A comprehensive performance evaluation of container migration strategies

A comprehensive performance evaluation of container migration strategies

연구 분야: Software Development

학회: Computing

Containerization has emerged as a transformative technology in modern data centers, offering significant advantages in efficiency and resource management across diverse applications. While it is often associated with platforms like Docker and Kubernetes, its applications extend further and have been widely embraced by leading cloud providers such as Amazon Web Services, Microsoft Azure, and Google Cloud. Effective software resource migration, particularly container migration, is essential in large-scale data centers for managing server downtime, resource consolidation, and disaster recovery, especially in edge mobile computing contexts. Checkpoint/Restore in Userspace enhances container migration by allowing applications to freeze and save their state, facilitating seamless relocation. The study addresses the high cost-both financial and resource-related-of conducting real-world experiments to predict the migration performance of container migrations in real-world scenarios. This paper proposes stochastic Petri net models featuring an absorbing state and the other without. These models assess container migration strategies-Cold, PreCopy, PostCopy, and Hybrid-focusing on metrics such as migration total time (MTT), mean migration time, utilization, discard probability, and migration rate. The models consider the number of elements migrated simultaneously and the system’s parallel migration capacity. The absorbing state model also calculates the cumulative probability distribution function. A sensitivity analysis using the design of experiment was also conducted for the hybrid migration policy within the absorbing state model. The results indicate that the cold policy presents lower MTT in scenarios of a high migration arrival rate. At the same time, PostCopy maintains the lowest discard probability, making it suitable for high-demand scenarios.