大规模固态存储阵列系统性能优化方法的研究与实现

发布时间：2018-06-06 10:18

  本文选题：存储系统 + SSD　； 参考：《国防科学技术大学》2013年硕士论文

【摘要】：人类已经进入了信息时代,每天会产生大量的数据需要处理和存储,尤其是那些数据密集型应用,对系统的性能提出了更高的要求。在整个计算机系统中,存储系统由于其性能提高跟不上CPU的速度成为一个瓶颈问题,这也使得存储成为目前计算机领域的研究热点。其中,使用RAID机制构建存储阵列系统是大规模、高性能、高可靠存储系统中的一个重要方法。闪存作为一种快速发展的新型存储介质,由于其非易失、低延迟、低功耗等优点,已经得到了广泛的应用。研究基于闪存的固态存储阵列系统成为一个新的热点。本研究主要面向大规模固态存储阵列系统,研究其性能优化方法,包括并行性与可扩展性两个方面。主要工作分为以下三个部分:首先,研究了固态存储并行技术,分析比较了传统磁盘阵列的多种可扩展算法。研究结果表明,现有的并行技术可以加以充分利用来提高系统的并行性,也可以对传统的可扩展算法进行改进和优化使其适用于固态存储阵列系统。其次,针对当前固态存储系统并行性挖掘度不高的问题,本文设计了一种多级并行体系结构,并研究了固态盘内部各层级并行性的实现机制,然后提出一种在层次化RAID基础上进一步优化并行性的新方法。实验结果表明,该方法考虑不同层级的并行性优先级,可以更好的实现并行效果,并行性最多提高17%,提高整个系统的性能。最后,在可扩展性研究方面,主要针对RAID5等校验阵列的可扩展性需求,提出一种基于校验信息动态重分布的扩展方法,本文结合实例分析、数学论证和实验测试等手段表明,新方法可以减少大量数据迁移,减少了多达82.75%的I/O操作,迁移时间最多减少71.7%,有效加速扩展过程,并且提高了扩展后的系统性能。
[Abstract]:Human beings have entered the information age, and a large number of data need to be processed and stored every day, especially those data-intensive applications, which put forward higher requirements for the performance of the system. In the whole computer system, the storage system can not keep up with the speed of CPU because of its performance improvement becomes a bottleneck problem, which makes storage become the research hotspot in the field of computer. It is an important method in large scale, high performance and high reliability storage system to construct memory array system with RAID mechanism. Flash memory, as a rapidly developing new storage medium, has been widely used because of its advantages of non-volatile, low delay and low power consumption. The research of solid state memory array system based on flash memory has become a new hotspot. This paper focuses on large scale solid state memory array systems and studies its performance optimization methods, including parallelism and scalability. The main work is divided into the following three parts: firstly, the parallel technology of solid state storage is studied, and several scalable algorithms of traditional disk array are analyzed and compared. The results show that the existing parallel techniques can be fully utilized to improve the parallelism of the system, and the traditional scalable algorithms can be improved and optimized to be suitable for solid-state memory array systems. Secondly, aiming at the low degree of parallelism mining in solid-state storage system, this paper designs a multi-level parallel architecture, and studies the implementation mechanism of parallelism in different levels of solid-state disk. Then a new method to optimize parallelism based on hierarchical RAID is proposed. The experimental results show that the parallel effect can be improved by considering the parallelism priority of different levels, and the parallelism can be improved by up to 17 percent and the performance of the whole system can be improved. Finally, in the aspect of scalability research, aiming at the scalability requirement of RAID5 and other checkout arrays, this paper proposes an extension method based on dynamic redistribution of check information. The new method can reduce the large amount of data migration, reduce the I / O operation by 82.75%, reduce the migration time by 71.7%, accelerate the expansion process effectively, and improve the performance of the extended system.
【学位授予单位】：国防科学技术大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP333
，

本文编号：1986181