A Comparison of Container Systems for Machine Learning Scenarios: Docker and Podman

With the rising needs of corporations and their users, as well as the ever-increasing complexity of applications, Container engines and technologies have become a pivotal point in every modern technological infrastructure. This has standardized execution runtimes and when combined with modern virtualization techniques, it has lowered costs and given access to the power of cloud computing to many more people. The aim of this paper is to compare two of the most popular container engines to see what differences exist in performance and architectural levels between the so-called “drop-in” replacements. To ensure consistency and replicability of testing, we standardize the benchmark environment with a custom-built tool that describes differences among container engines in the millisecond range. On this basis, we also present quantified results and compare differences with real-world metrics and pricing. From our results, we observe a small but not insignificant difference among the container systems, in favor of Docker. Our results are quantified with real-world pricing of cloud resources to figure out how much more costly Podman would be in a cloud deployment.