| Summary: | 碩士 === 國立臺灣大學 === 資訊管理學研究所 === 93 === As Internet grows and the capability of PC improves, peer-to-peer (P2P) systems become increasingly popular due to the potential for overcoming several limitations of client/server systems with respect to scalability, content availability and computing power. To tackle the challenges in scalability and the issues of search, various schemes have been proposed in the literature, such as Chord, CAN, Tapestry, and Pastry. These are distributed hash tables (DHTs). In order to make DHT based search e±cient, nodes use local information as routing guides to forward requests to appropriate intermediary nodes. The efficiency and correctness for all DHT overlays depends on the consistent maintenance of routing tables at each peers even under highly dynamical network conditions. Although DHT based P2P systems are praised for their e±cient routing performance, most released P2P systems do not adopt DHT algorithms and are still using unstructured P2P manners. The main problem in DHT algorithm implementation is the high traffic load for maintaining DHT structures.
The primitive DHT designs tend to involve the participating nodes equally, but the empirical studies have demonstrated the diversity between system nodes. Considering such diversity, we design to collect more resources from capable participants rather than treat all participants equally. In this thesis, we propose a super-peer based approach to reduce maintenance overhead in Chord P2P algorithm. Our design takes advantage of the heterogeneity of network peers. The experimental results show that our approach saves the meaningless maintenance overhead and makes Chord P2P system more practical.
|
|---|
