Trust - Peer-to-Peer Computing: Applications, Architecture, Protocols, and Challenges - page 148

6.9 GossipTrust
Recently Zhou and Hwang also proposed another novel trust management
scheme, called GossipTrust [Zhou and Hwang, 2007a], designed for unstruc-
tured P2P systems such as KaZaA. Figure 6.12(a) shows the architecture of
GossipTrust.

Storage Service

Gossip Steps

Global Reputation Scores V

Converged

<n1, v1> <n2, v2> <n3, v3> ...

1 2 ... k k+1 ... g

V(0)

Initial Reputation
Computation
Gossip-Based Reputation Aggregation

Reputation
Updating

V(1) Power Nodes

Local Trust Scores

.
.
.

V(t)

.
.
.

V(d)

Underlying P2P Network

(a) GossipTrust architecture (b) gossiping procedure

FIGURE 6.12: GossipTrust [Zhou and Hwang, 2007b, Zhou and Hwang,
2007a].

In GossipTrust, each peer keeps a row vector of trust matrix S based
on its outbound local trust scores. In addition, each peer also maintains a
global reputation vector V (t) at aggregation cycle t. Internally, this vector is
represented by a collection of <nodeID, score> pairs. At the first aggregation
cycle, V (0) is initialized with equal global reputation scores, i.e., v i (0) = 1/n,
for i = 1, 2, . . . , n.
To compute the successive reputation vectors, GossipTrust uses a gossip-
based protocol to perform the matrix-vector computation. Gossiping supports
lightweight communications among nodes during the aggregation process. In
GossipTrust, each aggregation cycle consists of several gossip steps as shown
in Figure 6.12(b). In a gossip step, each peer receives reputation vectors from
others, selectively integrates the vectors with its current reputation vector, and
then sends the updated one to a random peer in the network. This gossiping
process continues until the gossiped scores converge in g steps, where g is
determined by a set of gossiping error threshold ǫ. After the convergence
of gossip steps, GossipTrust continues the next aggregation cycle until the
global reputation vectors converge in d cycles, where d is determined by the
aggergation error threshold δ.
Simulation results indicated that GossipTrust generated good performance