 23-24, 2008

where Tij is the peer's i trust in peer j, n is the number of peeri's transactions with peer j and ae [0,1] is the learning rate[21].

Indirect Trust: The trust that an entity has on a serviceprovider through some other service provider or entity, isknown as indirect trust. Indirect trust that a service providerentity has in requestor entity based on recommendation, iscalculated as:

Where Iij is indirect trust of peer i in peer j, n is threshold,fixed by service provider domain or host, Tit denotes the directtrust of peer i in t and Ttj is the direct trust of peer t in j asshown in figure 1. [22].

Recommendation Trust: is the trust of a peer on anotherone based on direct trust and other peer's recommendation. In[7] a recommendation trust model is proposed. To detect andpunish malicious peers direct trust uses a time factor and apenalty function.. For recommendation a simple reputationpolling mechanism is used and recommendation trust iscalculated as:

p = 0; if DTx ( 0) ~ 0 and p = 1 ; otherwise

Dynamic Trust: Trust building with different intervals oftime is known as dynamic trust. Sudden change in peersbehavior requires to build trust on the basis of confidence andrecommendation after making transactions with peers atdifferent times. Both [23], [5] provide a dynamic model fortrust that can be calculated as D.Trust = Ti * Ci ; where Ti isthe recent trust calculated by peer a at the interval i about thetarget peer 0, i.e. Ti (a,O) =RTi (a,o).

(6)

where A is self confidence factor DTi denotes the directtrust built by experience, R is the set of referrer ,Cri denotesthe recommended credibility or trust that a has in theexperience reported by x.

If Ti is built from the history experience it can becalculated by history trust, as given below:

(2)KI ..lJ=

where N(x) denotes total no of votes and R(x) is accuracyfactor of peer x's voting, p is a factor related to direct trustDTx ofpeer x on other peer, can be calculated as:

(8)

(7)

Confidence is defined by the following equation

Ci(a, 0) =RrRi(a, 0) + RrH i (a,°).Dri (a, 0) (9)

maxH

I p J-1RI:_J (a,0)

1'; =HI: (a, 0) =-J=-I-m-ax

-H---­

I p J-l

J=1

where p (0 ~ p ~ 1) is forgetting factor, maxH is thenumber of intervals previously experienced. If a peer has boththe recent and history trust, she can calculate trust as:

I: (a, 0) = min{HI: (a, 0), R1'; (a, o)}

(3)

(4)

RcT (a ,0)=( T - T' )

Where RCT(a,o) is recommendation trust calculated byPeer a for Peer 0, T is the polling result from the perspective oftrust, can be defined as:

N(x)

IR(x) *PT=_i=_I _

R(x)

(10)

otherwise

RrRi(a,o)=

Jr((Ni(a,o) +aICri(a,x).Ni(x,o))sin( xd?

2.maxN

if(Ni(a,o) +aICri(a,x).N i (x, 0)) < maxNxd?

where Ci is the confidence that reflects the reliability of trustN(o) S(x 0) *M(x 0) *Z 1 calculated by peer a;

DTx(o)=I(' , +pen(i) -n)i=1 N x (0) 1+ e

(5)

Nx(0) is total number of interactions x does with 0, S(x,o)denotes satisfaction degree of x for 0, M(x,o) is the ratiobetween size of ith interaction and average size of interactionand pen(i) is the punishment factor (pen(i) =1 if ith interacting

1fails and p(i)=O if succeeds), --_- is the acceleration factor

l+e n

where n denotes the number of failures. Obviously it is used tomaintain trust values, when an interaction fails trust valuedrops. Now p can be described as

302

Figure 2. P2P System Architecture for Trust Management.

Where a,~ are weights such that a+~=1

n

LRatingWSrep =a* ;=1 + {J* CoWS (14)

n

C. Some Reputation Models

Peer-to-Peer (P2P) reputation systems are essential toevaluate the trustworthiness of participating peers and tocombat the selfish, dishonest, and malicious peer behaviors.WSrep[3] and PowerTrust [6] are the Reputation models todeal with dishonesty ofpeers.

WSrep reputation model depends on user rating on averagebasis and performance history ofprovided services

. Model uses MQ= {Latency, DomQ}to define MeasurableQuality of Services, Latency describes the delay in responseand DomQ denotes the domain specific service characteristics.Further both the attributes are calculated Like Latency byAveraging the user rating and DomQ by credibility to producea reputation.

