一种基于热数据识别技术的UBIFS优化方案

发布时间：2018-06-02 15:51

本文选题：闪存 + UBIFS文件系统　；参考：《计算机科学》2017年06期

【摘要】：NAND Flash介质具有特殊的物理性质,传统文件系统不能直接对NAND Flash进行管理,容易造成NAND Flash设备性能下降、磨损不均衡等负面影响。闪存文件系统将FTL与文件系统功能有机地结合起来,可更好地发挥NAND Flash的高性能。无序区块镜像文件系统(Unsorted Block Image File System,UBIFS)是一种被广泛使用的闪存文件系统,但其存在着写放大和垃圾回收操作频繁触发等问题。针对UBIFS中存在的问题,提出利用多哈希函数的哈希表对热数据进行识别,以降低热数据识别开销,提高热数据识别的准确率;采用多日志技术,将不同热度的日志、数据分开存放,以减少垃圾回收触发频率;采用热数据日志延迟提交技术来减少UBIFS中日志提交带来的元数据修改,进而减少了写放大产生的次数。测试与分析表明,与原UBIFS相比,优化后的UBIFS的系统内部写物理块的次数减少5%~10%,垃圾回收操作触发的次数减少了7%~13%,同时系统的IOPS提高了5%~18%,系统性能下降现象得到了有效缓解。
[Abstract]:NAND Flash media has special physical properties, traditional file system can not directly manage NAND Flash, which can easily lead to the deterioration of NAND Flash equipment performance, uneven wear and other negative effects. Flash file system combines the functions of FTL and file system organically, which can better play the high performance of NAND Flash. Unsorted Block Image File system UBIFS) is a widely used flash file system, but there are some problems such as write amplification and frequent trigger of garbage collection. Aiming at the existing problems in UBIFS, this paper proposes to use the hash table of multi-hash function to identify the thermal data in order to reduce the cost of thermal data identification and improve the accuracy of thermal data recognition. The data is stored separately to reduce the trigger frequency of garbage collection, and the delayed commit technology of thermal data log is adopted to reduce the metadata modification caused by log submission in UBIFS, thus reducing the number of write magnification. The test and analysis show that compared with the original UBIFS, the number of writing physical blocks in the optimized UBIFS system is reduced by 5 times and the number of triggered garbage collection operations is reduced by 7% and 13%, while the IOPS of the system has been improved by 518%, and the decline of system performance has been effectively alleviated.
【作者单位】：华中科技大学光电国家实验室;华中科技大学计算机科学与技术学院;
【基金】：国家863项目计划(2015AA016701,2015AA015301) 国家自然科学基金资助项目(61402189,61472153) 计算机体系结构国家重点实验室(CARCH201505) 武汉市应用基础研究计划项目(2015010101010004) 信息存储系统教育部重点实验室资助
【分类号】：TP333

【相似文献】