Alleviating Deduplication-oriented Fragmentation of SSDs by Considering File Hotness and Popularity

Data deduplication is a highly efficient method for data storage, providing significant benefits in terms of storage utilization and cost reduction in large-scale storage systems. This research introduces a novel approach to dynamic chunk allocation in deduplication systems, aiming to minimize fragmentation and enhance read performance by leveraging supported SSD commands. The write strategy is intelligently determined based on the popularity of chunks within each die, optimizing overall system performance. Additionally, we introduce a read/write unit based on the plane level to harness the internal parallelism of SSDs. Through extensive simulations, our proposed design demonstrates an average performance improvement of 18.8%, highlighting our approach’s effectiveness and impact.