COMPUTER SCIENCE REVIEW 6 (2012) 145–160

Available online at [Link]

journal homepage: [Link]/locate/cosrev

Survey

A survey on Security Issues of Reputation Management

Systems for Peer-to-Peer Networks

Chithra Selvaraj a,∗ , Sheila Anand b

a SSN College of Engineering, Anna University, Chennai, Tamil Nadu, India
b Rajalakshmi Engineering College, Anna University, Chennai, Tamil Nadu, India

A R T I C L E I N F O A B S T R A C T

Article history: The objective of this paper is to present a comprehensive survey of security issues in
Received 9 May 2011 Reputation based Trust Management system (RTMS) also known in short as Reputation
Received in revised form Management Systems for P2P networks. The wide adoption of P2P computing has enhanced
29 March 2012 content publishing, pervasive information collection, streaming of real-time sensed data
Accepted 7 April 2012 and information sharing on an enormous global scale. At the same time, the open and
anonymous nature of P2P makes it vulnerable to malicious attacks and the spread of
Keywords: malware. In this paper, we discuss in detail the different security attacks on P2P systems
P2P and have categorized them as network-related and peer-related attacks. RTMS helps to
Reputation Management System establish and evaluate Trust, which is the degree of belief that is established to prove
Trust that the right user is accessing the right resource. We have explained the different Trust
Security Management schemes used in P2P networks and have compared them on the basis of
trust establishment, security features, trust evaluation and weakness. We have surveyed
the RTMSs currently in use and have compared them on the basis of reputation collection,
aggregation, computation, storage and degree of centralization of reputation computation
and management. We also present a comparison of protection provided by RTMs against
the various security attacks discussed. Open research issues and challenges that have yet to
be addressed in the design of current RTMs have been presented in detail. This survey can
be used as a reference guide to understand Trust Management and RTMS for P2P networks
and to further research in RTMSs to make them efficient, reliable and scalable to enable
and promote the utilization of P2P systems for large communities and applications.
c 2012 Elsevier Inc. All rights reserved.
⃝

Contents

1. Introduction ................................................................................................................................................................................. 146

2. P2P networking paradigm ............................................................................................................................................................. 146
2.1. P2P architectures ............................................................................................................................................................... 146
2.1.1. Unstructured overlays .......................................................................................................................................... 147
2.1.2. Structured overlays............................................................................................................................................... 147

∗ Corresponding author.
E-mail addresses: chithras@[Link] (C. Selvaraj), [Link]@[Link] (S. Anand).
1574-0137/$ - see front matter ⃝c 2012 Elsevier Inc. All rights reserved.
doi:10.1016/[Link].2012.04.001
146 COMPUTER SCIENCE REVIEW 6 (2012) 145–160

2.2. Content search in P2P networks......................................................................................................................................... 148

2.2.1. Centralized Indexing in P2P networks .................................................................................................................. 148
2.2.2. Distributed Indexing in P2P networks................................................................................................................... 148
2.2.3. Hybrid Indexing in P2P networks .......................................................................................................................... 148
2.3. P2P Applications ................................................................................................................................................................ 148
3. Security attacks in P2P networks .................................................................................................................................................. 150
3.1. Network related attacks ..................................................................................................................................................... 150
3.2. Peer related attacks............................................................................................................................................................ 151
4. P2P Trust Management ................................................................................................................................................................. 151
5. Reputation based Trust Management Systems ............................................................................................................................. 152
5.1. Reputation Collection ........................................................................................................................................................ 153
5.2. Reputation Aggregation ..................................................................................................................................................... 153
5.3. Reputation Computation ................................................................................................................................................... 153
5.4. Reputation Exchange ......................................................................................................................................................... 154
6. Comparison of Reputation Management Systems ........................................................................................................................ 154
6.1. RTMS based on a centralized approach.............................................................................................................................. 154
6.2. RTMS based on decentralized approach............................................................................................................................. 154
6.3. Security and Reputation Management Systems................................................................................................................. 156
7. Open research issues in Reputation Systems................................................................................................................................ 156
8. Conclusion.................................................................................................................................................................................... 158
References .................................................................................................................................................................................... 158

1. Introduction The rest of the paper is organized as follows. Section 2

gives a brief introduction to P2P network architecture
Peer-to-peer (P2P) is a distributed network architecture that and discusses some popular P2P applications. Section 3
provides resource sharing and collaboration between peers categorizes various types of security attacks on P2P networks
that are equally privileged, equipotent participants without and classifies them based on the entity (network or
the need for a central server or authority for coordination and peer) that is attacked. Section 4 discusses the need for
control. The resources shared include processing power, data, trust and trust management systems in P2P networks.
content storage and network bandwidth. These resources Different types of trust management systems have been
are accessible by other peers directly, without the support discussed and classified on various parameters. Sections 5
of intermediary entities. P2P networks are self organizing and 6 summarizes and compares the different reputation
distributed systems where peers join and leave the network management systems. Section 7 discusses open research
dynamically. issues and challenges in designing reputation management
The open and unrestricted environment of P2P architec- systems and Section 8 gives the conclusion.
ture makes it an ideal environment for unauthorized access
to resources and information. P2P raises some serious se-
curity concerns for users as there is a complete lack of ac- 2. P2P networking paradigm
countability for the content a peer puts on the network. P2P
systems have to protect themselves from attacks by anony- 2.1. P2P architectures
mous malicious peers who infect the network with unreliable
resources that can potentially destroy data and infect pro- The two primary types of network architecture are client-
grams on a peer’s computer system [1]. server and peer-to-peer networks. In client-server networks,
This paper presents a comprehensive survey of security a central server handles all security and file transactions
issues in Reputation based Trust Management system (RTMS) while in peer-to-peer networks; each machine shares its own
also known in short as Reputation Management Systems resources and handles its own security. The comparison of
for P2P networks. Attacks by anonymous malicious peers client-server and peer-to-peer architecture models is given in
have become increasingly common in many of the popular Fig. 1 [2].
P2P networks. Peers must be able to determine whether the The primary advantage of P2P networks is that all clients
requesting peers are who they claim to be and also be able to contribute and share their own resources and all peers are
distinguish between reliable and malicious peers. Peers must both users and providers of resources. Hence in comparison
be able to establish trust with the requesting peers before to client–server networks, P2P networks are more robust and
allowing them access to their shared resources and services. scalable as there are no central authorities or servers. The
Many trust models have been proposed for P2P networks. P2P overlay network consists of all the participating peers as
This survey can be used as a reference guide to understand network nodes. There are communication links between any
Trust Management in general and RTMS in particular detail. two nodes that know each other: i.e. if a participating peer
In addition, this paper attempts to address open research knows the location of another peer in the P2P network, then
issues and challenges to further research in RTMSs that would there is a directed edge from the former node to the latter in
enable them to be efficient, reliable and dynamically scalable. the overlay network. Based on how the nodes in the overlay
 COMPUTER SCIENCE REVIEW 6 (2012) 145–160 147

Fig. 1 – Client–server and P2P models.

