We conducted joint work with Université de Neuchâtel on improving the handling of Docker container images in the increasingly heterogeneous hardware environments. We propose to (1) finer-grained incorporate hardware dependency information in the image metadata, (2) leveraging heuristic analysis techniques to populate such information at large scale (although of course preferring properly curated metadata), and (3) improving the tool support around container creation from images. The work has led to new tools like hdocker and heuristic analysis rules. Furthermore, to underline the need for such a solution, we have been conducting a long-term tracking over fourteen now seventeen months of selected subsets of registered Docker container images.
This work has been accepted by the 21st International Conference on Distributed Applications and Interoperable Systems (DAIS 2021). Ahead of the event we already provide the collected data and code. Have a look!