基于LSM-tree的键值存储系统的读写性能优化

Read And Write Performance Optimization Of LSM-tree Based Key-Value Storage System

摘要:

在写密集型工作环境中，日志结构合并树(log-structured-merge，LSM-Tree)已逐渐成为的主流存储系统，LSM-tree存在读操作速度慢、写操作成本高、范围查询操作效率低等问题；针对这些问题，为提升LSM-tree的性能进行了研究，提出了一种基于LSM-tree的键值存储系统的读写性能优化策略，通过键值分离策略设计vTree结构，并提出层内归并与消极的层间合并相结合的方法，以及范围查询优化合并的策略，从而优化系统的范围查询性能，在LSM-tree和vTree采用不同的压缩结构，以实现系统读写性能的提升；实验结果表明，与RocksDB相比读性能提升30%，与RocksDB-vTree相比范围查询性能提升10%。

Abstract:

In a write-intensive work environment, log-structured-merge (LSM-Tree) has gradually become the mainstream storage system, LSM-tree exists problems such as slow read operation speed, high cost of write operation, and low efficiency of range query operation, etc. In view of these problems, a study is carried out to improve the performance of LSM-tree, and a strategy to optimize the read and write performance of key-value storage system based on LSM-tree is proposed. a read and write performance optimization strategy for LSM-tree-based key-value storage system, designing the vTree structure through the key-value separation strategy, and proposing the combination of intra-tier subsumption and negative inter-tier merging, as well as the strategy of range query-optimized merging, so as to optimize the range query performance of the system, and adopting different compression structures in the LSM-tree and the vTree in order to achieve the system's read and write performance improvement; the experimental results show that the read performance is improved by 30% compared to RocksDB, and the range query performance is improved by 10% compared to RocksDB-vTree.