Research On Performance Optimization For Solid State Drive-based Storage Systems

With the rapid increase of the information, the demand for the capacity, performance and reliabilityof storage systemsbecomes more and more high. It is an urgent problem that how to optimize the existing system to provide a storage system withlarger storage capacity, higher reliability and higher performance..In general, the existing storage systems use the hard disk drives (HDDs) as the main medium to store data. The inherent mechanical properties of the HDDscannotradicallysolve the problems of the low performance and high power consumption. Thus, itis necessary to use other storage devices such as solid-state drives (SSDs) to replace or assist the HDDs. In this paper, we study the performance optimization techniques of SSD-based storage systems,including optimizingboth the single SSD and SSD-based disk arrays.The response times are linear with the request sizes for flash-based SSDs with the same request type. Moreover, the read performance and write performance of flash-based SSDs are asymmetric. Based on these characteristics, we propose a Size-based I/O Scheduler (short for SIOS) for flash-based SSDs to improve the I/O performance of SSD-based storage systems from the viewpoint of average response time. SIOS utilizes the asymmetric read and write performance characteristics of flash-based SSD and giveshigher priority to the read requests. Moreover, by first processing the small requests in the I/O waiting queue, the average waiting times of the requests are reduced significantly. The evaluation results show that compared with the state-of-the-arts, SIOS reduces the average response times significantly. Consequently, the I/O performance of the SSD-based storage systems is improved.A single flash-based Solid State Drive (SSD) can notsatisfy the performance, reliability and capacity requirementsof storage systems. Applying the RAID algorithm without any optimizationsto SSD-baseddisk arrays will cause the reliability and performance problemsdue to the parity update operations. In this paper, we proposea Log Disk Mirroring scheme (short for LDM) to improve the performanceand reliability of SSD-based storage system. LDM isa hybrid disk array architecture that consists ofseveral SSDsand two HDDs. In the LDM array, the two HDDs aremirrored as a write buffer that temporally absorbs the smallwrite requests. The small and random write data iswritten on the mirroringbuffer by using the logging technique that sequentially appends new data. The bufferedsmall write data is merged and destaged to the SSD-based disk array duringthesystem idle period. The reliability analysis shows that the reliabilityof the LDM array, in terms of MTTDL (Mean TimeTo Data Loss), is better than that of HDD-based disk arrays and SSD-based disk arrays. Our prototype implementation of the LDM arrayand the performance evaluations show that the LDM array significantlyoutperforms the HDD-based disk arrays and SSD-based disk arrays.