Prediction of HDD Failures by Ensemble Learning
Qiang Li, Hui Li, Kai Zhang · 2019
Storage is an important infrastructure in data center, and it is the carrier of user's data assets. In the storage system, hard disk is the component which has the highest fault rate. Disk failure can lead to loss of user data, slow system operation and unavailability of services. If we can accurately predict disk failures, we can backup and restore data when the user's service is not busy. And this will improve the reliability and availability of storage system. Based on hard disk's SMART data, this paper uses XGBoost, LSTM and ensemble learning algorithm to effectively predict disk faults. Our main contribution is to predict disk failures for weeks, not just on the day of failure. Experiments show that our method can effectively predict disk faults within 42 days with an accuracy of 78%.