失效盘优先的高速缓存替换方案的设计与实现

发布时间：2018-01-03 16:01

本文关键词：失效盘优先的高速缓存替换方案的设计与实现　出处：《华中科技大学》2013年硕士论文　论文类型：学位论文

【摘要】：高速缓冲存储器对于平衡上层计算组件和底层存储设备间的速度差异有着很重要的作用。有效的缓存管理方案不仅能够提升上层计算组件的计算速度，还能够降低访问底层存储设备的I/O次数。然而，现有的缓存替换算法仅考虑磁盘阵列在正常模式下的情况，而没有考虑如何提高磁盘阵列在失效模式下的性能，例如在基于条带的磁盘阵列存在失效盘的情况下，这些算法都不能使缓存发挥最优的性能。为了解决这一问题，提出了一种新型的缓存替换方案来提升由磁盘阵列和高速缓存构成的存储系统的可靠性和读写性能，并将其命名为VDF（Victim Disk First）。VDF的基本思想是：当存在失效盘的情况下，给予失效盘上的数据块更高的缓存优先级，进而减少对失效盘的I/O请求数。由于对失效盘的一次请求会被RAID算法层转化成为有效盘的多次请求，，因此减少对失效盘的I/O请求数，有利于降低失效模式下磁盘的总体访问次数。另外，一个磁盘阵列的带宽是有限的，通过VDF方案节省下来的带宽可用于重构过程或被其他应用占用，这样有利于提高磁盘阵列的可靠性和服务率。为了验证VDF的有效性，将VDF思想与LRU和LFU算法相结合，并分别命名为VDF-LRU和VDF-LFU。通过仿真实验可以发现VDF-LFU最多可以降低42.3%的磁盘I/O数，而VDF-LRU最多能够降低36.2%的磁盘I/O数。另外，将VDF算法运用到Linux内核中，使VDF与Linux内核中的页缓存替换算法相结合。通过实验可以看到，在重构模式下，VDF最多能够降低17.2%的重构时间，在降级模式下，VDF最多能够提升15.9%的系统吞吐率。
[Abstract]:Cache plays an important role in balancing the speed difference between the upper computing components and the underlying storage devices. An effective cache management scheme can not only improve the computing speed of the upper computing components. It can also reduce the number of I / O access to the underlying storage device. However, the existing cache replacement algorithms only consider the disk array in normal mode. However, no consideration is given to how to improve the performance of disk arrays in failure mode. For example, in the presence of disk diskettes based on strip arrays, none of these algorithms can make the cache perform optimally. In order to solve this problem, a new cache replacement scheme is proposed to improve the reliability and read and write performance of the storage system composed of disk array and cache. The basic idea of calling it VDF(Victim Disk First).VDF is to give a higher cache priority to the data block on the invalid disk when there is an invalid disk. And then reduce the number of I / O requests for the invalid disk. Because a request to the invalidation disk will be converted into a valid disk by the RAID algorithm layer, the number of I / O requests for the invalid disk will be reduced. In addition, the bandwidth of a disk array is limited, and the bandwidth saved by the VDF scheme can be used in the reconstruction process or occupied by other applications. This is helpful to improve the reliability and service rate of disk array. In order to verify the validity of VDF, the idea of VDF is combined with LRU and LFU algorithms. It is named VDF-LRU and VDF-LFU.Through simulation experiments, we can find that VDF-LFU can reduce the I / O number of disk by up to 42.3%. VDF-LRU can reduce the I / O number of disk by up to 36.2%. In addition, the VDF algorithm is applied to the Linux kernel. Combining VDF with page cache replacement algorithm in Linux kernel, we can see that in refactoring mode, the refactoring time can be reduced by 17.2% at most, and in downgrade mode. VDF can increase system throughput by up to 15.9%.
【学位授予单位】：华中科技大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP333

【共引文献】