Research On Performance Optimization And Test Technology Of SSD-based Storage Systems

Due to the inherent mechanical rotation limit of traditional Hard Disk Drivers(short for HDDs),the IO bandwidth and latency has became the main performance bottlenecks for large-scale storage systems.Flash-based Solid State Drivers(short for SSDs)have been widely used in personal computers and enterprise storage systems because of their high performance and low power consumption.However,since SSDs have different device characteristics from HDDs,the traditional storage software suitable to HHDs cannot fully exploit the performance advantages of SSDs.For example,straightfowardly applying Redundant Array of Inexpensive Disk(short for RAID)to SSDs will not only aggravate the adverse effects of Garbage Collection,but also decrease the performance of SSDs and shorten the life time.Traditional Key-Value(short for KV)database is directly deployed on the SSDs,which cannot achieve optimal performance by software and hardware co-optimization.Therefore,the research on performance optimization and evaluation technology of SSD-based storage systems can lay a better foundation for the next-generation large-scale storage systems.Firstly,due to the limitations of flash-based SSDs,flash-based SSDs have low write performance and limited erase counts,simply applying RAID algorithms to flash-based SSDs is challenging and promising.In this thesis,we propose a hybrid data layout scheme,called Hybrid-RAID,to improve the performance of SSD-based storage system.Hybrid-RAID divides the SSD-based Disk Arrays into two areas,RAID5 and RAIDI.Hybrid-RAID layouts data according to the different preferences of RAID5 and RAIDI for the size of load requests and the read/write intensity ofdata.Hybrid-RAID reduces SSDs overhead which caused by frequent update of verification information,and it also improves the performance of the SSD-based storage systems.The performance evaluation results show that Hybrid-RAID significantly reduces the average response time by an average of 41.5%and 38.8%,respectively,compared with the traditional RAID5 and RAID1.Secondly,we study the new Open Channel Solid State Drives,on the basis of fully understanding its basic structure and operating principle,we propose two evaluation schemes called comprehensive evaluation and applied evaluation,to test the Open Channel SSD-based storage platform.According to evaluation,we verify the effectiveness of hardware and software co-design,analyze the basic operation timing and the internal parallelism of Open Channel SSDs in detail,the evaluation results show that the parallelism between Channels is greater than the parallelism between Logical Unit Number.Finally,we compare two evaluation schemes,and summarize the evaluation technology for SSD,which provides reference data for the later development of Open Channel SSD-based storage systems.In summary,we deeply study the performance optimization and evaluation technology of SSD-based storage systems to fully exploit the characteristics of SSDs and improve the performance of SSD-based storage systems.