云平台主机资源负载预测分析研究  

Analysis and Prediction of Host Resource Load in the Cloud

作　　者：朱金灿 邓莉[1,2] 梁晨君 严明 谢同磊 任正伟 ZHU Jin-can;DENG Li;LIANG Chen-jun;YAN Ming;XIE Tong-lei;REN Zheng-wei(College of Computer Science and Technology,Wuhan University of Science and Technology,Wuhan 430065,China;Hubei Province Key Laboratory of Intelligent Information Processing and Real-time Industrial System,Wuhan 430065,China;Agency for Science,Technology and Research,Singapore 138632,Singapore)

机构地区：[1]武汉科技大学计算机科学与技术学院,武汉430065 [2]智能信息处理与实时工业系统湖北省重点实验室,武汉430065 [3]新加坡科技研究局,新加坡138632

出　　处：《小型微型计算机系统》2021年第12期2538-2544,共7页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61902285)资助;湖北省自然科学基金项目(2019CFB099)资助。

摘　　要：云平台主机资源负载预测对于提高系统资源利用率以及实现资源分配的优化至关重要,也是实现云平台服务水平协议的关键所在.有效的主机负载预测机制可促进主动作业调度,辅助主机负载平衡决策,这反过来可以提高主机资源利用率、改善作业性能、降低数据中心成本.具体来看,云平台中主机工作负载具有快速变化、波动大和长期信息依赖等特点,这使得负载预测工作变得复杂.为了解决上述预测问题,本文做了如下工作:(1)实现了适合主机平均负载预测的指数分段预测模式;(2)完成了主机实际负载多步预测模式;(3)在2个真实云平台数据集进行实验,并采用3种评价函数对实验结果进行评估.最终结果表明,相对于目前已经提出的主机负载预测模型,本文方法具有更好的预测性能.In the cloud host resource load prediction is very important to improve cloud system resource utilization and optimize resource allocation. It is also the key to realize cloud platform service level agreement. Effective host load prediction mechanism can promote active job scheduling and assist host load balancing decision,which in turn can improve host resource utilization,improve job performance and reduce data center cost. Specifically,the host workload in cloud platform has the characteristics of rapid change,large fluctuation and long-term information dependence,which makes the load forecasting work more complex. In order to solve the above prediction problems,the following work has been done in this paper: (1) The exponential segmentation prediction model suitable for the average load prediction of the host is realized;(2) The multi-step prediction model of host actual load is completed;(3) Experiments were conducted on two real cloud platform datasets,and three evaluation functions were used to evaluate the experimental results. The final results show that the proposed method has better prediction performance than the existing host load prediction model.

关 键 词：主机负载 BC-LSTM 时间序列 

分 类 号：TP319[自动化与计算机技术—计算机软件与理论]