节点文献

基于层次调度的网格结构模型及其资源预留技术的研究

Research and Design on Grid Structure Model and Its Resource Reservation Technology

【作者】 罗建鼎

【导师】 刘元安;

【作者基本信息】 北京邮电大学 , 电磁场与微波技术, 2011, 硕士

【摘要】 资源管理是网格系统的最主要的关键技术之一,对于开发计算网格系统具有重大的理论意义和实用价值。资源管理包括资源的组织、定位、发现、调度、分配、确认、进程创建以及准备所需资源的其它活动。它提供了管理的功能和概念,使集群能够被当作单一资源,系统管理员根据预先定义好的标准通过资源管理软件确保资源的合理分配和使用,以最终达到资源共享的目的。为了让用户分配到合适的资源,完成用户递交的任务、并达到用户所希望的QoS,资源预留技术成为资源管理中不可回避的一个重要组成部分。本文的主要工作是在传统的基于层次调度的网格模型—P2P-Grid之上,提出了一种新的网格结构模型Group-Grid,解决了传统网格结构模型中尚存的几个主要问题。并在该结构模型之上,提出了一种具体的资源预留机制。本文提出的网格结构模型Group-Grid,是把网格中所有的资源按照资源提供者分为一个个的小型网格。该模型通过借用服务发现组件来达到快速查找资源的目的。改进了P2P-Grid使用泛洪法所导致的网络负担增大,查找周期过长的缺点。而且还具有扩展性强,易于调整等特点。而基于该结构模型的资源预留策略,是在现有的一般性预留策略的基础上,提出了任务分块、预留取消以及预留级别等概念,从而保证了资源预留的灵活性和可靠性,并提高了预留成功率和资源预留客户的QoS。另外,针对传统资源预留算法无法均衡网络整体资源利用的不足,本文还对现有的兼顾即时任务的资源预留算法进行改进,提出了一种改进的资源预留算法,通过将即时任务转化为预留任务提交到整个网络来减少即时任务的平均等待时间,而且能实现整个网络的负载均衡,提高网格的资源利用率。仿真结果表明,该资源预留策略比旧有的策略在预留成功率上提高了15%,在网络资源利用率上提高了5%。

【Abstract】 Resource management is one of the most important key technologies, and it is of great theoretical practical value for the development of grid computing. Resource management includes resource organization, location, discovery, scheduling, allocation, confirmation, process creation, and other activities to prepare the necessary resources. It provides functions and concepts of management, so that clusters can be used as a single resource, and according to predefined criteria, the system administrator ensures the rational allocation and use of resources through resource management software, in order to eventually achieve the purpose of sharing resources. Resource reservation technology is an important part of resource management, so as to make users to have appropriate resources, complete the task which the users submitted, and achieve the QoS that users desired.The goal of this paper is to give a new model Group-Grid based on traditional grid hierarchical scheduling model-"P2P-Grid", and it can solve the main problems in traditional grid model. Group-Grid use service discovery component to achieve quickly find resources. Moreover, it improved the performance to avoid network burden caused by flooding algorithm and long searching period. It’s also scalable and easy to adjust. Based on the existing general strategy, this thesis put forward some new concepts such as task block; cancel reservation and the reservation level, for ensuring the flexibility and reliability of resource reservation. The resource reservation strategy improves reservation success rate as well as the entire grid resource reservation client QoS.In addition, the traditional resource reservation algorithm cannot balance the lack of overall network resource utilization, and the existing resource reservation which including the task queue has been improved in this paper. And the paper gives an improved resource reservation algorithm and it can reduce the average waiting time for queued tasks by reserved for the task instead the task queue. It also can achieve load balancing of the network and improve the utilization of grid resources. Simulation results show that the resource reservation strategy than the old reservation on the success rate increased by 15%, improve the utilization of network resources by 5%.

节点文献中: 

本文链接的文献网络图示:

本文的引文网络