network are linked to each other, we can classify the P2P 2.1.2. Structured overlays
networks as unstructured or structured. Structured Overlay is an architecture in which nodes
cooperatively maintain routing information about how to
reach all nodes in the overlay [7]. Structured overlay networks
2.1.1. Unstructured overlays let the set of cooperating peers act as a distributed data
An unstructured overlay is an overlay in which a node structure with well-defined operations. A Distributed Hash
relies only on its adjacent nodes for delivery of messages Table (DHT) is a structured overlay that uses key-based
to other nodes in the overlay. Example message propagation routing for indexing in which each peer is assigned to
strategies are flooding and random walk [3]. Unstructured maintain a portion of the DHT index. The address space is
P2P systems do not maintain network structure and establish virtualized and addresses are randomly assigned. Peers that
links arbitrarily. Unstructured overlay designs of P2P systems are neighbors in the overlay can be far-away in the underlying
include Freenet, Gnutella, FastTrack, Fast Freenet, Local network. Structured overlays are differentiated according to a
Minima Search (LMS). variety of dimensions such as: maximum number of hops for
routing a request, routing algorithm, node degree with size
Unstructured overlays are usually further distinguished
of overlay, overlay geometry and lookup type. Chord, Pastry,
by how search requests are propagated and by differences
Tapestry, Kademlia, HyperCup, P-Grid, Koorde, CAN are some
in link formation with neighboring peers. Flooding is the
of the P2P systems based on structured overlays [8].
fundamental approach of search where queries are forwarded
Chord [9] is a well known DHT-based distributed protocol
to all connected peer nodes to check whether they have that is aimed to efficiently locate the peer node that stores
the requested file. Gnutella adopts a flooding algorithm for a particular data item. Peer nodes are arranged in a ring
searching files [4]. that keeps the keys ranging from 0 to 2m − 1. A consistent
Random walk is another method of search which enables hashing is used to assign items to nodes, which provides
peer nodes to forward a query to a randomly chosen neighbor load balancing and only requires a small number of keys
rather than broadcast the query to all its neighbors. For to change when nodes join or leave the network. The
example, LMS is a P2P system which uses a Random Walk hash function assigns each node and each key an ID using
search in which the objects are proactively replicated using SHA-1 hashing algorithm.
consistent hashing of object identifiers to place objects at Pastry [10] is a prefix-based routing system using a
close node identifiers [5]. proximity metric. Each node in the Pastry network has a
unique identifier (nodeId). When presented with a message
Yang and Garcia [6] have presented an Iterative Deepening
and a key, a Pastry node efficiently routes the message to
technique for resource discovery in Unstructured P2P
the node with a nodeId that is numerically closest to the key,
systems. Iterative Deepening is a method of search which
among all the currently live Pastry nodes.
enables query originators to use successive Breadth First
Content-Addressable Network (CAN) [11] generalizes the
Search (BFS) queries with increasing depths, until the request
DHT methods used in Chord and Pastry. A CAN identifier
is satisfied or the maximum depth is reached.
space can be looked at as a d-dimensional version of
Guided Search is a search method which enables peer Chord and Pastry identifier space. Higher dimensions in
nodes to maintain additional information about other peer the identifier space reduce the number of routing hops but
nodes in the network like network topology and resource slightly increase the size of routing table saved in each node.
locations. For example, Napster uses a centralized server to Hence, it can be seen that structured P2P networks use a
maintain such additional information. KaZaa and JXTA utilize global protocol to route search requests among the peers in
a set of supernodes to maintain the additional information the network and in DHT based P2P systems, “a set of keys is
about the leaves and other supernodes. associated with addresses in the address space such that the
148 COMPUTER SCIENCE REVIEW 6 (2012) 145–160

nearest peer to an address stores the values for the associated

keys, and the routing algorithm treats keys as addresses” [12].
Structured Overlay networks provide a limit on the
number of messages needed to find any object in the overlay
as compared to unstructured overlays. To retrieve files that
are accessed infrequently, more structured overlay links
are required. Hence, topology-aware overlays use different
methods and metrics for determining the neighbor peers in
the overlay.

2.2. Content search in P2P networks

Searching is an important step in data access. P2P networks

take advantage of the distributed resources at peer nodes
to scatter and duplicate data contents among the peers.
The characteristics of the data content, the meta-data are
also distributed. The capability of Content search schemes
Fig. 2 – Centralized indexing in P2P networks.
is largely dependent on content indexing and management
schemes and also on the underlying P2P network [13]. Based
on content indexing, the P2P networks can be categorized
as Central Indexing, Decentralized Indexing and Hybrid into multiple groups. In each group, there is a special peer
Indexing P2P systems. called Super Peer to serve the regular peers within the same
group. Resource search query first searches locally at the
local peer and the super node that is connected to the local
2.2.1. Centralized Indexing in P2P networks
peer directly. If content object is not found in local indices,
Some P2P models use a central server to store peer
the query is propagated to other super nodes according to
information and respond to the request for information
the routing table until it is found or a predefined Time-To-
stored on the server. However, in contrast to the traditional
Live (TTL) threshold is reached. Super Peer networks strike a
client server model, this model uses server to store only
balance between the inherent efficiency of centralized search,
the metadata information about the files. In this model, a
and the autonomy, load balancing and robustness to attacks
server is approached first to obtain meta-information as given
provided by distributed search. This model is represented in
in Fig. 2, such as the identity of the peer and other stored
Fig. 4. Examples of hybrid P2P networks are KaZaa, JXTA etc.
information that may be used to verify security credentials.
From then on, the P2P communication for resource sharing
2.3. P2P Applications
happens directly between serving peer and requesting peer.
Examples of centralized P2P networks include Napster,
P2P networks are currently being used for a variety of
BitTorrent, Groove, Aimster, Magi, Softwax, and iMesh.
applications such as file sharing, digital library, video and
voice calls and video streaming.
2.2.2. Distributed Indexing in P2P networks
File sharing is perhaps one of the most commonly used
P2P networks are said to be pure P2P networks if a applications of P2P networks. Peers share files and allow other
central server is not used for managing the network [14]. peers to download these files. This differs from traditional
A query is forwarded to neighborhood peers based on a file downloading where centralized servers hold the resources
peer routing table until the target object index is found. A that users access to download the required data. Such
peer searches for resources by flooding the query through its servers must be capable of storing and retaining enormous
neighboring peers. The peer(s) with the appropriate resource amounts of data and have also to bear the high cost of
responds to the query. Resource sharing happens in a direct providing such a service. P2P networks are decentralized
connection established between the serving and requesting distributed systems that are highly scalable and enable peers
peer. Dynamic routing table mechanisms may be adopted to share and integrate their computing resources, data and
to avoid network flooding. This model is represented using services. Some of the popular P2P file sharing applications
Fig. 3. Most popular pure P2P models are Gnutella, eDonkey, include Napster, Gnutella, KaZaa, BitTorrent, eDonkey, eMule,
FastTrack, Freenet, OverNet etc. Limewire, FastTrack, Freenet, OverNet etc.
Napster was one of the first widely used P2P music sharing
2.2.3. Hybrid Indexing in P2P networks application. To participate in the Napster network, new users
Hybrid P2P networks follow a hierarchical architecture in needed to register with the Napster server and publish a
which certain peers known as Super peers or Super nodes are list of files that they were willing to share. To search for a
designated with the responsibility of providing service to the required file in the network, users could refer to the server
other peers in the network. A super node is a temporary index and retrieve the list of providers. File transfer takes place
server for other peer nodes. Peer nodes with high computing between the peers without the need for intervention by the
power and fast network connection automatically become Napster server. In contrast to Napster, the Gnutella network
super nodes. In this type of network, peers are organized is a decentralized P2P file sharing network used not only for
 COMPUTER SCIENCE REVIEW 6 (2012) 145–160 149

Fig. 3 – Decentralized Indexing in P2P networks.

Fig. 4 – Super peer based P2P system architecture.

file storage but also for content lookup and query routing. In and replicated across the participating computers [17]. KaZaa
the Gnutella network, each peer node uses a Breadth-First uses a hybrid P2P network in which it uses a two-level hierar-
search mechanism to search the network by broadcasting the chy with supernodes and leaves. A supernode is a temporary
query to all connected peer nodes. Each peer node receiving index server for other peer nodes. Any requesting peer can
the query will check the local file storage, and respond to the retrieve the index of the file from supernode. The supernode
query if at least one matched filed is found [15]. communicates with other supernodes if it does not have the
BitTorrent is currently one of the widely used applications requested file. Then the file is transferred directly between
designed to distribute large amounts of data. In order to the requester and provider.
share a file or a group of files, users need to create a small Digital Library applications have been developed for
(dot) .torrent file that contains the address of the tracker searching the relevant content on P2P networks. P2P-4-DL is
machine that launches the file distribution. The .torrent file is one of the widely used Digital Library applications that are
published on well known web sites and other users can find widely used [18]. A typical DL may store documents, images,
and download the .torrent file of interest. In order to facilitate sounds and video media. References to the documents like
efficient file download, a file is broken into smaller fragments. title, authors and keywords are then registered with the index
The client, attempting to download the file, simultaneously peer, allowing for a Napster style search system.
connects to these peer nodes that are participating in file Skype is a popular and widely used voice-over P2P (VoP2P)
transfer and downloads different pieces of the file from application. Skype provides services like P2P voice and video
different peer nodes [16]. calls, voice calls to PSTN endpoints, file transfer, instant
Freenet is a decentralized P2P data storage system de- messaging and video conferencing [19]. Like KaZaa, the
signed to provide electronic document exchange. In contrast Skype protocol is encrypted and the specifications of the
to Gnutella, Freenet acts as a P2P storage system that en- protocol have not been released for public consumption.
ables users to share unused local storage space for popular file Skype uses a super peer model, and the super peers support
replication and caching. The stored information is encrypted Network Address Translation (NAT) traversal for connecting
150 COMPUTER SCIENCE REVIEW 6 (2012) 145–160

