Real-Time, Cost-Aware Resource Scheduling in Multi-Cloud Systems using PPO-Based Reinforcement Learning

Shanmugam, Priya (2025) Real-Time, Cost-Aware Resource Scheduling in Multi-Cloud Systems using PPO-Based Reinforcement Learning. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (988kB) | Preview
Preview	PDF (Configuration Manual) Download (4MB) | Preview

Abstract

Scheduling and task placement strategies on heterogeneous platforms are most important in the age of cloud computing in order to optimize the use of cloud computing resources with the aim of achieving the Service-Level Agreements (SLAs) with the lowest possible operation costs. In this study, a real-time, intelligent scheduling system informed by Proximal Policy Optimization (PPO), a reinforcement learning algorithm is proposed in order to dynamically distribute computations in using multiple cloud providers including Amazon Web Services (AWS), Microsoft Azure, and Google Cloud Platform (GCP). The system emulates the workload in the real world on fog-cloud setting using iFogSim and the crucial workload figures to create the vectors of task states, such as the CPU requirements, memory usage, execution time, and SLA level. The PPO agent is trained with lifetime reward (contingent on SLA and cost-optimality) signals on how to select the optimal choices of cloud. It is compared with traditional (Round Robin (RR), First-Come-First-Serve (FCFS)) and reinforcement learning based scheduling policies such as Deep Q-Network (DQN), and Advantage Actor Critic (A2C). The outcomes demonstrate that PPO has increased SLA satisfaction score and a lower average CPU cost in every case, which indicates its efficiency in real-time decision-making frameworks. A single Streamlit dashboard was devised to process the results of the scheduling along with the performance of the system, whereas explainability aimed at SHAP was utilized to understand the choices of PPO. The study presents an explainable intelligent and modular solution that involves task scheduling in the cloud, providing a scalable philosophy to solving resource allocation in multi-clouds.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Arun, Shreyas Setlur UNSPECIFIED
Uncontrolled Keywords:	Cloud Optimization; Cost Prediction; AI Scheduling; Fault Tolerance; Cloud Simulation
Subjects:	Q Science > QH Natural history > QH301 Biology > Methods of research. Technique. Experimental biology > Data processing. Bioinformatics > Artificial intelligence Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Artificial intelligence T Technology > T Technology (General) > Information Technology > Cloud computing
Divisions:	School of Computing > Master of Science in Cloud Computing
Depositing User:	Ciara O'Brien
Date Deposited:	31 Mar 2026 09:17
Last Modified:	31 Mar 2026 09:17
URI:	https://norma.ncirl.ie/id/eprint/9266

Actions (login required)

View Item