Improved disk-drive failure warnings

摘要

Improved methods are proposed for disk-drive failure prediction.nThe SMART (self monitoring and reporting technology) failure predictionnsystem is currently implemented in disk-drives. Its purpose is tonpredict the near-term failure of an individual hard disk-drive, andnissue a backup warning to prevent data loss. Two experimental tests ofnSMART show only moderate accuracy at low false-alarm rates. (A rate ofn0.2% of total drives per year implies that 20% of drive returns would bengood drives, relative to ≈1% annual failure rate of drives). Thisnrequirement for very low false-alarm rates is well known in medicalndiagnostic tests for rare diseases, and methodology used there suggestsnways to improve SMART. Two improved SMART algorithms are proposed. Theynuse the SMART internal drive attribute measurements in present drives.nThe present warning-algorithm based on maximum error thresholds isnreplaced by distribution-free statistical hypothesis tests. Thesenimproved algorithms are computationally simple enough to be implementednin drive microprocessor firmware code. They require only integer sortnoperations to put several hundred attribute values in rank order. Somentens of these ranks are added up and the SMART warning is issued if thensum exceeds a prestored limit. These new algorithms were tested on 3744ndrives of 2 models. They gave 3-4 times higher correct predictionnaccuracy than error thresholds on will-fail drives, at 0.2% false-alarmnrate. The highest accuracies achievable are modest (40%-60%). Care wasntaken to test will-fail drive prediction accuracy on data independent ofnthe algorithm design data. Additional work is needed to verify and applynthese algorithms in actual drive design. They can also be useful inndrive failure analysis engineering. It might be possible to screenndrives in manufacturing using SMART attributes. Marginal drives might bendetected before substantial final test time is invested in them, therebyndecreasing manufacturing cost, and possibly decreasing overall fieldnfailure rates