peers behind NATs. Super peers also act as media relays. want to guarantee that data stored by a user will always be
Experimental studies of Skype have shown a significantly available to the user with high probability and that it will
higher node life time when compared to P2P file sharing persist in the network with a much higher probability. The
systems. dynamic nature of the systems may impose limitations on
P2PTV is an application software which is used for the data consistency and availability. If the rate of change of data
redistribution of video streams in real time on P2P network. is high, then the maintenance of globally accessible indexes
The distributed video streams are typically TV channels from becomes unaffordable as the number of peers in the system
all over the world. Various models are used including torrent- grows dynamically.
style distribution, application layer multicasting and hybrid The situation is made even more difficult when adver-
CDNs (Content Delivery Network). The architecture of many saries are actively attempting to corrupt the content that
P2PTV networks can be thought of as real-time versions of the peers provide. Objects may be modified by malicious
BitTorrent. If a user wishes to view a certain channel, the peers leading to Object Corruption. Object Identifier corrup-
P2PTV software contacts a “tracker server” for that channel tion, [22] exploits weakness in pervasive environment to
in order to obtain addresses of peers who distribute that generate multiple false identifiers for a single object. Decoy
channel. The user then contacts these peers to receive the insertions consist of inserting a polluted version of a file with
feed. Some of the P2PTV applications include TVUPlayer, the same metadata of the original file but with a different
PPLive, QQLive, PPStream, Abroadcasting, Zattoo, Octoshape, identifier.
LiveStation, Joost and Babelgum. A common integrity attack in P2P is related to pollution
which may lead to file-targeted DoS attack in which malicious
peer announces one or more copies of a corrupted version of
3. Security attacks in P2P networks the content [23]. Examples of such attacks include creating an
infinite loop in an application process, making it unavailable
P2P networks provide a powerful platform for the construc- by tying up essential services, and the use of Zip bombs. A
tion of a variety of decentralized services. P2P networks offer Zip bomb is a file containing multiple nested compressed files
many advantages to users but it becomes a challenge to re- that expand exponentially when unzipped. A zip bomb, also
sist and protect against various attacks on security. The secu- known as decompression bomb, is a malicious archive file
rity attacks on P2P systems have been broadly categorized and designed to crash or render useless the program or system
discussed as attacks on P2P networks and attacks on peers. reading it. In another integrity attack, the false attack reply, the
malicious peer intercepts a reply from a query and announces
3.1. Network related attacks itself as having the resource; if selected, the peer sends a
corrupted copy [23].
Sharing of resources and services is the principle application P2P overlay networks are constructed in order to route
of P2P networks. There are a number of challenges associated messages to destinations not specified by IP addresses. Each
with maintaining this shared feature because of the highly node maintains a set of links to other nodes (its neighbors or
unstable nature of P2P networks. This section covers a wide routing table). Communication with remote nodes is provided
range of network related issues starting from storage and by sending messages hop-to-hop across this overlay network.
indexing, replication and searching to security and privacy. Poisoning the routing table of the peers leads to a Bandwidth
The membership of P2P system is relatively unpredictable, Flooding DDoS attack. The incorrect lookup routing attack is the
ad-hoc and dynamic. It is necessary to maintain the location attack by which malicious nodes forward lookups to incorrect
of resources using indices for decentralized structured P2P or nonexistent nodes [19].
systems. An index in a P2P network is a set of mappings A slow peer is a peer that cannot keep up with the rate at
from keys to values [20]. For a P2P file sharing network, the which the routing software is generating update messages.
keys are file hashes and the values are locations at which the There are many reasons like less bandwidth, heavy CPU load
file corresponding to the file hash is present. Index poisoning for a peer to be slow peer. When a slow peer is present in
consists of inserting bogus records in P2P indices making the an update group, the number of formatted updates pending
system fail to locate an existing file [21]. transmission builds up [24].
Data placement is largely demand driven and must The frequency and magnitude of peer-to-peer (P2P)
be done in a distributed manner. Data Placement Problem enabled Denial-of-Service (DoS) attacks are increasing. For
highlights the need to place data in strategic locations example, DC++ is a P2P file sharing software which has
for easy and fast access [15]. Freenet is a P2P system created an attack emanating from hundreds of thousands of
which consists of a network of peers that host encrypted Internet protocol addresses (IPs), with many of the attacks
documents. Peers use keys to access the contents of the producing more than a gigabit of junk data every second. The
requested documents. Each node, in addition to forwarding sheer number of Internet addresses has caused problems for
the document towards its destination can maintain a local routers and firewalls [25].
copy of the document. The locally cached documents will not Peers join and leave the network dynamically in large
be encrypted and can be subjected to security attacks like numbers. Accelerated arrival and departure may lead to
unauthorized modification. excessive churn. For example, Rhea et al. [26] make use of a
For most of the applications, users require certain FreePastry implementation to discover that most lookups fail
guarantees on the reliability or availability of system services. to complete when there is excessive churn. They claimed that
For example, a distributed data storage application would short-lived peers leave the overlay with lookups that have
 COMPUTER SCIENCE REVIEW 6 (2012) 145–160 151

not yet timed out. A few Plaxton-based [27] schemes would can launch an eclipse attack, control must be gained
perform well under churn and it has been proven that peer over a certain amount of nodes along strategic routing
departures and arrivals would affect a logarithmic number of paths. If successful, the attacker can mediate most or all
peers. communication to and from the victim and can isolate the
Peer Anonymity is an attractive feature of P2P systems. node which is attacked [34]. Further, in scenarios where a
Each peer connects to a small number of known nodes. large number of peers refuse to collaborate, honest peers may
Only the direct neighbors of a node know its IP address. be eclipsed by malicious ones and the effects can be very
However, anonymity adds to the security vulnerabilities damaging.
and also enables malicious users to shelter and disguise
themselves [28].
4. P2P Trust Management
3.2. Peer related attacks
P2P systems are decentralized in nature and hence it is
One of the common problems is non-cooperative peer or selfish difficult to implement security protections as compared
peers who act as free-riders. Most of the current P2P systems to centrally administered systems. The absence of a well
assume equal participation and that all peers are willing defined defensible border means that it is hard to distinguish
to contribute their resources. Such selfish peers frequently malicious peers from honest peers. Strategies for tackling the
access and use the available shared resources but are not so security challenges also have to be decentralized in nature.
forthcoming with sharing their own resources [29]. Trust can serve as the metric to decide the accessibility
The risks associated with P2P unsolicited messaging, the of peers to shared resources. Trust in P2P systems is a peer’s
speed with which they spread and the extent of potential belief in another peer’s identity, reliability and capability
damages are staggering, and increasing exponentially. based on its own experiences [35]. Trust Management entails
Spammers, who can only survive if their messages get to the collecting the information necessary to establish a trust
maximum number of people at the lowest possible cost, have relationship and dynamically monitoring and adjusting the
started to use P2P networks to send unsolicited messages or existing trust relationship [36].
turn PCs into spam gateways [30]. Trust Management is a successful approach for encour-
A Sybil attack is one in which an attacker subverts the aging honest and cooperative behavior among peers. Trust
reputation system of a peer-to-peer network by creating a management systems can be broadly classified into three
large number of pseudonymous entities, using them to gain categories, policy-based trust systems, social network-based
a disproportionately large influence. Entities in peer-to-peer trust systems and reputation-based trust systems. The cate-
networks use multiple identities for purposes of redundancy, gories are represented in Fig. 5. Various Trust Models based
resource sharing, reliability and integrity. A faulty node or on these trust systems have been introduced to evaluate peer
an adversary may present multiple identities to a peer- trustworthiness and reliability of the shared resource.
to-peer network in order to appear to be functioning as In credential and policy-based trust management systems,
distinct nodes. By becoming part of the peer-to-peer network, peers use credential verification to establish a trust relation-
the adversary may then overhear communications or act ship with other peers [37]. The primary goal of such systems
maliciously [31]. is to enable access control; their concept of trust management
Peers may misbehave in the network and may gain a is limited to verifying credentials and restricting access to
bad reputation. But those peers may leave the system and resource according to application-defined policies. Resource
join with new identities leading to a White Washing attack. sharing is based on the trust established from the credentials
Whitewashers leave the system and rejoin with new identities of the peers.
to avoid reputational penalties [32]. PolicyMaker [38] is a trust management system that
Some peers who acquire a good reputation may try to facilitates the development of security features including
misuse it and this may lead to a Traitor attack. This technique privacy and authenticity for different kinds of network
is effective when increased reputation gives a peer additional applications. It provides each peer with local control to
privileges, thus allowing malicious peers to do extra damage specify its policies: using PolicyMaker a peer may grant
to the system when they defect. An example of traitor another peer access to its service if the providing peer can
attack are eBay merchants that participate in many small determine that the requesting peer’s credentials satisfy the
transactions in order to build up a high positive reputation, policies. Other Policy based Trust systems like SPKI/SDSI [39],
and then defraud one or more buyers on a high-priced KeyNote [40], DelegationLogic [41] use credential verification
item [33]. to establish trust relationships for access control. These
In many situations multiple malicious peers acting to- systems are based on the notion of delegation, whereby one
gether can cause more damage than each acting indepen- entity gives some of its authority to other entities.
dently. This is termed as a Collision attack. The malicious Social network based trust systems uses the social rela-
peers acting together would try to provide false testimony to tionship between peers for computing trust and reputation
honest peers. This is especially true in peer-to-peer reputa- values. These systems are community based systems which
tion systems, where covert affiliations are untraceable and provide access to the resource based on the analysis of the
the opinions of unknown peers impacts one’s decisions [31]. social relationship that the peer possesses within its com-
In an Eclipse attack, a set of malicious peers coordinates munity. Marsh [42] is among the first to try to give a formal
to isolate one or more honest peers. Before an attacker treatment of trust that could be used in computer science.
152 COMPUTER SCIENCE REVIEW 6 (2012) 145–160

Fig. 5 – Categorization of trust management systems.

