Betweenness Centrality Approximations for an Internet Deployed P2P Reputation System

In the Barter cast reputation mechanism of the BitTorrent-based P2P client Tribler, peers compute local, subjective reputations of other peers by applying a flow-based algorithm to a locally maintained Barter cast graph with peers as nodes and bandwidth contributions as edges. We have previously shown that the computed reputations are more accurate when a peer takes the node with the highest Betweenness Centrality (BC) in its local Barter cast graph as the initial point in this algorithm rather than itself. BC is a powerful metric for identifying central nodes in complex network analysis, but its computation in large and dynamic networks is costly, %Unfortunately, BC computation in large and dynamic networks is costly, and previously proposed approximation methods are only designed for static networks. In this paper, first we assess the stability of the nodes with the highest BC values in growing synthetic random and scale-free, and Barter cast graphs. Next, we evaluate three BC approximation methods proposed in the literature in terms of their ability to identify the top-most central nodes. We show that these approximations are efficient and highly accurate in scale-free and Barter cast graphs, but less so in random graphs. Finally, we integrate the three BC approximations into Barter cast, and we evaluate the quality of the reputations they yield.