对等网络环境下高效资源共享机制研究-智能与分布计算实验室

对等网络环境下高效资源共享机制研究

姓名	宋伟
论文答辩日期	2008.06.03
论文提交日期	2008.06.06
论文级别	博士
中文题名	对等网络环境下高效资源共享机制研究
英文题名	Research on the Mechanism of Efficient Resource Sharing in the Peer-to-Peer Network
导师1	卢正鼎
导师2	李瑞轩
中文关键词	对等网络;资源搜索;自适应备份;二阶矩;多维空间;无标度特性;信任管理
英文关键词	Peer-to-Peer;resource search;adaptive replication;second moment;multi-dimensional space;scale free;trust management
中文文摘	对等网络（Peer-to-Peer Network，简称P2P）是一种互联网环境中出现的新型计算模式。与传统Client/Server（C/S）计算模式不同，对等网络中的每个节点彼此之间地位是完全平等的，拥有对等的功能和责任，每个节点在对等网络中既做为服务器，为网络中其它节点提供资源和服务，同时也做为客户机享用其它节点提供的各种网络资源和服务。对等网络中，节点之间直接建立对等连接共享资源，无需依赖集中式服务器的支持，这种分布式的业务模式，有效解决了传统C/S环境下的单点失效问题。同时，对等网络中的节点具有高度的自治性，任何节点可以随时自由地加入和退出，形成了一个真正的动态、分布式的网络环境。与传统的C/S计算模式相比，P2P在可扩展性、自组织性、动态性、鲁棒性以及系统资源利用率等方面都有着明显的优势。P2P技术蕴含着巨大的商业和技术价值，从它出现的那一刻起，就吸引了产业界和学术界的极大关注。针对对等网络所具有的自治性、动态性和大规模分布的特点，研究对等网络环境下的高效资源共享机制所涉及到的网络拓扑结构、资源高效搜索、冗余备份、信任管理等问题。随着对等网络应用的不断发展，目前在各种流行的文件共享应用中所采用的支持资源单维属性（文件名、关键字等）的资源共享方式已经越来越不能满足应用的需求。但是目前还没有一个很好的支持多维资源属性的对等网络拓扑结构。结合目前已有的各种结构化对等网络的拓扑特点，给出了一种支持多维资源属性的对等网络资源共享框架模型FAN（Flabellate Addressable Network）。FAN网络中，节点被统一映射到多维空间坐标系中，采用节点相对坐标原点的二阶矩进行资源路由和网络拓扑结构的维护。FAN提供了一种全新的支持多维资源属性描述的结构化对等网络资源共享框架，为网络中的有效资源组织和高效资源搜索奠定了基础。网络中资源分布和用户对资源的访问都不是均匀分布的，目前的各种结构化对等网络都是基于分布式哈希表（Distributed Hash Table）设计的，其设计初衷是将网络资源均匀分布在整个对等网络中，实现负载均衡。但是资源分布的不均匀性使得网络中节点需要进行的处理也是不均匀的，因此某些维护热点资源的节点成为网络服务的瓶颈。针对对等网络中资源分布和用户对资源访问所具有的这种不均匀性，基于FAN网络，研究了一种自适应热点资源备份策略为热点资源构建VRN（Virtual Replica Network），提供网络热点资源的高效访问。VRN利用FAN网络中资源访问所具有的方向性，优先将网络中的热点资源在热点访问路径上进行自适应冗余备份，有效地降低了对等网络中热点节点的处理压力和查询响应延迟。同时利用对等网络所具有的小世界特性，采用幂率分布在VRN中有效传播节点负载信息，实现了网络访问的优化配置。目前的对等网络资源共享都是基于资源描述匹配开展的，越来越多的应用要求可以基于资源内容进行高效资源共享。探讨了对等网络环境下基于资源语义信息的资源共享机制并分析了对等网络环境下的资源语义描述和传播机制，给出了一种对等网络中的语义资源搜索框架SemSearch，利用语义相似度对共享资源和查询请求进行语义分析。在资源搜索过程中，查询请求向语义最大相似方向进行传播，从而实现对等网络中基于语义的资源共享。同时基于SemSearch设计了基于资源语义信息的冗余备份策略，选择热点内容的共享资源，将热点资源均匀合理地分配在整个网络中，提高资源共享性能。对等网络与传统C/S网络具有不同的资源访问特点，对等网络中节点具有高度自治性，用户采用匿名方式访问网络资源，这些都给对等网络带来了重大的安全隐患和风险。目前针对对等网络的相关研究发现，各种欺诈、抵赖、free riding等行为在对等网络中大量存在，极大地影响了对等网络的服务质量。探讨了对等网络资源共享中的安全问题，研究了一种基于节点兴趣的对等网络综合信任评估模型。有效地抑制各种安全隐患在对等网络中的传播，为对等网络资源共享提供了一个有力的安全保障。
英文文摘	Peer-to-Peer (P2P) network is a new computing model came forth in the Internet. Differenced with the traditional Client/Server (C/S) computing model, every node in the P2P network is peer-to-peer in logic, and has the equivalent functionalities and responsibilities. A node in the P2P network acts both as a server to service for any other nodes also as a client to access the shared resources and services from other nodes. Furthermore, the interaction among the peers can be direct and symmetric without a central server. It makes the P2P networks well solve the problem of single point failure. At the same time, the peers in a P2P network have a high autonomy. And any peer can join and leave the network at any time. So, it formats a full dynamic and distributed network environment. Comparing with the traditional C/S computing model, P2P has the obvious advantages of scalability, self-organization, dynamicity, robustly, and utilization of network resources. Therefore, there is a great potential value at commerce and technology to apply the P2P technology. Since its appearance, the P2P technology has attracted the attentions from academe and industry. This paper is devoted to the issues of resource sharing in the P2P environment. We have studied the network topology, semantic searching, redundancy replication, and reputation management for the P2P network based on its self-organization, dynamicity, and large scale. The main contributions are as follows: (1) With the continuous development of P2P applications, the resource