Table 1 – Comparison of trust management systems.

Trust management Policy based trust systems Reputation based trust Social network based trust
systems systems systems
Trust establishment Based on credential verification Based on recommendation of Based on social relationships
other peers between peers
Primary security Access control, privacy, Identification of malicious peers Anomaly detection, malicious peer
features authentication using recommendation detection based on behavioral
measures
Trust evaluation Local peer Local and global trust evaluation Local peer
Weakness Trust worthiness if checked only for Malicious peers can provide false The agents cannot collectively build
the service providers and the provided recommendation and can also a network of trust because the trust
services. Trust of service requesters collude with other peers to falsify establishment depends on every
not taken into consideration the recommendation peers’ own experiences
Examples PolicyMaker, SPKI/SDSI, KeyNote, DMRep, EigenRep, P2PRep, XRep, Marsh, Regret and NodeRanking
DelegationLogic NICE, PowerTrust, PeerTrust

His model is based on social properties of trust and presents 5. Reputation based Trust Management Sys-
an attempt to integrate all the aspects of trust taken from so- tems
ciology and psychology. Several limitations exist in his simple
trust model: too strong a sociological foundation makes the Reputation management has come into wide use with the
model rather complex and it cannot be easily implemented; recent advent of widespread P2P computing. Reputation is a
the agents cannot collectively build a network of trust be- measure that is derived from direct or indirect knowledge on
cause the model puts emphasis on an agent’s own expe- earlier interactions with peers and is used to assess the level
riences. Regret [43] and NodeRanking [44] are some of the of trust a peer places on another peer [46].
social network based trust systems. Regret establishes peer
Most P2P applications happen between users who are
communities and peer members are considered trustworthy.
virtual strangers to each other. It is very difficult to figure
NodeRanking provides trust score ranking based on the social
out the ‘trustworthiness’ of other users and the chances of
network behavior of peers.
getting cheated are very high. In such a scenario, reputation
Reputation is the measure of trust collected from other
systems help in making an informed decision regarding the
peers through direct or indirect knowledge of earlier
‘trustworthiness’ of other users based on their reputation [47].
transactions with them. These recommendations serve as the
metric to decide on the accessibility that is to be provided The trust evaluation can be done via two approaches.
on the resources being shared. A reputation system in a fully In the first approach, only the direct transaction partners
decentralized environment must possess the ability to collect of a peer can express their opinion on the reputation of
and aggregate the opinions of the users on the quality of the the peer [48]. A practical example is the eBay reputation
resources and services received by other users [45]. RTMs have system. After each transaction at eBay, the buyer and the
been explained in greater detail in the next section. seller rate each other with a positive, negative and neutral
The comparison of various Trust Management Systems is feedback. The reputation is calculated at a central server by
represented in Table 1. assigning 1 point for each positive feedback, 0 point for each
 COMPUTER SCIENCE REVIEW 6 (2012) 145–160 153

neutral feedback and −1 point for each negative feedback. how to determine the accuracy of the collected opinions
The reputation of a participant is computed as the sum of its and how to aggregate the conflicting opinions to yield a
points over a certain period. global reputation. RTMSs use various predefined criteria for
In the second approach, the reputation of a peer is processing complex data to report reputation. A reputation
computed based on the opinion of its direct transaction system in a fully decentralized environment must possess the
partners as well as some third-party peers [49]. In this ability to collect and aggregate the opinions of the users on
approach, a peer A that wishes to know the reputation of the quality of the resources and services received by other
another peer B, can ask some peers (e.g., its neighbors) to users [28]. Various techniques available for aggregating the
provide their opinion on B. A then combines the opinion from
recommendations received from other peers.
the peers to calculate B’s reputation. Clearly, this model is
more like our real social networks, where third-party peers • Keeping track of past recommendations and weight the
besides transaction partners can express their opinion on a feedback according to the credibility of the recommender
peer. But it takes more cost to collect and aggregate third- peers.
party opinion. In a fully distributed P2P system involving • Use of suspicious transactions to measure the credibility
numerous peers, a peer often cannot assess another peer’s of recommender peers.
reputation effectively, but rather must rely on collective
• Use of trust and reputation values as credibility metrics for
opinions from other peers.
the recommender peers as in EigenTrust, Fuzzy Trust.
RTMSs typically perform the trust evaluation in four
phases of Reputation Collection, Reputation Aggregation, • Use of different score managers to compute the trust value
Reputation Computation and Reputation Exchange. and using the majority vote to eliminate the false reports
by malicious score managers.
5.1. Reputation Collection • Multivariate Outlier detection technique—this technique
is used to detect liar peers as in FineGrainedTrust [50].
The most common ways of collecting the reputation values • Use of Beta distribution based on previous recommenda-
are by the Transitive trust method, Collecting trust values tions.
from the neighbors, the Clustering method and Trusted Third • Enforcing policies—in eBay, the net effect on seller’s
Party. feedback score is based on the number of negatives,
• Collecting from Neighbors—the requester peer can use neutrals and positive received [51]; when the seller
its neighbors in the overlay network to get reputation receives multiple feedbacks from the same buyer within
information using polling algorithms as in the P2Prep the same week.
protocol.
• Transitive Trust method—Any trustworthy peer would
5.3. Reputation Computation
query the trustworthy peers that they have interacted
with. In the transitive trust method, a peer tends to trust
those peers who have a high reputation in the opinion After filtering the feedback and getting rid of dishonest
of trustworthy peers. For better reputation values, the reports, the data gathered from different recommender
transitive trust technique can also combine negative and peers along with the local trust data available at the
positive opinions expressed by peers to reach a global requester peer is used for reputation computation. Different
consent on trust for each peer of the network [31]. approaches have been proposed to aggregate and synthesize
The reputation information is available as metadata the trust values received from the recommender peers for the
in the downloaded files. A peer sends a query to its generation of reputation value for a provider peer.
neighboring peers to obtain the reputation information
from files download by those peers. The requesting peer • Deterministic approach—A peer’s reputation is based on
can then analyze the reputation information to decide the simple summation or average of collected ratings. In
trustworthiness of peers [32]. BinaryTrust, reputation is computed based on the number
• Clustering Method—In clustering method the peers are of negative complaints. The reputation scheme used in
named as servents and Super Peers. The opinion about all eBay is based on the sum of the number of positive and
the servent peers is collected and maintained by the Super negative ratings, in Amazon, the reputation is computed
Peer. The servents can collect the reputation about other based on the average of all the ratings.
peers from the Super Peer before processing a resource • Probabilistic approach:
request [45]. ◦ Bayesian approach: The Bayesian approach uses a
• Trusted Third Party—Trusted Third party (TTP) maintains
probabilistic approach which is based on Bayes formula.
the reputation about every peer. It has to update the
Bayesian systems take binary ratings as input, and
reputation periodically by querying the peers. Any peer
compute reputation scores based on statistical update of
that is interested in knowing the reputation of a peer can
beta probability density functions. Power Trust [52] uses
retrieve it from the TTP.
Bayesian method to generate local trust scores.
5.2. Reputation Aggregation ◦ Maximum Likelihood Estimation (MLE): MLE [53] uses
a probabilistic approach to compute the reputation
Reputation is hard to quantify because many dynamic factors value based on the probability of recommender peers to
are involved. This also introduces other challenges such as provide inaccurate information.
154 COMPUTER SCIENCE REVIEW 6 (2012) 145–160

