面向绿色云计算的资源配置及任务调度研究

发布时间：2018-04-30 21:17

本文选题：绿色计算 + 云计算　；参考：《南京邮电大学》2014年硕士论文

【摘要】：随着云计算技术的迅速发展，使得云数据中心服务器的规模每年都在不断的扩大，产生巨大的能源开销。不合理的调度策略同样导致能源浪费严重，使得云数据中心的运营成本不断增加。因此，绿色云计算的概念应运而生。针对目前云数据中心的高能耗、低效率问题，，本文围绕着绿色云计算系统模型的构建、资源配置和任务调度算法的设计及验证等工作展开了一系列的研究与探索，主要包括：首先，从实现绿色云计算的内部和外部因素出发，设计了一种高效可靠的绿色云计算系统模型。该模型重点着手于系统内部的资源配置模块和任务调度模块，针对这两个子模块的实际需求，给出了具体的实施方案。资源配置模块使得系统资源的分配更加合理，提高了系统资源的利用率，降低了云计算系统整体的能源消耗；任务调度模块能够保证云计算系统具有合理的任务调度机制。其次，从全局角度出发，抽象任务调度问题为资源配置问题。通过预测用户请求的负载大小，并结合当前系统状态和资源分布，采取保守控制策略，计算下一个周期内任务对系统资源的需求量。然后，建立满足多目标约束的能耗模型，提出基于概率匹配的资源配置算法，实现低能耗资源配置。在此基础上，提出基于改进型模拟退火的资源配置算法，进一步降低系统能耗。实验结果表明，采取的预测与控制策略能够有效避免资源配置滞后于用户请求的问题，提高系统的响应比和稳定性；提出的资源配置算法能够激活更少主机，实现激活主机集合之间更好的负载均衡和资源的最大化利用，降低云计算系统能耗。最后，为了设计合理的任务调度机制，提出一种面向云计算平台任务调度的多级负载评估方法。该方法为绿色云计算系统的任务调度提供一种合理的负载评估方法。在此基础上，提出一种基于动态负载调节的自适应云计算任务调度策略，任务节点能够自适应负载的变化，按照计算能力获取任务，实现各个节点自调节，同时避免因采用复杂的调度算法，使得管理节点承载巨大的系统开销，成为系统性能瓶颈。实验结果表明，基于动态负载调节的自适应云计算任务调度策略高效可靠，适用于异构云计算集群的计算环境，性能优于现有的任务调度策略。
[Abstract]:With the rapid development of cloud computing technology, the scale of cloud data center server is expanding every year, which brings huge energy cost. Unreasonable scheduling strategy also leads to serious energy waste, which makes the operating cost of cloud data center increasing. Therefore, the concept of green cloud computing came into being. Aiming at the problem of high energy consumption and low efficiency in cloud data center, this paper focuses on the construction of green cloud computing system model, the design and verification of resource allocation and task scheduling algorithm, etc. Firstly, an efficient and reliable green cloud computing system model is designed based on the internal and external factors of green cloud computing. The model focuses on the resource allocation module and task scheduling module in the system. According to the actual requirements of the two sub-modules, the concrete implementation scheme is given. Resource allocation module makes the allocation of system resources more reasonable, improves the utilization of system resources, and reduces the overall energy consumption of cloud computing system. Task scheduling module can ensure that cloud computing system has a reasonable task scheduling mechanism. Secondly, from the global point of view, the abstract task scheduling problem is a resource allocation problem. By predicting the load size of the user request and combining the current system status and resource distribution, conservative control strategy is adopted to calculate the demand for system resources for the next cycle. Then, the energy consumption model satisfying multi-objective constraints is established, and a resource allocation algorithm based on probability matching is proposed to achieve low energy resource allocation. On this basis, a resource allocation algorithm based on improved simulated annealing is proposed to further reduce the energy consumption of the system. Experimental results show that the proposed prediction and control strategy can effectively avoid the problem of resource allocation lagging behind user requests, improve the response ratio and stability of the system, and the proposed resource allocation algorithm can activate fewer hosts. In order to reduce the energy consumption of cloud computing systems, better load balance and maximum utilization of resources can be achieved between the active host sets. Finally, in order to design a reasonable task scheduling mechanism, a multi-level load assessment method for task scheduling in cloud computing platform is proposed. This method provides a reasonable load assessment method for task scheduling of green cloud computing systems. On this basis, an adaptive cloud computing task scheduling strategy based on dynamic load regulation is proposed. The task node can adapt the load change, acquire the task according to the computing power, and realize the self-adjustment of each node. At the same time, it is avoided that the management node carries huge system overhead and becomes the bottleneck of system performance because of the use of complex scheduling algorithm. The experimental results show that the adaptive cloud computing task scheduling strategy based on dynamic load regulation is efficient and reliable, and it is suitable for computing environment of heterogeneous cloud computing cluster, and its performance is superior to the existing task scheduling strategy.
【学位授予单位】：南京邮电大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.01

【参考文献】