The total estimated energy bill for data centers in 2010 was \$11.5 billion, and experts estimate that the energy cost of a typical data center doubles every five years. On the other hand, computational developments have started to lag behind storage advancements, therein becoming a future bottleneck for the ongoing data growth which already approaches Exascale levels. We investigate the relationship among data throughput and energy footprint on a large storage cluster, with the goal of formalizing it as a metric that reflects the trading among consistency and energy. Employing a client-centric consistency approach, and while honouring ACID properties of the chosen columnar store for the case study (Apache HBase), we present the factors involved in the energy consumption of the system as well as lessons learned to underpin further design of energy-efficient cluster scale storage systems.
翻译:2010年数据中心能源支出总额估计为11.5亿美元,据专家估计,典型数据中心的能源成本每五年翻一番。 另一方面,计算发展开始落后于存储进展,从而成为当前数据增长的未来瓶颈,而目前数据增长已经接近超尺度水平。我们调查了大型存储集群数据输送量和能源足迹之间的关系,目的是将数据输送量和能源足迹正规化为反映一致性和能源交易的一种衡量标准。采用了以客户为中心的一致性方法,在为案例研究(Apache HBase)尊重选定专栏仓库的ACID特性的同时,我们介绍了系统能源消耗所涉因素以及进一步设计节能集束储量系统的经验教训。