Contain This, Unleashing Docker for HPC

Containers are a lightweight virtualization method for running multiple isolated Linux systems under a common host operating system. Container-based computing is revolutionizing the way applications are developed and deployed. A new ecosystem has emerged around the Docker platform to enable container based computing. However, this revolution has yet to reach the HPC community. In this paper, we provide background on Linux Containers and Docker, and how they can be of value to the scientific and HPC community. We will explain some of the use cases that motivate the need for user defined images and the uses of Docker. We will describe early work in deploying and integrating Docker into an HPC environment, and some of the pitfalls and challenges we encountered. We will discuss some of the security implications of using Docker and how we have addressed those for a shared user system typical of HPC centers. We will also provide performance measurements to illustrate the low overhead of containers. While our early work has been on cluster-based/CS-series systems, we will describe some preliminary assessment of supporting Docker on Cray XC series supercomputers, and a potential partnership with Cray to explore the feasibility and approaches to using Docker on large systems. Keywords-Docker; User Defined Images; containers; HPC systems