Research On Performance Optimization For SSD-based Storage Systems

With rapidly increasing of data,computer storage systems encounter serious performance challenge.As new storage media,Solid State Drives(short for SSDs),have been widely used in enterprise storage systems and data centers.However,simply applying traditional storage software to SSDs will bring some new problems,such as the performance variability issue caused by Garbage Collection(short for GC)operations on Redundant Array of Independent SSDs(short for RAIS)and the low storage efficiency issue caused by the dual-blind design in traditional key-value stores(short for KV stores)which are deployed on SSDs.From the point of view of black-box and white-box,this paper conduct researches on RAIS and Open-Channel SSDs based K-V store.First,we study the optimal chunk size of RAIS.The theoretical analysis and performance evaluations on real devices showed that the chunk size of RAIS have a great impact on the read and write performance of RAIS.Based on the analysis of the optimal chunk size and the study on the read/write intensity of real workload,we propose a Multi-Chunk RAIS(short for MC-RAIS),which layouts data according to the workload characteristics and takes advantages of multiple chunk sizes,to improve the performance of the SSD-based storage systems.Performance evaluation result shows that MC-RAIS outperforms the existing fix-chunk-size RAIS in the I/O performance measure by 10%-20%.Second,we conduct study on SSD-based KV store from the point of view of white-box.Open-Channel SSDs(short for OCSSDs)are new kind of SSDs which can solve the internal problems of SSDs,such as garbage collection and wear-leveling,by making SSDs be white boxes to the upper software.In OCSSDs,the NAND Flash block management,data layout,wear leveling,garbage collection and other mechanisms can be independently customed by the upper application software.In LightNVM based Open-Channel SSDs,we propose an Open-Channel based File System middleware,short for OC-FS,to provide some file interface functions and enable application characteristics based customization.By connecting the upper-layer KV database and the lower-layer storage device together,OC-FS reduces the redundant I/O stack and integrates the GC operations of both the upper-layer and lower-layer software,thus improving the storage efficiency and performance.Performance evaluation result shows that OC-FS outperforms the existing traditional KV store system by approximately 50%.From the perspective of black-box and white-box,we deeply study two key technologies of SSDs to improve the performance of SSD-based storage systems.Extensive performance evaluations demonstrate the effectiveness and usability of our proposed optimization technologies,thus providing thinking and inspiration for industries to solve similar problems.