Multi-Objective Reinforcement Learning Based Algorithm for Dynamic Workflow Scheduling in Cloud Computing

It is essential to consider the infrastructures of workflows as a critical research area where even slight optimizations can significantly impact infrastructure efficiency and the services provided to users. Traditional workflow scheduling approaches using heuristics may not be efficient due to the dynamic workloads and diverse resources of cloud infrastructure. Additionally, the resources at any given time have different states that must be considered during workflow scheduling. The emergence of artificial intelligence has made it possible to address the dynamics and diverse resources of cloud computing during workflow management. In particular, reinforcement learning enables understanding the environment at runtime with an actor and critic approach to make well-informed decisions. Our paper introduces an algorithm called Multi-Objective Reinforcement Learning based Workflow Scheduling (MORL-WS). Our empirical study with various workflows has demonstrated that the proposed multi-objective reinforcement learning-based approach outperforms many existing scheduling methods, especially regarding makespan and energy efficiency. The proposed method with the Montage workflow demonstrated superior performance compared to scheduling 1000 tasks, achieving a least makespan of 709.26 and least energy consumption of 72.11 watts. This indicates that the proposed method is suitable for real-time workflow scheduling applications.