Resilient and energy-efficient multi-cloud service grid for 5G edge applications

本研究项目旨在设计、实施和验证用于说明天空计算范式的边缘到云连续统的控制器框架。控制器框架将优化多云、多接入、多核网络，称为多云服务网格。分布式网格将异类计算、数据和智能的结构编织在一起，以随时随地按需实现情景感知、沉浸式和分布式现代应用程序。这样的框架面临着专业基础设施、多租户服务和分布式控制的挑战。这一框架的验证将通过在仓库中操作一群机器人的应用程序来完成；这需要新的分析模型。对于这些模型，我们将重点介绍深度强化学习和卷积神经网络。关键输出之一将是具有连续和离散动作的静态最优控制策略，利用控制器框架实现所需的优化。该框架以及模型、人工智能方法和策略将部署在提供云制造服务的ETS的CEOSNet基础设施上。此部署应展示集成解决方案的可行性、优势和潜在限制。该项目将为VMware的多云和网格平台做出贡献，并促进加拿大新一类创新的5G应用。

This research project aims to design, implement, and validate a controller framework for an edge-to-cloud continuum that illustrates the Sky Computing paradigm. The controller framework will optimize the multi-cloud, multi-access, multi-core network, named Multi-Cloud Service Grid. A distributed Grid weaves together a fabric of heterogeneous compute, data, and intelligence to enable context-aware, immersive, and distributed modern applications anytime, anywhere, on-demand. Such a framework faces challenges of specialized infrastructure, multi-tenant services and distributed control. The validation of this framework will be accomplished via applications that operate a swarm of robots in a warehouse situation; requiring new analytical models. For these models, we will focus on deep reinforced learning and convolutional neural networks. One of the key outputs will be a stationary optimal control policy with continuous and discrete actions, achieving the required optimization using the controller framework. This framework, along with the models, AI methods, and policies, will be deployed on the CEOSNet infrastructure at ÉTS that provides cloud manufacturing services. This deployment should demonstrate the feasibility, advantages, and potential limitations of an integrated solution. The project will contribute to VMware's multi-cloud and grid platform as well as facilitating a new class of innovative 5G applications in Canada.