Conceptes and Methods for Edge Computing

快速发展的物联网（IoT）渗透到工业和社会。未来的智能医院、生产工厂或整个城市可能包含数千甚至数百万个传感器和智能物体。越来越多的数据必须在真实的时间限制下被有效地处理，以提供被认为是智能的功能。在广泛使用的两层设置中（参见移动云计算），计算密集型或多设备/用户任务从资源受限设备卸载到云，具有如下严重缺点。(1)在实时条件下，无限的通信延迟波动和间歇性连接是不可容忍的。(2)数据预处理不充分会浪费网络资源并加剧拥塞。(3)隐私和安全问题比在本地或数据源附近更难解决。该项目的主要目标是进一步推进最近传播的边缘计算范式，将两层设备-云设置进一步推向三层设备-边缘-云计算设置。虽然它有可能消除上述缺点，但仍然存在相当大的研究挑战。我们将开发和推进算法、协议和机制，以及统一模型，作为集成开发和运行时支持的基础。主要目标分为四个次级目标;这些次级目标的简要形式和计划的贡献如下。(1)虚拟化：用于虚拟化的新型轻量级容器的方法，这些方法能够将应用程序模块映射到计算资源上并在高度变化的能力下进行迁移，从而为边缘计算提供高度动态的运行时支持。(2)对照品：基于软件定义网络（SDN）和网络功能虚拟化（NFV）的最新进展而构建的新配置和控制协议及算法，实现了边缘环境中异构设备的灵活管理以及设备-边缘-云应用的可扩展部署。(3)最佳化：- 将（启用虚拟化的）应用模块有效地在线基于图映射到服从多个优化目标的资源上，具有对位置、负载和资源成本变化的动态适应。(4)沟通：具有统一通信框架和高效协议和机制的通信层，可根据相关内容类别（可执行文件、媒体流、复杂事件流）、跳数类别（边缘内、设备边缘、边缘云）、服务质量要求和网络接入技术进行调整。项目成果将帮助开发人员通过统一模型从边缘计算的技术问题中抽象出来，并通过其新颖的算法，协议和机制以运行时自适应的方式实现对这些问题的复杂自动处理。我们认为它对促进大规模智能物联网应用具有开创性意义。

The fast-growing Internet-of-Things (IoT) penetrates industry and society. Future smart hospitals, production plants or entire cities may comprise thousands, even millions of sensors and smart objects. Ever larger amounts of data must be processed efficiently, increasingly under real time constraints, to provide the functionality deemed smart. In the widely used two-tier setup (cf. mobile-cloud computing), computationally intensive or multi-device/user tasks are offloaded from resource-constraint devices to the cloud, with serious shortcomings as follows. (1) Unbounded communication delay fluctuations and intermittent connectivity are intolerable under real-time conditions. (2) Insufficient pre-processing of data wastes network resources and aggravates congestion. (3) Privacy and security issues are harder to solve than on premise or in proximity of the data source. Key goal of the project is the furthering of the recently propagated edge computing paradigm which furthers the two-tier device-cloud setup towards a three-tier device-edge-cloud computing setup. While it has the potential to remove the above shortcomings, considerable research challenges remain. We will develop and advance algorithms, protocols and mechanisms, as well as unifying models as a basis for integrated development and runtime support. The key goal is divided into four sub-goals; the short-form of these sub-goals and planned contributions are as follows. (1) Virtualization: approaches to novel lightweight containers for virtualization that enable the mapping of application modules onto computing resources and their migration despite highly varying capabilities, providing highly dynamic runtime support for edge computing. (2) Control: new configuration and control protocols and algorithms built upon recent advancements in Software Defined Networking (SDN) and Network Function Virtualization (NFV), enabling flexible management of heterogeneous devices in the edge environment and scalable deployment of device-edge-cloud applications. (3) Optimization: efficient online graph based mapping of (virtualization-enabled) application modules onto resources subject to multiple optimization objectives, with dynamic adaptation to location, load, and resource cost changes. (4) Communication: a communication layer with a unified communication framework and efficient protocols and mechanisms, adjusting to relevant content classes (executables, media streams, complex event streams), hop classes (intra-edge, device-edge, edge-cloud), quality-of-service requirements, and network access technologies. The project outcome will help developers abstract from technical issues of edge computing by means of the unifying model, and it will enable the sophisticated automatic handling of these issues in a runtime-adaptive manner with its novel algorithms, protocols and mechanisms. We deem it seminal for fostering large-scale smart IoT applications.