网格资源自治管理和任务自主调度的设计与实现

【作者】 王涛；

【导师】 杨志义；

【作者基本信息】 西北工业大学 ， 计算机应用技术， 2005， 硕士

【摘要】 网格就是将广域范围的各类计算资源(包括CPU、存储器、数据库等)通过高速的互联网组成共享的资源集成,提供一种高性能的计算、管理和服务的资源能力。使用这些资源就像用电力资源一样,不必关心其来源和负载情况。 资源管理、任务调度和任务管理是网格计算中需要具备的三种基本功能,同时它们也是网格计算中最具有研究价值的三个方面。本文以校园网格计算应用为背景,介绍了网格计算在国内外的发展与现状,网格计算系统的结构、特点和主要功能。并通过对网格计算环境下所采用的任务凋度与资源信息服务的分析和研究,指出了目前的网格任务调度机制并不完善,没有综合考虑用户的个性化服务要求和网格资源的状况,不利于实现网格计算系统的全局负载共享、全局的相对负载均衡,网格计算资源的优化使用。 针对以上问题,本文利用JAVA语言设计实现了一种自适应的网格任务调度算法、策略和机制。它是根据用户的个性化服务需求、网格资源的负载信息、资源处理能力、资源可用数量以及最大化资源效益等因素,综合权衡进行任务调度。显然要实现对网格任务的自适应调度,就必须要求有一个强大的资源信息服务模型来提供支持。于是,本文采用了基于轻量目录访问协议—LDAP模型的目录结构,设计实现了网格计算环境下的on-demand的资源信息服务模型。该模型实现了对网格资源的统一表示,隐藏了网格资源在各个层次上的异构性,最终实现了对网格计算环境下全局资源的有效聚合与充分共享。 本文的研究丰富了网格任务调度和资源信息服务的理论和方法,对于网格计算在实际中的应用具有理论意义和参考价值,设计和实现的网格任务自适应调度机制和on-demand资源信息服务模型也具有一定的推广价值。更多还原

【Abstract】 The grid is a resource ability that use high-speed networks to integrate all kinds of computing resource such as super-computers large databases, archival storage devices and so on to form a sharing resources to provide a kind of high-performance computing management and service. People use these resources like using electrical sources, and do not care for the origin and load of these resources.The job management, job schedule and resources management are three basic functions of the grid computing, at the same time they are have the most values for grid computing research. With Campus Grid computing as the application background, this paper introduces the development and status of grid computing, the structure, characteristic and main functions of the grid computing system. Throughout analysis and researched the job schedule and the resource information services of the grid system, this paper points out the shorting in the grid job schedule mechanism that be used at the present time.Towards these questions, this paper using the JAVA programming language design and finished a job self-adapt schedule mechanism. It is according the user’s services request and the status of grid resources to decide how to schedule one grid job. Obviously the self-adapt job’s schedule must need one good grid resource information service to support. Therefore, this paper also design and finished a on-demand resource information services model, that is using a directory structure that based model of Lightweight Directory Access Protocol. This model finished the unified description method for the geographically distributed, heterogeneous resource, At last it implemented the high-efficiency converge and best share for the all resources of grid computing system.The research achievements of this paper will enrich the theory and methodology for the grid job scheduling and the grid resources information servicing. At same time the grid job’s self-adapt schedule mechanism and the on-demand resource information services model have some value for spreading.更多还原