Jia et al. / Front Inform Technol Electron Eng 2017 18(4):559-569 561
can address various malicious behaviors in a P2P
community. However, the computation convergence
rate is not guaranteed in large-scale P2P systems. In
addition, the five factors used in their trust model
are costly to retrieve. PET (Liang and Shi, 2005) is
a model (Chang and Kuo, 2009) for sharing files in
P2P networks, and evaluates reliability and risk of
suppliers as resource shares. SFTrust (Zhang et al.,
2009) is a trust model based on the topology adap-
tation protocol, which was proposed in an unstruc-
tured P2P system. SFTrust separates trust between
the service provided and feedback. The reliability of
service and feedback was calculated respectively.
A robust distributed reputation and trust man-
agement scheme was proposed in M-Trust (Qureshi
et al., 2012). M-trust incorporates distributed trust
rating aggregation algorithms that acquire trust rat-
ings from direct and witness recommendations from
distant peers. The scheme uses confidence in reputa-
tion, based on interactions among peers, to decrease
the time required in computing trust ratings and
reduce the space for storing trust ratings. However,
setting threshold limits for selecting highly trustwor-
thy recommenders in a dynamic MANET environ-
ment is not an easy task. Threshold limits need to
be adjusted according to the network situation.
2.3 Group-based trust management
Sun and Tang (2007) proposed a multilayer and
grouping P2P trust model. This model avoids the
infinite iterations of global reputation. However,
the groups are based on physical distance cluster-
ing. Thus, its performance cannot be guaranteed in
MP2P networks. Tian HR et al. (2006) proposed
a group-based reputation system GroupRep to infer
the direct trust relationship between peers in P2P
networks. However, such a reputation system is not
suitable for large-scale MP2P networks due to the
message overhead of global reputation aggregation.
Largillier and Vassileva (2012) argued that many dif-
ferent contexts and groups could be formed based on
a user’s criteria or using methods that match user
desires. Al-Oufi et al. (2012) extended the Advogato
trust metric (Leskovec et al., 2010) so that trust-
worthy users can be identified. Their approach can
discover reliable users and unreliable users. How-
ever, their approach is based on some specific envi-
ronments. Easa et al. (2012) considered two factors
in their trust model, i.e., intermediate group confi-
dence and group confidence between two groups. Wu
(2011) proposed a stable group based trust manage-
ment scheme (SGTM) to construct sufficient and re-
liable trust relationships. Though their experimental
results illustrated that the model can handle peers
joining and leaving the network, they did not present
the methods to validate it. SuperTrust is a trust
model for P2P networks based on a super peer (Tian
C et al., 2010). A feedback filtering algorithm was
proposed to effectively filter fake, misleading, and
unfair feedback in the referral, but peers joining or
leaving networks/groups were not considered.
3 Dynamic grouping management
strategy
In an MP2P network, each peer has high mo-
bility, which means that the availability of peers fre-
quently varies. We first define three roles of peers,
basedonwhichtransactioninformation is effectively
managed. Then we propose a dynamic grouping
method to add a peer to the best-fit group in an
MP2P network. As the topology of the MP2P
network changes, we consider peer joining, leaving,
and the movement velocity of peers in our grouping
method.
3.1 Roles of peers
In each group, we define three roles for peers:
super peer, ordinary peer, and relay peer.
Super peer (SP): An SP is a peer that maintains
a trust table and a file list of all the peers in a group.
A trust table records the trust information of all
the peers in the group. When a peer requests some
files, it can send the request to the super peer. The
super peer advises the requesting peer which peers
are trustworthy according to the trust table. The
file list records the files of all peers. When a peer
requests some files, it can send the request to the
super peer. The super peer tells the requester which
group member has the requested files.
Relay peer (RP): An RP is an ordinary peer
who connects two adjacent groups. The transaction
information between peers from different groups is
stored in the relay peers. In Fig. 1, we present an
example of transactions among different roles in an
MP2P network. First, P7 requests a file owned by
P9. P7 sends a query to the super peer P1. If P1
or one of the other peers in P1’s group (i.e., P4 or