一种固态盘存储阵列的设计与实现

发布时间：2018-09-04 09:13

【摘要】：爆炸式增长的数据量对存储系统的性能和可靠性都提出了更高的要求,以磁性物质为介质的硬盘很难达到现有的对存储系统的要求。基于闪存的固态盘(solid state disk,SSD)是采用闪存芯片作为存储介质构建成的数据存储设备,具有抗震性好、能耗低、存取速度高等特点。随着固态盘容量的提高和价格的降低,固态盘相对于磁盘的优势也越来越明显。但是其容量的提升是以降低可靠性为代价的,随着闪存单元中多比特错误变得越来越常见,能够修复单比特错误的传统纠错码已经不再全面有效。此外,固态盘存在着随机写性能差和使用寿命有限等问题,单块固态盘难以满足人们对存储系统高可靠性、高容量、高性能的要求。正因为如此,基于闪存的存储应该利用那些应用于RAID配置的策略,但固态盘提供了许多不同于传统磁盘的功能,将RAID算法直接应用于固态盘又会遇到很多挑战。基于此,本文的主要工作有:1)对现有的几种典型的固态盘阵列技术Diff-RAID、I-CASH、FRA,及寿命感知可靠性方案进行了方法介绍,并对各个方法的性能从小写、随机写、寿命、可靠性等几个方面进行了分析和比较。2)针对中小型企业普遍存在的问题,参与设计了一种新型固态盘阵列RAID-CP(RAID with Co-located Parity Home),RAID-CP用多个RAID共享校验盘,将其组合成一个校验盘组(Parity Home),解决了负载比较忙的RAID校验信息访问成为瓶颈的问题,Parity Home用磁盘存储校验信息避免了数据频繁写入对固态盘的损伤,享受了校验信息带来的容错却没有损伤固态盘的寿命。进行数据恢复时,相比于其他RAID系统,RAID-CP优先恢复最热的数据,降低了用户响应延迟和数据恢复时间,提高了系统性能。3)为了验证RAID-CP的性能,利用Device Mapper框架在Linux内核中增加Parity Home设备驱动模块,同时修改内核块设备驱动MD模块,在版本为3.14.35的linux内核中实现了RAID-CP系统。4)测试RAID-CP和其他RAID系统的正常IO延迟、数据重建时间和在线恢复IO延迟。实验结果表明:RAID-CP的平均正常IO延迟比其他RAID系统的平均正常IO延迟低4%~8%。在数据恢复上,RAID-CP的数据重建时间比其他RAID系统低5.78%~17.48%,在线恢复IO延迟比其他RAID系统低9.84%~50.35,同时数据恢复带来校验信息的频繁读取操作并没有对RAID-CP上其他的卷造成负面影响。
[Abstract]:The explosive growth of data demands higher performance and reliability of the storage system. Hard disks based on magnetic materials can hardly meet the existing requirements of the storage system. With the increase of the capacity and the decrease of the price of the solid disk, the advantages of the solid disk relative to the disk become more and more obvious. However, the increase of the capacity is at the cost of reducing reliability. As the multi-bit errors in the flash memory unit become more and more common, it can repair the traditional correction of single-bit errors. Error codes are no longer fully valid. In addition, SSDs suffer from poor random write performance and limited service life. Single SSDs are difficult to meet the high reliability, high capacity, and high performance requirements of storage systems. Based on this, the main work of this paper is as follows: 1) The methods of Diff-RAID, I-CASH, FRA and life-aware reliability schemes for several typical solid-state disk array technologies are introduced, and the performance of each method is lowercase. Random write, life, reliability and other aspects are analyzed and compared. 2) Aiming at the common problems of small and medium-sized enterprises, a new type of solid-state disk array RAID-CP (RAID with Co-located Parity Home) is designed. RAID-CP uses multiple RAID to share the check disk, which is combined into a parity Home to solve the load comparison. Busy access to RAID checksum information becomes a bottleneck problem. Parity Home uses disk storage checksum information to avoid the damage caused by frequent data writing to the solid disk, and enjoys the fault tolerance brought by checksum information without damaging the life of the solid disk. In order to verify the performance of RAID-CP, Parity Home device driver module is added to Linux kernel using Device Mapper framework, and kernel block device driver MD module is modified. RAID-CP system is implemented in version 3.14.35 Linux kernel. 4) RAID-CP and other RAI systems are tested. The experimental results show that the average normal IO delay of RAID-CP is 4%~8% lower than that of other RAID systems. In data recovery, the data reconstruction time of RAID-CP is 5.78%~17.48% lower than that of other RAID systems, and the online recovery IO delay is 9.84%~50% lower than that of other RAID systems. 35. Meanwhile, frequent reads of check information brought about by data recovery do not negatively affect other volumes on RAID-CP.
【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP333

【相似文献】