SLA: A Cache Algorithm for SSD-SMR Storage System with Minimum RMWs

Abstract

AbstractTo satisfy the low-cost and massive data storage requirements imposed by data growth, Shingled Magnetic Recording (SMR) disks are extensively employed in the area of high-density storage. The primary drawback of SMR disks is the write amplification issue caused by sequential write constraints, which is becoming more prominent in the field of non-cold archives. Although a hybrid storage system comprised of Solid State Disks (SSDs) and SMRs may alleviate the aforementioned issue, current SSD cache replacement algorithms are still limited to the management method of Least Recently Used (LRU). The LRU queue, on the other hand, is ineffective in reducing the triggers of Read-Modify-Write (RMW), which is a critical factor for the performance of SMR disks. In this paper, we propose a new SMR Locality-Aware (SLA) algorithm based on a band-based management method. SLA adopts the Dual Locality Compare (DLC) strategy to solve the hit rate reduction problem caused by the traditional band-based management method, as well as the Relatively Clean Band First (RCBF) strategy to further minimize the number of RMW operations. Experiments indicate that, compared to the MSOT method, the SLA algorithm can maintain a similar hit rate as the LRU, while reducing the number of RMWs by 77.2% and the SMR disk write time by 95.1%.KeywordsShingled Magnetic RecordingReplacement algorithmHybrid storage systemSpatial locality