基于细粒度键值分离策略的键值存储引擎方案的设计与实现
首发时间:2023-02-20
摘要:现有的键值存储引擎为了获得极致的写性能大多采用了日志结构合并树作为核心的数据组织方式来管理键值对。但是LSM-Tree在实现高性能写吞吐的同时,却引入了高额的读、写放大。为了解决这一问题,本文设计了一个新的键值存储引擎,在采用细粒度的键值分离的基础上,使用传统的完全排序的LSM-Tree来管理键;参考跳表设计了部分排序的数据组织方式来管理值,同时针对新的数据组织方式设计了数据压缩和垃圾回收机制等方案。实验结果表明,本文的键值存储引擎在保持读、写操作高性能的同时,有效降低了读、写放大的问题。
关键词: 键值数据库 细粒度键值分离 部分排序的数据结构 数据压缩 垃圾回收
For information in English, please click here
The Design and Implementation of KV Storage Scheme based on Fine-Grained KV Separation
Abstract:In order to achievehigh write performance, most of the existing key-value storages adopt the Log Structured Merge Tree as their core data organization to manage the KV pairs. However, LSM-Tree offers excellent write throughput, but suffers high write amplification. To remedythis problem, this paper presentsa new key-value storage basing on the fine-grained key-value separation.we adopt thetraditional LSM-Tree to manage the keysand design a partially sorted data organization to manage the values. This paper alsopresents the data compaction and garbage collection mechanism for the newpartially sorted data organization. The experimental results show that the key value storage engine in this paper effectively reduces write amplification while maintaining the high performance of read and write operation.
Keywords: Key-Value Storage Fine-Grained KV Separation Partially-SortedDataOrganization Data Compaction Garbage Collection
基金:
引用
No.****
同行评议
勘误表
基于细粒度键值分离策略的键值存储引擎方案的设计与实现
评论
全部评论