HiKV: A Hybrid Index Key-Value Store for DRAM-NVM Memory Systems

Hybrid memory systems consisting of DRAM and Non-Volatile Memory are promising to persist data fast. The index design of existing key-value stores for hybrid memory fails to utilize its specific performance characteristics: fast writes in DRAM, slow writes in NVM, and similar reads in DRAM and NVM. This paper presents HiKV, a persistent key-value store with the central idea of constructing a hybrid index in hybrid memory. To support rich key-value operations efficiently, HiKV exploits the distinct merits of hash index and B+-Tree index. HiKV builds and persists the hash index in NVM to retain its inherent ability of fast index searching. HiKV builds the B+-Tree index in DRAM to support range scan and avoids long NVM writes for maintaining consistency of the two indexes. Furthermore, HiKV applies differential concurrency schemes to hybrid index and adopts ordered-write consistency to ensure crash consistency. For single-threaded performance, HiKV outperforms the state-of-the-art NVM-based key-value stores by reducing latency up to 86.6%, and for multi-threaded performance, HiKV increases the throughput by up to 6.4x under YCSB workloads.