论文标题:基于对等计算的信息共享相关技术研究
The Research on Information Sharing Based on Peer-to-Peer Computing
论文作者
论文导师 张维明,论文学位 博士,论文专业 管理科学与工程
论文单位 国防科学技术大学,点击次数 105,论文页数 135页File Size2933K
2006-04-01论文网 http://www.lw23.com/lunwen_1787052/
Information Sharing; Peer-to-Peer Computing; Community; Informatation Interaction; Distributed Search; Self-Organization; Self-Adaptation
对等计算(Peer-to-Peer Computing,P2P)是在互联网上实施网络计算的一种新的计算模型。P2P打破了传统的Client/Server(C/S)模式,网络中的每个节点是逻辑对等的,拥有对等的功能与责任,每个节点既充当服务器,为其它节点提供资源,同时也享用其它节点提供的资源;节点之间通过直接交互共享资源,无需依赖集中式服务器的支持;任何节点可以随时自由地加入或离开,形成一个真正动态的网络环境。与传统的C/S模型相比,P2P在可扩展性、自组织性、动态性、鲁棒性、系统资源利用率等多方面有明显的优势。因此,P2P蕴涵着巨大的商业和技术潜在价值。 本文研究P2P环境下的信息共享问题。主要针对P2P环境下系统的自治性、动态性和大规模分布性的特点,探索基于P2P信息共享所牵涉到的系统结构、信息交互、网络自组织构造、分布式搜索等技术问题。主要取得以下研究成果: (1)结合传统集中式网络易于管理与分布式网络具有良好的区域自治、负载平衡以及健壮性的优点,从有效资源组织的角度,提出系统结构REC。针对节点能力的不对称性,将系统中的节点进行分层,赋予高性能节点更多职责,能够利用节点的差异提高网络的性能;针对节点参与资源共享的不同动机,对拓扑连接进行分类,按节点资源需求和共享目的组织拓扑结构,为网络中的有效资源组织、高效资源搜索奠定基础。 (2)进行了语义信息交互机制的研究。提出了资源元数据本体模型MO,对资源元数据的共性进行建模,增加了资源元数据描述的清晰度和资源元数据间的语义联系。在此基础上,考虑丰富的语义和上下文信息,应用模糊集理论刻画语义相似程度,提出一种语义模糊匹配方法SFM,确保用户获取语义上相关联的、更多的数据以及有效辅助用户进行选择和决策,支持语义丰富的信息共享。 (3)类人类社会组织,提出从节点服务它人和满足自身的双重视角自组织构造对等网络的方法。建立在节点个体根据自身理性追求个体利益最大化分析的基础上,将节点利己和利他的双重动机有机地统一为节点的理性(兴趣),根据节点的交互历史,提出了动态拓扑进化模型DTL。借助自组织拓扑进化算法SOTE,各节点周期性地利用空闲时间调整拓扑连接,适应节点资源和需求的不断变化。由于将资源语义相关的节点动态自组织聚集在一起,实现了节点的有序组织,优化了网络整体性能。相应实验结果验证了该方法具有很好的伸缩性,支持大规模的应用。 (4)在分析现有非结构化对等网络搜索技术的基础上,从尽量减少搜索时经过的节点数以及增强资源搜索请求的针对性入手,提出了基于拓扑进化的自适应分布式搜索机制SAS。节点根据搜索内容选择最有可能包含结果的资源聚集发送查询,通过消息转发的智能性,及时反映资源分布以及搜索内容的动态变化。该方法最大的优点在于查询能够迅速定位答案提供节点,通常情况下可以在近邻找到结果。由于资源搜索的路径缩短,搜索范围缩小,节省了网络带宽,因而可以从多方面提高系统的搜索性能。实验结果表明,SAS在保证搜索效果的前提下大大提高了搜索的效率,具有很好的自适应特性。
Peer-to-Peer computing (P2P) is a new network model that has been sweeping through the computing industry over the past year or so. Being supported by distributed computing technology, P2P has several special features. In a P2P-based system, each peer has equal functionalities and responsibilities: each peer can act both as server to supply resource, and as client to utilize resource provided by other peers. Furthermore, the interaction among peers can be direct and symmetric without relying on the centralized server. In addition, peers can join in or leave from the system at any time to form dynamic network environment. Compared with the traditional Client/Server model, P2P has many advantages, such as scalability, self-organization, dynamicity, the elimination of bottleneck caused by centralized servers and the higher utilization of network resources. For these reasons, there is great potential to apply P2P technology for information sharing. This paper is devoted to the issues of information sharing in P2P environment. It studies system architecture, semantic information interaction, self-organizing network topology, and distributed search techniques for supporting information sharing in dynamic P2P environments with autonomic and large-scale distributed peers. Detailed research works have been done on the above issues. The main contributions are as follows: 1. For the sake of effective resources organization, the system architecture, named REC, is proposed by combining the strength of traditional centralized network (ex. easy to manage) and distributed network (ex. good regional autonomy, load balance and robustness). REC endows high capability peers with more responsibilities to deal with the problem of peer capability heterogeneity, which enhances the performance of the network. REC differs itself from existing system architecture in that, its topology connections classification mechanism enables the topology construction according to peers’resources and query requirements, reflecting the altruism and selfish nature of peers and providing a foundation to effective resources organization and efficient search. 2. The mechanism of semantic information interaction is studied. First, ontology-based resource metadata model, called MO, is proposed. Based on ontological descriptions of the metadata characterizing the resources to be shared, a semantic fuzzy matching algorithm, named SFM, is proposed by exploiting the linguistic and contextual characteristic of concepts. Through computing the semantic affinities between different ontological descriptions, and classifying the semantic association, resource fuzzy matching is supported, which gives the possibility for users to obtain more semantic-related data for decision-making. 3. Referring to the human society organization, a method of self-organizing P2P network based on interest is proposed. Based on the introduced dynamic topology model, called DTL, the P2P network is organized by clustering peers with similar interests into a community and choosing important communities as logical neighbors for individual peers in view of query requirement. The subsequently proposed self-organizing topology evolution algorithm, named SOTE, ensures that the connections between peers are dynamically reconfigured as peers’interests change. Therefore, the self-organization network topology can provide an efficient searching space from the viewpoint of each peer. The simulation result indicates SOTE can automatically optimize the network so that each peer can maintain near neighborship with peers with similar resources or provided resources. Also, the experiment proves that the employment of SOTE leads to adaptative as well as scalable P2P network construction technique. 4. A self-adaptative distributed search technique, named SAS, is investigated. Existing searching mechanisms in unstructured P2P network are not scaling well because of the mechanism of peers randomly choosing logical neighbors without any knowledge about where to forward a query message, which greatly limits the performance gain from various search or routing techniques. Under the assumption of“peer maintaining near neighborship with peers in similar interests by virtue of topology evolution”, SAS enables peers to select the most possible peers that have the answers to send queries, dynamically reflecting the change of resource distribution and requirement. Owing to the intelligent message forward technique based on content of request, the search request can quickly reach the destination peer, mostly nearby, which shortens the search path length, decreases response time, saves network bandwidth and improves the search recall. Simulation demonstrates that SAS can greatly improve the search efficiency and meanwhile guarantee the search effectiveness.
|
