Policy-based Autonomic Management in Data Centers

随着数据中心和系统的增长、变得更加异构化、支持日益多样化的应用程序和计算环境，系统管理员在管理系统和应用程序以满足运营要求(例如，体现在服务级别协议(SLA)中)方面面临着艰巨的挑战。与这些挑战相伴而来的是，数据中心运营商面临着计算和环境系统的电力成本不断增加以及环境立法不断增加的问题。因此，运营商必须努力优化几个相互竞争的目标：最大限度地降低功耗、减少排放/碳足迹、管理热输出、满足性能和可用性限制、确保安全性等。应对这些挑战是多方面的，需要：新的硬件方法、改进的软件、保护策略和提高数据中心管理的自动化。*拟议的研究重点是开发和评估数据中心管理的自动化方法。在本文中，自动化是指有一个或多个自主或半自主的计算代理，它们控制数据中心中的元素和行为，并在满足或管理操作员目标之间的权衡时减轻或增加人工管理员的负担。这项研究的长期目标是识别、开发和评估用于管理数据中心或系统集合的自动化方法。特别是，我们将研究基于策略的自主管理作为一种自动化管理多个异类系统组成的数据中心的方法的使用。然后，数据中心由许多不同的元件组成，例如处理器(核)、计算节点、网络设备、系统、应用等，这些元件需要不同的控制手段并需要在不同时间尺度上的管理动作，例如处理器的微秒到整个计算系统的数小时。那么，一位经理人就足够了，甚至不太可能是现实的。因此，这种规模的自动化方法必须考虑多个经理、他们的互动以及他们如何协调。这项研究还将评估和评估为数据中心和系统提出的各种能源感知技术，衡量替代方案的影响和开销，评估不同管理战略和技术的权衡，并提出新的、新的方法。*预计到2020年，数据中心将消耗全球8%的电力供应，贡献全球2.6%的碳排放。我们预计拟议的研究将对未来数据中心的管理和运营产生直接影响。鉴于加拿大对数字经济的关注，采用这些方法对加拿大具有成本效益的数据中心至关重要，并有助于使加拿大在数据中心专业知识和技术方面处于领先地位。

As data centers and systems grow, become more heterogeneous, support increasingly diverse applications and computing environments, system administrators face daunting challenges in managing the systems and applications in order to meet operational requirements, e.g., as embodied in service level agreements (SLAs). Concomitant with these challenges, data center operators face increasing costs of power for both computing and environmental systems as well as increasing environmental legislation. Consequently, operators must try to optimize several competing objectives: minimize power consumption, reduce emissions/carbon footprint, manage thermal output, meet performance and availability constraints, ensure security, etc. Addressing these challenges is multifaceted, requiring: novel hardware approaches, improved software, conservation strategies, and increased automation of data center management.******The proposed research focuses on the development and evaluation of automated methods for data center management. "Automated" in this context means that there are one or more autonomous or semi-autonomous computational agents that control elements and behavior in the data center and relieve or augment human administrators in meeting or managing tradeoffs among the objectives of the operator. The long term goal of the research is to identify, develop and evaluate automated approaches for the management of data centers or collections of systems. In particular, we will investigate the use of policy-based autonomic management as a means of automating the management of data centers comprised of multiple heterogeneous systems. The data center, then is comprised of many different elements, such as, processors (cores), compute nodes, network devices, systems, applications, etc., that require different means of control and require management actions at different time scales, e.g. in microseconds for a processor to hours for an entire computing system. It is unlikely then that a single manager would suffice or even be realistic. Therefore, automated approaches at this scale must consider multiple managers, their interactions and how they coordinate. The research will also evaluate and assess various energy aware techniques that have been proposed for data centers and systems, measure the impact and overhead of alternatives, evaluate the trade-offs of different management strategies and techniques, and propose new, novel methods.******Data centers are predicted to consume 8% of the worldwide electricity supply and contribute 2.6% of the global carbon output by 2020. We expect the proposed research to have a direct impact on future management and operations of data centers. Given Canada's focus on a digital economy, adoption of these approaches are central to cost efficient data centers in Canada and can help make Canada a leader in data center expertise and technology.