This document summarizes findings from analyzing failure trends in a large population of disk drives used in Google's computing infrastructure over a period of several years. Some key findings include: 1) Contrary to previous reports, there was little correlation found between failure rates and elevated temperature or high activity levels. 2) Some SMART parameters like scan errors, reallocation counts, and offline reallocation counts showed a strong correlation with failure probability. 3) Despite correlations with some SMART parameters, models based on SMART data alone are unlikely to accurately predict individual drive failures due to many failed drives not exhibiting predictive SMART signals.