根據使用者興趣來增進非結構性P2P網路的搜尋效能

Abstract

P2P檔案分享系統是現在消耗最多網路資源的應用之一，如何有效率的搜尋是一個挑戰。在非結構性的P2P系統中， 以Gnutella為例子，搜尋是透過洪流(Flooding)的機制來找尋資源。雖然這是個簡單且穩固機制, 但是這會產生很大的網路流量，當系統擴規模大的時候，會非常的沒有效率。 在本篇論文中，我們提出了一個有效率找尋資源的方法。 系統的成員會在Gnutella的網路上，逐漸的自我組織成結構。 本篇提出的方法核心為，我們假設，發出相同搜尋的成員們，他們可能會有類似的興趣或喜好。我們將這些成員們組織在一起，搜尋的訊息會先送給這些成員，因為他們有可能在過去搜尋的時候，已經找到所需要的資源。本篇論文，將一般系統視為是負擔的搜尋訊息，當成資源來運用。我們也思考搜尋的本質。一般搜尋，都是先有需求，然後尋找資源。然而，是否有可能反過來，讓擁有資源的成員，找出誰有需求呢？本篇提供了一種解答。我們也透過模擬，加以證實我們提出的方法，可以增進搜尋效率並且降低網路流量。

P2P file-sharing systems consume significant bandwidth nowadays. A fundamental challenge is how to locate resources efficiently. In unstructured P2P systems, take Gnutella for example, search is performed by flooding to find resources. Although simple and robust, flooding generates tremendous traffic, and is also not scalable. In this paper, we propose an efficient method to tackle this problem. Peers will gradually self-organize into an interest-based overlay on top of Gnutella. The basic concept is that peers request the same files are likely to have similar interests. Query messages are first sent to those peers that have issued the same query. Chances are the requested resources may have been retrieved through previous searches. We turn query messages, while other researches regard them as merely overheads, into useful resources. Furthermore, we also take an eye on the nature of search. In common practices, we want resources, and we locate them via search and pull it back. However, is it possible that we have the resources, and push them to those in need? We provide an answer as well. Our algorithm, called D□j□ vu, can improve search efficiency and alleviate network overheads, as confirmed by simulation.