ITR: Energy and Thermal Management for Data Centers

互联网基础设施的趋势推动了向基于服务的计算的转变。基于互联网的服务提供商正在为各种服务增加新的数据中心容量，包括Web托管、应用程序服务、外包存储、电子市场和其他网络服务。在这些大规模和快速发展的数据中心中，最重要的挑战之一是能源和热问题。数据中心通常具有非常高的电力需求，这显著增加了部署和扩展数据中心的难度，因为需要大型发电基础设施。此外，高耗电量可能导致高额电费和对环境的负面影响。电力消耗的另一个后果是散热。最近的一份报告显示，数据中心的每个产品占地面积的热密度每年增加约28%，增加了空调的冷却成本。如果温度没有得到有效的冷却，过高的热量会导致许多设备出现故障，甚至导致整个数据中心关闭。更具体地说，我们研究了数据中心的创新能源和热管理方案，重点是后端存储，这是数据中心最大的能源消耗之一。为了节约能源和减少散热，我们探索了能量和热感知存储组织，并研究了跨层和跨节点、节能、热安全和高性能的数据中心管理方案。此外，我们重新审视和重新设计了现有的高度性能优化的控制策略，以实现能源和热感知。

Trends in Internet infrastructure drive a shift toward service-basedcomputing. Internet-based service providers are adding new data centercapacity for a variety of services including Web hosting, applicationservices, outsourced storage, electronic markets, and other networkservices. In these large scale and rapidly evolving data centers, one ofthe most important challenges is the energy and thermal problem. Datacenters typically have very high power requirement, which significantlyincreases the difficulty of deploying and expanding a data center due tothe need for a large power generation infrastructure. Moreover, highpower consumption can result in high electricity bills and negativeenvironmental implications. Another consequence of power consumption isheat dissipation. A recent report shows that heat density per productfootprint in a data center increases around 28% per year, increasing thecooling costs for air conditioning. If the temperature is not cooleddown effectively, excessive heat can cause malfunctions of many devicesand even shutdown of the entire data center.The proposed research addresses the above problem. More specifically, weinvestigate innovative energy and thermal management schemes for datacenters with an emphasis on back-end storage, which is one of thebiggest energy consumers in data centers. To conserve energy and reduceheat dissipation, we explore energy and thermal aware storageorganizations and investigate cross-layer and cross-node,energy-efficient, thermally safe and performance-effective managementschemes for data centers. Moreover, we revisit and redesign existinghighly-performance-optimized control policies to be energy and thermalaware.