【作者】 夏莹杰；

【导师】 郑耀；

【作者基本信息】 浙江大学 ， 计算机科学与技术， 2009， 博士

【摘要】 网格技术和对等(Peer-to-Peer,简称P2P)技术是当今分布式计算领域的两个研究热点,它们为各种高性能计算应用提供了基础计算架构。网格技术通过网络大规模地集成地理上广泛分布的各种资源,为各种科学研究和工程应用提供强大的计算支持,但在网格环境中节点缺乏自由度,系统的可扩展性不强。而P2P技术与网格不同,它的每一个节点都有平等的地位,能够自由地加入和离开,因此具有较强的可扩展性和鲁棒性,但现有的P2P技术节点缺少安全策略,容易受到恶意攻击,安全性较差。本文提出了一种P2P网格模型,它融合了网格技术和P2P技术的优点,实现了一个低成本、高效、稳定、安全的科学计算架构。本文通过详细介绍如何实现基于P2P网络和网格模型的P2P网格架构,如何对P2P网格进行优化以实现节点负载均衡,以及P2P网格在科学计算和协同应用领域的实例,全面分析P2P网格的实现、优化和应用,并通过多个实验证明该混合式计算架构在计算性能、安全性、可扩展性、鲁棒性等方面相比于传统的P2P结构、网格模型以及相关应用的优势。研究工作取得以下创新成果:(1)利用Globus Toolkit(GT)和Sun Grid Engine(SGE)构建了层次型组合中间件架构的校园网格。GT能够高效快速地在网格节点间传输资源和计算任务,而SGE可以根据节点的负载情况进行自动任务调度。这种架构在计算性能、安全性方面都具有较大的优势,具有一定的新颖性和实用性。(2)实现了一种基于角色和信誉访问摔制的P2P网络模型——R2P,并对它进行改进,提出了SW-R2P模型。SW-R2P模型利用零知识交互式证明算法和Bayesian信誉网实现一种可信小世界P2P网络模型。SW-R2P模型把Bayesian信誉评估测量和小世界P2P网络进行结合,不仅能够保证网络的可信性和安全性,还能够利用拓扑结构提高网络的资源查找性能。(3)实现了基于SW-R2P模型实现的网格架构,把网格的安全、稳定的特点和P2P的可扩展、灵活的特点进行结合,提出了一种新型的科学计算基础架构。它利用P2P网络把所有节点划分成组,并使用中间件软件组织其中部分节点构建成网格。这种层次型的计算架构能够综合利用两种网络模型的优点,具有较强的创新性。它能够利用节点间的交互历史对P2P网格节点的负载情况进行预测,以此影响资源请求节点选择服务节点的过程,有效均衡P2P网格环境的负载,从而能够确保资源利用的高效性,节点服务的可靠性以及整个网络的稳定性。该模型是对P2P网格的一个重要优化。此外,利用本文的研究成果P2P网格,改进原有的层次型校园网格架构,实现一种新型的协同应用框架。大量实验证明,P2P网格能够较大程度地提高校园网格的科学计算性能和协同应用的效率,具有较强的实用性。更多还原

【Abstract】 Grid and Peer-to-Peer (P2P) are two state-of-the-art research spots in the area of distributed computing, and they are capable of providing cyberinfrastructures for various applications in high performance computing. Grid technology utilizes the network to integrate geographically distributed resources, supporting the scientific research and engineering applications by large computing cycles. However, the grid network harnesses the nodes over-tightly, which leads to a lower system scalability. In contrast, each peer in P2P network has an equal position, joining and leaving freely, therefore P2P has a much higher scalability and robustness. However, it has less security strategy, which causes to be fragile to malicious attacks.This dissertation specifies a P2P grid model, which integrates the grid technology with P2P network to utilize both their advantages and to avoid their disadvantages. This combined model can be used to implement a low-cost, efficient, stable and secure cyberinfrastructure. The dissertation also introduces how to implement the specified P2P network and grid model to construct this P2P grid architecture, how to optimize the P2P grid to implement the load balance, and how to apply the architecture into the scientific computation and collaborative applications. Through multiple test cases for parallel computing and simulation experiments, the P2P grid cyberinfrastructure is explicitly proved to be more excellent than the traditional P2P, grid and their related applications, in terms of computing performance, security, scalability and robustness. The main creative contributions of the whole research work are listed as follows.(1) The campus grid is constructed by using the hierarchical combination of the middleware GT and SGE, which can make use of both their advantages. GT facilitates the resource transfer and computing tasks between the grid nodes efficiently, while SGE schedules the tasks dynamically based on the node loads. This novel architecture is much excellent in terms of computing performance and security.(2) Stemming from the P2P network with role and reputation based access control policies, we propose a SW-R2P model, which utilizes the zero knowledge interactive proof and Bayesian trust model to construct a trusted small world overlay P2P network. As we know, SW-R2P is the first work to synthesize the Bayesian trust evaluation and small world P2P network topology. This architecture not only keeps the whole network trustability and security, but also uses the topology to improve the resources locating.(3) The P2P grid architecture is implemented based on the SW-R2P network. It inherits the security and stableness of grid, and scalability and flexibility of P2P, to provide a novel cyberinfrastructure for scientific computation. The P2P grid uses the P2P network to group all the peers, and deploys the grid middleware on some of them to construct the upper grid layer. This hierarchical cyberinfrastructure makes full use of the advantages of both network models.(4) The load balance model for the P2P grid can be implemented by adding some kinds of load measurements in the Bayesian trust network. It uses the interactions history to probably predict the load of the candidate peers, therefore affecting the server peers selection. This model facilitates to make full use of P2P grid resources, which keeps the services robust and the whole network stable. This model is an optimization strategy for the P2P grid.Moreover, by using the proposed P2P grid, we are able to improve the traditional campus grid architecture and implement a new collaborative application framework. The experiments show that the P2P grid is capable of significantly improving the performance of the scientific computation in the campus grid and the efficiency of the collaborative applications, and further imply that the architecture has much practicability.更多还原