Autonomous monitoring of data centre operations

我们社会的计算需求，其中大部分是在数据中心（DC）中执行的，每五年翻一番。然而，尽管发展中国家的生产率不断提高，但它们仍然面临着（1）高的建设资本成本，以及（2）由于能源使用效率低下而导致的高运营成本。高成本是由过度设计和过度供应的广泛行业实践引起的，过度设计和过度供应源于运营商不具有性能度量的实时可见性的事实，例如，利用率和能耗，或健康状况，例如，设施内各种组件的即将发生的故障和失效的指示器。实现实时可见性的主要瓶颈是每个设施都由数千个组件组成，这些组件因供应商、设计和操作原理而异。在此，我们将开发三种使能技术，以实现以最少的人为干预监测DC操作的通用平台，即（1）可以记录和无线传输DC中各种组件的健康状况和性能指标的廉价监测硬件模块，（2）用于监测数据和命令的实时通信的无线联网协议，（3）通过收集的实时数据和计算效率高的热建模实现的用于完全自主监测和诊断的算法。这些技术将允许DC运营商（1）自适应地优化其运营，以及（2）预测和触发维护工作，以最大限度地减少服务中断。这些技术将被我们的合作伙伴Cinnos使命关键公司（CMCI，www.cinnos.com）转化为一个整体的数据中心基础设施管理（DCIM）产品线，CMCI是一家快速发展的年轻加拿大公司。CMCI将与我们的努力协同合作，与主要客户进行广泛的beta测试，并提供有价值的反馈，以引导技术开发实现最大的商业影响。除了为我们的行业合作伙伴创造就业机会和新的收入来源外，一个自然和值得注意的结果是发展中国家的能源消耗大幅减少，占全球能源生产的3%左右。

The computing needs of our society, much of these performed in data centres (DCs), double every five years. However, despite their increasing productivity, DCs are fraught with (1) high capital costs for establishment, and (2) high operating costs due to inefficient energy usage. The high costs arise from the widespread industry practice of over-design and over-provisioning stemmed from the fact that operators do not have real-time visibility of the metrics of performance, e.g., utilization percentage and energy consumption, or health conditions, e.g., indicators for imminent faults and failures, of the various components within their facilities. The primary bottleneck in enabling real-time visibility is that each facility consists of thousands of components, which differ by vendors, design and operating principles. Herein, we will develop three enabling technologies towards a universal platform for monitoring DC operations with minimal human intervention, namely, (1) inexpensive monitoring hardware modules that can record and wirelessly transmit health conditions and performance metrics of various components in a DC, (2) wireless networking protocols for real-time communications of monitoring data and commands, (3) algorithms for fully autonomous monitoring and diagnosis enabled by the real-time data collected and computational efficient thermal modelling. These technologies will allow DC operators to (1) adaptively optimize its operations, and (2) predict and trigger maintenance efforts to minimize service interruptions. The technologies will be translated into a holistic data centre infrastructure management (DCIM) product line by our partner, Cinnos Mission Critical Incorporated (CMCI, www.cinnos.com), a fast growing young Canadian company. CMCI will perform extensive beta testing with lead customers synergistically with our efforts, and provide valuable feedback to steer the technology development towards maximum commercial impact. In addition to creating jobs and new sources of revenue for our industry partner, a natural and noteworthy consequence is the significant reduction in energy consumption of DCs, which account for ~3% of the energy produced worldwide.