RangeKV: An Efficient Key-Value Store Based on Hybrid DRAM-NVM-SSD Storage Structure

Abstract

Persistent key-value (KV) stores are an integral part of storage infrastructure in data centers. Emerging non-volatile memory (NVM) technologies are potential alternatives for future memory architecture design. In this study, we use NVM to optimize the KV store and propose RangeKV, an LSMtree based persistent KV store built on a heterogeneous storage architecture. RangeKV uses RangeTab in NVM to manage L0 data and increases L0 capacity to reduce the number of LSM tree levels and system compactions. RangeKV pre-constructs the hash index of RangeTab data to reduce NVM access times and adopts a double-buffer structure to reduce LSM-tree write amplification due to compactions. We implement RangeKV based on RocksDB and conduct a comparative test and performance evaluation with RocksDB and NoveLSM. The test results show that the overall random write throughput is improved by 4.5× to 5.7× compared to RocksDB. In addition, RangeKV has a significant performance advantage over NoveLSM.

Key-Value Stores on Flash Storage Devices: A Survey

“…SplitKV [58] uses a cache/split design for a B+-tree, storing hot and small items in NVM, and cold and large values on flash (reducing Q1 and Q4 in flash). NoveLSM [70] and RangeKV [145] both resort to splitting LSM-trees. NoveLSM keeps the mutable and immutable memtables in NVM and RangeKV stores a modified L0, referred to as rangetabs, in persistent memory.…”

The Embedded IoT Time Series Database for Hybrid Solid-State Storage System

“…HiKV is a KV storage engine based on NVM [24], which established and retained a hash index in NVM to maintain its inherent fast index search capabilities and built a B+ tree index in DRAM to support range queries, thereby avoiding long-term NVM write and guaranteeing the index consistency. RangeKV is a persistent KV storage engine based on RocksDB, which is built on a heterogeneous storage architecture [25]. It used RangeTab in NVM to manage L0 layer data and increased L0 capacity to reduce LSM tree level and compression time overhead.…”