New Research by TSYS School's Yi Zhou Proposes New Data Hierarchy for Optimizing KV Stores

The rapid growth of unstructured data has driven the widespread adoption of LSM-tree-based key-value stores (KV stores). The write amplification resulting from compaction in LSM-trees causes a performance bottleneck. Existing solutions attempt to address this issue through key-value separation strategies. However, these studies fail to optimize the memory components of LSM-trees or provide efficient garbage collection (GC) strategies that achieve high performance while minimizing CPU overhead. These limitations motivated TSYS School computer scientist Yi Zhou and researchers from Anhui University, Zhongguancun Laboratory and Auburn University to propose a GPGPU-empowered gradient data hierarchy and key-value separation for optimizing KV stores, named GDH+. In a study set to appear in a forthcoming issue of ACM Transactions on Architecture and Code Optimization, Zhou et al. utilize GPGPU acceleration for sorting and lushing operations, optimizing the memory components of the LSM-tree. Additionally, they enhance read performance with an LRU-based memory component that distinguishes between hot and cold data, in combination with an adaptive migration strategy. Furthermore, they propose an in-place GC strategy to reduce CPU overhead while maintaining high performance. Their GDH+ achieves a 2×, 1.5×, and 40% improvement in write performance, read performance, and CPU utilization, respectively, compared to state-of-the-art KV stores.