Poster 112: Building Complex Software Applications Inside Containers
TimeThursday, 21 November 20198:30am - 5pm
DescriptionHigh performance computing (HPC) scientific applications require complex dependencies to operate. As user demand for HPC systems increases, it becomes unrealistic to support every unique dependency request. Containers can offer the ability to satisfy the users’ dependency request while simultaneously offering HPC portability across systems. By “containerizing” Model for Prediction Across Scales (MPAS, a large atmospheric simulation suite), we show that it is possible to containerize and run complex software. Furthermore, the container can be run across different HPC systems with nearly identical results (21 bytes difference over 2.1 gigabytes). Containers have the possibility to bring flexibility to code teams in HPC by helping to meet the demand for user defined software stacks (UDSS), and giving teams the ability to choose their software, independently of what is offered by the HPC system.