Multi-objective mechanisms for dynamic resource adaptation in virtualized network environments

Abstract

The growth of virtualization technologies and cloud solutions has fundamentally transformed the management and utilization of computing resources. These advancements have empowered cloud providers to offer scalable, flexible, and cost-effective services to their customers. However, with the ever-increasing demand for cloud services, effective resource management has become a critical concern. To address this challenge, various strategies, such as workload consolidation, resource utilization prediction, and resource scaling and migration techniques, have been proposed to optimize resource management in virtualized environments. This thesis presents a pioneering set of adaptive multi-objective resource management techniques designed to maximize resource utilization, reduce energy consumption, and ensure compliance with Service Level Agreement (SLA) requirements. Considering the inherent challenges of workload variability, application diversity, and conflicting optimization goals, the research encapsulates several significant contributions, each focusing on a specific aspect of resource management. First, the dynamic resource adaptation problem within Network Function Virtualization (NFV)-cloud environments is explored, incorporating resource scaling and migration strategies for service function chains (SFCs). This resource allocation problem is tackled from a novel perspective, formulated as Integer Linear Programming (ILP) model and developed to generate optimal solutions. Second, innovative multi-objective decision-making metaheuristic algorithms, based on NSGAII, CRO, and PSO, are proposed to enable real-time resource adaptation with sub-optimal solutions. Third, proactive resource reallocation is investigated through the development of a multi-resource and multi-step-ahead workload prediction model. By integrating the Kalman filter and support vector regression, this model accurately anticipates host resource utilization, including CPU, memory, and bandwidth. Fourth, building upon this predictive capability, an optimized consolidation approach is introduced, incorporating proactive host state estimation strategies for overload and underload detection. To validate the effectiveness of the proposed techniques, extensive experiments are conducted employing diverse datasets such as Planetlab, Materna, and Bitbrains, coupled with the Cloudsim simulator. The experimental results demonstrate the potential of these techniques to enhance resource management in virtualized environments.

Date	8 Aug 2023
Original language	American English
Awarding Institution	École de technologie supérieure
Supervisor	Nadjia Kara (Supervisor) & Aris Leivadeas (Co-supervisor)