Research And Implementation Of Approach To Optimize The Performance For Large-scale Solid-state Storage Array System

Mankind has entered the information age, there are large amounts of data produced every day to be processed and stored, especially those data-intensive applications, which puts forward higher requirements for the performance of the system. In the entire computer system, the storage system becomes a bottleneck problem, because the raise of its performance can’t keep up with the CPU speed, which also makes storage become a research hotspot of computer industry at the present time, in which the storage array using RAID mechanism as an important way to build large-scale, high-performance, high-reliability system. Flash memory, as a fast development of new storage media, has been widely used, because of its advantages such as nonvolatile, low latency and low power consumption. It’s a new topic to study the Flash-based solid-state storage array systems.This study focuses on large-scale solid-state storage array systems, we study its performance optimization methods, including parallelism and scalability aspects. The main work is divided into three parts:First of all, we study the parallel technology of solid-state storage, analyze and compare a variety of traditional disk array scalable algorithms. Existing parallel technology can be fully utilized to improve the system parallelism. Traditional scalable algorithms can also be improved and optimized to make it suitable for solid-state storage array systems.Then, in the study of parallel, we design a multi-level parallel system architecture and study the internal SSD parallelism at all levels of implementation mechanisms. We also propose a new method to optimize the parallelism of system based on level-RAID. The validation experiments show that this method takes different levels of parallelism priority into account, we can better achieve parallel effects, increase the parallelism by 17% and improve overall system performance.Finally, in the study of scalability, mainly for such parity RAID5 array scalability requirements, we propose a parity-based dynamic redistribution approach to accelerate RAID5 scaling. Combining with case studies, mathematical proof and experimental testing, the new approach can reduce the data migration largely, reduce the I/O number by 82.75%, shorten the scaling time by 71.7%, speedup the scaling process effectively and improve the system performance after scaling.