SSD Failure Prediction Based on Classification Models and Data Engineering
Ziyao Wang, Jie Xu · 2022 IEEE Intl Conf on Dependable, Autonomic and Secure Computing, Intl Conf on Pervasive Intelligence and Computing, Intl Conf on Cloud and Big Data Computing, Intl Conf on Cyber Science and Technology Congress (DASC/PiCom/CBDCom/CyberSciTech) · 2022
As the increasing demand of data storage in recent years, solid state drives (SSDs) have developed rapidly. However, the frequent occurrence of SSD failures is affecting the stability and reliability of the systems and the IT infrastructures which use SSDs. SSD failure prediction would help to improve the system reliability, but the current method is mainly based on the design rules of SSD, leading to weak performance and poor prediction accuracy. In this paper, we study methods for predicting SSD failures based on various classification models and investigate how to improve the accuracy of failure prediction and reduce the O&M costs. The publicly available S.M.A.R.T. (Self-Monitoring, Analysis and Reporting Technology) data-set is analysed, pre-processed, and then used as a basis for examining three classification models which could be employed to predict SSD failures. Moreover, we develop two failure prediction methods, both with the highest macro average F1 score amongst the methods and techniques considered. The experimental results show that both methods improve the performance and accuracy of SSD failure prediction subject to certain environmental and testing conditions.