Reproducible research: Recording environments

Containers

• Containers can be built to bundle all the necessary ingredients (data, code, environment).
• A container provides operating-system-level virtualization, sharing the host system’s kernel with other containers.
• Popular container implementations are Docker and Singularity.

Docker

• Available for most common operating systems.
• A mechanism to “send the computer to the data” when data is too large or too sensitive to travel over network.
• DockerHub is a platform to share Docker images (stored in repositories - similar to a Git repository).
• Public Docker images available on DockerHub.

Singularity

• Singularity is aimed at scientific community and to run scientific workflows on HPC resources.
• Docker images can be converted into Singularity images.

Container vs. image vs. recipe (Dockerfile)

• Image is like a blueprint. It is immutable.
• Container is an instance of an image.
• Dockerfile is a recipe which creates a container based on an image and applies small changes to it.

Pros and cons of containers

Containers are popular for a reason - they solve a number of important problems:

• Allow for seamlessly moving workflows across different platforms.
• Much more lightweight than virtual machines.
• Eliminates the “works on my machine” situation.
• For software with many dependencies with it turn own dependencies possibly the only (?) way to preserve the computational experiment for future reproducibility.

However, containers may also have some drawbacks:

• Containers can have security vulnerabilities which can be exploited.
• Can be used to hide away software installation problems and thereby discourage good software development practices.
• It may not be clear whether to record the environment in the image part or the recipe part.

Discussion: reproducibility aspects of container images

• Do you think containers contribute to reproducible research?
• Do you see a use case for your own work?