A Reinforcement Learning Based Load Balancing Approach for Energy Efficiency and QoS Optimization in Live Cloud Streaming

Shanmugam, Ganesh Ram (2025) A Reinforcement Learning Based Load Balancing Approach for Energy Efficiency and QoS Optimization in Live Cloud Streaming. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (802kB) | Preview
Preview	PDF (Configuration Manual) Download (510kB) | Preview

Abstract

Traditional load balancing methods work with fixed rules or set limits, which makes it difficult to handle unpredictable workloads in real time. This results in high latency, resource wastage, and a poor user experience. To address these challenges, this research focused on using Reinforcement Learning (RL) to make smarter autoscaling decisions in live cloud streaming environments. Two RL methods—Proximal Policy Optimization (PPO) and Deep Q-Network (DQN)—were used to dynamically allocate virtual machine (VM) resources based on live workload metrics. The live cloud streaming simulation scenario was developed using the CloudSim Plus tool with Java. Three different types of video streaming scenarios (light, medium, and heavy workloads) were simulated to collect the dataset. The standard Gym API was used to design a compatible environment in Python for training RL agents with the synthetic dataset generated from the simulation. The trained model was integrated with a prediction server to make real-time decisions by selecting the best VM configurations during workload execution. For evaluation, three types of workloads were simulated, and the RL agents’ results were compared with a random autoscaling model. The results showed that both RL models, PPO and DQN, significantly outperformed the random baseline, reducing latency by 74%, saving energy by 67%, and increasing throughput by 88%. RL mainly improved response speed and data delivery, while only minor improvements were observed in jitter. These results demonstrate the potential of RL-based load balancing as an energy-efficient strategy for live cloud streaming without affecting the QoS for users.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Kazmi, Aqeel UNSPECIFIED
Uncontrolled Keywords:	Cloud computing; Reinforcement Learning; PPO; DQN; CloudSim Plus; Quality of Service; Live video streaming
Subjects:	T Technology > T Technology (General) > Information Technology > Cloud computing Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning
Divisions:	School of Computing > Master of Science in Cloud Computing
Depositing User:	Ciara O'Brien
Date Deposited:	31 Mar 2026 09:10
Last Modified:	31 Mar 2026 09:10
URI:	https://norma.ncirl.ie/id/eprint/9265

Actions (login required)

View Item