一个对等计算模型及其在网格中的应用研究

【作者】 陈博；

【导师】 迟学斌； 张怀宇；

【作者基本信息】 中国科学院研究生院（软件研究所） ， 计算机软件与理论， 2004， 硕士

【摘要】 随着互联网的兴起，大型分布式计算从集中在一地发展为分布到若干在地理上分散、由高速网络互连的高性能站点，甚至分布到位于网络边缘的成千上万台桌面计算机。分布式计算的这一转变趋势明显地体现在两个快速发展的研究领域中：网格计算(Grid Computing)和对等计算(Peer to Peer Computing)。这两个领域的研究有一个共同的关键需求——需要建立一个广域网范围的虚拟平台，容纳的计算资源分布在不同地点，网络条件各异，计算能力则上至超级计算机下至普通个人计算机。 作为这两个领域其中之一的网格计算已经成为了当前研究的热点。网格计算是伴随着互联网技术而迅速发展起来的，起初是专门针对复杂科学计算的新型计算模式，其研究目的是把整个网络整合成一台巨大的超级计算机，实现计算资源、存储资源、数据资源、信息资源、知识资源、专家资源的全面共享。然而随着网格计算的发展，网格已经从单一的面向高性能计算的计算网格转向为面向服务的商业网格了。而且随着网格计算的发展，已经逐步显现出标准化趋势、大型化趋势和技术融合的趋势(如与Web Service技术融合)。 类似于网格计算，对等计算(P2P)技术将点到点(Point to Point)的层次从主从式(Client/Server)架构连接进一步变为Peer之间的连接。对等计算希望能够充分利用每个对等节点自身的资源，从而提供更强的整体性能。由于没有主从式结构中的服务器，因此降低了单一失败的风险，比传统的主从式架构有更高的可靠性。由于这些特点，对等计算在文件共享、分布式计算、网络安全、在线交流甚至是企业计算与电子商务等应用领域都有很强的技术优势。随着对等计算的发展，对等计算除了需要解决自身的一些问题之外，也出现了技术融合的趋势。 本文是在863和中科院十五信息化建设项目“超级计算应用环境建设”背景下展开的。超级计算网格节点建设是其中的一个重要组成部分。在本文中，介绍了一个对等计算模型的设计与实现，通过借用并改进一个IP网络中的路由算法，实现了基本的通信功能。同时根据网格的需要设计了一个应用层的多播系统。本文的另一个目标是探索将网格计算和对等计算二者融合起来的可能性，为此本文另辟一章来分析二者之间的区别与联系，并通过应用实例给予证明。更多还原

【Abstract】 With increasing popularity of the Internet, large-scale distributed computing, previously concentrated on only one place, has developed into multiple high performance sites that are distributed geographically and connected by high speed networks. Further, it has distributed over millions of desktop computers locating on the edge of grid. This trend for distributed computing is obviously reflected in two rapid growing research areas: Grid Computing and Peer to Peer Computing. And a common key pre-requisite exists between them: a virtual platform of wide area network (WAN) is need, where the accommodated computing resources distribute over different places, dissimilar network situations, varying computing capability from supercomputers to PCs.Now grid computing has become a very active research area, which develops rapidly with the technique of World Wide Web (WWW). Initially it was a new computing mode aimed at complex scientific computing with the purpose of combining the whole computer network into a huge supercomputer so as to achieve a full-scale share including computing, storage, data, information, knowledge and expert resources. Later, with the advances of grid computing, grid has grown from high-performance-oriented computing grid to service-oriented commercial grid, and it reveals the tendency of standardization, large-scale and technique integration such as integration with web services.Similarly, peer to peer computing changes the point-to-point communication from the client/server architecture to the interconnection of peers. It aims to make fully use of each peer’s resources to provide higher performance. Without the server role in client/server architecture, the failure due to one server’s fault has been greatly reduced, so it is more reliable than the traditional client/server architecture. Because of these, peer to peer computing has its strong technical superiority in application areas like file sharing, distributed computing, network security, online communication, and even enterprise computing or electronic commerce. With the developing of peer to peer computing, it not only needs to make itself more perfect but also need to integrate with other technologies.This thesis is under the background of high-tech project 863 and the project "Supercomputing Environment and its Application (SEA)", one of the CAS projects for formation infrastructure construction during the 10th Five-Year Plan Period (2001--2005). The grid node of supercomputing is a main component. In this theis, we introduced how to design and implement a model of peer to peer computing. To implement the basic function of communication, we used and modified a routing algorithm of IP networks. And to satisify the need of grid, we design a multicast system on application layer. The other aim of this thesis is to find the possibility of integration of grid computing and peer to peer computing. To do that, we analyse thesimilarity and dissimilarity between these two technologies in a charpter and give examples for validation.更多还原