The structure of the ambient solar corona and solar wind plays a key role in space weather at Earth and throughout the solar system. Solar eruption impacts can cause disruptions in radio transmissions, damage satellites, harm astronauts, and severely impact power transmission grids. Therefore, accurate and fast modeling is critical to the study and forecasting of such solar activity.
CORona-HELiosphere (CORHEL) is a state-of-the-art modeling suite used to generate solar wind and solar eruption simulations from the Sun out into the heliosphere for research and space weather applications. It contains a multitude of model options from inexpensive potential fields to computationally heavy thermodynamic magnetohydrodynamic (MHD) simulations.
We propose to improve performance and reduce the cost of running CORHEL by adding GPU-acceleration to its model codes using OpenACC. OpenACC allows the GPU implementations to be done through the use of specialized comments, leading to single source codes that can be compiled to run on either multiple GPUs or CPUs. When completed, the GPU-enabled CORHEL suite will be able to run low- to medium-resolution simulations in a single multi-GPU workstation, and perform the largest runs on a few multiple multi-GPU compute nodes. This is expected to greatly reduce costs, improve efficiency, and be a significant step forward for performing solar wind and solar eruption modeling in a timely manner.
At the completion of Phase I, all models in CORHEL will be GPU-accelerated with the exception of the thermodynamic MHD model of the corona. New installation scripts and documentation will be added, and extensive validation performed. Phase II will complete the GPU implementation by adding GPU support to the thermodynamic model, and culminate in a validated, ready-to-deploy, GPU-enabled CORHEL package capable of modeling the solar corona, solar wind, and solar eruptions.
Adding GPU-acceleration to CORHEL (currently hosted at the Community Coordinated Modeling Center (CCMC)) would reduce costs and improve performance for solar wind and solar eruption simulations. Runs on a multi-GPU workstation could be deployed at remote stations (e.g. sea vessels, space stations, etc.) and used for space weather applications independent of ground-locked computing centers. This project also serves as an example of GPU-accelerating legacy science codes, opening the door for adding GPU-computing to other such codes used at NASA.
Solar wind and eruption models are of interest to many government and commercial organizations that rely on satellites, aircraft, and space travel. Besides CCMC, CORHEL has been delivered to AFRL and receives support from NOAA. Organizations may show great interest in obtaining a pre-configured CORHEL multi-GPU workstation, allowing low cost simulations independent of supercomputer allocations.