【作者】 兰明敬；

【导师】 黄永忠；

【作者基本信息】 解放军信息工程大学 ， 计算机科学与技术， 2013， 博士

【摘要】 高性能计算广泛应用于情报分析、武器系统模拟、气象预报等国防和国民经济建设的重要领域，对保障国家安全、促进科技进步、推动经济发展具有不可替代的重要作用。研究高性能计算服务系统的服务组织、发现和访问具有现实意义。集中式服务组织存在单点故障和性能瓶颈等问题，基于结构化P2P（Peer-to-Peer）网络的服务系统受到广泛关注。传统分布式哈希表技术不支持复杂查询，并且无法有效满足高性能计算环境下服务调度、任务分配、协同处理需求。在此背景下，本文提出一种新的结构化P2P网络模型并将其成功应用于高性能计算服务组织，对服务覆盖网络建立、支持复杂查询的两阶段服务发现算法、结构化覆盖网环境下的协同计算模式等问题进行了深入研究。主要研究内容和创新点如下：1．分析P2P技术在高性能计算领域的研究现状，指出了当前存在的主要问题，明确了研究范围和基本思路。2．将结构化P2P网络模型引入高性能计算领域，建立了以服务为结点的服务覆盖网络，设计了路由表结构、路由算法、加入算法、更新算法。3．将语义信息融入到服务编址、路由表和路由算法中，实现了语义聚集的服务分布和语义相关的路由过程。相对Kademlia，具有更高的路由效率和更低的带宽消耗。4．利用高性能计算服务丰富的语义信息，建立多重索引机制。提出两阶段服务发现算法，实现了模糊关键字检索、混合检索等多种复杂查询功能。提出语义反馈机制，进一步优化路由选择。相对Kademlia，具有更好的查询效果。5．在语义树中植入计算结点，通过服务聚集形成计算域，在此基础上提出结构化覆盖网环境下的并发、调度等协同计算模式，满足了高性能计算系统去中心化的需求，拓展了结构化P2P技术的应用领域。6．设计并实现了Soverlay原型系统。给出了Soverlay的体系结构、功能单元操作接口和路由消息格式。更多还原

【Abstract】 The high performance computing (HPC) is widely applied in some significant fields of national defense and national economic construction (such as information analysis, weapon system simulation, weather forecast and more), playing an important role in safeguarding national security, promoting the development of science and technology and propelling economic development. The research on the service organization, discovery and access of HPC services system has important practical significance.Due to the problems in centralized services system such as single-point failure and single point of failure, the structured P2P-based (Peer-to-Peer) services system has received widespread attention. However, the traditional Distributed Hash Table (DHT) does not support complex query, and can not meet the demand for service dispatch, task allocation and cooperative processing in the context of HPC.In this context, this paper comes up with a new structured P2P network model which is applied in HPC service organization successfully and conducts a further research on the construction of service overlay network, two-stage service discovery algorithm for complex query, the collaborative computing model in the context of and structured overlay network, and other issues. The main research content and innovation points are as follows:1. Analyze the current research of P2P technology in HPC field, point out main existing problems, and define the research range and basic ideas.2. Introduce structured P2P network model to HPC field, build service overlay network with service as nodes, and design routing table structure, routing algorithm, embedding algorithm and updating algorithm.3. Incorporate semantic information into service addressing, routing table and routing algorithm to realize the service distribution of semantic congregation and semantically related routing processing. It has higher routing efficiency and lower bandwidth consumption than Kademlia.4. Use the rich semantic information of HPC service and build multiple index mechanism. The two-stage service discovery algorithm is proposed to realize many functions for complex query (such as fuzzy keyword search, mixed search and etc.). Besides, the semantic feedback mechanism is proposed to further optimize routing. It creates better query results than Kademlia.5. Implant semantic tree with computing nodes and form computational domain through service aggregation. The concurrence, dispatch and other cooperative computing models in the context of structured overlay network are proposed based on this to meet the demand of HPC system for decentration and expand the application fields of structured P2P technology.6. Design and realize Soverlay prototype system. And the structure, functional unit operation interface and routing message format of Soverlay system are provided in this paper.更多还原