Enhancing Convergence of Live VM Migration with Dynamic Workloads

Abstract

Abstract Live Virtual Machine (VM) migration is a critical process in cloud computing, enabling workload balancing, resource optimization, and fault tolerance. However, selecting the most suitable migration technique is challenging due to the dynamic nature of workloads. Traditional migration methods—pre-copy, post-copy, and hybrid approaches—often struggle with increased Total Migration Time (TMT), downtime, and performance degradation, especially under varying workload conditions. This research proposes ProMig, a novel migration decision framework that leverages Markov models to predict future workload behavior and select the optimal migration technique while adapting to the dynamic nature of workloads. CPU usage, memory usage, and network usage are used as key resource indicators to define system states and anticipate workload transitions. The evaluation reveals that ProMig consistently minimizes TMT by adapting to both stable and fluctuating workload patterns. It has 87% accuracy for the total migration time reduction. Furthermore, certain workloads, such as those with high dirty page rates but low memory usage, present additional migration challenges. Future work will explore integrating dirty page rate as a resource metric and extending the model to multi- VM migration scenarios for enhanced scalability and e!ciency. Through intelligent prediction and adaptive decision-making, ProMig improves migration e!ciency, reduces downtime, and ensures optimal resource utilization in cloud environments.