Distributed Serverless Computing in Cloud-Edge Environments

“无服务器”计算范式被认为是云计算的下一个演进，它促进了按需计算，而无需最终用户配置服务器。“无服务器”范例是使用功能即服务（FaaS）模型实现的，在该模型中，云提供商通过对每个功能调用或功能持续时间收费来实现使用货币化。FaaS与传统云模型的不同之处在于，终端用户只需为他们消耗的资源付费，而不是为他们分配的资源付费（即使这些资源是空闲的）。虽然“无服务器”以其对云计算的影响而闻名，但FaaS的许多好处也可以使Edge受益。边缘物联网（IoT）应用通常是分布式和资源受限的，因此，“无服务器”功能初始化和代码执行提供了一种计算模型，可以最大限度地减少资源和能源消耗。这个项目有三个目的和目标，每个目的和目标都是为期一年的研究，目的是发表研究结果。这三个目标将形成在第三年的最后论文的基础。第一个目标是开发一个“无服务器”边缘平台，以响应边缘物联网应用需求。特别是，它将研究为边缘物联网部署所呈现的各种工作负载提供实时响应的方法。主要研究领域包括边缘工作负载的趋势分析，边缘节点资源的智能配置和调度，以及“无服务器”边缘平台的网络/路由功能。第二个是研究在异构边缘节点上减少边缘机器学习推理的延迟。这一目标涉及到将“无服务器”原则映射到边缘上下文，其中网络中所有边缘节点的推理延迟应该是一致的。第三是为边缘上的“无服务器”执行开发足够的性能安全性，其中硬件是共享的，资源是受限的。项目研究方法包括继续研究和审查边缘和“无服务器”计算的相关文献。“无服务器”方法的分析和基准测试，包括收集定量数据，如延迟、内存和能耗，以及各种工作负载下的容量吞吐量。分析和识别现有研究和用例中的当前挑战。为合成数据生成和“无服务器”边缘系统的开发创建计算测试平台。为现实世界领域应用开发创新系统和服务。该项目与EPSRC的众多数字经济战略（如“可持续数字社会”研究主题）直接一致。“无服务器”原则与物联网设备要求协同作用，在需要计算时仅消耗有限的设备资源（例如，当传感器定期读取数据时），因此，电池供电和节能的物联网部署受益于较低的电力成本和延长的设备寿命。该项目与EPSRC的另一个研究主题相一致，即加强个人隐私，作为“超越数据驱动经济”的一部分。边缘计算使计算能力更接近数据源，因此，敏感信息不会传输到云服务器。存储在集中数据中心的数据更少，这些数据更有可能因安全漏洞泄露或被利用来获取商业利益。该项目将与行业合作伙伴，即日本乐天移动（Rakuten Mobile）合作开展。他们将提供技术方面的建议，并将建议的研究应用于实际用例。在东京的一个实验室实习可以是远程的，也可以是亲自的，这取决于何时国际旅行更加顺畅。

Considered as the next evolution in Cloud computing, the 'Serverless' computing paradigm facilitates on-demand computation without an end-user provisioning a server. The 'Serverless' paradigm is realised using a Functions-as-a-Service (FaaS) model in which Cloud providers monetise usage by charging per function invocation or function duration. FaaS differs from conventional Cloud models as end-users only pay for the resources they consume as opposed to paying for the resources they allocate (even when these resources are idle). Although 'Serverless' is known for its Cloud impact, there are numerous benefits of FaaS that could benefit the Edge. Edge Internet of Things (IoT) applications are often distributed and resource-constrained, as such, 'Serverless' function initialisation and execution of code provide a model of computation that minimises resource and energy consumption. This project has three aims and objectives that map to one year of study each with the intent to publish findings. The three objectives will form the basis for the final thesis in the third year. The first objective is to develop a 'Serverless' Edge platform that is reactive to Edge IoT application demand. In particular, it will investigate methods of providing real-time responsiveness to the varied workloads that Edge IoT deployments present. Main areas of research include trend analysis of Edge workloads, intelligent provisioning and scheduling of Edge node resources, and networking/routing capabilities of a 'Serverless' Edge platform. The second is investigating reducing latency for Edge Machine Learning inference on heterogeneous Edge nodes. This objective involves mapping 'Serverless' principles to an Edge context where inference latency should be consistent for all Edge nodes in the network. The third is developing sufficient and performative security for 'Serverless' execution on the Edge where hardware is shared and resources are constrained. The project research methodology involves continued study and review of related literature in Edge and 'Serverless' computing. Profiling and benchmarking of 'Serverless' methods including gathering quantitative data such as latency, memory and energy consumption, and volume throughput under various workloads. Analysis and identification of current challenges in existing research and use cases. Creation of computation testbeds for synthetic data generation and development of 'Serverless' Edge systems. Development of innovative systems and services for real-world domain applications. This project directly aligns with numerous EPSRC digital economy strategies such as the 'Sustainable Digital Society' research theme. 'Serverless' principles synergise with IoT device requirements by only consuming limited device resources when computation is necessary (e.g. when a sensor takes periodic readings), therefore, battery operated and energy-conscious IoT deployments benefit from lower electricity costs and extended device longevity. Another EPSRC research theme this project aligns with is enhancing individual privacy as part of the 'Beyond a Data-Driven Economy'. Edge computing brings computational ability closer to the data source, as such, sensitive information is not transported to Cloud servers. Less data is stored in centralised data centres that are more likely to be leaked from security vulnerabilities or exploited for commercial gain. This project will be carried out in collaboration with an industry partner, namely Rakuten Mobile, Japan. They will provide advice on technical aspects and on applying the proposed research to real-world use-cases. An internship with one of their labs in Tokyo is expected either remotely or in person, which is subject to when international travel is more seamless.