More detailed version of the Model is as follows:

Credibility of Web Service (CoWS) shows the overallcredibility of all the attributes and can be calculated usingCredibility of attributes (CoA)that expresses the credibility ofa single attribute defined in MQ; the CoA is computed in thetrustworthy third-party with the objective feedbacks collectedfrom different users.

The PoweTrust presents a trust overlay network model toanalyze the feedback reputation by collecting real-life datafrom eBay and confirmed the power law connectivity by TON(Trust Overlay Network) graph .Using Look-ahead RandomWalk (LRW) strategy and Locality Preserving Hash (LPH)functions fast global reputation aggregation , ranking andupdating is provided. In [8]

(11)

maxH

LRrR;_j(a,o)RrH;(a,o) =_J_·=_I _

maxH

_ {AT;_1 (a,o) + HT; (a, 0) - RT;(a,o),ij

AT;(a,o)- RT;(a,o)-HT;(a,o)<-£

AT;-1 otherwise

{

o ATi(a,o»maxAT

DR. = trATI 1- sine ;) Otherwise

2.maxAT (13)

Proceedings ofthe 12th IEEE International Multitopic Conference, December 23-24,2008

Where a (0 ~ a ~ 1) is the discount weight of referrer'sexperience for the maxN (fixed threshold) interactions andNi(a,o) are present number of interactions within interval i .Above given Rating reliability for recent trust is used to findout rating reliability for history trust.

Where E represents the tolerated margin of error due tonoise while taking experiences. Deviation reliability forinterval i is

To observe the rational behavior, deviation reliability iscalculated using misusing trust accumulation denoted by ATi.

(12)

B. Trust management using Game Theoretic Models

Every trust framework relies on self experience andreputation to calculate trustworthiness of a peer contribution inthe network Game theory provides a wisest way to do sotherefore, Game Theory has widely been applied in trustmanagement.

M. Harsh et al. [12] proposes different strategies for Gametheoretic model. System architecture for trust managementusing Game theoretic approach is given in figure 2 [12].Architecture consists of a group manager that takes care ofleaving and joining peers in the network, a transaction managerthat has two components job submitter that distributes the jobsand after job completion evaluate the result, the job doerreturns the job's result to job submitter. A trust store isimplemented as a DHT (Dynamic Hash Table), is used toaggregate feedbacks given by peers. Trust evaluator reads thereputation information from trust store and calculate reputationbased trust. To evaluate a member's trustworthiness RedundantJob Submission Gob is submitted to various nodes and result isconcluded after comparing all values) and Auditing StrategyGob is submitted to one node) is used. To select a candidatethere are tit For tat strategy (Decision is based on previousmove), Self Trust Based Strategy (direct trust ), DynamicStrategy (trust aggregation and dynamic trust based) and GameTree Strategy (maximization ofpeer's utility) are suggested.

303

(16)

File ID

UI Evaluation

U2 Evaluation

U3 Evaluation

p * (M,N) arg_maxp {ER(P,M,N)}

Figure 3. Framework in DHT based overlay

,1 ] denotes the accumulatative data loss rate and (x, ~ areweight factors.

Another virtual money based scheme is presented in [20].In the CAIM- (Contingent Auction Incentive Mechanism)incentive mechanism all resources are marked price in virtualmoney. The peers bid for resources in an auction; it is todetermine who can get the resources by pay for it with virtualmoney. The seller has M unit resources and there are N buyersare willing to buy one unit resource. If Pr( p, k, N) is theprobability function for the possibility of k users participate inauction for price p then the expected revenue of the modelcould be calculated as follows:

N

ER(p, M, N) = L prep, k, N)1CA,B (p, M, k)k=l (15)

with the supposition that user's evaluation ~ is uniformlydistributed 1tA (P,M,k) is calculated for p> 0.5 and 1tB iscalculated for the probability for p~ 0.5. Obviously, theincentive mechanism is price discrimination mechanism, andthe optimal Price p* with respect to N ,M is:

Peers contributing the resources can maXImIze theirrevenue and therefore are encouraged to participate in thesystem. The peers wanting resources have to pay so they arealso required to earn virtual money.

Differentiated Service based incentive mechanism.

