【作者】 王菁；

【导师】 杨寿保；

【作者基本信息】 中国科学技术大学 ， 计算机系统结构， 2007， 博士

【摘要】 在过去的几年中，P2P系统规模不断扩大，应用不断增长。高效的资源管理机制成为P2P系统的关键技术。目前，P2P叠加网络拓扑的构建大多数没有考虑底层物理网络的实际情况，产生了叠加网络拓扑和底层物理网络不匹配的问题，导致了P2P节点定位资源的延迟和开销增加，严重影响了系统的效率，而P2P网络的动态性和匿名性又无法保证所有的响应节点都提供诚实良好的服务和可靠资源。同时，结构化P2P系统在提高良好可扩展性的同时存在的最大问题是仅限于精确关键字的匹配查询，而不支持基于内容的复杂查询。这些问题严重阻碍了P2P系统的迅猛发展。如何减少搜索成本，降低定位延迟，并提供支持基于内容查询、安全有效的P2P系统成为当前P2P研究领域的重要课题。P2P系统的资源管理机制分为物理传输层、P2P叠加层、文件管理层以及检索模型四个部分，本文的研究重点为后三个部分。P2P叠加层，负责P2P协议的构建和优化。目前P2P系统中存在搜索成本高，定位延迟大以及由匿名性带来的安全隐患等问题，如何构建一个快速、高效、安全的叠加网络成为研究的新热点。文件管理层，负责管理数据仓库，按照关键字集合对文件进行有效地分类，同时提供文件到P2P叠加层的合理映射。实现从语义层到P2P层映射的最关键问题就是维数不匹配问题，因此采用有效的降维算法需要进一步的深入研究。检索模型，负责文本的插入和删除操作，并负责计算文本之间的相似性。“维灾”和高维稀疏矩阵是文本处理过程中急待解决的主要问题。本文从P2P资源管理机制的三个方面入手，针对上述问题展开深入研究，主要取得如下研究成果。针对目前大多数P2P系统的构建没有考虑到物理因素，提出了基于延迟敏感的双向查询Chord系统，该系统根据节点和参考点之间的延迟对节点进行分组，减少平均相应时间；通过实现Chord上的双向查询，缩短搜索路径长度。针对P2P系统中由节点匿名性所带来的恶意欺诈服务问题，提出了基于信誉感知的资源发现算法，基于Directed-BFS算法，融入信誉的概念，有效的抑制了P2P网络中恶意节点的欺诈行为，降低了系统的消息负载，保证了节点获得服务的可靠性和安全性。为了减少女巫攻击对于P2P系统的影响，利用P2P的小世界特性，通过引入虚拟节点，依据最短增广路径算法寻找系统中的女巫攻击团体，有效地降低了女巫攻击节点在整个系统中的百分比。针对当前结构化P2P文件存储系统存在仅支持单关键字的精确匹配，而缺乏支持内容查询的局限性，本文提出一种基于快速映射的结构化P2P存储系统FCAN。FCAN利用FastMap实现从语义空间到P2P空间的映射，在保证文本矢量语义完整性的同时对其进行有效降维，从而实现基于内容的查找；通过对质心法确定的最近邻居进行聚类，有效地减少了冗余信息和查询时间。对于高维语义空间引起的文本相似度计算复杂，以及完全分布式非结构化P2P系统中广播算法带来的大量冗余消息，本文提出利用集合差异度实现基于内容聚类的P2P搜索模型。该模型利用集合差异度定义文本相似度，将文本相似性的计算复杂度控制在线性时间内而有效地减少了查询时间；利用节点之间的集合差异度实现基于内容的聚类，可以降低查询时间，减少冗余消息。模拟试验表明，本文提出的资源管理机制可以有效减少P2P系统中的搜索成本，降低资源定位延迟，并且在保证安全性的情况下提供基于内容的有效查询。更多还原

【Abstract】 The past several years witness the rapid development of P2P systems. Efficient resource management scheme becomes the key technology of P2P systems. On one hand, the fact that the construction of P2P overlay network doesn’t consider the real network situation of physical layer leads to the mismatch of overlay layer and physical layer, which makes the latency and the cost of locating resources increase. At the same time, the characteristics of dynamics and anonymity of P2P can’t ensure that all the peers will provide honest services and reliable resources. On the other hand, although structured P2P systems have good scalability, the main issue exists in them is that they are stricted to exact match and don’t support content based query. All these problems prohibit the rapid development of P2P systems. How to reduce searching cost and decrease the latency of locating resources and how to construct a safe efficient P2P system that supports content based query becomes an important subject of the P2P system research field.Resource management scheme in P2P system has four componets: physical transport layer, P2P overlay network layer, file management layer and information retrieval model. The main topics in this dissertation are the last three parts. P2P overlay network layer focuses on the construction and optimization of P2P protocols. Currently, there exist many problems in P2P system, which are high searching cost, long locating latency and hidden troubles that are brought by anonymity. Then, how to construct a fast, efficient and safe overlay network is challenging. File management layer aims at managing data depository, dividing the files efficiently according to keywords set and providing efficient mapping from file to P2P overlay network layer. The main issue to implement mapping from semantic layer to P2P layer is the mismatch of dimensions. How to reduce dimensions still needs further investigation. Information retrieval model mainly deals with the insertion and deletion of files and calculates the similarity between documents. The curse of dimensionality and the high dimension sparse matrix are the main issues to resolve when dealing with the documents. Detailed research works have been done on the above issues. The main contributions are as follows.As current construction of most P2P systems doesn’t take the physical elements into considerations, a bidirectional Chord system based on latency-sensitivity is proposed to decrease the latency and cost of resources locating. This system partitions the nodes according to the latencies among these nodes and the landmarks to decrease the average response time. And by the means of bidirectional query algorithm on Chord, the search path length is reduced.To the problem of hostile service brought by node anonymity in P2P system, this disertation proposed a reputation aware resource finding algorithm. This algorithm is based on Directed-BFS algorithm and adopts the conception of reputation to efficiently restrain the deceiving behavior of hostile nodes in P2P system and reduce the messages load in the system, which can ensure the reliability and safety of the services.To reduce the affection of Sybil attack on P2P system, a Sybil defense mechanism based on small world is proposed. Inducing the difficulty of finding Sybil group to max-flow/min-cut problem, this dissertation uses crawler to find Sybil group by adopting an artificial vertex, which can effectively isolate the Sybil attacker nodes from P2P system.To solve the problem of existing DHΥ-based file storage systems being restricted to exact match and lacking semantic-based query, a DHT-based file storage system named FCAN is proposed in this paper. Mapping a document vector from semantic space to P2P space by FastMap, FCAN can effectively reduce the dimensions of document vector to implement semantic-based query without affecting the semantic integrality of this document. By clustering the nearest neighbors based on centriod, FCAN is further improved to reduce the redundant messages and searching time.As the main issues that affect query efficiency and searching cost in content-based P2P search system are the complexity of computing the similarity of the documents brought by high dimensions and the great deal of redundant messages coming with flooding. This dissertation defines the document similarity by the way of set distance. This method restrains the complexity of computing the document similarity in linear time. Also, this dissertation clusters the peers based on content by their set distance to reduce the query time and redundant messages.Simulation results show that the resource management schemes presented in this dissertation not only reduce the searching cost and decrease the latency to locate resources in P2P system effectively, but also provide content-based query and ensure the safety of the whole system.更多还原