AI-Based Workload Balancing for Distributed Systems

Abstract

The Artificial Intelligence (AI) has become a disruptive enabler in the distributed systems workload balancing addressing the long-term issue related to the distribution, the heterogeneity, the uncertainty, and the availability of the resource dynamically. The traditional methods of workload balancing, including fixed partitioning, round-robin, and load-sharing based on heuristics, are not usually applicable to the modern distributed systems based on the cloud computing architectures, edge/fog systems, Internet of Things (IoT) systems, and other systems with the intensive usage of data sizes. In this paper, workload balancing methods in distributed systems based on AI are undertaken systematically and in depth as experienced in their theoretical background, architectures, algorithms and performance implications. The paper expounds on machine learning, deep learning, reinforcement learning, and hybrid AI techniques adopted in the intelligent workload allocation, task migration, and adaptive resource management. An elaborate literature review is performed, including the traditional load balancing approaches and their development to the AI-based ones. The suggested approach describes an AI-based workload balancing model and model that includes the workload prediction, model of system state, smart decision-making, and continuous learning. Thousands of experiments and discussions prove that AI-based workload balancing is effective regarding a shorter response time, better resource consumption, a higher scale, and fault tolerance. Lastly, the paper identifies open research opportunities and future directions, including the topics of explainable AI, federated learning, and energy-conscious scheduling in the next-generation distributed systems.