面向云测试服务的资源分配策略研究

发布时间：2019-06-19 21:00

【摘要】：在云计算环境下软件测试面临许多新的挑战。本文通过对云测试服务研究现状的调查,发现利用云测试服务进行测试时,需要可靠的资源分配策略作为保障。目前,对于云资源分配策略的研究具备一定的理论研究和实际操作,但缺乏一个同时站在云测试用户和云服务提供商需求角度考虑的整体资源分配架构。本研究在参考云服务模型的基础上,设计了一个云测试服务模型。模型中包含两个组件:一是为云测试用户提供面向资源可用性分配策略的资源分配器1;另一个是在满足用户的可用性需求基础上,为云服务提供商提供面向资源有效性分配策略的资源分配器2。然后使用预测分配的方法对资源分配策略进行研究,并对两个组件实现的算法展开阐述。资源分配器1采用BP神经网络对云测试虚拟机的CPU利用率、可用内存大小进行预测,从而为虚拟机资源的分配提供一种策略。在预测过程中,为提高资源预测精度,引入改进的智能算法即粒子群算法对BP神经网络的初始阈值和权值进行优化,最后采用对比实验证明了改进算法的有效性。在资源分配器1分配的多台云测试虚拟机已满足资源可用性的基础上,资源分配器2采用遗传算法,以最小化云服务器内存为目标,实现云测试虚拟机到云服务器的安置,从而为虚拟机资源的分配提供另一种策略。在分配过程中,传统的遗传算法易求得不可行解,引入单点交叉修复、旋转变异以及外部罚函数理论对遗传算法进行改进。最后采用实验证明了改进算法寻优能力的可行性。
[Abstract]:Software testing faces many new challenges in cloud computing environment. Based on the investigation of the research status of cloud test services, this paper finds that reliable resource allocation strategy is needed to ensure the use of cloud test services for testing. At present, the research on cloud resource allocation strategy has some theoretical research and practical operation, but there is a lack of an overall resource allocation architecture from the perspective of cloud test users and cloud service providers at the same time. Based on the reference of cloud service model, a cloud test service model is designed in this paper. The model contains two components: one is to provide resource allocation policy-oriented resource allocators for cloud test users; the other is to provide cloud service providers with resource allocation policy-oriented resource allocators for resource efficiency allocation policies on the basis of meeting the availability needs of users. Then the resource allocation strategy is studied by using the method of predictive allocation, and the algorithm implemented by the two components is described. Resource allocator 1 uses BP neural network to predict the CPU utilization of cloud test virtual machines and the size of available memory, thus providing a strategy for the allocation of virtual machine resources. In the process of prediction, in order to improve the accuracy of resource prediction, an improved intelligent algorithm, particle swarm optimization algorithm, is introduced to optimize the initial threshold and weight of BP neural network. Finally, the effectiveness of the improved algorithm is proved by comparative experiments. On the basis that multiple cloud test virtual machines allocated by resource allocator 1 have satisfied the availability of resources, resource allocator 2 adopts genetic algorithm to minimize cloud server memory to realize the placement of cloud test virtual machines to cloud servers, thus providing another strategy for the allocation of virtual machine resources. In the process of allocation, the traditional genetic algorithm is easy to obtain the infeasible solution, and the single point cross repair, rotation mutation and external penalty function theory are introduced to improve the genetic algorithm. Finally, the feasibility of improving the optimization ability of the algorithm is proved by experiments.
【学位授予单位】：重庆邮电大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP393.09;TP311.53

【参考文献】