現代同儕網路上查詢服務的複製品平行存取延伸之設計與實作

Abstract

隨著近年來網路基礎建設的進步，家家戶戶都能擁有很高的網路頻寬，網路生態將因此漸漸由傳統的Client/Server架構轉變成Peer-to-Peer架構，但從中也衍生出不少新的問題。新一代的Peer-to-Peer技術為了避免使用廣播訊息造成網路的負擔，所以依賴Distributed Hash Table Algorithm (DHT)在網路上來找尋所需的資源。 而目前的DHT Algorithms中有一個很大的問題，它無法很有效地利用replica。它們只是將replica當作提高可用性的手段，沒有充份利用讓它提高網路利用率及降低各peer node的負載。在本篇論文中，我們提出了一個演算法的改良，讓現在的DHT Algorithm可以充份地利用replica。這個改良，可以讓資源要求者的需求由多個回應者同時平行回應，不再是由單一回應者回應，如此一來可以達到分散負載的效果。為了驗證這個方法的正確性及可行性，我們在Linux上實作了一個Chord的模擬器來實驗。實驗的結果顯示在新增文件時能有193%的改良，擷取文件方面平均也能有341%的改進。

The Internet infrastructure is progressing by leaps and bounds. Every family can have wide bandwidth Internet connection at cheap price rate. The network ecosystem will change from traditional Client/Server architecture to Peer-to-Peer architecture. But this phenomenon also produces many new problems. New generation of Peer-to-Peer technology try to use Distributed Hash Table algorithm (DHT) to avoid using flooding broadcast message. However, DHT algorithms can not use replicas efficiently. Most algorithms just treat replicas as a solution to raise availability, but ignore the possibility to raise network usage and to lower the load of other peer nodes. In this thesis, we provide an extension to recent DHT algorithms that enhance the capability to respond requests by many peer nodes in parallel. We implemented a Chord simulator under Linux to verify the feasibility and correctness of that extension. The experiment results show that document insertion speed increases 193%, and the document retrieval also increases 341% in average.