We now analyze the failure behavior of our fleet of disk
drives using detailed monitoring data collected over a
nine-month observation window. During this time we
recorded failure events as well as all the available environmental
and activity data and most of the SMART
parameters from the drives themselves. Failure information
spanning a much longer interval (approximately five
years) was also mined from an older repairs database.
All the results presented here were tested for their statistical
significance using the appropriate tests.