基于Flashcache的混合存储方案设计与实现

发布时间：2018-01-05 09:23

本文关键词：基于Flashcache的混合存储方案设计与实现　出处：《华中科技大学》2013年硕士论文　论文类型：学位论文

【摘要】：计算机技术在飞速地发展，而磁盘的存取速度却提升缓慢，成为了影响计算机系统性能的主要因素。新型的存储介质固态盘(Solid State Driver, SSD)的迅猛发展给存储系统注入了新的活力，SSD读写速度快，特别是随机I/O能力很优秀，只是容量较磁盘小很多，而且价格昂贵。在磁盘向SSD过渡期间，混合存储技术能发挥磁盘和SSD各自的优点，平衡成本和性能需求。开源项目flashcache是混合存储技术的一种实现，它使用SSD作为磁盘和内存之间的缓存，以达到提升性能的目的。Flashcache是一个完整可用的混合存储方案，但其缓存调度方式过于简单，不能充分利用SSD的高性能，可以作进一步改进。在分析了flashcache具体实现的基础上，结合SSD的特性，从性能和数据可靠性两方面考虑，提出了对flashcache的改进方案。使用基树组织管理缓存，对SSD上的缓存块进行全局的调配，提高SSD的利用率和缓存命中率，缩短检索时间。利用双LRU链表的方式识别热点缓存块，当缓存块要被替换时同时回收地址相邻的非热点缓存块，保留了最有缓存价值的缓存块，同时使得对SSD的写入接近于大块写，更好的发挥SSD的性能和维护其写入寿命。引入NVRAM以日志的方式实时保存缓存块元数据改变，，一方面可以保证在系统故障恢复后的数据有效性，另一方面避免了元数据实时写入到SSD给混合存储系统带来的性能影响。在flashcache的框架基础上，实现了对flashcache的改进方案，并进行了对比测试。测试结果表明，相对于纯磁盘方案，混合存储方案可以有效的提升数据访问速度、减小访问响应时间，对flashcache进行改进后，性能得到进一步提升。
[Abstract]:Computer technology is developing rapidly, while disk access speeds up slowly. It has become the main factor that affects the performance of computer system. The solid State Driver, a new storage medium, is introduced in this paper. The rapid development of SSDs has injected new energy into the storage system. The speed of reading and writing of SSD is fast, especially the random I / O is excellent, but the capacity is much smaller than the disk. And expensive. During the transition from disk to SSD, hybrid storage technology can take advantage of both disks and SSD. Balancing cost and performance requirements. The open source project flashcache is an implementation of hybrid storage technology that uses SSD as a cache between disk and memory. To achieve the purpose of improving performance. Flashcache is a completely available hybrid storage scheme, but its cache scheduling is too simple to take full advantage of the high performance of SSD. Further improvements can be made. Based on the analysis of the implementation of flashcache and the characteristics of SSD, the performance and data reliability are considered. This paper proposes an improved scheme for flashcache, which uses the base tree to organize and manage the cache, and allocates the cache blocks on the SSD globally, so as to improve the utilization of SSD and the hit ratio of the cache. The retrieval time is shortened. The hot cache block is identified by using double LRU linked list. When the cache block is to be replaced, the non-hot cache block with adjacent address is reclaimed at the same time, and the most valuable cache block is retained. At the same time, the writing of SSD is close to that of bulk writing, so that the performance of SSD and the maintenance of its write life can be better brought into play. NVRAM is introduced to save the cache block metadata changes in real time by the way of log. On the one hand, it can ensure the validity of the data after the system fault recovery, on the other hand, it can avoid the impact of real-time metadata written to SSD on the performance of hybrid storage system. Based on the framework of flashcache, the improved scheme of flashcache is implemented, and the test results show that it is relative to the pure disk scheme. The hybrid storage scheme can effectively improve the speed of data access, reduce the access response time, and improve the performance of flashcache.
【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】