基于内容感知型去重Cache的磁盘能耗管理策略研究

On-board disk cache can effectively reduce the number of I/O requests arriving at the disk platters, thus increasing the time length that the disk drives can stay in a low-power state. However, due to the cost of SRAM, the capacity of on-board disk cache is normally smaller than 128 MB. This project proposes to take advantage of the data characteristics in the on-board disk cache, perform a data deduplication to increase the effective cache size, thus improving the hit ratio of disk cache and increasing the time intervals between the I/O requests arriving at the disk platters. Meanwhile, the power state transitions are leveraged to reduce the energy consumption when the disk experiences long idle intervals. The main research contents are as follows: (1) Investigate the characteristics,distribution,and semantic correlations of data blocks residing in disk cache,explore the mechanisms behind the data behaviour, design a real time and on-line method to identify the data behaviour; (2) Design an effective on-board disk cache architecture, a cache based lightweight deduplication system, a power-aware prefetching for the disk cache, and a collaborative memory flushing scheme to enhance the energy saving of the deduplication based disk cache; (3)Investigate the different factors that impact the power management policy of disk drives, understand the correlations among these factors and build a cost model. A prototype will be implemented to evaluate the theoretical research results. The research achievements will further facilitate and improve the approaches and theories of energy saving in the field of disk storage systems.

磁盘上的板载Cache 能有效减少到达盘片的数据请求，延长其处于低功耗状态的时间。但受SRAM 的价格影响，其容量一般小于128MB。传统的基于地址的Cache 替换策略导致该Cache中存在较多来自不同地址但内容相同的数据块。本项目提出对磁盘Cache 进行重复数据删除以增加其有效存储空间，提高命中率并加大到达盘片的请求间的时间间隔，再结合磁盘功耗状态的切换以降低能耗。具体内容包括：（1）探明磁盘Cache 中的数据特点、分布规律和语义关联，明确其数据行为的形成机理，设计实时在线的方法来识别其数据行为；（2）研究并设计有效的去重磁盘Cache 结构，高效轻载的去重方法，能量感知的Cache 预取策略，以及协作性内存刷新机制；（3）理清影响该能耗管理策略的各种影响因素，了解它们之间的相互作用和影响趋势，构建其代价模型。相关研究成果，将进一步推动和完善磁盘存储系统在节能降耗方面的方法和理论。

磁盘上的板载Cache 能有效减少到达盘片的数据请求，延长其处于低功耗状态的时间。但受SRAM 的价格影响，其容量一般小于128MB。传统的基于地址的Cache 替换策略导致该Cache中存在较多来自不同地址但内容相同的数据块。本项目提出对磁盘Cache 进行重复数据删除以增加其有效存储空间，提高命中率并加大到达盘片的请求间的时间间隔，再结合磁盘功耗状态的切换以降低能耗。主要完成内容包括：（1）研究了磁盘Cache 中的数据特点、分布规律和语义关联，明确其数据行为的形成机理，设计了一种在线的方法来识别其数据行为；（2）研究并设计了有效的去重磁盘Cache 结构，高效轻载的去重方法，能量感知的Cache 预取策略，以及协作性内存刷新机制；（3）研究了影响该能耗管理策略的各种影响因素，构建其代价模型。实现了一个仿真的原型系统来验证该项目的理论研究结果。实验数据证明了该项目所提思想的有效性。