A file evaluation and user's reputation method is used in[1]. A voting technique is proposed for calculation the userreputation. The evaluation information of file and user is keptin DHT( Dynamic Hash Table) based overlay .For servicedifferentiation, user can give different services to other userswith different reputations. In figure 3 [1], U4 requests otherusers to download a file from them. Other users can calculateU4's reputation and give U4 a suitable service which includes abandwidth quota and the position in the waiting queue. A

Proceedings ofthe 12th IEEE International Multitopic Conference, December 23-24, 2008

considering the fact that most recent information's are usefulto calculate trust , a technique of sliding window is used. Atrust value , finally is calculated using both direct and indirecttrust. Reputation is calculated using witness voting's and actualvalue. Three types of behavior patterns of Service Providersare considered: good peers, fixed malicious peers and dynamicmalicious peers The behaviors of peers as raters can be one ofthe three types: honest peers, fixed dishonest peers anddynamic dishonest peers.

To compute global reputation values of peers in P2Psystems based on the transaction between peers in the past [4]proposes a model. This model minimizes the impact of EvilPeers on he performance of a P2P system. The trust vectorsstore the transaction between peers in the past. The systemcomputes a global reputation value for a peer by compute thesimilarity between peers and compute the credibility with theinformation that the peer provides. In [10] the TBRM (Trust­Based Reputation Model) is proposed, in which each node hastwo global values, which are reputation and trust. T hereputation data is stored, the trust value and the reputation valeare updated according to each transaction. And then the trustvalue serves to the reputation computing.

III.INCENTIVE MECHANISMS FOR P2P NETWORKS. TRUST ANDREpUTATION BASE

In recent years the popularity of P2P networks has beenincreased very rapidly due to their co-operative nature as theparticipating peers share resources equally. In real systemshowever, a trust mechanism without incentive would face lackof user's enthusiasm and thus cause sparse relationship ofdirect trust while an incentive mechanism without trust couldinduce user's bad behavior [4]. Therefore, P2P networksrequire reputation system combined with trust and incentivemechanism.

Price based incentive mechanism. A Price-basedincentive scheme is discussed in [15]. Data Stream is dividedinto sub streams with a unit bandwidth; each sub stream isallocated incoming bandwidth slots (in slot) and outgoingbandwidth slots (out slot). For a fixed time period say Tk a

peer "i" has to pay price p;f to its parent j. A parent peer earns(virtual money) points from child and bonus from the paymentsystem. Moreover child peers can bid for desired parents foreach time period; parents obviously choose the highest bidders.The probability of resource hunger or resource starvation forpoor peers is no higher than non incentive network.

Definition: To maximize the peers own benefit in term ofmedia quality (in a free market) the given utility function isused for peer's sub stream.

sVi =In(l +LU ij ) ,S denotes number of sub streams.

j=I

Individual utility can be calculated as

In{I + max[O,I- a.dij ]}U - ; lij >1 denotes the

ij - 1n2.(lij)P

service latency in milliseconds of sub stream ofpeer I, dij E [ 0

304

Problem. A malicious request responder, if selected as aservice provider can attack on the system.

Solution. For open networks in [11] the proposed modeluses trust and reputation information to choose both requestresponders and servers. System is partitioned into clustershaving selected trustworthy header nodes. Intra-cluster trust iscalculated to choose request responders and inter-cluster trustis for Service provider. Intra--eluster trust is managed by headeror super node, and centralized therefore help to detectmalicious node very quickly, while inter-cluster trust isdecentralized

Problem. Malicious raters, easily attack Reputation modelsbased on subjective user rates.

Solution. The dishonest providers often oscillate theirreputation between building and milking to mislead thecustomers; they also collude with malicious raters who alwaysprovide low feedbacks to their opponents and high feedbacksto themselves. In WSrep [3], the reputation model depends onuser rating on average basis and performance history ofprovided services, which help to detect effectively thedishonest providers

V.CONCLUSION AND FUTURE WORK

In this paper we surveyed Trust and Reputation buildingtechniques and their usage in P2P networks. We also discusssome incentive mechanisms to show the significance ofincentives in attracting the honest and resource providingentities for a domain. This survey also has showed that with theincrease in popularity of P2P networks there are some dangersfrom Evil peers , who can consume a lot while contributingnothing, or by sharing fake files or doing some group attackgiving false rating to desired peer. All these issues and theirprovided solutions have been discussed.

The future work in this area demands optimizations tomost of the standardized and proposed techniques, schemesand algorithms. How to verify the reliability of the reportedtrust and how to select the trust related factors in differentscenarios is very complex problem for the trust based P2Psystem. One method to cope the challenge is to use statisticalapproach and assumption checking way to select the receivedtrust reports from neighbors peer node and to manage the trustrating [14]. To design a strategy to approximate the rational

Problem. Reputation building based feedback is difficult,due to dynamic changes in open networks

Solution. A probabilistic computational reputation basedtrust model having the property of separating service andfeedback using a topology adaptation algorithm can solve theproblem [13].Coupon based incentive mechanism.To encouraged the

participants to share data over wireless medium and in theenvironments such as P2P networks [16] introduces a couponbased incentive mechanism. The basic concept behindCoupons is to distribute a given piece of information through amobile network. The incentive proposed, is based on anordered list of unique IDs appended to a message. The idea isthat, once the information/coupon is used, users contained inthe ordered list receive some sort of benefit. The higher a useris on the list the more the reward value.

