Mondrian Conformal Prediction for Disk Health Scoring and Scrubbing Optimization

Traditional disk scrubbing improves data integrity but drains performance and energy. This paper introduces a selective, intelligent scrubbing method powered by Mondrian conformal prediction, assigning health scores to individual drives to determine which disks to scrub and when to do it. By aligning maintenance cycles with idle periods and using predictive models to detect latent or “fail-slow” issues, the approach enhances reliability while minimizing resource waste—offering a data-driven path toward smarter, greener data centers.