Malicious peers eviction for P2P overlays
Hatem Ismail, Daniel Germanus, Neeraj Suri · 2016
P2P networks constitute the foundation for many scalable and fault-tolerant applications. These networks may consist of millions of peers due to their decentralized design. While each peer advertises the same service interface, the actual data provided by common large-scale P2P applications often yields an imbalance, i.e., particular peer subsets are more frequently contacted than the rest of the overlay. Such subsets may refer to peers that replicate critical or popular content. As a consequence, Localized Attacks (LA) target these subsets and despite using a relatively small amount of attacking malicious peers, LAs severely impair the overall network's reliability. Hence, we propose a new two-fold LA countermeasure to detect and evict malicious peers. Our countermeasure has been evaluated in a comprehensive simulation experiment study using a generic LA model that covers a wide range of known LAs such as Sybil, Eclipse or different poisoning attacks. The study shows reliability improvements of up to 97% in the presence of LAs, as well as successful evictions for up to 99% of the cases.