Proceedings ofthe 12th IEEE International Multitopic Conference, December 23-24, 2008

payment based scheme for service differentiation is also A distributed trust-based reputation model is required forproposed in [19]. Proposed bandwidth allocation algorithm can p2p system to avoid false reputation feedback. Such reputationprovide service differentiation .If the source node Ns has a management system should record not only trust but alsofinite upload bandwidth, which is denoted as Ws(in bits/s), and distrust for the target node for collusive attack, model mustseveral peers concurrently download the files from the source detect dishonest recommendations. A reputation-based modelnode, one can allocate the bandwidth among the competing is introduced in [9] to prevent the spread of malicious contentnodes appropriately. If a node contributes far more than his in the open community using a resource chain modelconsumption, he may earn more money than that he needed. Toget service, a node can accumulate tokens through either toprovide more files or to provide files, which are more valuable(i.e. asks for higher price). So, the bandwidth allocationalgorithm can incentivize nodes to share and provide files totheir maximum extent.

IV. SOME COMMON PROBLEMS AND THEIR SOLUTIONS

The Peer-to-Peer (P2P) paradigm of computing has beengrowing dramatically in popularity over the last decade.Consequently, large amounts of data and resources are beingshared co-operatively among P2P users on a global-scale, thatis a good sign but also there are some problems related to suchsystems some are discussed.

Problem. A number of Users do not want to share files,data, or resources rather desire to free ride on others.

Solution. Incentives to motivate peers to provide their filesare essential. An incentive model that enable users to getservice differentiation in a P2P network, based on how muchthe peers are willing to pay for a file or other resources, cansolve this problem.

Problem. Some participants consume more resources thanthey contribute

Solution. To control free-riding, an incentive model basedon peer credit or based on payment system can resolve suchproblem by getting rewards for contributing the resources andpaying for using the resource..

Problem. Open networks are not completely secure

Problem: Some malicious behavior can't be punished dueto open nature ofP2P networks

Problem. A worst condition for open networks is when agroup of malicious peers make collusive attempts tomanipulate the ratings.

Solution. In this regard, the concept of introducer can behelpful. A Peer can send a query to neighboring node about thetarget peer node and each peer node independently cancompute the trust rating of target peer node and can select the"best" peer node for downloading the target content.

305

Proceedings ofthe 12th IEEE International Multitopic Conference, December 23-24,2008

behavior due to environmental changes is the requirement ofopen networks..

Future work is to secure transactions malfunctioningavoiding inflation and deflation and provide security againstmalicious attack by implementing an honest and secure votingscheme keeping history information as well as presentcredibility. Another problem to be approached is how to build adistributed trust mechanism in a domain where sub domainexists, each peer is free to join a sub domain or a group to dobusiness for a certain time period, in such environment peermanagement frame work and security issues will be the majorproblems.

REFERENCES

[1] M. Yang, Q. Feng, Y. Dai and Z. Zhang. Multi dimension Reputationsystem combined with Trust and Incentive Mechanism in P2P FileSharing systems. In27th International Conference on DistributedComputing Systems Workshops (ICDCSW'07), 2007.

[2] S. Y. Lee, O. H. Kwon, 1. Kim and S. J. Hong. A ReputationManagement System in Structured Peer-to-Peer Networks. In 14thIEEE International Workshops on Enabling Technologies: Infrastructurefor Collaborative Enterprise (WETICE'05), 2005.

[3] Z. Li, S. Su, and F. Yang. WSrep: A Novel Reputation Model for WebServices Selection. N.T. Nguyen et al. (Eds.): KES-AMSTA 2007,LNAI vol 4496, pp. 199-208, Springer Heidelberg, 2008.

[4] H. Liul,Y. Qiu. A Reputation Model base on Transactions in Peer-to­Peer networks. In Third International Conference on Semantics,Knowledge and Grid, 2007.

