云平台中多宿主等待队列动态预测调度算法  

Dynamic Prediction and Scheduling Algorithm for Multi Host Waiting Queue in Cloud Platform

作　　者：吴菲[1] 徐平平[2,3] WU Fei;XU Ping-ping(Pujiang Institute,Nanjing Tech.University,Nanjing Jiangsu 210000,China;College of Information Science and Technology,Southeast University,Nanjing Jiangsu 210000,China;Key Laboratory of Mobile Communication,Southeast University,Nanjing Jiangsu 210000,China)

机构地区：[1]南京工业大学浦江学院,江苏南京210000 [2]东南大学信息科学与技术学院,江苏南京210000 [3]东南大学移动通信国家重点实验室,江苏南京210000

出　　处：《计算机仿真》2022年第1期451-455,共5页Computer Simulation

基　　金：国家自然科学基金项目(61702229);国家自然科学基金项目(61571111);江苏省自然科学基金项目(19KJB520037)。

摘　　要：由于云平台中多宿主数据流在调度过程中容易发生阻塞和负载失衡,为此,提出一种等待队列动态预测调度算法。首先对请求包流入队的情况进行预测,搜索出请求包流存在的可用等待队列,并得到每个等待队列中的最大请求包数量预测边界,比较确定当前时刻入队的请求包流。然后通过负载情况确定当前等待队列的忙闲,再根据马氏迁移概率预测出最佳迁移队列,从而完成请求任务的迁移,使云服务任务能够得到及时有效的处理。为了准确判断等待队列的忙闲状态,采用流量等级作为判定依据。最后为了使任务迁移适应云平台的分布式集群架构,在预测调度时加入了对虚拟机状态的衡量。通过对最大输入率、负载均衡性,以及响应延时的仿真,验证了等待队列预测调度算法能够快速有效的处理云平台中多宿主请求数据包流,并且具有良好的响应延时和负载均衡性,避免发生调度阻塞。Because the multi host data flow in the cloud platform is prone to block and load imbalance in the scheduling process, this paper proposes a waiting queue dynamic prediction scheduling algorithm. First of all, we predicted the flow of request packets into the queue, searchedd out the available waiting queues of the request packet flow, and get the prediction boundary of the maximum number of request packets in each waiting queue, compared and determined the request packet flow in the queue at the current time. Then the free time of the current waiting queue was determined through the load situation, and the best migration queue was predicted according to the migration probability of Markovian, so as to complete the migration of the request tasks, so that the cloud service tasks can be processed in a timely and effective manner. In order to accurately determine the free and busy status of the waiting queue, the traffic level was used as the judgment basis. Finally, in order to adapt the task migration to the distributed cluster architecture of cloud platform, the measurement of virtual machine state was added to the prediction and scheduling. Through the simulation experiments of the maximum input rate, load balance and response delay, it was verified that the waiting queue predictive scheduling algorithm can deal with the multi host request packet flow in the cloud platform quickly and effectively, and has good response delay and load balance to avoid scheduling blocking.

关 键 词：云平台 多宿主请求 等待队列 迁移概率 动态预测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]