Research On In-Memory Database Persistence Optimization Based On Open Channel Solid State Drive

The in-memory database,which makes data resident in main memory,is widely used in the field of the real time database for its excellent performance.And Redis is an open source,network-enabled,high-performance and memory-persistent Key-Value(K-V)inmemory database.The present usage of Redis is often required to make RDB(Redis DataBase)or AOF(Append-Only-File,)to ensure that data can be stored in external storage devices reliably.However,the persistence process will adversely affect the performance of the database.Among them,the I/O performance and refresh rate of AOF will directly affect Redis itself's real-time response ability.In view of the contradiction between the performance of Redis database and the cost of the data persistence,Redis' s efficient and reliable persistence project(OC-Redis)based on the open-channel SSD(OC-SSD)is proposed.Applications directly read and write the OC-SSD,bypassing the file system and block layer to shorten the I/O path for database persistence and loading,and improve I/O performance.At the same time,the data address mapping is designed,and the space management and garbage collection of OC-SSD are realized.In order to balance the different hardware wear rate of RDB and AOF,a blocklevel wear-leveling method is implemented.Aiming at the problem that the high-frequency of data write back of AOF affects the performance of database,the strategy that the data of AOF is stored in the form of linked list is proposed to incorporate metadata and data storage.This strategy reduces one write operation by one per persistence and can reduce write operations by nearly half overall.Besides,small-write strategy is optimized and combined with the short path I/O project of OC-SSD to improve the performance of small-write.The improvement of small-write can reduces the ratio of time spent per persistence to time interval between AOF and decreases lost rate of data.Experimental tests show that the OR-Redis based on open channel SSD performs better than the traditional Redis in persistence.Also,compared with Redis using the block layer I/O method based on Open-channel SSD,the OC-Redis based on open channel SSD can improve the performance in persistence and reduce the probability of data loss.RDB performance in persistence improved by 40%~60% in different situations.Each small-write operation time delay of AOF is reduced to 8.8% of the traditional Redis.