基于SSD-SMR混合存储的LSM树键值存储系统的性能优化

Performance Optimization of LSM Tree Key-value Storage System Based on SSD-SMR Hybrid Storage

大数据对存储系统的可扩展性、性能和成本等方面提出了更高的要求。瓦记录(Shingled Magnetic Recording,SMR)硬盘由于存储密度高、价格便宜,正逐步被广泛应用于大数据存储系统。但是,SMR硬盘的随机写性能较差,与快速的基于闪存的固态硬盘(Solid State Drive,SSD)一起构成混合存储时可以显著提升性能。同时,基于写优化的日志结构合并(Log-Structured Merge,LSM)树的键值存储已被广泛应用于许多NoSQL系统,如BigTable,Cassandra和HBase等。因此,如何基于新型的SSD-SMR混合存储构建出高性能的LSM树键值存储系统是一个具有很大研究价值的问题。首先建立基于SSD-SMR混合存储的LSM树键值系统的性能模型,然后针对SSD和SMR的硬件特征以及LSM树键值存储的软件特点,设计了一套面向SSD-SMR混合存储进行性能优化的LSM树键值存储系统,并基于LevelDB实现了该系统。在仅仅使用0.4%~2%空间的SSD的情况下,所提方法可以使SSD-SMR混合存储方案比普通磁盘方案的随机写性能提高20%,随机读性能提高5倍。

Because of the higher requirements on the scalability,performance,and cost for storage systems proposed by big data,Shingled Magnetic Recording（SMR）disks are widely used in big data storage systems due to the high storage density and low cost.However,since the random write performance of SMR disks are usually weak,the hybrid storage consisted of both SMR disks and the fast Flash-based Solid State Drives（SSDs）can promote the performance significantly.Meanwhile,the write-optimized Log-Structured Merge（LSM）Tree-based key-value storage system have been widely used in many NoSQL systems,such as BigTable,Cassandra,HBase,etc.Therefore,how to construct a fast LSM tree key-value storage system based on SSD-SMR hybrid storage is a research problem with great practical significance.This paper first modeled the performance model of LSM tree key-value storage system based on SSD-SMR hybrid storage,and then designed a performance-optimized LSM tree key-value storage system and implemented it based on LevelDB.The evaluation results indicate that the system based on SSD-SMR hybrid storage improves the random-write performance by 20% and improves random-read performance by 6 times coupled with only a very small SSD（i.e.,0.4%~2% of disk capacity）compared with the HDD-based solution.