Abstract

A distributed resource sharing system enables multiple participants of the system to share their resources and solve complex problems with improved resource reliability and availability. Examples include Grids, Peer-to-peer (P2P) systems, and pub/sub systems. This dissertation identifies two important research problems in data sharing P2P systems, and proposes novel approaches to address these problems.

Structured P2P systems excel in content location but suffer maintenance over-head under node churn, while unstructured P2P systems can well accommodate node dynamic activities but require efficient content location schemes. This dissertation addresses both issues by fully exploiting node behavior patterns and locality properties observed in real-world P2P system traces. For the structured overlay maintenance problem, we study the node join and departure events in real-world traces, identify the round-trip pattern in many node dynamic behaviors, and thus propose a routing indices caching scheme to simplify node rejoin procedures and improve the routing performance. For the content location problem, this dissertation proposes two novel schemes with different rationale and design. The first scheme, called Foreseer, takes advantage of geographical and temporal locality in P2P systems and constructs a two-dimensional overlay. The second scheme, named ASAP, employs a content pushing technique and improves search performance by distributing and saving content indices beforehand.