固态盘闪存转换层优化技术研究
发布时间：2018-10-18 08:17
【摘要】：基于NAND Flash的固态盘凭借其低能耗、低故障率、高性能等优势,正逐步应用到企业级服务器和高性能计算领域。为保持主机对固态盘的访问接口与磁盘兼容,并有效解决NAND Flash的重写前擦除和擦除次数有限等固有特性引起的各类问题,提高固态盘的整体性能,固态盘在文件系统和NAND Flash之间引入闪存转换层,以处理地址映射、垃圾回收和磨损均衡等问题。其中,地址映射将请求的逻辑地址映射为固态盘内物理地址,根据映射粒度的不同可分为页映射、块映射和混合映射。页映射方式存在映射表过大的问题,限制了其在固态盘中的实际应用。当前主流的DFTL算法在页映射的基础上选择性地缓存最近常用的页映射关系,一定程度上缓解了页映射表过大的问题,但因映射信息未命中缓存而产生的额外开销严重影响了固态盘的整体性能。针对上述问题,本文提出了一种有效利用负载时空局部性的地址映射方法ECAM,ECAM将多个物理页号和逻辑页号均连续的页映射关系用一条带有SIZE域的地址映射关系表示,并在对页映射信息的缓存管理策略中赋予SIZE值较大的地址映射信息更高的优先级。ECAM根据负载的局部性原理对缓存内地址映射信息进行管理,在缓存空间一定的情况下缓存更多有用的地址映射信息,提高缓存命中率,减少缓存失效产生的额外读写操作,提高固态盘整体性能。同时,针对NAND Flash读写性能与负载特性相关的特点,本文在ECAM基础上提出了一种基于负载预测的自适应闪存转换层算法WAPFTL。WAPFTL通过对负载特征的预测动态调整缓存管理策略,避免缓存随机写请求的地址映射信息时造成的空间浪费,在各类负载下提高有限空间缓存的命中率,进一步提高固态盘整体性能。测试结果表明本文提出的ECAM和WAPFTL可以有效提高地址映射信息的缓存命中率,减少缓存未命中而产生的额外开销,有效提高固态盘的整体性能。
[Abstract]:With the advantages of low energy consumption, low failure rate and high performance, solid state disk based on NAND Flash is gradually applied to enterprise server and high performance computing field. In order to keep the host's access interface to the solid disk compatible with the disk, and to effectively solve all kinds of problems caused by the inherent characteristics of NAND Flash, such as erasing before rewriting and limited erasure times, the overall performance of the solid state disk can be improved. Flash transfer layer is introduced between file system and NAND Flash to deal with address mapping, garbage collection and wear balance. Address mapping maps the requested logical addresses to solid-state in-disk physical addresses, which can be divided into page mapping, block mapping and hybrid mapping according to the mapping granularity. The problem of large mapping table in page mapping mode limits its practical application in solid-state disk. The current mainstream DFTL algorithm selectively caches the recently commonly used page mapping relationships on the basis of page mapping, which to some extent alleviates the problem of excessive page mapping. However, the additional overhead caused by the missed cache of mapping information seriously affects the overall performance of the solid-state disk. In order to solve the above problems, this paper presents an effective address mapping method based on the spatio-temporal localization of load, ECAM,ECAM, which represents multiple physical page numbers and logical page numbers as an address mapping relation with SIZE domain. In the cache management strategy of page mapping information, the address mapping information with larger SIZE value is given higher priority. ECAM manages the address mapping information in the cache according to the local principle of load. In the case of fixed cache space, more useful address mapping information is cached, the cache hit rate is improved, the additional read and write operations caused by cache failure are reduced, and the overall performance of solid state disk is improved. At the same time, according to the characteristics of NAND Flash read and write performance and load characteristics, this paper proposes an adaptive flash conversion layer algorithm based on load prediction based on ECAM, WAPFTL.WAPFTL dynamically adjusts buffer management strategy by predicting load characteristics. It avoids the space waste caused by caching the address mapping information of random write requests, improves the hit rate of limited space cache under various loads, and further improves the overall performance of solid-state disk. The test results show that the proposed ECAM and WAPFTL can effectively improve the cache hit rate of address mapping information, reduce the extra overhead caused by cache miss, and effectively improve the overall performance of solid-state disk.
【学位授予单位】：国防科学技术大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP333
，
本文编号：2278538
[Abstract]:With the advantages of low energy consumption, low failure rate and high performance, solid state disk based on NAND Flash is gradually applied to enterprise server and high performance computing field. In order to keep the host's access interface to the solid disk compatible with the disk, and to effectively solve all kinds of problems caused by the inherent characteristics of NAND Flash, such as erasing before rewriting and limited erasure times, the overall performance of the solid state disk can be improved. Flash transfer layer is introduced between file system and NAND Flash to deal with address mapping, garbage collection and wear balance. Address mapping maps the requested logical addresses to solid-state in-disk physical addresses, which can be divided into page mapping, block mapping and hybrid mapping according to the mapping granularity. The problem of large mapping table in page mapping mode limits its practical application in solid-state disk. The current mainstream DFTL algorithm selectively caches the recently commonly used page mapping relationships on the basis of page mapping, which to some extent alleviates the problem of excessive page mapping. However, the additional overhead caused by the missed cache of mapping information seriously affects the overall performance of the solid-state disk. In order to solve the above problems, this paper presents an effective address mapping method based on the spatio-temporal localization of load, ECAM,ECAM, which represents multiple physical page numbers and logical page numbers as an address mapping relation with SIZE domain. In the cache management strategy of page mapping information, the address mapping information with larger SIZE value is given higher priority. ECAM manages the address mapping information in the cache according to the local principle of load. In the case of fixed cache space, more useful address mapping information is cached, the cache hit rate is improved, the additional read and write operations caused by cache failure are reduced, and the overall performance of solid state disk is improved. At the same time, according to the characteristics of NAND Flash read and write performance and load characteristics, this paper proposes an adaptive flash conversion layer algorithm based on load prediction based on ECAM, WAPFTL.WAPFTL dynamically adjusts buffer management strategy by predicting load characteristics. It avoids the space waste caused by caching the address mapping information of random write requests, improves the hit rate of limited space cache under various loads, and further improves the overall performance of solid-state disk. The test results show that the proposed ECAM and WAPFTL can effectively improve the cache hit rate of address mapping information, reduce the extra overhead caused by cache miss, and effectively improve the overall performance of solid-state disk.
【学位授予单位】：国防科学技术大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP333
，
本文编号：2278538
本文链接：https://www.wllwen.com/kejilunwen/jisuanjikexuelunwen/2278538.html
