SSD-Based LSM-Tree Key-Value Storage System

Facing the explosive growth of massive data, key-value storage systems have gradually become a key component to support large-scale applications. Solid-State-Drive and log-structured merge-tree has been widely used to build key-value store. However, Deploying the LSM-Tree key-value storage engine directly on a general-purpose SSD has great software redundancy, which leads to cascaded write amplification and poor read performance. In this paper, we design and implement a key-value solid-state drive (KVSSD), which embeds the LSM-Tree structure into the SSD device, provides services directly through the key-value interface, and algorithms of FTL are optimized in combination with the characteristics of software and hardware. The experimental results show that the overall throughput of KVSSD is improved by a factor of about 2 and the write amplification is reduced by about 50% under different workloads compared to a key-value storage system using conventional SSDs.