面向云计算的高性能计算柔性服务平台
发布时间:2018-03-22 22:19
本文选题:云计算 切入点:高性能计算柔性服务平台 出处:《国防科学技术大学》2012年硕士论文 论文类型:学位论文
【摘要】:高性能计算机作为提供云服务的基础设施,在云计算核心业务中具有不可取代的地位。随着云用户规模不断扩大,层次不断提高,传统高性能计算机服务模式已不能满足和适应云用户的服务需求。本文基于云计算应用模式,研究设计了面向云计算的高性能柔性服务平台,该平台可以按照用户需求灵活配置运行环境,提供计算和软件服务,满足用户对易用性、安全性、高效性、可定制性的服务需求。 针对云用户对高性能计算资源的需求,设计了高性能计算柔性服务平台架构,该平台共包含三层。其中,基础设施层主要是将拥有不同结构计算阵列的高性能计算机抽象为整体高性能计算服务,并且按照任务并行处理需要,根据现有系统资源,动态构建高性能用户容器(HPUC)。平台支持层主要是软件的开发和运行环境,用户通过该层提供的软件开发工具,使用高性能计算机提供的各项远程应用服务。用户开发层主要为云用户提供操作简单、易用性好的胖客户端开发平台,并为用户提供基于eclipse的网络化图形开发环境,从而提高云用户的开发效率。 针对云用户共享计算资源所带来的安全性和可配置性问题,本文研究设计了高性能用户容器(HPUC)技术,该技术基于操作系统层虚拟化技术,向用户提供构筑独立运行环境的支持,通过该运行环境,操作系统支持用户可以按照自己的意愿灵活配置系统。同时,通过隔离相互环境隔离,,使一个HPUC中的错误操作不会影响其余HPUC运行。HPUC技术在操作系统层之上实现系统虚拟化,对系统性能的损失小,,整体性能损失小于3%,满足了高性能并行计算机提供高效云计算服务的需求。 最后,本文从满足HPUC负载均衡需求入手,研究实现了基于复合容器名的负载均衡技术,该技术按照任务并行处理需要,依据系统现有资源,动态构建高性能用户容器,实现了系统服务效率最大化。同时,通过并行负载调度,用户只需提出并行处理能力的要求,而无需关心提供服务的到底是哪个服务结点、计算阵列和存储阵列,实现了将高性能计算以服务方式提供给云用户的设计要求。
[Abstract]:As an infrastructure to provide cloud services, high-performance computers play an irreplaceable role in the core business of cloud computing. The traditional high performance computer service mode can not meet the needs of cloud users. Based on the cloud computing application model, this paper studies and designs a high performance flexible service platform for cloud computing. The platform can flexibly configure the running environment according to the user's needs, provide computing and software services, and meet the user's service requirements for ease of use, security, efficiency and customization. In order to meet the demand of cloud users for high performance computing resources, a flexible service platform architecture for high performance computing is designed, which consists of three layers. The infrastructure layer abstracts the high-performance computers with different computing arrays into the overall high-performance computing services, and according to the needs of task parallel processing, according to the existing system resources, The platform support layer is mainly the software development and running environment, through which users provide software development tools. Using various remote application services provided by high-performance computers, the user development layer mainly provides cloud users with a platform for developing fat clients with simple operation and good usability, and provides users with a networked graphics development environment based on eclipse. In order to improve the development efficiency of cloud users. In order to solve the security and configurable problem caused by cloud users sharing computing resources, this paper studies and designs a high performance user container (HPUC-based) technology, which is based on operating system layer virtualization technology. Provide users with the support to build an independent running environment, through which the operating system supports the user to configure the system flexibly according to his own wishes, at the same time, By isolating each other, the error operation in one HPUC will not affect the other HPUC running. HPUC technology to realize system virtualization on the operating system layer, and the loss of system performance is small. The overall performance loss is less than 3, which meets the demand of high performance parallel computers to provide efficient cloud computing services. Finally, starting with the requirement of HPUC load balancing, this paper studies and implements the load balancing technology based on the compound container name. According to the need of task parallel processing, this technology dynamically constructs the high performance user container according to the existing resources of the system. At the same time, through parallel load scheduling, users only need to put forward the requirement of parallel processing ability, and do not care which service node, computing array and storage array provide the service. The design requirements of providing high performance computing to cloud users are realized.
【学位授予单位】:国防科学技术大学
【学位级别】:硕士
【学位授予年份】:2012
【分类号】:TP38
【参考文献】
相关期刊论文 前2条
1 江务学;张t
本文编号:1650659
本文链接:https://www.wllwen.com/kejilunwen/jisuanjikexuelunwen/1650659.html