High Performance Overlay File Distribution by Integrating Resource Discovery and Service Scheduling

Abstract

Many existing studies in overlay networking have focused on resource discovery or server selection. This work studies the integrated performance of peer-to-peer (P2P) file sharing systems using location-aware resource discovery and service scheduling schemes. A novel file capacity amplification (FCA) model is first presented to capture the problem of the file distribution problem. Then two novel service scheduling schemes and protocols, Capacity Amplification (CA) and its variant CA with Penetration (CAP), are presented to enhance the performance of file distribution in overlay networks. The CA scheme represents a greedy approach in selecting clients for services disregarding the effect of delay latency on transport capacity. The CAP scheme, on the other hand, adopts the semantics of small world networks to reduce effectively delay latency among peer servers and clients. Consequently, the effective transport capacity can be increased efficiently leading to fast file distribution. The analytical results indicate that traditional scheduling schemes such as FCFS perform poorly compared with CA and CAP schemes. Furthermore, a high-performance P2P file distribution system needs both an efficient resource recovery scheme and a good service scheduling scheme.