• Difficulty of managing dependencies
• Maximizing isolation and transparency
• Portability of computational environment
• Make extendibility and reuse easy
• Ease of use generally

• Size: VMs are large files which makes them impractical to store and transfer
• Performance: running VMs consumes significant CPU and memory

• A shipping container for the online universe: hardware-agnostic and platform-agnostic
• A tool that lets programmers neatly package software and move it from machine to machine.
• Released as open source in March 2013, a big deal on github: 18.6k stars, 3.8k forks

• dockerfiles: plain-text instructions to automatically make images
• containers: the active, running parts of Docker that do something
• images: pre-built environments and instructions that tell a container what to do.
• registry: open online repository of images, including many 'trusted builds'

• Security: it is possible for a hosted image to be written with some malicious intent
• Limited to 64-bit host machines, making it impossible to run on older hardware
• Does not provide complete virtualization but relies on the Linux kernel provided by the host
• On OSX and Windows this means a VM must be present (boot2docker installs VirtualBox for this)

• Install & start boot2docker
• docker pull <username>/<image_name> gets an existing image from registry
• eg. docker pull ubuntu notice there's no username here, because this is an 'official repo'
• after docker pull then docker run
• or simply docker run, which will pull, create and run in one step

• docker run -it ubuntu
• gets ubuntu and gives us a terminal for interaction
• docker run -dp 8787:8787 rocker/rstudio
• gets R & RStudio and opens port 8787 for using RStudio server in a web browser at localhost:8787 (linux) or 192.168.59.103:8787 (Windows, OSX)

• It is possible to use docker commit <container> to commit a container's file changes or settings into a new image
• But it is better to use Dockerfiles & git to manage your images in a documented and maintainable way
• A Dockerfile is a short plain text file that is a recipie for making a docker image

• FROM specifies which base image your image is built on (ultimately back to Debian)
• MAINTAINER specifies who created and maintains the image.
• CMD specifies the command to run immediately when a container is started from this image, unless you specify a different command.
• ADD will copy new files from a source and add them to the containers filesystem path
• RUN does just that: It runs a command inside the container (eg. apt-get)
• EXPOSE tells Docker that the container will listen on the specified port when it starts
• VOLUME will create a mount point with the specified name and tell Docker that the volume may be mounted by the host

• To build an image from a dockerfile: docker build --rm -t <username>/<image_name> <dockerfile>
• simple and moderately complex examples
• To send an image to the registry: docker push <username>/<image_name> You need to be registered at the hub bfore pushing

• Automated image build testing on a new commit to the Dockerfile
• Analogous to the travis-ci service, has a shield
• Requires a .circle.yml file in github repo, eg. https://github.com/benmarwick/1989-excavation-report-Madjebebe/blob/master/circle.yml
• Pushes new image to hub on successful complete of test
• And gives a badge to indicate test status

• The rocker project provides images that include R, key packages and other dependencies (RStudio, pandoc, LaTeX, etc.), and has excellent documentation on the github wiki
• I run RStudio server in the browser, with host folder as volume, very easy to use
• I store scripts on host volume because VC is simpler this way, but do development and analysis in container for isolation

docker run -dp 8787:8787 -v /c/Users/marwick/docker:/home/rstudio/ -e ROOT=TRUE rocker/hadleyverse

• -dp 8787:8787 gives me a port for the web browser to access RStudio
• -v /c/Users/marwick/docker:/home/rstudio/ gives me read and write access both ways between Windows (C:/Users/marwick/docker) and RStudio
• -e ROOT=TRUE sets an environment variable to enable root access for me so I can manage dependencies
• I can access the docker (Debian) shell via RStudio for file manipulation, etc. (or docker exec -it <container-id> bash)

• Choose your favourite from the registry
• the IPython project have a few images, and there are many user-contributed ones

• Amazon EC2 Container Service: docker clusters in the cloud (no registry)
• Google Compute Engine: has container-optimized VMs
• Google container registry: secure private docker image storage on google cloud platform
• Microsoft Azure supports docker containers (docker hub is integrated)

• http://arxiv-web3.library.cornell.edu/pdf/1410.0846v1.pdf
• http://sites.duke.edu/researchcomputing/tag/docker/
• https://rc.duke.edu/duke-docker-day-was-great/
• https://github.com/LinuxAtDuke/Intro-To-Docker
• http://reproducible-research.github.io/scipy-tutorial-2014/environment/docker/
• http://ropensci.org/blog/2014/10/23/introducing-rocker/
• https://github.com/wsargent/docker-cheat-sheet

Presentation written in R Markdown using ioslides

Compiled into HTML5 using RStudio & knitr

Source code hosting: https://github.com/benmarwick/UW-eScience-docker-for-reproducibility

ORCID: http://orcid.org/0000-0001-7879-4531

Licensing:

• Presentation: CC-BY-3.0

• Source code: MIT