| Abstract: |
Efficient resource allocation and task scheduling are fundamental challenges in cloud computing because of dynamic workloads, heterogeneous virtualized resources and competing requirements for performance, cost, energy efficiency and Quality of Service (QoS). Traditional static and heuristic scheduling techniques frequently fail to adapt effectively to rapidly changing cloud environments. Artificial Intelligence (AI), particularly Machine Learning (ML), Deep Learning (DL) and Deep Reinforcement Learning (DRL), offers an opportunity to develop adaptive scheduling systems capable of learning from workload patterns and continuously optimizing resource allocation decisions. The present study develops a conceptual AI-driven adaptive resource allocation framework for cloud-based systems. The proposed framework combines workload monitoring, state representation, predictive analytics and reinforcement-learning-based scheduling to dynamically allocate computing resources according to workload conditions. The optimization objective incorporates execution time, resource utilization, energy consumption, cost and Service Level Agreement (SLA) violations. Recent studies demonstrate increasing effectiveness of DRL and hybrid AI models in multi-objective cloud-resource optimization. The paper identifies the limitations of conventional approaches, reviews recent developments and proposes an integrated architecture suitable for experimental validation using CloudSim/CloudSim Plus and real-world cloud workload traces. The proposed approach contributes to the broader development of an AI-driven analysis and optimization framework for cloud-based systems. |
