CSD-CoKV: Host-CSD Collaborative Offloading for High-Performance LSM-Tree Based KV Stores

LSM-tree-based key-value stores are widely used due to their write efficiency and scalability. However, their background compaction process requires substantial data movement between the host and storage devices, resulting in inefficient resource usage, frequent write stalls, and degraded overall throughput. Existing optimizations fail to address the performance bottlenecks caused by excessive system interconnect traffic. Although in-storage computing offers a promising way to mitigate data transfer overhead, prior works typically offload all compaction tasks blindly to computational storage devices. We have performed plenty of preliminary experiments and observed that such naive offloading leads to performance inefficiency due to CSD resource constraints. In this work, we propose CSD-CoKV, a novel framework that harmonizes CSD arrays with the host to accelerate LSMtree compaction more intelligently. We propose a semantic-aware offloader to selectively pick appropriate compaction jobs to CSD arrays and deploys a lightweight high-speed compaction kernel inside each CSD. Additionally, we design a CSD-friendly data partitioning scheme that leverages both KV store characteristics and CSD storage locality to achieve balanced workload distribution and improved efficiency in CSD arrays. Evaluations show that, with the same number of compaction threads, CSD-CoKV improves write throughput by up to $2.39 \times$ and boosts compaction speed by up to $3.28 \times$.