磨损均衡在提高SSD使用寿命中的应用与改进
发布时间：2018-08-19 12:09
【摘要】：固态硬盘是当下存储界很热的一个词,在存储界的同仁都清楚,未来的存储硬盘将由现在的机械硬盘转到固态硬盘上面。从最开始的20MB的容量到现在几TB的容量,固态硬盘正在发生着翻天覆地的变化.每一年都有新的产品的推出,未来PCIE固态硬盘的出现,读写数据的速度将会是一个几何级的增长.固态硬盘(Solid State Drive)是一种非易失性的存储设备,可以长时间在固态闪存介质中的存储数据。固态硬盘实际上没有像传统硬盘那样有很多固定的机械形式,例如没有机械硬盘那样的磁头和磁盘。换句话说,固态硬盘只是一组半导体存储阵列的组织的硬盘,存储介质使用的是集成电路一样的NAND性介质,而不是磁性介质。由于固态硬盘有更高的读写性能,所以导致开发和采用固态硬盘的方案达到了快速扩张的需求。另外固态硬盘在随机读写访问数据方面比传统硬盘的响应延时要小很多,这样固态硬盘对于大规模的读或者随机访问的工作量有着巨大的优势。这种低延时的特性就是因为固态硬盘可以直接有能力从闪存NAND中直接立刻的读取出数据来。但是如果读写擦除的数据总是坐落于NAND闪存介质中的某一特殊块中,这样这一特殊数据块将会比其他所有的NAND数据块很快磨损坏掉,这样将会过早的导致整个固态硬盘的使用寿命的结束。出于这个原因的考虑,固态硬盘的控制器,也是本篇论文重点地陈述观点,将会使用一种称之为磨损均衡的技术,将写和擦除尽可能的将所写的数据均匀的分布在所有的SSDNAND存储介质的每个数据块中。对于这个完美的场景的实现,将会使每一个NAND存储数据的块达到最大的生命周期,然后会在某一时刻同时坏掉,以达到最大的固态硬盘使用周期。但是不幸的是,这一过程存在着很多不可定得因素,包括一些经常使用读写的数据我们称之谓热数据,和一些很少被读写到的数据称之为冷数据。冷数据和热数据将导致存储在的NAND介质数据块的读写擦除计数将会严重不平衡,这就促使找到一个关键的算法解决或者缓和这种问题。磨损均衡算法正是在这种情况下面提出来优化改进NAND存储数据块的读写擦除计数生成的.但是磨损均衡算法也会相应的带来一些不必要的写放大损耗,于是本文创新的双池磨损均衡算法可以使优化性能与损耗性能达到一个平衡点,使固态硬盘相对达到一个最大的使用寿命,而性能相对没有明显的降低。
[Abstract]:Solid state drives are a hot term in storage, and everyone in the storage world knows that future storage drives will be switched from mechanical drives to solid state drives. From the beginning of the 20MB capacity to the current several terabytes of capacity, solid-state drives are changing dramatically. Every year there are new products coming out, and the future of PCIE solid-state drives, the speed of reading and writing data will be a geometric growth. Solid-state hard disk (Solid State Drive) is a non-volatile storage device that can store data in solid-state flash media for a long time. Solid state drives do not actually have the same fixed mechanical forms as traditional hard drives, such as magnetic heads and disks without mechanical hard drives. In other words, the solid-state disk is just a set of semiconductor memory arrays of the organization of the hard disk, storage media using integrated circuit like the NAND medium, rather than magnetic media. Due to the higher read and write performance of solid state hard disk, the development and adoption of solid state hard disk scheme meet the requirement of rapid expansion. In addition the random read and write access data in the solid-state disk is much smaller than the traditional hard disk response delay so the solid-state disk for large-scale read or random access workload has a huge advantage. This low latency feature is due to the ability of solid-state drives to read data directly from flash NAND. But if the read and write erasure data is always in a particular block in the NAND flash media, this particular block will wear out faster than any other NAND data block. This will prematurely lead to the end of the life of the entire solid-state disk. For this reason, the solid-state hard disk controller, which is also the focus of this paper, will use a technique called wear equalization. Write and erase the data evenly distributed as much as possible in each block of all SSDNAND storage media. The implementation of this perfect scenario would allow each block of NAND to store data for the maximum life cycle, and then break down at some point at the same time to achieve the maximum lifetime of the solid-state disk. Unfortunately, there are many uncertainties in this process, including some data that is often read and write, which we call hot data, and some data that is rarely read or written, called cold data. Cold data and hot data will lead to a serious imbalance in the read and write erasure count of stored NAND data blocks, which leads to finding a key algorithm to solve or alleviate this problem. It is in this case that the wear equalization algorithm is proposed to optimize the generation of read and write erasure counting for NAND storage data blocks. However, the wear equalization algorithm will also bring some unnecessary write amplification losses, so the novel two-cell wear equalization algorithm can achieve a balance between the optimal performance and the loss performance. Make solid-state hard disk relatively reach a maximum service life, but the performance is relatively not significantly reduced.
