EdgeWorkflow: One click to test and deploy your workflow applications to the edge

Abstract

In recent years, edge computing has become the ideal computing paradigm for various smart systems, such as smart logistics, smart health and smart transportation. This is due to its advantages including fast response times, energy efficiency and cost effectiveness over conventional cloud computing platforms. However, running complex computational scientific workflow tasks is still a very challenging issue at the edge, due to its typical three-layered computing environment consisting of an end device layer, an edge server layer, and a cloud server layer. A large number of recent studies have proposed different solutions for optimizing such computing resource management problems in an edge computing environment. However, since evaluation of most such studies is conducted through simulation, the effectiveness cannot be guaranteed in a real world environment. Therefore, to advance research on efficient execution and deployment problems for real world workflow applications using edge computing, an open-source edge workflow management system with comprehensive empirical evaluation capabilities is urgently required. This paper presents the first edge workflow system (named EdgeWorkflow) that is able to deploy user-created workflow applications to a real-world edge computing environment with “one-click” after optimizing the configuration with the simulation tool. With the aid of EdgeWorkflow, the user can automate the generation of specific edge computing environments, easily model and generate executable workflow applications with a visual modelling tool, effectively select various resource management methods included in the systems or apply their own resource management and task scheduling algorithms, efficiently monitor the statuses of computational tasks and obtain comprehensive reports on the execution results (such as those regarding time, cost and energy). We use an edge computing-based unmanned aerial vehicle (UAV) last-mile delivery system as a real-world case study, and a number of representative scientific workflows are employed for our experiments. Our experimental results show that EdgeWorkflow can effectively evaluate the performance of different resource management and workflow task scheduling algorithms and efficiently deploy and execute user-defined scientific workflow applications to user-specified edge computing environments.