面向集群管理的“虚拟操作环境”系统的设计与实现

【作者】 雷波；

【导师】 赵政文；

【作者基本信息】 西北工业大学 ， 计算机软件与理论， 2004， 硕士

【摘要】 由于超级计算机的价格十分昂贵,其应用基本局限于一些特定领域;近年来,随着计算机技术的发展,集群系统以其极高的性能价格比逐渐赢得了超级计算机的广大市场,如电子商务系统中的服务器、大型工业企业的生产服务器等。为此,国际上很多大型服务器厂商(如IBM,NEC等)都争先推出自己的集群服务器产品。但是针对集群系统的管理监控软件的发展还远远不能满足人们的需求,能全面监控集群系统中的所有计算资源,并高效、可靠地运行的软件产品极少。针对这一问题,本文对集群管理系统进行了研究,并且设计和实现了一种集群服务器的管理监控软件:“虚拟操作环境”系统。该软件主要以基于策略的管理方式和GUI的操作方式对集群资源进行管理和监控,旨在通过软件的操作,使集群系统具备较高的RAS(可靠性、可用性和可维护性)性能。文中,作者首先对集群管理软件的理论进行了分析和研究。在仔细分析传统计算体系结构的基础上,选择了系统的集中式和分层、分布相结合的体系结构;并根据系统需要,选择了Java开发语言、MySQL数据库和SNMP网络管理协议,使得系统达到了较为理想的开发效果。系统通过提供动态资源重组、自动故障恢复、全面用户监控等方法,达到了拥有较高RAS特性的设计目标。据此,作者利用分层式结构设计了系统的整体结构,使得系统的开发达到了较高的健壮性、扩展性和独立性。最后,论文着重论述了系统中作者负责担当的三部分内容的实现要点,其中,系统采用基于策略的管理方式,实现了系统管理的自动化和系统运行的高可用性等;采用Java的JDBC技术实现了系统数据库的管理,使系统其它模块对数据库的访问变得简洁、高效,增加了代码的可复用性;利用FACADE设计模式实现了系统的事件处理接口,降低了子系统之间访问的复杂性,提高了访问的可靠性。本课题来源于一个国际合作研发项目,其研究成果己得到了成功的应用,所研制的集群服务器系统已推向市场,客户反映良好。更多还原

【Abstract】 In the last decades, since super computer is expensive, its application has been limited in a few special fields. But with the development of computer technology, cluster is winning the market of super computer by its high rate of performance to price, for example the server of big technology company and e-business. Therefore, many international server manufacturers, such as IBM and NEC, are trying to rank the first position for server products. However, the management and monitoring system of cluster is far insufficient for utility. There are a few softwares that can fully monitor the resource of cluster and make them running in a high performance and reliability way. This problem is addressed in this paper. The system of cluster management is studied. Then a cluster management software, Virtual Operation Environment system, is devised and implemented. This software manages and monitors cluster resources by Policy-Based management and GUI operation Our aim is to make the cluster system has higher RAS features.In the paper, the theory of cluster management software is firstly studied on the basis of analyzed conventional computer architecture which is centralized accompanying with layered and distributed system. And then, according to system requirement, Java programming language, MySQL DBMS and SNMP protocol are decided. All of these make the system have ideal development results The system provides active resource reconstruction, auto fault recover, and all-around monitoring, so it achieves the aim of having higher RAS features. Based on these features, a layered architecture is developed to design the over-all architecture of the system, which makes the system more robust, extensible, and independent. Finally, the paper emphasized on essential three parts which were mainly done by the writer. First, policy-based management was adopted to achieve automatic management and high availability. Second, JDBC of Java was adopted to achieve concision and effectiveness when data base was visited, and to increase reusability of code. Third, FACADE design pattern was adopted for event interface to decrease complexity and enhance reliability in subsystem calls.This task originates from an international cooperation research project; its production has reached successful utility; and the cluster server has been sold on market, which has good responds from customers.更多还原