Forecast-Aware Versus Threshold-Based Capacity Expansion in Thin-Provisioned Storage: A Trace-Driven Simulation

Abstract

Threshold-based expansion is simple to operate but can react too late when thin-provisioned storage experiences rapid growth. We compared fixed-threshold expansion with a forecast-aware policy using 12 months of anonymised capacity traces from 48 enterprise volumes. A rolling forecasting model estimated seven-day utilisation trajectories and triggered expansion when projected headroom breached a risk-adjusted limit. Both policies were replayed under identical workload traces with expansion delay, demand volatility, and forecast error varied in sensitivity analyses. Forecast-aware expansion reduced critical-capacity events by 63% and lowered the number of emergency expansions from 94 to 31. It also decreased unused allocated capacity by 12.7% relative to a conservative 80% fixed threshold. Benefits were greatest in workloads with recurring growth patterns, while highly irregular volumes generated more false-positive expansions. Forecast errors above 20% materially reduced the advantage of the predictive policy. A hybrid strategy combining minimum safety thresholds with short-horizon forecasts provided the most stable performance across workloads. These results indicate that capacity expansion can be made more efficient when predictive signals complement, rather than replace, deterministic safety controls.