EigenTrust

EigenTrust is an incentive system based on traffic history of each peer. Uploading data to peer A increases the reputation of peer B from peer A's perspective. The reputation between all peers can be pictured using a trustmatrix $M$ with elements $c_{i,j}$. $c_{i,j}$ is peer j 's rank from i 's perspective related to normalized download volumes that i has received from j during a certain time period. In order to increase its local view of reputation, each peer takes into account his friends view ($M^{2}$). Thus the coverage of the rank matrix gets larger and it becomes less sparse. Continuing this step to its extreme ($M^{\infty }$), each column of $M$ consists of the same values (= EigenTrust-vector), meaning that each peer has the same global view to other peers.

Downloadtraffic s mapped to a Trustmatrix M

A peer with large global reputation (EigenRank) gets higher download bandwidth and less latency. Moreover, peers with large EigenRank are preferred to download from.

Simulation Results

Examined in Maze-P2P-Network, EigenTrust helped to punish colluding peers by branding them with low EigenRank. However, EigenTrust leaves some room for misinterpretations. High reputation peers (super peers), which randomly download from "spam"-accounts, boost the EigenRank of these colluding peers. On the other hand, peers inside satellite clusters like university networks are unfairly punished. A university network as a whole consumes by downloading much more than it uploads. Neglecting and underestimating internal cluster traffic, EigenTrust underrates cluster peers EigenRank.