基于遗传算法的云计算环境下银行实时预测系统任务调度的研究与实现

发布时间：2018-04-24 13:19

本文选题：云计算 + 实时任务调度　；参考：《北京邮电大学》2016年硕士论文

【摘要】：云计算通过将计算资源、存储资源和服务资源等通过网络连接起来,形成一个资源池,然后根据用户的需求,对资源进行统一的调度和管理。如何对资源池中的资源进行及时、高效地调度,以满足用户对云服务质量的多方面的需求,是云计算研究中很重要的一环。当前的实时任务调度算法只关注于满足用户任务的实时性需求并且这些算法不够灵活,对于实时变化的异构系统,不能较好的适应。本文借助遗传算法全局优化搜索的特点,从用户的实时性需求和系统的总体吞吐量出发,设计出基于实时性和总体吞吐量的适应度函数,进而可以将基于遗传算法的任务调度应用到实时任务调度的环境当中。不仅能够更好地适应实时变化的云计算环境,更能够保证任务的总体运行效率。针对遗传算法存在的收敛速度慢的问题,本文提出了基于资源感知的优化策略。根据任务负载的大小、资源使用情况以及任务的类型,引导遗传算法的收敛过程,从而加速收敛过程。然后在模拟环境中进行了实验,通过与传统的实时任务调度算法在不同的参数条件下进行比较,验证了改进后的遗传算法在任务完成率、资源利用率等方面都具有较好的效率。银行实时预测系统通过机器学习的方法预测资源的阈值以及使用趋势等,对于及时发现系统故障,瓶颈具有很好的参考价值。本文提出的调度算法对于这种数据量、运算量较大,并且需要较好的实时性的系统具有较好的适用性。最后将文中提出的改进算法和传统的任务调度算法分别对该系统进行任务调度,通过任务完成率、系统总体运行时间及系统能耗等方面的对比,证明本文提出的算法在实际应用中也有较高的效率。
[Abstract]:Cloud computing connects computing resources, storage resources and service resources through the network to form a pool of resources. Then according to the needs of users, resources are uniformly scheduled and managed. How to schedule the resources in the resource pool in a timely and efficient manner to meet the needs of users in many aspects of cloud service quality is an important part of cloud computing research. The current real-time task scheduling algorithms only focus on meeting the real-time requirements of user tasks and these algorithms are not flexible enough to adapt to real-time changing heterogeneous systems. In this paper, according to the characteristics of global optimization search based on genetic algorithm, a fitness function based on real-time and total throughput is designed based on the real-time requirements of users and the overall throughput of the system. Then the genetic algorithm based task scheduling can be applied to the real-time task scheduling environment. It can not only adapt to the real-time changing cloud computing environment, but also ensure the overall operational efficiency of the task. Aiming at the problem of slow convergence rate of genetic algorithm, an optimization strategy based on resource awareness is proposed in this paper. According to the size of task load, resource usage and task type, the convergence process of genetic algorithm is guided and the convergence process is accelerated. By comparing with the traditional real-time task scheduling algorithm under different parameter conditions, the improved genetic algorithm is proved to be more efficient in terms of task completion rate and resource utilization. The bank real-time prediction system predicts the threshold value and the usage trend of the resource by the method of machine learning, which has a good reference value for detecting the system fault in time. The scheduling algorithm proposed in this paper is suitable for the system with large amount of computation and good real-time performance. Finally, the proposed improved algorithm and the traditional task scheduling algorithm are compared in terms of task completion rate, system total running time and system energy consumption. It is proved that the proposed algorithm has high efficiency in practical application.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP311.52;TP18

【参考文献】