Machine Learning Models for SSD and HDD Reliability Prediction
Riccardo Pinciroli, Lishan Yang, Jacob Alter, Evgenia Smirni · 2022 Annual Reliability and Maintainability Symposium (RAMS) · 2022
This paper compares hard disk drives (HDDs) and solid-state drives (SSDs), the two most used storage devices in data centers, which frequently fail and are among the main causes of data center downtime. Using a six-year field data of 100,000 HDDs from the Backblaze dataset and a six-year field data of 30,000 SSDs from a Google data center, we characterize workload conditions that prompt drive failures. % and show that they differ from common expectation. We develop machine learning models that accurately predict the drive failure state several days in advance and provide highly interpretable results that are useful to identify the causes and symptoms of drive failures.