A predictive energy-aware scheduling strategy for scientific workflows in fog computing

Abstract

Fog computing paradigm provides diverse processing resources and storage close to the edge of Internet of Things (IoT) networks. Workflow scheduling is an open issue in fog computing as it addresses the deployment of IoT workflow tasks on fog cells. Deploying workflow tasks on fog cells in an inefficient manner can have negative consequences. It can lead to bandwidth waste, resource depletion, and significant operating expenses. Additionally, the majority of scheduling evolutionary algorithms suffer from premature convergence and sticking in local optima. In this paper, a predictive energy-aware scheduling framework is proposed as a MAPE-K control model consisting of the four Monitor, Analyzer, Planner and Executer components with a shared Knowledge base in fog computing. First, a prediction method applying an Adaptive Network-based Fuzzy Inference System (ANFIS) into an analyzer component is introduced to predict future resource load. Second, a resource management strategy based on predicted resource load is presented to reduce energy consumption. Third, the Improved Ant Lion Optimizer (ALO) and weighted Grey Wolf Optimizer (GWO) are combined into a planner component called I-ALO-GWO for workflow scheduling. In the end, the decisions made in the previous steps are executed on the fog cells. The effectiveness of the I-ALO-GWO evolutionary algorithm has been tested on the IEEE CEC2019 benchmark functions. In addition, the proposed approach is evaluated practically under several famous scientific workflows using the iFogSim tool. Experimental results indicate that I-ALO-GWO improves the makespan up to 18 %, the energy consumption up to 17 %, the total execution cost up to 11 % and 26 % in terms of efficiency in comparison with the second-best results.