[5] B. Li ,M. Xing, 1. Zhu, T. Che. A Dynamic Trust Model for the Multi­agent. In 2008 International Symposiums on Information Processing,2008

[6] R. Zhou, K. Hwang. Power Trust: A Robust and Scalable ReputationSystem for Trusted Peer-to-Peer Computing. In IEEE Transactions onParallel and Distributed Systems, YOLo 18, NO.4, APRIL 2007

[7] X. Wu, J. He, F. Xu. An Enhanced Trust Model Based on Reputation forP2P Networks. In 2008 IEEE International Conference on SensorNetworks, Ubiquitous, and Trustworthy Computing, 2008.

[8] 1. Chang, H. Wang, G. Yin, and Y. Tang. A New Reputation MechanismAgainst Dishonest Recommendations in P2P Systems. B. Benatallah etal. (Eds.): WISE 2007, LNCS vol 4831, pp. 449-460, SpringerHeidelberg, 2007.

[9] S. Lee, S. Zhu and Y. Kim

[10] . P2P Trust Model: The Resource Chain Model. In ACIS InternationalConference on Software Engineering, Artificial Intelligence,Networking, and Parallel Distributed Computing (SNPD'07) ,2007.

[11] Y. Liu, S. Yang, L. Guo, W. Chen and L. Guo. A Distributed Trust­based Reputation Model in P2P System.In Eighth ACIS InternationalConference on Software Engineering, Artificial Intelligence,Networking, and Parallel Distributed Computing (SNPD'07), 2007.

[12] Y. Jin, Z. G. and Z. Ban. Using Trust and Reputation Information toChoose Both Request Responders and Servers in Peer-to-Peer Networks.In Eighth ACIS International Conference on Software Engineering,Artificial Intelligence, Networking, and Parallel/Distributed Computing,2007.

[13] M. Harish, G.S. Mahalakshmi and T.Y. Geetha. Game Theoretic ModelFor P2P Trust Management. International Conference on ComputationalIntelligence and Multimedia Applications (ICCIMA'07), 2007.

[14] C. Niu, 1. Wang, R. Shen. A Trust-Enhanced Topology AdaptationProtocol for Unstructured P2P Overlays. In Third InternationalConference on Semantics, Knowledge and Grid (SKG'07), 2007.

[15] H. Wu, C. Shi 1, H. Chen, C. Gao. A Trust Management Model for P2PFile Sharing System. In 2008 International Conference on Multimediaand Ubiquitous Engineering (MUE'08), 2008.

[16] G. Tan and S. A. Jarvis. A Payment-Based Incentive and ServiceDifferentiation Scheme for Peer-to-Peer Streaming Broadcast. In IEEE

306

TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS,YOLo 19, NO.7, JULY 2008.

[17] A. Garyfalos a nd K. C. Almeroth. Coupons: A Multilevel IncentiveScheme for Information Dissemination in Mobile Networks.

[18] In lEE Transactions on Mobile Computing (TMC'08), YOLo 7, NO.6,JUNE 2008.

[19] Z. Guan , M. H. Durad, Y. Cao and L. Zhu. An Efficient Hybrid P2PIncentive Scheme. In Eighth ACIS International Conference on SoftwareEngineering, Artificial Intelligence, Networking, andParallellDistributed Computing (SPDN'07),2007.

[20] J. Suomalainen, A. Pehrsson and J. K. Nurminen. A Security Analysis ofa P2P Incentive Mechanisms for Mobile Devices. In The ThirdInternational Conference on Internet and Web Applications and Services(ICIW'08), 2008.

[21] Q. Huang, S. Huang and C. Gao. A

[22] Differentiated Service Based Incentive Mechanism in P2P File-sharingSystems. In IFIP International Conference on Network and ParallelComputing - Workshops, 2007.

[23] H.T. Liu, Z. X. Huang, Y. Bai and Y. H. Qiu. Auction IncentiveMechanism in P2P. In International Conference on Multimedia andUbiquitous Engineering(MUE'07), 2007.

[24] H. Tran, M. Hitchens, Y. Yaradharajan and

P. Watters. A Trust based Access Control Framework for P2P File­Sharing Systems. In Proceedings of the 38th Hawaii InternationalConference on System Sciences - 2005

[25] S. Singh and S. Bawa . A Privacy, Trust and Policy based AuthorizationFramework for Services in Distributed Environment. In IJCS vol 2,ISSN1306-4428,2007.

[26] 1. Chang and H. Wang and Y. Yang. A Dynamic Trust Metric for P2PSystems. In the proceedings of the Fifth International Conference onGrid Cooperative Computing Workshops (GCCW'06), 2006.