Identifying malicious peers before it's too late: A decentralized secure peer sampling service

Abstract

Many unstructured peer to peer (P2P) systems rely on a Peer Sampling Service (PSS) that returns randomly sampled node s from the population comprising the system. PSS protocols are often implemented using "gossiping" approaches in which connected nodes exchange their links in a randomized way. However, such services can be defeated easily by malicious nodes executing "hub attacks"which distort the PSS such that all nodes in the network, ultimately, only gain access to malicious nodes. From this leading status - i.e. being a "hub" - the malicious nodes can affect the overlay in several ways, ranging from total network disruption to obtaining an application dependent advantage. We present a completely distributed defense against such attacks and give results from simulation experiments. The approach is generic as it is independent of the adopted PSS implementation. © 2007 IEEE P2P, overlay, security, gossiping.