Drive Failure Prediction Using I/O-Aware Learning Models

A learning model serves as an intermediary between raw response performance data and failure prediction results. The learning model receives response performance data along with command information (I/O patterns, block lengths, command counts) and processes them to generate accurate failure predictions, thereby resolving the contradiction by introducing a sophisticated processing layer that handles workload variations without requiring fundamental changes to the storage system architecture

The system changes the parameters used for failure prediction from simple response time comparisons to a multi-parameter approach that includes response performance data combined with command information (I/O patterns, transfer block lengths, command counts). The learning model processes these multiple parameters to generate accurate failure predictions, thereby resolving the contradiction by transforming the prediction approach from simple comparison to multi-parameter analysis

The learning model is applied locally to each drive apparatus within the RAID group, processing only that specific drive's response performance data and command information. This localized approach allows accurate failure prediction for individual drives without requiring a centralized complex system that processes all drives' data simultaneously, thereby reducing overall system complexity while maintaining high detection accuracy

The system uses response performance data and command information that are already being collected and processed by the storage control apparatus for other purposes. By reusing this existing data as input for the learning model, the system avoids duplicating data collection mechanisms and reduces implementation complexity while still achieving accurate failure prediction