Task Scheduling in Cloud Computing Using Predictive Proximal Policy Optimization
Year: 2026
Type: conference-paper
Abstract: Task scheduling in cloud computing is a persistent problem driven by dynamic workloads, heterogeneous virtual machines (VMs), and stringent Quality of Service (QoS) requirements. Conventional heuristic and metaheuristic schedulers often lack adaptability, leading to increased makespan, load imbalance, and Service Level Agreement (SLA) violations. This paper presents a Proximal Policy Optimization (PPO)–based, machine learning–assisted task scheduling framework for cloud environments. The scheduling problem is modeled as a Markov Decision Process (MDP), where a PPO agent learns an adaptive policy using state representations enhanced by supervised predictors for execution time, energy consumption, and VM reliability. A feasibility-aware action space and a multi-objective reward function are employed to ensure scalable and balanced scheduling decisions. The proposed approach is evaluated using trace-driven workloads in a discrete-event simulation environment. Experimental results show that the proposed method consistently outperforms Genetic Algorithm (GA) and Cuckoo Particle Swarm Optimization (PSO) baselines in terms of makespan, SLA violation rate, and load imbalance under increasing workload intensity.
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Keywords: task scheduling, cloud computing, proximal policy optimization, reinforcement learning, Markov decision process, virtual machines, quality of service, service level agreements, makespan, load balancing, execution time prediction, load forecasting, multi-objective reward function, dynamic workloads
Open Access status: hybrid
Funder: European Commission
OpenAlex ID: W7167242572
