Paper Info
Title
Disk Failure Prediction Based on SW-Disk Feature Engineering
Abstract
With the rapid development of cloud storage, the number of disks has increased dramatically, and the absolute number of failed disks has also continued to increase. Since the impact of these failures on the quality of cloud services cannot be ignored, the demand for disk failure prediction is also increasing. In addition, the accuracy of the disk failure prediction algorithm largely depends on the input data set. Therefore, according to the large difference in the feature morphology in the disk tracking dataset, we analyze the change trend of features in different dimensions, and propose a new method. The SW-Disk feature engineering method for disk failure prediction constructs features from both horizontal and vertical dimensions. Including feature crossover between indicators and indicators to construct higher-order features. In the time dimension, the time series features under the multi-scale sliding window are constructed. We use the random forest prediction model to predict whether the disk will fail in the next 30 days, so as to reserve more time to take measures on the disk to achieve the purpose of improving the reliability of the cloud storage system. We validated our SW-Disk feature engineering method and prediction model on the Backblaze2019 disk dataset, and the experimental results demonstrate the effectiveness and superiority of our proposed method.
Year
DOI
Venue
2022
10.1109/BigDataSecurityHPSCIDS54978.2022.00027
2022 IEEE 8th Intl Conference on Big Data Security on Cloud (BigDataSecurity), IEEE Intl Conference on High Performance and Smart Computing, (HPSC) and IEEE Intl Conference on Intelligent Data and Security (IDS)
Keywords
DocType
ISBN
disk failure,feature engineering,sliding window,random forest
Conference
978-1-6654-8070-3
Citations 
PageRank 
References 
0
0.34
18
Authors
5
Name
Order
Citations
PageRank
Chenjun Liang100.68
Li Deng200.68
Jincan Zhu300.68
Zhen Cao400.68
Chao Li5525110.37