The Impact of Containerization Techniques on AI Training and Inference Performance

Evaluate how different containerization techniques affect the performance, resource utilization, and reproducibility of AI training and inference workloads.

The candidate will: 1. Review containerization approaches used for AI workloads. 2. Select representative container technologies and configurations. 3. Define training and inference benchmarks. 4. Implement and execute controlled experiments. 5. Measure performance, overhead, and resource utilization. 6. Analyze the results and identify relevant trade-offs.

3 to 9 months according to the student's background.

• Knowledge of Linux
• Good knowledge of Python
• Familiarity with containers and benchmarking

1. Sun, M., Guo, Y., Zhang, H., Cao, W., & Yuan, M. (2021, June). Performance comparison of multiple containers running artificial intelligence applications. In Journal of Physics: Conference Series (Vol. 1948, No. 1, p. 012005). IOP Publishing.