Serverless computing has emerged as a new cloud computing model which in contrast to IoT offers unlimited and scalable access to resources. This paradigm improves resource utilization, cost, scalability and resource management specifically in terms of irregular incoming traffic. While cloud computing has been known as a reliable computing and storage solution to host IoT applications, it is not suitable for bandwidth limited, real time and secure applications. Therefore, shifting the resources of the cloud-edge continuum towards the edge can mitigate these limitations. In serverless architecture, applications implemented as Function as a Service (FaaS), include a set of chained event-driven microservices which have to be assigned to available instances. IoT microservices orchestration is still a challenging issue in serverless computing architecture due to IoT dynamic, heterogeneous and large-scale environment with limited resources. The integration of FaaS and distributed Deep Reinforcement Learning (DRL) can transform serverless computing by improving microservice execution effectiveness and optimizing real-time application orchestration. This combination improves scalability and adaptability across the edge-cloud continuum. In this paper, we present a novel Deep Reinforcement Learning (DRL) based microservice orchestration approach for the serverless edge-cloud continuum to minimize resource utilization and delay. This approach, unlike existing methods, is distributed and requires a minimum subset of realistic data in each interval to find optimal compositions in the proposed edge serverless architecture and is thus suitable for IoT environment. Experiments conducted using a number of real-world scenarios demonstrate improvement of the number of successfully composed applications by 18%, respectively, compared to state-of-the art methods including Load Balance, Shortest Path algorithms.

中文翻译：

---

一种深度强化学习方法，适用于云-边缘连续体中基于分布式函数即服务 （FaaS） 的边缘应用程序编排


无服务器计算已成为一种新的云计算模型，与 IoT 相比，它提供对资源的无限且可扩展的访问。这种模式提高了资源利用率、成本、可扩展性和资源管理，特别是在不规则的传入流量方面。虽然云计算被认为是托管 IoT 应用程序的可靠计算和存储解决方案，但它并不适合带宽有限、实时和安全的应用程序。因此，将云边缘连续体的资源转移到边缘可以减轻这些限制。在无服务器架构中，作为功能即服务 （FaaS） 实现的应用程序包括一组链式事件驱动微服务，这些微服务必须分配给可用实例。由于 IoT 是动态的、异构的、大规模的、资源有限的环境，因此 IoT 微服务编排仍然是无服务器计算架构中一个具有挑战性的问题。FaaS 和分布式深度强化学习 （DRL） 的集成可以通过提高微服务执行效率和优化实时应用程序编排来改变无服务器计算。这种组合提高了整个边缘-云连续体的可扩展性和适应性。在本文中，我们提出了一种基于深度强化学习 （DRL） 的新型微服务编排方法，用于无服务器边缘-云连续体，以最大限度地减少资源利用率和延迟。与现有方法不同，这种方法是分布式的，在每个间隔中需要最少的真实数据子集，以便在建议的边缘无服务器架构中找到最佳组合，因此适用于物联网环境。 使用大量真实场景进行的实验表明，与包括负载均衡、最短路径算法在内的最先进方法相比，成功组合的应用程序数量分别提高了 18%。