Multi-Agent Reinforcement Learning based Predictive Scheduling and Resource Allocation in AWS Edge-Cloud Kubernetes Environments

Kantipudi, Radha (2025) Multi-Agent Reinforcement Learning based Predictive Scheduling and Resource Allocation in AWS Edge-Cloud Kubernetes Environments. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (2MB) | Preview
Preview	PDF (Configuration Manual) Download (2MB) | Preview

Abstract

The optimization of task scheduling and placement in Kubernetes based containerized cluster environments presents serious difficulties imposed by the heterogeneous constraint on resources, the dynamic nature of workloads, and compounding requirements on performance goals. Although the performance of the existing single-agent deep reinforcement learning (DRL) methods such as Proximal Policy Optimization with Least Response Time (PPO-LRT) on the issue of load balancing has improved, it does not satisfactorily consider a distributed decision-making environment, nor does it support the multi-objective optimization requirements of modern containerized systems. This research presents a multi- agent deep reinforcement learning (MADRL) system that extends to the existing traditional single-agent systems to add additional specific agents with the purpose of optimising response time, balancing load and privacy protection. The framework incorporates LSTM-based predictive scaling to undertake proactive resource management and performs intelligent node selection decisions depending on workload characteristics. The synthetic workload evaluation using PolybenchC-inspired benchmarks demonstrates that the experimental performance is better than the baseline Kubernetes scheduler and current PPO-LRT techniques, in terms of providing average response-time competitiveness, increased resource utilization efficiency (achieving 100% node utilization vs 75% for baseline), and better load balancing across cluster nodes while without violating the privacy requirements of sensitive task placement.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Gupta, Shaguna UNSPECIFIED
Uncontrolled Keywords:	deep reinforcement learning; edge-cloud computing; predictive resource management; scheduling; Kubernetes
Subjects:	T Technology > T Technology (General) > Information Technology > Cloud computing Q Science > QA Mathematics > Electronic computers. Computer science > Computer Systems > Computers > Electronic data processing > Electronic data processing--Distributed processing > Edge computing T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science > Computer Systems > Computers > Electronic data processing > Electronic data processing--Distributed processing > Edge computing T Technology > T Technology (General) > Information Technology
Divisions:	School of Computing > Master of Science in Cloud Computing
Depositing User:	Ciara O'Brien
Date Deposited:	26 Mar 2026 14:18
Last Modified:	26 Mar 2026 14:31
URI:	https://norma.ncirl.ie/id/eprint/9225

Actions (login required)

View Item