Formulation and heuristic for the container image reconstruction problem

Using container images is a popular technique to share and deploy software by packaging it with all its dependencies. Software deployment with containers was an important enabler for cloud native computing and changed the way software development teams deliver software. Docker and other tools uses content-addressable storage to share identical layers across images and reduce storage usage. However, in practice some layers can be very similar but not equal, which reduces potential storage savings. In this article we propose the reconstruction of sets of container images to increase layers similarity and optimize storage usage. We formulate the image reconstruction problem as an integer linear model and propose an off-line constructive heuristic to optimize storage usage within the context of a software development team. We analyze our solution with 4 open-source applications with multiple containers and compare the results with a solver.