以小世界理論為基礎之中繼資料輔助同儕網路搜尋系統

Abstract

近幾年來，由於計算機的能力和網路的頻寬都快速地成長，同儕網路系統正蓬勃的發展現今最熱門的應用非檔案分享莫屬。而同儕網路搜尋技術也不斷的在改進當中，從同儕網路系統的始祖Napster，到後來慢慢發展出來的Gnutella、Kazaa，一直到現在最熱門的BitTorrent、Emule等等。近年來，許多的研究引入``小世界理論'，以社群網路來模擬同儕網路，並獲得不錯的成效。但是``小世界理論'只考慮使用者之間的相似度，並未考慮到實際網路的狀況。因此，在資料下載時，並不能保證可以獲得最佳的效率。有鑑於此，我們提出一個以``小世界理論'為基礎的同儕網路搜尋系統，並考慮各種的網路因素，期使能同時保證搜尋時的效率以及下載時的品質。 更進一步地，我們更深入地探討中繼資料對於同儕網路搜尋系統的影響。在某些狀況之下，使用者可以使用中繼資料來協助搜尋的工作。比方使用者需要搜尋一些未知名字的資料，或是想利用中繼資料來加快搜尋的速度，例如由不同歌手手演唱的同一首歌曲，或是兩首歌名相同的歌曲等等。因此，我們將中繼資料對於搜尋效能的影響作深入的探討，以及討論哪些中繼資料對於同儕網路搜尋最有幫助。 我們透過系統模擬的方式來評估本文所提出方法之效能。除了本文提出的方法外，我們另外比較其他幾種的同儕網路搜尋架構。實驗結果顯示我們的機制可以同時提升搜尋的效率，並改善資料取得之傳輸品質。

In recent years, as the network capacity becomes wider and the computation capability becomes more powerful, Peer-to-Peer applications have become very prevalent. The most popular application for P2P today is file sharing. P2P file sharing approaches have been constantly improved. From the earliest search approach - Napster, to the recent Gnutella, Kazaa, and BitTorrent. More recently, much research works put ``Small World' theory into P2P search systems to simulate the social networks. But ``Small World' theory builds the structure only based on similarity between peers, and does not reflect the network conditions. Similar peer may not provide satisfactory quality of service as content downloading. Thus, we want to incorporate network factors into ``Small World' theory to improve the efficiency of both content search and content retrieval. Furthermore, we propose the search structure allows users to search by metadata of content. In some circumstances, users could adapt metadata to speed up P2P search. For example, users may want to search a song, but he does not know it name, or sometimes two different song with the same name. Thus, we evaluate the effect of metadata on the efficiency of P2P search, and discuss which metadata is appropriate for P2P search. We evaluate the performance of the scheme via simulations. Results show that our approaches can improve content search and retrieval efficiency, and the structure built by the metadata with moderate dimension can achieve better performance.