RASP: Region-Aware Single Use Page Predictor for SSD Workloads

Single use pages degrade SSD performance by causing cache pollution, write amplification, and flash wear. To address this, we propose theRegion-Aware Single Use Page Predictor(RASP), a lightweight firmware level mechanism that exploits spatial locality by grouping logical block addresses into fixed size regions. Each region is associated with a confidence score, and a history in region tracking table (RTT). On MSR and Systor workloads, RASP achieves high accuracy (97.67% recall and 80.21% F1-score) in single use page prediction compared to perfect predictor. We leverage RASP to enable the DRAM cache bypassing and single use page aware cache evictions, thereby reducing cache pollution, decreasing flash writes by 8.04% and reducing end to end latency by 11.71% over the baseline, improving SSD performance and lifespan.