Storage Capacity Forecasting via Machine Learning

The patent segments the capacity prediction problem from pool-level to storage object-level forecasts. Instead of applying a single rule-based threshold at the pool level, the system divides the pool into individual storage objects (LUNs, file systems) and generates capacity forecasts for each object separately using machine learning. These individual forecasts are then aggregated to determine pool-level capacity status, thereby improving prediction accuracy while maintaining manageable complexity through automated processing.

The patent implements self-service through automated machine learning models that independently forecast capacity for each storage object without requiring manual configuration or intervention. The system automatically collects historical capacity data, trains forecasting models, generates predictions, and aggregates results. This automation eliminates the need for manual rule configuration at multiple levels, allowing the system to handle the increased complexity of object-level forecasting while maintaining ease of operation through self-service capabilities.

The patent transitions from monitoring capacity at a single pool level to forecasting capacity at the individual storage object level, effectively adding a dimensional breakdown to the monitoring approach. By examining capacity trends for each LUN and file system separately and then aggregating these forecasts, the system captures the nuanced impact of oversubscription on individual objects. This dimensional change allows the system to maintain management efficiency through automation while improving prediction reliability by accounting for object-specific capacity behaviors.