PheonixLSM:高性能低空间开销的分布式键值存储本地引擎  

PheonixLSM:Local Key-Value Engine for Distributed Key-Value Storage with High Performance and Low Spatial Consumption

作　　者：李润辉 古亮 喻之斌[1] LI Runhui;GU Liang;YU Zhibin(Shenzhen Institute of Advanced Technology,Chinese Academy of Sciences,Shenzhen 518055,China;Sangfor Technologies Inc.,Shenzhen 518071,China)

机构地区：[1]中国科学院深圳先进技术研究院,深圳518055 [2]深信服科技股份有限公司,深圳518071

出　　处：《集成技术》2022年第2期28-40,共13页Journal of Integration Technology

摘　　要：分布式键值存储将数据复制到多个存储服务器的本地引擎中,并通过一致性协议保证各副本数据的一致性。其中,以日志结构合并树为核心数据结构的实现方式最为常见。然而,面向通用业务模式设计的日志结构合并树,并不适合一致性逻辑的特殊业务模式,会引发增删改性能的降低,并在全量修复过程中造成空间放大。针对上述问题,该文提出了一种新型本地引擎PheonixLSM,通过增加增删改操作和回刷操作的约束,消除了分布式键值存储增删改流程中的双写问题,提升了引擎性能。通过重构日志结构合并树底层的SST文件布局,支持删除实时回收空间,消除了全量修复时的额外空间放大。实验结果显示,与原生本地引擎相比,使用PheonixLSM的分布式键值存储系统,增删改性能提升90.7%,全量修复的空间放大从65.6%降至6.4%,并减少了72.3%的修复时间。Distributed key-value storage is a key component in the distributed storage system,which replicates the key-value pairs to local engines in different storage servers and uses the consensus algorithm to keep replicas consistent.The log-structured merge tree based local key-value engine is the most popular storage algorithm that designed for general purpose.However,the original LSM-tree structure is not suitable for the specific workload of the upper-layer consensus logic.It usually causes performance loss in the write operations and extra spatial amplification during full-node repair.To solve this problem,a local engine named PheonixLSM is designed for distributed key-value storage.PheonixLSM boosts performance by eliminating the double-sync problem for write operations.It also reorganizes the SST file layout to eliminate extra write amplification during full-node repair.Experimental results showed that,compared with distributed key-value storage using default local engine,that PheonixLSM can achieve up to 90.7%write performance gain and reduce the extra write amplification from 65.6%to 6.4%,and the repair time also can be reduced by 72.3%.

关 键 词：分布式系统 键值存储 一致性协议 空间放大 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]