• Fuzzy Logic—Reputation can be represented as Fuzzy peers that have provided authentic files to it. The global
values which are imprecise and not accurately quantified. reputation of each peer is given by the local trust values
Different factors can be represented by fuzzy sets and assigned to the peer by other peers, weighted by the global
membership functions are used. In FuzzyTrust, fuzzy reputation of the assigning peers. This trust algorithm
inferences are used to produce local trust values and requires the aggregation of local trust values for computing
aggregate them to global reputation values. P2Prep uses the global trust score which assumes that a central server
fuzzy values for local trust. knows all local reputation values. Kamvar and Schlosser [60]
• Flow Models—Systems that compute trust or reputation projected the idea of computing a global reputation value
based on transitive iteration or arbitrarily long chains as for a peer by calculating the left principal eigenvector of
in EigenTrust and FineGrainedTrust. a matrix of normalized local reputation values for a P2P
file sharing application. This approach has been applied to
5.4. Reputation Exchange satisfy the distributed nature of P2P systems and uses polling
mechanism for reputation collection. Distributed Hash Table
In Reputation Exchange phase, the reputation information (DHTs) are used to store the trust values. However, for new
about a peer has to be exchanged with other peers in peers, a centralized authority is required to assign a position
a secured manner. Many reputation systems are built on in the hash space.
positive reputation only [54], where false accusations are not Adrian Alexa [66] proposed a RTMS using an Eigen Trust
an issue since no negative information is kept. However, algorithm to identify the sources of inauthentic files and
the incorrect disseminated information may result in a disseminate the information to other peers. This is done
good reputation for misbehaving nodes. Some reputation by assigning the peers global trust values based on the
systems add privileges to accumulated good reputation like previous peer’s behavior. The global trust value computation
auctioning [55] and Beta reputation system [56] and these is centralized but it can compute the global trust values in
systems use a centralized approach for exchanging the a distributed way, where every peer has to compute its own
reputation information. global trust value. This may lead to malicious peers lying
For distributed systems, Jurca and Faltings [57] aim about their global trust value. This system does not fully
for an incentive-compatible mechanism by introducing support peer anonymity.
payment for reputation. The peers pay to receive reputation Shanshan Song et al. [67] have proposed a P2P reputation
ratings from so-called R-agents, which in turn pay peers system based on fuzzy logic inferences, for better handling
providing the information. To encourage the exchange of of uncertainty, fuzziness, and incomplete information in
reputation information, Pinocchio [58] rewards participants peer trust reports. It was tested with eBay transactions
that advertise their experience to others and uses a which have a centralized reputation system. Peers perform
probabilistic honesty metric to detect dishonest users and
fuzzy inference on local parameters to generate the local
deprive them their rewards. However, this method does not
scores. The system uses fuzzy inference to obtain the global
provide protection against conspiracies or bad-mouthing.
reputation aggregation weights. The fuzzy trust aggregation
reduces the message overhead when compared to Eigen trust
because of the usage of fuzzy inferences.
6. Comparison of Reputation Management
Systems
6.2. RTMS based on decentralized approach
Reputation system uses different techniques for evaluating
trust based on the reputation information. A number of With decentralization, each peer will act as an agent and
RTMS have been suggested for P2P networks and these be allowed to take responsibility for trust evaluation based
include DMRep [59], EigenRep [60], XRep [61], P2PRep [62] and on trust policies. A peer’s trust policy is individualistic and
NICE [63]. The role of the recommender in the domain of need not be communicated to other peers, so this could
the target being recommended plays a major role in trust lead to ambiguity in trust evaluation. Each agent then makes
evaluation [64]. decision for itself, on its own policies. The disadvantage of
decentralization is that more responsibility and expertise
6.1. RTMS based on a centralized approach is required from the agent for managing trust policies.
However this responsibility can be assigned to their trusted
The RTMS based on a centralized approach needs a central authority if needed. Decentralization does not completely
server to compute the trust score based on the local replace current centralized approaches, but it gives agents a
reputation values stored with every peer. The model proposed choice of managing their own trust policies.
by Gupta et al. [65] uses a centralized approach for tracking a Loubna Mekouar et al. [68] have put forward an effort
positive peer’s contribution to the system using a credit–debit to develop a new and simple reputation management
mechanism. Each peer computes the reputation based on scheme for partially decentralized peer-to-peer systems. The
its activity with other peers and stores it locally and the reputation scheme helps to build trust between peers based
Reputation Computation Agent (RCA) periodically collects on their past experiences and the feedback from other
reputation from peers. peers. They also proposed two selection advisor algorithms
Eigen Trust [48] makes use of a transitive definition of for helping peers to select the right peer for download,
trust where any peer will have high opinion about those by collecting the reputation from neighboring peers. In
 COMPUTER SCIENCE REVIEW 6 (2012) 145–160 155

this technique, the system overhead will be high and the determine object authenticity, the degree to which an object’s
anonymity of the peer may not be maintained. data matches with its advertised description. It computes
Li Xiong, and Ling Liu [59] have proposed a trust model the reputation scores based on the statistical measure of
that uses community based reputations which can be the reliability of past voting habits. The peers can learn
computed through feedback received from other peers in relationships even in the absence of direct observations or
the community about a peer’s transaction history. Trust interactions since it uses a trust computation mechanism
context can vary from community to community and based on the flow. This system provides incentives for peers
from transaction to transaction. They have introduced two to participate honestly in voting.
adaptive factors, transaction context factor and community Runfang Zhou and Kai Hwang [52] have designed a P2P
context factor as metrics to allow the system to adapt to reputation system that is based on the Power Trust system.
different domains. However, this model does not address The system selects few nodes to be power nodes and assumes
collision attacks and sudden and malicious attacks. Also, those nodes to be trustworthy. Those nodes will help in
this model requires prior building of the community before collecting locally-generated peer feedbacks and aggregate
transactions can happen and hence will not provide full them to yield the global reputation scores. This paper uses
anonymity. a structured trust overlay network (TON) to model the trust
Ernesto Damiani [62] addresses the problem of spreading relationships among peers and a power-law distribution in
malicious information due to peer anonymity. A self regu- user feedbacks. However, in a large P2P system with frequent
lating system is proposed to implement a reputation mech- peer joining and leaving, we cannot assume that there always
anism. Reputation sharing is realized through a distributed exist some static and predetermined power nodes.
polling algorithm by which the resource requestors can Runfang Zhou and Kai Hwang [55] have extended their
assess the reliability of the resource. XRep [33] is a protocol work on unstructured P2P systems and designed a novel
introduced by Damiani which combines the resource and ser- mechanism of GossipTrust which resorts to gossip protocols
vents’ reputation. This work was proven to be secure against to aggregate global reputation scores. Each peer repeatedly
various security attacks like pseudo spoofing, Id stealth and contacts others at random, and exchanges reputation data
Shilling. However, it requires the reputation information to with them. The power nodes are dynamically chosen after
be exchanged among peers in a secure manner. every reputation aggregation. There is a tradeoff that exists
Natalia Stakhanova et al. [69] have proposed a fully de- between gossip error and convergence overhead.
centralized approach that allows computing peers’ reputation Debora Donato et al. [45] proposed a new approach to
based on the traffic between peers. This model does not em- the design of fully decentralized reputation mechanisms that
ploy a centralized storage of reputation score but computes it combine negative and positive opinions expressed by peers
on demand. This approach relies entirely on a peer’s reputa- to reach a global consensus on trust and distrust values for
tion and depends on the persistence of Peer ID which affects each peer of the network. This system concentrates on quick
anonymity. It provides a mechanism of assigning minimum retrieval of the trust ratings of the peers and for an update of
average trust for new comers which is based on the assump- ratings based on feedback information. This system is able
tion that new comers are good peers. to detect malicious peers with a high degree of malicious
Shalendra Chhabra et al. [70] have proposed a reputation activity. However, there is the danger of assuming peers with
management protocol for Super peer based P2P networks. low malicious activity to be honest peers.
This approach uses clustering technique for reputation Ali Aydın et al. [73] have proposed a reputation system
collection which is an extension of P2Prep protocol. They also to identify malicious peers and to prevent the spread of
have proposed the use of repeaters among Super Peers to malicious content. The protocol is based on the query–
facilitate interaction between servents behind the firewalls response architecture in which the user evaluates the
and also studied the case of malicious peers, Super Peers and outcome of its past transactions and shares this information
repeaters. This system raises the issue of exposure of polling with other peers when requested. The system relies on the
to security violations. judgment of the users and is effective only against attacks
Vladislav Jumppanen [71] has introduced the concept of that the users perceive. This protocol does not distinguish
combining file reputation with peer reputation. Combining between malicious and careless peers.
peer reputation with file reputation increases the efficiency Jianli Hu et al. [74] have introduced a secure and effective
of the reputation system and consequently increases the reputation based distributed P2P global trust management
efficiency of a P2P network with a reduced number of model (DSRM), and presented its corresponding distributed
malicious peers, malicious transactions and malicious files. storage mechanism of reputation information, and security
This system allows the peers to store their own reputation protection protocol. This work is based on the clustering
and also collect from the neighbors to obtain the reputation technique to collect the reputation values. It uses a terrace
values of other peers. Since file reputation is used, the based distributed reputation storage mechanism in which
system overhead will be high when compared to a system a uni-hash function is used to provide the advantage of
that uses only peer reputation. Also the issue of secure anonymity. This system provides protection against various
reputation exchange and other security measures has not malicious behaviors and collusions, suppresses Sybil attacks
been addressed. and trust information tampering.
Kevin Walsh [72] introduced Credence, a decentralized RVVSV Prasad et al. [75] in their work have introduced a
object reputation and ranking system for large-scale peer- reputation management system based on the similarities be-
to-peer file sharing systems. This system enables peers to tween the peers. Credibility factor was assigned to the peers
156 COMPUTER SCIENCE REVIEW 6 (2012) 145–160

Table 2 – Comparison of reputation management systems.

