提高MLC NAND Flash存储系统可靠性的方法研究

发布时间：2018-03-23 01:28

本文选题：MLC　切入点：NAND　出处：《哈尔滨工业大学》2016年博士论文　论文类型：学位论文

【摘要】：NAND flash作为一种非易失性存储器件,相对于磁盘存储器具有功耗较低、访问速度更快以及良好的抗震特性等优点,因此在信息存储领域尤其是测试信息存储领域取得了广泛应用。其中,MLC(Multi Level Cell)NAND flash由于存储密度为SLC(Single Level Cell)NAND flash数倍,成本较低,已经成为主流的NAND flash存储介质。但是,随着MLC NAND flash存储容量及存储密度增加,基于MLC NAND flash存储系统可靠性问题也日益突出。目前,MLC NAND flash存储系统可靠性问题主要集中在以下三个方面:错误率升高、使用寿命降低及发生意外掉电时数据丢失问题。本论文首先从MLC NAND flash结构及工作原理角度分析了错误率升高、使用寿命降低及对意外掉电敏感问题的内在机理,然后针对这三个可靠性问题,提出了提高存储系统可靠性方法,主要工作如下:首先研究了MLC NAND flash存储系统错误率升高的问题。本文通过分析文件系统不同请求的重要程度,针对传统冗余备份方法会给存储系统带来大量存储空间开销的缺陷,提出了元数据冗余方法,通过仅对系统元数据进行备份降低元数据错误率;为了降低冗余备份方法对系统I/O性能造成的开销,利用MLC NAND flash错误率非对称分布的特性,通过将系统关键数据合理分布于物理存储空间,最终达到降低存储系统错误率,同时不牺牲存储系统I/O性能的目的;设计并实现了实际的硬件实验平台,本文所有实现均基于此硬件实验平台。其次研究了延长MLC NAND flash存储系统使用寿命的方法。NAND flash存储系统的使用寿命与垃圾回收次数直接相关,本文通过对MLC NAND flash存储系统垃圾回收操作及并行访问方法进行分析,针对基于MLC NAND flash的固态盘,将并行访问方法与垃圾回收方法结合,提出了一种垃圾回收相关的并行访问方法。垃圾回收相关的并行访问方法将顺序访问请求队列的请求顺序根据存储系统并行程度重新组织,以实现与串行访问方法相近的垃圾回收开销同时不牺牲存储系统I/O性能,有效地降低了存储系统垃圾回收次数,从而延长了使用寿命。最后研究了MLC NAND flash存储系统意外掉电时数据丢失问题。本文对采用电容作为临时电源的固态盘建立意外掉电模型,分析了固态盘各部分耗能情况,针对以电容作为临时电源的MLC NAND flash存储系统,提出了贪婪备份方法。贪婪备份将电容放电过程与器件特性进行结合,通过动态通道选择策略和快速安全编程策略,避免了存储系统意外掉电时出现的低页破坏现象,同时有效提高了备份效率。
[Abstract]:As a non-volatile memory device, NAND flash has the advantages of lower power consumption, faster access speed and better seismic performance than disk memory. Therefore, it has been widely used in the field of information storage, especially in the field of test information storage, in which MLC / Multi Level Cell)NAND flash has become the mainstream storage medium of NAND flash because of its low cost and several times the storage density of SLC(Single Level Cell)NAND flash. With the increase of storage capacity and density of MLC NAND flash, the reliability problem of storage system based on MLC NAND flash is becoming more and more prominent. At present, the reliability problem of NAND flash storage system based on MLC NAND is mainly focused on the following three aspects: the increase of error rate. In this paper, the inherent mechanism of the increase of error rate, the decrease of service life and the sensitivity to accidental power loss are analyzed from the point of view of the structure and working principle of MLC NAND flash. Then, aiming at these three reliability problems, a method to improve the reliability of storage system is proposed. The main work is as follows: firstly, the problem of increasing the error rate of MLC NAND flash storage system is studied. In view of the defect that the traditional redundant backup method will bring a large amount of storage space overhead to the storage system, a metadata redundancy method is proposed, which can reduce the metadata error rate by only backing up the system metadata. In order to reduce the overhead caused by redundant backup method to the I / O performance of the system, by using the asymmetric distribution of MLC NAND flash error rate, the key data of the system can be reasonably distributed in the physical storage space to reduce the memory system error rate. At the same time, without sacrificing the performance of the storage system I / O, the hardware experimental platform is designed and implemented. All the implementations of this paper are based on this hardware experimental platform. Secondly, the method of prolonging the service life of MLC NAND flash storage system is studied. The service life of NAND flash storage system is directly related to the number of garbage collection. By analyzing the garbage collection operation and parallel access method of MLC NAND flash storage system, this paper combines the parallel access method with the garbage collection method for the solid-state disk based on MLC NAND flash. This paper proposes a parallel access method related to garbage collection, which reorganizes the request order of sequential access request queue according to the degree of parallelism of storage system. In order to realize the garbage collection cost similar to the serial access method without sacrificing the I / O performance of the storage system, it can effectively reduce the garbage collection times of the storage system. Finally, the problem of data loss when the MLC NAND flash storage system is accidentally power down is studied. In this paper, an accidental power loss model is established for the solid-state disk with capacitance as the temporary power supply, and the energy consumption of each part of the solid-state disk is analyzed. In this paper, a greedy backup method is proposed for MLC NAND flash storage system with capacitors as temporary power supply. Greedy backup combines capacitive discharge process with device characteristics, and adopts dynamic channel selection strategy and fast security programming strategy. It can avoid the low page damage when the storage system power down unexpectedly, and improve the backup efficiency effectively at the same time.
【学位授予单位】：哈尔滨工业大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TP333

【相似文献】