杨明伟, 李亚晖, 赵益民, 蔡晓乐. AFTL: 一种基于高级指令的闪存转换层[J]. 微电子学与计算机, 2017, 34(1): 15-20.
引用本文: 杨明伟, 李亚晖, 赵益民, 蔡晓乐. AFTL: 一种基于高级指令的闪存转换层[J]. 微电子学与计算机, 2017, 34(1): 15-20.
YANG Ming-wei, LI Ya-hui, ZHAO Yi-min, CAI Xiao-le. AFTL: A Flash Translation Layer Based on Advanced Operations[J]. Microelectronics & Computer, 2017, 34(1): 15-20.
Citation: YANG Ming-wei, LI Ya-hui, ZHAO Yi-min, CAI Xiao-le. AFTL: A Flash Translation Layer Based on Advanced Operations[J]. Microelectronics & Computer, 2017, 34(1): 15-20.

AFTL: 一种基于高级指令的闪存转换层

AFTL: A Flash Translation Layer Based on Advanced Operations

  • 摘要: 为了对多层次并行固态盘提供的多种并行性进行高效利用, 提出一种基于冷热数据的FTL算法: AFTL, 通过对高级指令多分组指令(multi-plane)和交错指令(interleave)的使用, 达到并行性利用的最大化, 冷热数据识别与内部数据迁移指令(copy-back)、多分组擦除指令的结合则大大降低了垃圾回收的成本, 两者的组合最终使得整体性能获得进一步提升.仿真结果显示, 在多种负载、多种不同通道数的情况下, 相对于同样综合利用多种并行性的动态FTL算法, 写响应时间平均减少15.3%, 相对于DLOOP算法, 写响应时间平均减少78.9%.

     

    Abstract: In this research, we proposed a new page mapping FTL called AFTL in order to fully take advantage of several parallelism provided by multi-level parallel solid-state disk. Through using multi-plane operations and interleave operations, AFTL achieves the goal to exploit the parallelism effectively, meanwhile the combination of identification of hot-cold data, copy-back operation and multi-plane-erase operation decreases the cost of garbage collection a lot, all the factors make the performance get largely promoted. The experimental results show that under the several real world workloads with different channel number, AFTL decreases the write response time 15.3% and 78.9% respectively compared with existent FTL which also exploit multi-level parallelism and DLOOP.

     

/

返回文章
返回