Reputation based trust Reputation collection Reputation aggregation Reputation computation Reputation
management systems storage
Global trust model Agent based mechanism Statistical data analysis of Binary trust evaluation Decentralized
former transactions based on normalization of P-Grid
complaints
EigenRep Polling mechanism Use of trust and reputation Eigenvector of a matrix of DHT
values as credibility metrics normalized local reputation
for the recommender peers values
Fuzzy trust Query–response with the Fuzzy logic inferences of Fuzzy inference on local DHT
peers that meet a threshold local trust scores parameters
XRep Distributed polling algorithm Use of combined reputations Vote evaluation using binary Every peer’s
of servents and resources value repositories
P2Prep Enhanced polling Servent selection based on Ordered weighted averaging Every peer’s
mechanism to selected its own past experiences of votes repositories
servents
SupP2Prep Polling mechanism of super Vote aggregation based on Weighted average of positive Every peer’s
peers credibility of a peer votes repositories
Object reputation system Local trust score collection Statistical measure of Flow based voting Every peer’s
from neighbors or reliability of the peer’s past mechanism repositories
community members voting habits
PowerTrust Distributed ranking of power Applies power law Bayesian approach DHT
nodes and applies look distribution on user’s
ahead random walk strategy feedback
Gossip based trust Gossip protocol Normalization of local Matrix–vector computation Bloom filters
scores using gossip
aggregation protocol
PeerTrust Public key infrastructure Weighted sum of feedback Normalized rate on each Decentralized
factors transaction P-Grid
NICE Signed cookies Decentralized trust Weighted sum of strongest Every peer’s
inference scheme disjoint paths on trust graph repositories

based on its similarity. The reputation computation is based The various Reputation based Trust Management Systems
on feedback similarity, common vendor similarity, interac- are compared based on its functionality and tabulated using
tion similarity and age of transactions. The trustworthiness Table 2.
of any peer is viewed as the expectation of cooperative behav-
ior from that peer. It is a community based reputation system 6.3. Security and Reputation Management Systems
which requires the cooperation of the peers in the community
to which they belong. Most P2P systems work on the assumption of honest
William Conner et al. [76] have presented a reputation- cooperation of peers. But in an anonymous P2P networks,
based trust management framework that enables services to it is difficult to make the participating peers cooperate.
make customized trust level assessment of feedback from They may be selfish and unwilling to upload data to others.
More seriously, some peers may launch attacks to disrupt
many entities. This also provides a way of applying different
the service or distribute viruses in the overlay network. We
scoring functions to each entity over the same feedback data.
call these uncooperative, abnormal or attacking behavior
The trust framework provides a way of storing the feedback
malicious actions and the associated peers as malicious
on previous service interactions with clients. However, the
peers. In Table 3 we present a comparison of protection
trust framework development is based on many assumptions
provided by the various reputation systems against different
which include the absence of a Sybil attack and that secure
types of security attacks.
communication exists between the services and the trust
As seen from the table, many of the reputation systems
management service instances. It only considers the attacks
are able to differentiate malicious peers from normal peers.
characterized by negative feedback.
The P2P reputation systems that use Distributed hashing
Dewan and Dasgupta [77] have suggested a cryptographic
mechanism are able to solve the problem of Data Placement.
protocol for ensuring secure and timely availability of the The reputation systems, which are distributed and use
reputation data of a peer to other peers at extremely low hashing mechanism for indexing is secure against Index
costs. The past behavior of the peer is encapsulated in its Poisoning attack.
digital reputation, and is subsequently used to predict its
future actions. The cryptographic protocol is coupled with
self-certification and cryptographic mechanisms for identity 7. Open research issues in Reputation Sys-
management and countering Sybil attack. The approach in tems
this paper is based on relative ranking of the peers. There
might be some systems that need absolute values, which the Reputation systems provide a way for building trust through
proposed approach does not support. social control by using community based feedback about
 COMPUTER SCIENCE REVIEW 6 (2012) 145–160 157

Table 3 – Comparison of protection provided by RTMs against various security attacks. ✓—Secure, ×—Vulnerable.

Reputation Sybil Collision Index Data Identifier DoS White Traitor Free Malicious
based trust attack attack poisoning place- corruption attack washing attack riding peer
management ment attack detection
systems problem
Global trust model × × × ✓ × × × ✓ × ✓
EigenRep ✓ ✓ × ✓ × × × × ✓ ✓
Fuzzy trust × ✓ ✓ ✓ × × × × × ✓
XRep ✓ × ✓ × ✓ ✓ × × × ×
P2PRep × × ✓ × × × × × × ✓
SupP2Prep ✓ × × × ✓ ✓ × × × ✓
Object reputation × × × ✓ × ✓ × × × ✓
PowerTrust × × × ✓ × × × × × ✓
Gossip based trust × × × ✓ × × × ✓ × ✓
PeerTrust × × × ✓ × × × × ✓ ✓
NICE × × × ✓ × ✓ × × × ✓

past experiences of peers to help make recommendation proper services. Hence right peer selection is an important
and judgment on quality and reliability of the transactions. aspect of any reputation system that remains still an area for
The challenge of building such a reputation based trust further research focus.
mechanism in a P2P system is to effectively cope up with • Combining peer reputation with resource reputation
various malicious behaviors of peers such as providing fake or It has been seen that combining peer reputation with file
misleading feedback about other peers, collusion attack etc. reputation increases the effectiveness of the reputation sys-
Most existing reliable reputation mechanisms require a tem and reduces malicious peers, resources and transactions.
central server for storing and distributing the reputation However, present systems have high overhead for trust in-
information. A number of RTMSs for distributed and formation exchange and computation. Hence, this is an area
decentralized environment have been discussed. But it still where research could be focused on improving the efficiency
remains a challenge to build a decentralized P2P trust of trust aggregation and evaluation.
management system that is efficient, scalable, reliable, and • Decentralized reputation storage
secure in both trust computation and trust data storage Presently much of the work done on reputation storage
and distribution. We present some issues that need to be uses a centralized database for storing the reputation
addressed in greater detail in the design of RTMSs. information. Super peers or agents are used to collect and
• Malicious reputation information share the reputation information with other peers. However,
There is a challenge in preserving anonymity while P2P systems are essentially dynamic in nature with peers
maintaining the privacy about a peer. The problem of securing joining and leaving at random intervals. So, it is not feasible
hosts on P2P network while keeping the openness of the to assume some static and predetermined peer agents for
system has been studied extensively over last couple of years maintaining reputation information. While there has been
but still remains an open research issue that has not yet some work carried out on decentralized storage of reputation
been fully resolved. Existing solutions based on reputation information, there is still plenty of scope for research for
management either employ centralized algorithms or rely on secure storage and distribution of reputation information. As
peers’ cooperation in the network. To decrease the number of global reputation scores are aggregated from local feedbacks,
downloads of inauthentic files in a peer-to-peer file-sharing the proper distribution of feedbacks play a significant role in
network each peer can be assigned with a unique global the design of an efficient reputation system.
reputation value, based on the peer’s history of uploads. • Reputation exchange
However, this also gives rise to the difficult problem of The manner in which reputation exchange is carried
preserving the integrity of file metadata and to prevent false out is a major aspect of any reputation system. Existing
modification from malicious peers. techniques like Eigen Trust and Power Trust mechanisms
• Free riding aggregates trust information from peers by having them
Free riders are peers who just download files but do not perform a distributed calculation for global trust reputation.
share anything to other peers. They significantly destroy the However, peer anonymity is not fully maintained. Protocol
philosophy of P2P file-sharing networks. Research has to be based reputation systems use secure protocols for reputation
focused on the development of acceptable mechanisms for exchange. But these protocol based approaches depend
rewarding good peers as a motivation for peers to share their on traditional security mechanisms which are not fully
resources and also to deter free riders. appropriate for decentralized P2P systems. Hence, there is a
• Right peer selection need for research focus on secure exchange of distributed
A reputation system should provide effective mechanisms reputation information.
for identifying good peers who share reliable resources. As the • Witness anonymity
reputation scheme helps to build trust between peers based A peer’s reputation should be associated with an opaque
on their past experience and transactions, it should also be identifier rather than with an externally associated identity
able to identify low performing peers that fail to provide such as a peer’s IP address. There is a need for research
158 COMPUTER SCIENCE REVIEW 6 (2012) 145–160

