A cooperative multi‐agent offline learning algorithm to scheduling IoT workflows in the cloud computing environment

Abstract

Regarding the problem of workflow scheduling in cloud environments, users want the workflow to be processed at a suitable time while cloud providers want to increase resource utilization. This article proposes a cooperative multi-agent offline learning algorithm called CMOL for minimizing makespan and energy consumption. This algorithm schedules a workflow that is represented by a directed acyclic graph (DAG) and assigns them to virtual machines (VMs). Multiple parallel agents interact and cooperate based on an algorithm in three steps of research, improvement, and selection to meet the imposed constraints of deadline and energy. Depending on the number of DAG levels, there is the same number of specialist agents who use strategies to create a Pareto feasible solution and simultaneously gain experience in the first two steps. The parallel agents exploit the extracted knowledge to improve the solution obtained by ensembling their experience in the selection step. To compare the efficiency of CMOL, two algorithms based on multi-agent systems and one algorithm based on single-agent are developed. The performance of the four algorithms is investigated on different real-world workflows and compared on various sizes. Computational results reveal the competitiveness of CMOL and its relative superiority compared with others.