Spin: A Docker-Based Platform for Deploying Science Gateways at NERSC

This poster accompanies our demonstration of Spin, a Docker-based, on-premise cloud platform at NERSC that enables researchers to design, build, and manage their own science gateways and other services using container technology.

Spin is designed to help users overcome some of the most common challenges in science gateway development and maintenance: poor portability and unreliable reproducibility of application code; inflexibility of science gateway hosting infrastructure that results in slow initial provisioning and cumbersome processes for making configuration changes; various performance and scaling issues; and inconvenient access to HPC resources, to name just a few.

With Spin, researchers can architect their science gateways with lightweight, portable Docker containers and deploy them to managed servers instantly. The Docker Build-Ship-Run philosophy depicted provides a coding, testing, and release workflow model that eases the software development and release cycle and paves the way for continuous integration and deployment of complete service architectures. Docker container technology directly addresses the portability, reproducibility, and flexibility needs, ensuring that applications behave consistently between an engineer's laptop and the robust server environment. Problems of performance and scaling and the difficulty of integrating with HPC resources are addressed through Spin's technical architecture, which is shown within the larger contect of the NERSC environment. The key Docker concepts underlying all of these capabilities - container images, orchestration systems such as Kubernetes, and the idea of the running container itself - are listed and defined.

The ease and speed with which science gateways can be instantiated using Docker container technology opens exciting possibilities for what's next. Given compatible infrastructure at facilities, it becomes practical (and even trivial) to share common code bases, trigger on-demand deployment from schedulers or workflow managers, or create broadly-distributed systems.