基于零星请求的低能耗磁盘调度策略研究

Energy consumption of disk storage systems is one of the most important chalenges in the community. The non-bursty periods contained in disk I/O traffic offer potential opportunities for reduce the energy consumption of disk drives. However, the sporadic requests distributed within the non-bursty periods incur significant time and energy penalty.Based on the semantic patterns of block correlations, this project proposes to locate the sporadic requests by leveraging a frequent sequence mining approach. The sporadic requests are re-arranged to extend the time length so that the disk drives can stay in a low power state longer, thus saving energy. This project will expore the following topics: (1)This project will explore the non-bursty behaviour contained in disk I/O traffic, disclose the internal patterns and construct the corresponding mathematical models, and investigate how to leverage the bursty behaviour to locate those sporadic requests. (2) This project will employ multi-objective optimization theory to design the cost model of a power-aware disk scheduler based on the located sporadic requests, analyze the impact of this scheduler on the on-board disk cache, design an adaptive strategy to allocate the space of disk cache, thus minimizing the energy consumption of disk drives. (3)This project will explore and design a mechanism to guarantee the quality of service of disk drives when using the power-aware disk schedule. A prototype will be implemented to verify the theoretical results. This research will offer useful insights into how to reduce the energy consumption of disk based storage systems, distributed storage systems, data centers, etc.

基于磁盘的不同能耗状态切换，磁盘I/O中的非突发时段为降低其能耗提供了可能。但非突发时段之间的零星请求会导致高昂的时间和能量开销。本项目提出利用I/O的块关联性和频繁序列挖掘技术来定位该零星请求并进行调度，以尽可能延长磁盘处于低能耗状态的时间，从而在保证服务质量的前提下最大限度地降低其能耗。本项目将研究磁盘I/O中的非突发性数据访问行为，揭示其内在规律并构造合适的描述模型，探索该行为和零星请求的潜在联系并阐明其机理，设计零星请求的定位方法；利用多目标优化理论来研究并设计基于零星请求的低能耗调度策略的代价模型，明确磁盘Cache对该调度策略的影响，设计自适应的磁盘Cache分配策略以最大化该策略的能量节省；研究并设计基于该调度策略的磁盘服务质量保证机制。实现一个基于该策略的原型系统来验证理论成果。相关研究结果，对于降低多盘存储系统、分布式存储系统、数据中心等的能耗都将具有重要的参考作用。