focus on witness anonymity which combines the seemingly [10] A. Rowstron, P. Druschel, Pastry: scalable, distributed object
conflicting requirements of anonymity for honest peers who location and routing for large-scale peer to peer systems, in:
report on the misbehavior of other peers and accountability IFIP/ACM International Conference on Distributed Systems
Platforms, Heidelberg, Germany, 2001.
for malicious peers that attempt to misuse the anonymity
[11] Sylvia Ratnasamy, Paul Francis, Mark Handley, Richard Karp,
feature to deprecate honest peers.
Scott Shenker, A scalable content addressable network, in:
SIGCOMM’01, August 2001, San Diego, California, USA.
[12] J. Buford, H. Yu, E.K. Lua, P2P Networking and Applications,
8. Conclusion Morgan Kaufmann, 2008.
[13] A. Crespo, H. Garcia-Molina, Routing indices for peer-to-
In this survey, we have presented an overview of P2P peer systems, in: International Conference on Distributed
networks and reviewed currently available Reputation Trust Computing Systems, Vienna, Austrian, 2002.
Management Systems for P2P. P2P networks are open to [14] In-suk Kim, Yong-hyeog Kang, Young Ik Eom, An efficient
contents discovery mechanism in pure P2P environments,
various forms of attacks, break-ins, espionage, and malicious
in: M. Li, et al. (Eds.), GCC 2003, in: LNCS, vol. 3032, Springer,
mischief. Trust Management Systems provide a proven, 2004, pp. 420–427.
efficient and feasible mechanism for identifying malicious [15] Srinivas Raaghav Kashyap, Algorithms for data placement,
peers and provide security against malicious attacks. RTMSs reconfiguration and monitoring in storage networks, Doctor
have been studied and discussed in detail with respect Dissertation, 2007, University of Maryland ACM. ISBN: 978-0-
to mathematical representation of reputation collection, 549-45087-0.
aggregation, storage and exchange of reputation information [16] Johan Pouwelse, The BitTorrent P2P file-sharing system,
and trust evaluation. We have presented a comparison Detailed Measurement Study, Published Saturday 18th
December 2004.
of protection provided by RTMSs against various security
[17] I. Clarke, O. Sandberg, B. Wiley, T.W. Hong, Freenet: a
attacks. Open research issues that need to be addressed in the distributed anonymous information storage and retrieval
design of RTMS to make them more reliable and robust have system, Lecture Notes in Computer Science (2001).
been discussed. These issues, properly addressed can enable [18] James Walkerdine, Paul Rayson, P2P-4-DL: digital library over
P2P networks to be secure, reliable and trustworthy so that peer-to-peer, in: Proceedings of the Fourth International
the benefits of scalability, efficiency, and resilience to failures Conference on Peer-to-Peer Computing, P2P’04, 0-7695-2156-
8/04 $20.00 ⃝ c 2004.
and dynamics can be fully realized. The review can also be
[19] Anil Saroliya, Vishal Shrivastava, Security problems and their
considered as a source for future research directions in RTMS
upshots in routing protocols of DHT based overlay networks,
for P2P systems.
Journal of Theoretical and Applied Information Technology
(2005).
REFERENCES [20] Eleni Koutrouli, Aphrodite Tsalgatidou, Reputation-based
trust systems for P2P applications: design issues and
comparison framework, in: TrustBus, LNCS, vol. 408,
[1] Dan S. Wallach, A survey of peer-to-peer security issues, in: pp. 152–161, 2006.
International Symposium on Software Security, Tokyo, Japan, [21] Xin Sun, Ruben Torres, Sanjay Rao, DDoS attacks by
November 2002. subverting membership management in P2P systems, in: 3rd
[2] Robin Jan Maly, E.T.H. Zurich (Switzerland), Comparison of IEEE Workshop on Secure Network Protocols, 2007, NPSEC
centralized (client–server) and decentralized (peer-to-peer) 2007, p. 1–6.
networking, Semester Thesis, March 2003. [22] Cristiano Costa, Jussara Almeida, Reputation systems for
[3] J. Buford, H. Yu, E.K. Lua, P2P Networking and Applications, fighting pollution in peer-to-peer file sharing systems, 0-
Morgan Kaufmann, 2008, p. 415. 7695-2986-0/07. [Link]
[4] S. Zhao, D. Stutzbach, R. Rejaie, Characterizing files in [23] Mudhakar Srivatsa, Ling Liu, Countering targeted file attacks
the modern gnutella network: a measurement study, in: using location guard, in: Proceedings of the 14th Conference
SPIC/ACM, Multimedia Computing and Networking, San Jose, on USENIX Security Symposium, 2005—SSYM’05, vol. 14.
CA, 2006. [24] Detecting and mitigating a BGP slow peer Cisco IOS and NX-
[5] R. Morselli, B. Bhattacharjee, A. Srinivasan, M. Marsh, OS software configuration guide, Updated July 2010.
Efficient lookup on unstructured topologies, in: Proceedings [25] [Link]
of the Twenty-Fourth Annual ACM Symposium on Principles problems-with-dc.
of Distributed Computing, Las Vegas, NV, USA, July 17–20, [26] S. Rhea, et al. Handling churn in a DHT, in: Proc. 2nd Int’l.
PODC’05, ACM Press, New York, NY, pp. 77–86, 2005. Wksp. Peer-to-Peer, IPTPS 2003, February 2003.
[6] B. Yang, H. Garcia-Molina, Efficient search in peer-to- [27] Eng Keong Lua, Jon Crowcroft, Marcelo Pias, A survey and
peer networks, in: International Conference on Distributed comparison of peer-to-peer overlay network schemes, in:
Computing Systems, Vienna, Austria, 2002. IEEE Communications—1553-877X 72 IEEE Communications
[7] John F. Buford, Heather Yu, X. Shen, et al., Peer-to-peer net- Surveys & Tutorials, Second Quarter, 2005.
working and applications: synopsis and research directions, [28] Sergio Marti, Hector Garcia-Molina, Identity crisis: anonymity
in: Handbook of Peer-to-Peer Networking, ⃝ c Springer Science vs. reputation in P2P systems, in: Proceedings of the Third
+Business Media, LLC, 2010, [Link] International Conference on Peer-to-Peer Computing, P2P’03.
387-09751-0 1. 0-7695-2023-5/03.
[8] Sameh El-Ansary, Seif Haridi, An overview of structured P2P [29] Krisztina Lója, Paolo Giaccone, Nash equilibria in bandwidth
overlay networks, Thesis, Royal Institute of Technology— allocation for non-cooperative peer-to-peer networks, Jour-
IMIT/KTH, Sweden, July 19, 2004. nal of Systems Architecture 54 (2008) 81–96.
[9] I. Stoica, R. Morris, D.H. Karger, “Chord: a scalable Peer- [30] Lin Wang, Attacks against peer-to-peer networks and
to-Peer lookup service for Internet applications”, in: ACM countermeasures, in: TKK T-110.5290 Seminar on Network
SIGCOMM, San Diego, 2001. Security 2006-12-11/12.
 COMPUTER SCIENCE REVIEW 6 (2012) 145–160 159

