The Reputation System For Robust, Structured P2p Systems
Dhadphale, Apurv Ashok
MetadataShow full item record
Structured peer-to-peer systems are distributed communication systems that typically use Distributed Hash Table (DHT) indexing to efficiently locate the resources. These networks are highly scalable and can route the messages correctly even for the extremely dynamic environment. But these networks are vulnerable; even a small fraction of malicious nodes can bias the lookup results when they are present on a lookup path. In our thesis, we address this problem of reliably searching the insecure p2p networks. We propose a reputation system to reduce the number of failed lookups and make the networks more robust. For our study, the concept is applied to the Salsa peer-to-peer communication system. Since Salsa is a DHT-based structured p2p system, it is possible to approximate the actual overlay network and lookup path which is not possible with the other unstructured p2p networks like BitTorrent. Each node builds its own reputation tree using the look-up results. The closest results are considered to be good. The reputations are then used to select the nodes for lookup. Since we use the redundancy that is inherent in Salsa, there is no separate communication overhead to collect reputation. Also since the working of reputation system does not depend on peers, it is not subject to the attacks like bad-mouthing. We first study the effectiveness of the reputation system for a static Salsa environment where the peers are fixed and then adapt the approach to dynamic environment where the peers join and leave randomly. We modified the existing continuous time Salsa simulator to include reputation system module. Using the simulation results, we show that the number of failed lookups reduces by up to 90%. The experimental results also demonstrate how the different system parameters can be changed to control reputation score and further decrease the number of failed lookups.