02:05 PM - 02:45 PM (40 min)
Matt Anderson (Idaho National Laboratory)
Abstract: Containers in scientific computing can make it much easier to deploy an application to multiple different supercomputing systems without having to rebuild the application. They can also reliably reproduce archival simulations that use codes that are no longer supported or have changed so much that they can’t even be compiled against a modern software stack. This talk discusses how to prepare a container where mpirun is used to launch the container for running a simulation at system scale. In this modality, a user simply downloads a prepared container and can run a simulation at scale on any supercomputer without having to build the application for that system. Archival capabilities of this operation modality are also discussed.