[31] Eng Keong Jua, Jon Crowcroft, Marcelo Pias, University— [51] Paul Resnick, Richard Zeckhauser, Trust among strangers in
A survey and comparison of peer-to-peer overlay network Internet transactions: empirical analysis of Ebay’s reputation
schemes, IEEE Communications Surveys & Tutorials 7 (2) system, in: Working Paper for the NBER Workshop on
(2005) 1553–1877. Second quarter. Empirical Studies of Electronic Commerce, 2001.
[32] Michal Feldman, Christos Papadimitriou, John Chuang, [52] Runfang Zhou, Kai Hwang, PowerTrust: a robust and scalable
Ion Stoica, Free-riding and whitewashing in peer-to-peer reputation system for trusted peer-to-peer computing, IEEE
systems, IEEE Journal on Selected Areas in Communications Transactions on Parallel and Distributed Systems 18 (4) (2007)
24 (5) (2006) 1010–1019. [Link]
[33] Sergio Marti, Hector Garcia-Molina, Taxonomy of trust: [53] Z. Despotovic, K. Aberer, P2P reputation management:
categorizing P2P reputation systems, Computer Networks 50 probabilistic estimation vs. social networks, Computer
(2006) 472–484. Networks 50 (4) (2006) 485–500.
[34] Marlom A. Konrath, Marinho P. Barcellos, Rodrigo B.
[54] Sonja Buchegger, Jean-Yves Le Boudec, A robust reputation
Mansilha, Attacking a Swarm with a Band of Liars: evaluating
system for P2P and mobile ad-hoc networks, EPFL IC
the impact of attacks on BitTorrent, in: Seventh IEEE
Technical Report IC/2003/50.
International Conference on Peer-to-Peer Computing, 0-7695-
[55] Runfang Zhou, Kai Hwang, Gossip-based reputation aggrega-
2986-0/07, 2007, IEEE. [Link]
tion for unstructured peer-to-peer networks, in: IEEE Inter-
[35] Riidiger Schollmeier, A definition of peer-to-peer networking
national on Parallel and Distributed Processing Symposium,
for the classification of peer-to-peer architectures and
IPDPS-2007. 1-4244-0910-1/07.
applications, 0-7695-1503-7102, 2002, IEEE.
[36] Huaizhi Li, Mukesh Singhal, Trust management in dis- [56] Audun Josang, Roslan Ismail, The beta reputation system,
tributed systems, IEEE Journal (2007). in: Proceedings of the 15th Bled Electronic Commerce
[37] Matt Blaze, Joan Feigenbaum, Jack Lacy, Decentralized trust Conference, Bled, Slovenia, June 2002.
management, in: Proceedings of the 1996 IEEE Symposium [57] R. Jurca, B. Faltings, An incentive compatible reputation
on Security and Privacy, IEEE Computer Society Press, 1996, mechanism, in: Proceedings of the IEEE Conference on E-
pp. 164–173. Commerce, Newport Beach, CA, USA, June 24–27, 2003.
[38] Matt Blaze, Joan Feigenbaum, Martin Strauss, Compliance- [58] Alberto Fernandes, Evangelos Kotsovinos, Sven String, Boris
checking in the PolicyMaker trust management system, Dragovic, Incentives for honest participation in distributed
in: Proceedings of Second International Conference on trust management, in: Proceedings of iTrust 2004, Oxford, UK,
Financial Cryptography, FC’98, in: Lecture Notes in Computer March 2004.
Science, vol. 1465, Springer, 1998, pp. 254–274. [59] Li Xing, Ling Liu, A reputation-based trust model for peer-
[39] D. Clarke, J. Elien, C. Ellison, M. Fredette, A. Morcos, R.L. to-peer ecommerce communities, in: 2003 IEEE International
Rivest, Certificate chain discovery in SPKI/SDSI, Journal of Conference on E-Commerce Technology, CEC’03, 2003, p. 275.
Computer Security 9 (4) (2001) 285–322. Proc. of the IEEE [60] D. kamvar, Mario T. Schlosser, Eigen trust algorithm for
Symposium on Security and Privacy, May 2002, pp. 114–130. reputation management in P2P networks, May 2003. ACM
[40] Ninghui Li, John C. Mitchell, William H. Winsborough, Design 1-58113-680-3/03/0005.
of a role based trust management framework, in: Proc. of
[61] F. Cornelli, E. Damiani, S.C. Vimercati, S. Paraboschi, P.
LMW02, 2004.
Samarati, A reputation-based approach for choosing reliable
[41] Ninghui Li, Benjamin N. Grosof, Joan Feigenbaum, A
resources in peer-to peer networks, in: CCS’02, USA, 2002.
practically implementable and tractable delegation logic, in:
Proceedings of the IEEE Symposium on Security and Privacy, [62] E. Damiani, S.D.C. di Vimercati, S. Paraboschi, P. Samarati,
IEEE Computer Society Press, pp. 27–42, 2000. Managing and sharing servents’ reputations in P2P systems,
[42] S. Marsh, Formalising trust as a computational concept, Ph.D. IEEE Transactions on Knowledge and Data Engineering 15 (4)
Thesis, University of Stirling, 1994. (2003) 840–854.
[43] J. Sabater, C. Sierra, Reputation and social network analysis in [63] S. Lee, R. Sherwood, Cooperative peer groups in NICE, in: IEEE
multi-agent systems, in: First International Joint Conference Infocom, San Francisco, USA, 2003.
on Autonomous Agents and Multi-Agent Systems, Bologna, [64] Yan Wang, Vijay Varadharajan, Role-based recommendation
Italy, 2002. and trust evaluation, in: 4th IEEE International Conference on
[44] J. Pujol, R. Sanguesa, Extracting reputation in multi agent Enterprise Computing, E-Commerce and E-Services, CEC-EEE
systems by means of social network topology, in: First 2007.
International Joint Conference on Autonomous Agents and [65] M. Gupta, P. Judge, M. Ammar, A reputation system for peer
Multi-Agent Systems, Bologna, Italy. to peer networks, in: Proceedings of NOSSDAV, 2003.
[45] Debora Donato, Stefano Leonardi, Mario Paniccia, Combining [66] Adrian Alexa, Reputation management in P2P networks: the
transitive trust and negative opinions for better reputation eigentrust algorithm, Thesis, Supervised by Anja Theobald.
management in social networks, in: Procs of SNAKDD, Las [67] Shanshan Song, Kai Hwang, Runfang Zhou, Trusted P2P
Vegas, Nevada, 2008, p. 10. Transactions with Fuzzy Reputation Aggregation, Published
[46] Karl Aberer, Zoran Despotovic, Managing trust in a peer-2-
by the IEEE Computer Society, 2005, 1089-7801/05.
peer information system, 2001. ISBN: 1-58113-436-3.
[68] Loubna Mekouar, Youssef Iraqi, Raouf Boutaba, A reputation
[47] B.S. Jyothi, D. Janakiram, Robust sybil detection strategy for
management and selection advisor schemes for peer-to-peer
P2P reputation systems built over structured overlays, 2006.
systems.
[48] Sepandar D. Kamvar, Mario T. Schlosser, Hector Garcia-
Molina, EigenRep: reputation management in P2P networks, [69] Natalia Stakhanovism, Sergio Ferrari, Johnny Wong, Ying
2003. ISBN: 1-58113-680-3. Cai, A reputation-based trust management in peer-to-peer
[49] F. Cornelli, E. Damiani, S. Vimercati, S. Paraboschi, P. network systems, 2004.
Samarati, Choosing reputable servents in a P2P network, in: [70] Shalendra Chhabra, Ernesto Damiani, Sabrina De Capitani
Proc. ACM WWW’02, 2002, pp. 376–386. di Vimercati, Stefano Paraboschi, Pierangela Samarati,
[50] Yanchao Zhang, Yuguang Fang, A fine-grained reputation SupP2PRep: a protocol for reputation management via polling
system for reliable service selection in peer-to-peer net- in P2P networks with superpeers, 2008.
works, IEEE Transactions on Parallel and Distributed Systems [71] Vladislav Jumppanen, File reputation in decentralized P2P
18 (8) (2007) 1134–1145. reputation management, HUT T-110.551, 2005.
160 COMPUTER SCIENCE REVIEW 6 (2012) 145–160

[72] Kevin Walsh, Emin Gun Sirer, Experience with an object [75] R.V.V.S.V. Prasad, Vegi Srinivas, V. Valli Kumari, K.V.S.V.N.
reputation system for peer-to-peer filesharing, in: USENIX Raju, An effective calculation of reputation in P2P networks,
Association NSDI ’06: 3rd Symposium on Networked Systems Journal of Networks 4 (5) (2009).
Design & Implementation. [76] William Conner, Arun Iyengar, Thomas Mikalsen, Isabelle
[73] Ali Aydın Selçuk, Ersin Uzun, Mark Reşat Pariente, A Rouvellou, Klara Nahrstedt, A trust management framework
reputation-based trust management system for P2P net- for service-oriented environments, WWW 2009, April 20–24,
works, International Journal of Network Security 6 (3) (2008) 2009, Madrid, Spain. ACM 978-1-60558-487-4/09/04.
235–245. [77] Prashant Dewan, Partha Dasgupta, P2P reputation manage-
[74] Jianli Hu, Quanyuan Wu, Bin Zhou, Secure and distributed ment using distributed identities and decentralized recom-
P2P reputation management, Journal of Communication 3 (7) mendation chains, IEEE Transactions on Knowledge and Data
(2008). Engineering 22 (7) (2010).