快速检测数据中心机械硬盘异常的方法  

The Method for Quickly Detecting Abnormality of Mechanical Hard Disks in the Data Center

作　　者：魏坤 沈记全[1] WEI Kun;SHEN Jiquan(School of Computer Science and Technology,Henan Polytechnic University,Jiaozuo Henan 454000,China)

机构地区：[1]河南理工大学计算机科学与技术学院,河南焦作454000

出　　处：《北京联合大学学报》2022年第2期47-51,共5页Journal of Beijing Union University

基　　金：国家自然科学基金项目(61972134)。

摘　　要：针对性能异常的机械硬盘在数据读写时出现的带宽不稳和时延增大,导致数据中心存储集群性能下降及不稳定的问题,提出一种基于高斯分布算法检测性能异常硬盘的方法。通过滑动平均算法对健康硬盘性能的样本数据进行预处理,以确定健康硬盘性能数据集,然后采用高斯分布异常检测算法对预处理的数据集进行预测故障模型的训练与建立。实验表明,该方法能快速检测性能异常的硬盘,对于拥有大规模存储集群的数据中心,以及对负载均衡和性能要求较高的工程领域均有较好的应用价值。An algorithm based on Gaussian bandwidth is proposed to detect the abnormal performance of the hard disk when the read/write performance of the hard disk is unstable.The moving average algorithm is used to preprocess the performance sample data of the healthy hard disk to determine the performance data set of the healthy hard disk,and then the Gaussian distribution anomaly detection algorithm is used to train and establish the predictive fault model for the preprocessed data set.Experiments show that this method can quickly detect abnormal hard disks,and it has good application value for data centers with large-scale storage clusters and engineering fields with high requirements for load balancing and performance.

关 键 词：硬盘 滑动算法 高斯异常检测 存储集群 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]