PrEstoCloud: A Novel Framework for Data-Intensive Multi-Cloud, Fog, and Edge Function-as-a-Service Applications

1. Introduction

Fog computing extends multi-cloud computing by enabling services or application functions to be hosted close to their data sources, which are typically Internet of Things (IoT) sensors. Hosting functions close to data sources can reduce the latency and cost of delivering sensor-generated data to a remote cloud and can improve the Quality of Service (QoS; Hao et al. 2017). A key challenge for fog computing is auto-scaling, i.e. the autonomous capacity for continuous adaptation and control of the computing infrastructure through the recognition of insights and knowledge in the data. Insights from the analysis of sensor-generated data can adapt the computing resources to meet current or predicted computing needs, save cost, increase performance and reliability, and meet environmental concerns.

To realize the deployment of applications and services and take advantage of the adaptive capabilities of fog computing, two new software engineering paradigms have emerged: Serverless and Function-as-a-Service (FaaS), which are seen as two enabling technologies for next-generation fog computing (Van Eyk et al. 2018). The serverless paradigm exploits functions or microservices as the unit of deployment and is hence more efficient than using a virtual machine (VM) or a container since their inherent complexity becomes transparent to the application owner (Castro et al., 2019; Trihinas et al., 2018). FaaS is a facet of serverless computing where applications can run server-side logic in stateless compute containers that can be event-triggered, ephemeral (may only last for one invocation), and fully manageable by a third party. Such a serverless paradigm is desirable for many event-based IoT applications, especially mission-intensive applications, as well as applications requiring energy efficiency and data delivery reliability (Gusev et al., 2019). The increased complexity of heterogeneous fog computing infrastructures poses management challenges to the DevOps of IoT applications, however (Chiang et al., 2016). In response, advanced cloud management tools and methods have started to emerge in an effort to automate infrastructure performance tuning and anomaly detection (Di Martino et al., 2019; Mahesh et al., 2011).

This paper reviews prominent fog frameworks that deploy and monitor applications that span over multiple clouds, fog and edge resources. It discusses associated challenges and proposes a novel fog architecture and framework for managing dynamically multi-cloud and edge resources in order to cope with the requirements of FaaS-enabled applications. Our research objective focuses on the development and evaluation of a framework to support the seamless deployment of fog computing applications on heterogeneous cloud, fog, and edge resources independently of underlying infrastructures while supporting the FaaS paradigm and providing auto-scaling capabilities.