searching method supporting a single resource attribute (file name, keyword, etc.) used in various file sharing applications is more and more difficult to satisfy the requirements of applications. However, there is not a P2P topology well supporting multi-dimensional resource attributes. Therefore, based on the topology characters of exist structured P2P network, we have proposed a P2P resource sharing framework supporting multi-dimensional resource attributes named FAN (Flabellate Addressable Network). The peer in a FAN is mapped into a multi-dimensional space. The resource routing and maintenance of network topology in FAN are both based on the second moment of the node to the coordinate origin. FAN provides a novel structured P2P resource sharing framework supporting multi-dimensional resource attributes and lays the foundation for efficient resource searching. (2) The resource distribution and users’ accesses both are unbalanced. The current structured P2P networks are all improved based on the DHT (Distributed Hash Table) technology. The original intention of using DHT is to balance the network load by putting the resources evenly by hash computing. However, the imbalance of the resource distribution makes the imbalance of the peer’s processing. So, some nodes who host the popular resources become the bottle neck for the service quality. For the unbalanced of resource distributions and user accesses, we have proposed an adaptive popular resource replication algorithm VRN (Virtual Replica Network) over FAN to improve the access efficiency of the popular resources. To release the hot peer and decrease the query delay, VRN analyzes the hot accessing path and adaptive creates the replicas of the popular resource on the hot accessing path. Moreover, to optimize the network accesses, based on the small world character of the P2P network, VRN creates the power law links to inform the load information among the peers in the VRN. (3) The existing P2P resource sharing methods all are designed based on the match of the resource description, however more and more P2P applications requires the efficient resource search over the contents of shared resources. Therefore, we have worked on the resource semantic search in the P2P network and proposed a P2P semantic resource sharing framework named SemSearch. Moreover, we have reviewed the related work on the resource semantic description and spreading mechanism and propose a framework of describing resource semantic information for the P2P network. In the SemSearch, we use the semantic similar degree to analyze the semantic information of shared resources and searching requests. To achieve the semantic resource search, the query messages always forward along the most semantic similar path. At the same time, to improve the efficiency of resource search, we improve a replication algorithm over SemSearch to distribute the popular resource in the SemSearch network. (4) Comparing with the C/S model, the P2P network has different resource searching pattern. For example, the peers in the P2P have a high autonomy, and anonymous user access the resources. This brings a great security problem for the P2P network. Some statistical results over the P2P networks discover that there are many cheat, disavow, and free riding actions exist in the P2P network, these vicious actions influence the P2P service quality. To restrain the vicious actions, we propose a P2P integrated trust model over peer’s interests named InterestTrust. As the experimental results shown, InterestTrust provides a powerful safeguard for the P2P network.