ETCHING: Edge-To-Cloud Heterogeneous Infrastructure NetworkinG

蚀蚀，边缘到云异构基础设施网络测试平台将使至少三个完全集成的nserc资助的异构云计算、云扩展和新兴云计算范例的研究行动成为可能。******ETCHING构成了一个全面的实验边缘到云异构计算平台，允许研究团队进行低级系统修改、可重构网络、云内和边缘FPGA和gpu驱动的加速以及自定义云管理扩展的实验，这些扩展由于涉及风险和对其他租户的潜在干扰而无法部署在公共云中。“蚀刻”的隐喻是指在物理层和物理层对平台的计算和网络子系统进行动态修改和干预，以实现所需的探索。蚀刻允许研究团队在现实条件下追求他们的研究目标，而无需求助于模拟，从而导致易于转移到生产优势，最先进的框架和软件堆栈的贡献。******异构基础设施由研究云、可通过收发器热插拔的加速设备和边缘设备组成。云将以一种允许研究团队动态更新物理资源的方式运行（例如，提供它们的进出，修改它们的功能，即，包括有或没有加速的节点），通过云管理层进行实验，以实现自主缩放算法的实验，并尝试设计和评估“无服务器计算”范式。所有这些研究行动都需要直接访问正在运行的云的整个硬件和软件堆栈，出于操作安全性和可靠性的原因，这在公共云和共享研究计算环境中是不可能的。******结果将是新颖的扩展技术，以更好地平衡稀缺的物理资源，提供支持高效和隔离执行所需的解决方案，以启用“无服务器计算”范式，以及加速云和边缘数据处理的原则。******拟议的基础设施将使研究团队能够在最先进的设备上进行现实的实验，并在加拿大进行具有国际竞争力的研究，达到世界上最好的水平

ETCHING, the Edge-To-Cloud Heterogeneous Infrastructure NetworkinG test bed will enable at least three fully integrated NSERC-funded research actions on heterogeneous cloud computing, scaling in clouds, and emerging cloud computing paradigms. ******ETCHING constitutes a comprehensive experimental edge-to-cloud heterogeneous computing platform that allows the research team to experiment with low-level systems modifications, reconfigurable networking and in-cloud and on-edge FPGA- and GPU-driven acceleration, and custom cloud management extensions that are not feasible to deploy in public clouds due to the involved risks and potential interference for other tenants. “Etching” metaphorically refers to the quintessential requirement to dynamically modify and intervene at the physical level and physical layer of the computing and networking subsystems of the platform to enable the required exploration. ETCHING allows the research team to pursue their research goals under realistic conditions without needing to resort to simulations, leading to contributions that are readily transferable to production-strengths, state-of-the-art frameworks and software stacks. ******The heterogeneous infrastructure is comprised of a research cloud, acceleration devices hot-pluggable via transceivers, and edge devices. The cloud will be operated in a manner that allows the research team to dynamically update physical resources (e.g., provision them in and out, modify their capabilities, i.e., include nodes with or without acceleration), to experiment with the cloud management layer to enable experimentation on autonomous scaling algorithms, and to experiment with designing and evaluating the "serverless computing" paradigm. All these research actions require direct access to the entire hardware and software stack of a running cloud, which for operational safety and reliability reasons is not possible in public clouds or shared research computing environments.******The outcome will be novel scaling techniques to better balance scarce physical resources, offer solutions to support the efficient and isolated execution required to enable the "serverless compute" paradigm, and acceleration principles for data processing in clouds and on the edge.******The proposed infrastructure will allow the research team to conduct realistic experiments on state-of-the-art equipment and to carry out internationally competitive research made in Canada that measures up to the best in the world.**