Performance Evaluation of an MCT-Based Peer-to-Peer Architecture

Abstract

The Mesh-Connected Tree (MCT) is a hierarchical network topology formed by joining two layers of network structures. The top layer is a mesh with each node being the root of a binary tree. Previous study has shown that a P2P network based on the MCT topology is easy to expand and able to distribute search messages through out the whole structure efficiently. However, if an MCT-based P2P network assigns location for each new peer node by using a distributed hash table (DHT) based algorithm, then search efficiency will be hard to improve, because two physically nearby peers may be allocated positions far apart in the MCT hierarchy. To solve this problem, we use ping response time to estimate network distance such that peers in proximity will not waste time going through a long search path to exchange information with each other just because they have been unwittingly allocated too far apart. Furthermore, we propose a binary search method for a requester to locate the potential provider of files of interest; a search time is thus achieved. We compare the performance of the proposed MCT-based P2P network with two other previously proposed P2P architectures, the Chord and the Grapes. Simulation results show that the proposed system not only has shorter average delay and lower jitter for packet transfer, but also has fewer packet losses and higher throughput. These results also indicate the potential of the MCT topology to be the foundation of a P2P network that is both efficient and reliable.