【作者】 刘航；

【导师】 孔祥维；

【作者基本信息】 大连理工大学 ， 信号与信息处理， 2011， 博士

【摘要】 随着云计算数据中心的规模和数量的不断扩大,以及内部单位面积能耗的不断增加,制冷能耗已经成为了制约云计算数据中心发展的主要瓶颈之一。在美国,所有数据中心在2006年的能耗约占全美当年电力消耗的1.5%；在我国,据推算在2009年约占全国全年电力消耗1%。然而,这些能耗的40～50%被用于对数据中心的温度控制。因此,提高数据中心制冷效率,是对数据中心在节能减排方面研究的主要途径之一,由于其可以节省大量能源,IBM、HP等数据中心研发机构均在积极研究相关内容。本文针对云计算中心内部热量分布不平衡的问题,基于无线多媒体传感器网络(WMSNs),实时监测局部高温(即热点)现象,并利用任务迁移等方式降低热点区域的热负荷,达到平衡热量分布、提高制冷效率的目的。文中基于WMSNs提出了云计算中心局部温度管理新体系,通过对云计算中心室内热量分布实时监测,以“热点发现—热点定位—特征提取—热点消除”为主线,实现了消除热点、提高制冷效率的目的。本文具体在以下几个方面取得了一定成果：基于WMSNs的动态频谱分配技术。局部温度管理新体系中的热成像传感器网络使用了通用的无线多媒体传感器模块,在必要时需要实时地连续传输热点的特征序列。鉴于无线多媒体传感器模块受计算能力和通信能力的限制,本文利用网内各节点的接收信号强度(RSSI),提出了具有轮换机制的分布式频谱感知和管理方法,利用认知无线电技术解决了系统内各节点之间及其与其它无线设备之间的共存问题。基于热成像传感器网络的热点联合定位技术。文中首先基于形态学原理提出了单节点在单幅温度图像中检测真实热点的方法。然后,提出节点在发现真实热点后,如何利用热点相对位置信息引导其它节点搜索该热点,以及利用多个节点对热点进行联合定位的方法。所有方法在满足传感器节点的有限计算能力和存储能力的同时,达到了满足应用要求的准确性。基于传感器节点的热点特征提取及重建技术。通过热点特征,可以获取热点状态、热点变化和热点产生原因,并自动选择最有效方式消除热点。本文针对热点二维形态特征的数据量较大问题,提出了非对称热点形态特征提取及热点二维重建算法,并提出了利用热点特征判别热点产生原因的方法和消除热点的任务迁移策略。通过实验证明,本文提出的局部温度管理新体系在发现热点后,通过对其的后续处理可以达到消除热点、平衡热量分布的目的,使得在CRAC低速供气的条件下,服务器可以处于较低的温度环境,从而节省一定制冷成本。更多还原

【Abstract】 Energy consumption in cooling has been one of the main factors restricting the development of cloud computing centers due to their increasing, both quantity and scale, and the increasing in heat dissipated per unit surface area. In the U.S., the power consumption of data centers is 1.5% of total national electrical power consumption of the full year in 2006. In China, the power consumption of data centers account for about 1% of total national electrical power consumption of the full year in 2009. However,40% to 50% of the power consumption is cost to cool data centers. Therefore, a means of energy saving and emission reduction is to improve cooling efficiency of data centers, that is the current research of IBM, HP and other R&D Institutions of data centers.This dissertation focus on the imbalanced distribution of heat inside the data center, using the Wireless Multimedia Sensor Networks (WMSNs) to test the local hot spots instantly and cool the hot spots by job migration etc. to balance the distribution of heat in data center. This dissertation addresses the cloud computing center’s local temperature management system based on WMSNs, which performs the real-time monitoring heat distribution to optimize datacenter performance in terms of energy consumption and throughput. The main tasks of the system are to detect and localize hotspots, and extract their characteristics for remove them. The specific contributions are as follows:The first is the dynamic spectrum allocation technology based on WMSNs. The thermal camera network in the local temperature management system uses the general wireless multimedia sensor platforms. It need real-time transmit characteristics series of hotspots in a hardware constrained environment. Thus this dissertation proposes the spectrum sensing and management method to prevent interference between WMSNs and other wireless devices.The second is cooperative localizing hotspots using the thermal camera network. This dissertation proposes the method to detect the real hotspot in the thermal using one thermal camera node, and the method to guide other thermal camera nodes find the hotspot by its information. Also this dissertation proposes the algorithm to cooperatively local the hotspot by multiple thermal camera nodes. The simulations prove that the location accuracy meets application requirements, and the algorithms can be used in hardware constrained environments. Last but not least, the characteristics of hotspots are extracted and compressed by nodes. The characteristics can indicate state, change and cause of the hotspot. It is conducive to selective effective way to remove hotspot. In addition, an asymmetrical compression method is proposed to extract two dimensional morphological characteristics of hotspots and to reconstruct it.Experiments prove that the local thermal management system proposed in this dissertation can remove hotspots and imbalance the distribution of heat, so that the servers in the cloud computing center can run in low temperature environment with low airflow from CRAC, and some energy can be saved.更多还原