An increasing amount of data is being injected into the network from IoT (Internet of Things) applications. Many of these applications, developed to improve society's quality of life, are latency-critical and inject large amounts of data into the network. These requirements of IoT applications trigger the emergence of Edge computing paradigm. Currently, data centers are responsible for a global energy use between 2% and 3%. However, this trend is difficult to maintain, as bringing computing infrastructures closer to the edge of the network comes with its own set of challenges for energy efficiency. In this paper, we propose our approach for the sustainability of future computing infrastructures to provide (i) an energy-efficient and economically viable deployment, (ii) a fault-tolerant automated operation, and (iii) a collaborative resource management to improve resource efficiency. We identify the main limitations of applying Cloud-based approaches close to the data sources and present the research challenges to Edge sustainability arising from these constraints. We propose two-phase immersion cooling, formal modeling, machine learning, and energy-centric federated management as Edge-enabling technologies. We present our early results towards the sustainability of an Edge infrastructure to demonstrate the benefits of our approach for future computing environments and deployments.

翻译：物联网应用正将越来越多的数据注入网络。许多旨在提升社会生活质量的应用对延迟敏感，并会向网络注入大量数据。这些物联网应用的需求催生了边缘计算范式的出现。目前，数据中心消耗了全球2%至3%的能源。然而，这一趋势难以持续，因为将计算基础设施移至网络边缘在能效方面面临一系列独特挑战。本文提出了一种面向未来计算基础设施可持续性的方法，旨在实现：(i) 能效与经济可行的部署方案，(ii) 具备容错能力的自动化运维，以及(iii) 提升资源效率的协作式资源管理。我们指出了在数据源附近应用基于云的方法所面临的主要局限，并阐述了由此产生的边缘计算可持续性研究挑战。我们提出将两相浸没式冷却、形式化建模、机器学习和以能源为中心的联邦管理作为边缘赋能关键技术。通过展示面向边缘基础设施可持续性的初步成果，我们论证了该方法对未来计算环境与部署模式的潜在优势。