【学位授予单位】：华东理工大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP333.35
本文编号：2191610
[Abstract]:Solid state drives are a hot term in storage, and everyone in the storage world knows that future storage drives will be switched from mechanical drives to solid state drives. From the beginning of the 20MB capacity to the current several terabytes of capacity, solid-state drives are changing dramatically. Every year there are new products coming out, and the future of PCIE solid-state drives, the speed of reading and writing data will be a geometric growth. Solid-state hard disk (Solid State Drive) is a non-volatile storage device that can store data in solid-state flash media for a long time. Solid state drives do not actually have the same fixed mechanical forms as traditional hard drives, such as magnetic heads and disks without mechanical hard drives. In other words, the solid-state disk is just a set of semiconductor memory arrays of the organization of the hard disk, storage media using integrated circuit like the NAND medium, rather than magnetic media. Due to the higher read and write performance of solid state hard disk, the development and adoption of solid state hard disk scheme meet the requirement of rapid expansion. In addition the random read and write access data in the solid-state disk is much smaller than the traditional hard disk response delay so the solid-state disk for large-scale read or random access workload has a huge advantage. This low latency feature is due to the ability of solid-state drives to read data directly from flash NAND. But if the read and write erasure data is always in a particular block in the NAND flash media, this particular block will wear out faster than any other NAND data block. This will prematurely lead to the end of the life of the entire solid-state disk. For this reason, the solid-state hard disk controller, which is also the focus of this paper, will use a technique called wear equalization. Write and erase the data evenly distributed as much as possible in each block of all SSDNAND storage media. The implementation of this perfect scenario would allow each block of NAND to store data for the maximum life cycle, and then break down at some point at the same time to achieve the maximum lifetime of the solid-state disk. Unfortunately, there are many uncertainties in this process, including some data that is often read and write, which we call hot data, and some data that is rarely read or written, called cold data. Cold data and hot data will lead to a serious imbalance in the read and write erasure count of stored NAND data blocks, which leads to finding a key algorithm to solve or alleviate this problem. It is in this case that the wear equalization algorithm is proposed to optimize the generation of read and write erasure counting for NAND storage data blocks. However, the wear equalization algorithm will also bring some unnecessary write amplification losses, so the novel two-cell wear equalization algorithm can achieve a balance between the optimal performance and the loss performance. Make solid-state hard disk relatively reach a maximum service life, but the performance is relatively not significantly reduced.
【学位授予单位】：华东理工大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP333.35
【参考文献】
相关期刊论文 前4条
1 孙丰;张福新;;YAFFS文件系统的研究与改进[J];计算机工程;2008年05期
2 李桂良,刘发贵;JFFS2文件系统的关键技术及其在嵌入式系统的应用[J];计算机应用;2003年07期
3 张骏;樊晓桠;刘松鹤;;一种Flash存储器静态负载平衡策略[J];计算机应用;2006年05期
4 韦鹏;岳丽华;;基于预测数据失效期的flash存储分配回收算法[J];中国科学技术大学学报;2007年12期
，本文编号：2191610
本文链接：https://www.wllwen.com/kejilunwen/jisuanjikexuelunwen/2191610.html
