基于热循环的大规模分布式存储系统数据放置策略研究

能耗是当前数据中心所面临的最重要的挑战之一。其中，制冷系统的能耗可达总能耗的30%-50%。传统的数据放置策略忽略了数据中心中气流组织对其影响。本项目提出在I/O密集型数据中心环境下，利用分布式文件系统中的多数据副本，针对同一热点数据按不同的访问行为构造不同的数据放置，以便始终选择热循环最小的数据放置来提供服务，从而降低制冷系统的能耗。具体内容包括：（1）探明I/O密集型数据中心的气流组织、热循环、数据访问行为、以及数据放置策略之间的关联关系和基本规律，设计一体化的能耗模型;（2）建立数据访问行为的提取、表达、和评估机制，重新审视并设计基于访问行为的热循环感知型数据放置策略以及副本管理机制，研究并设计轻载的数据放置策略的进化方法；（3）理清影响该放置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。相关研究成果将推动和完善分布式存储系统在节能降耗方面的方法和理论。

Energy consumption is one of the most important challenges for the modern data centers. The energy consumed by the cooling system takes 30% -50% of the overall energy consumption. This project proposes to leverage the multiple data replicas employed by the distributed storage systems to design multiple different data layouts for the same data in I/O intensive data centers. Meanwhile, the data layout which generates the minimal heat re-circulation is selected to offer data service, thus reducing the energy consumption of the cooling system in data centers. The main research topics cover: (1) Investigate the correlations among air flow，heat re-circulation, data access pattern, and data layout in the I/O intensive data centers, design a unified power model for the data center; (2) Construct approaches to identify and evaluate data access patterns, design data layout strategies and replica management methods based on the data access pattern and heat re-circulation; (3) Explore the factors that impact the data layout strategy, understand the correlations among these factors and build a cost model. The related research achievements will further facilitate and improve the approaches and theories of energy saving in the field of distributed storage systems.