A Proactive Drive Reliability Model To Predict Failures In The Hard Disk Drives
Impending failure detection of hard disks is very important task because it helps computer systems to prevent
performance degradation and loss of data. In this work, different types of data from the disk were used to predict the
impending failure of hard disk drives. Events track the progress of actions, both normal and exceptional, that occur on the
array. A controller event is a normal or exceptional action on any hardware component or logical object within the array. The
performance data provides information on activity within the storage system that can be useful in determining saturation
levels or assessing capacity for additional workload. A model to predict the failure of a single or multiple hard disk drives
within a computer system was developed using these event, performance data and some historical data available from the
drives. This model provided balance between the false alarm rate and the failure detection rateso as to provide to the users,
an advanced warning of hard disk drive failures, so that the users can back up their data before the failure actually occurs.
The developed model was applied on a test data set in order to prove its effectiveness on predicting failures. This method
was also applied on a real-life hard disk drive data set to demonstrate its practical usefulness. The result analysis shows that
our method could achieve 77.14% failure detection rate as compared to using only S.M.A.R.T. parameters which has the
detection rate of 50% to 60%.