Adaptive parallel time integration improves computing resource use
Adaptive Parallel-in-Time Integration with Dynamic Resource Management
Computational Engineering, Finance, and Science
Summary
Simulations like weather forecasting need lots of computing power and often run into limits when trying to do many calculations at once in space. The authors worked with a method called PFASST that allows several time steps to be processed at the same time, but it usually needs changing how many steps it works on as the simulation runs. They combined this with dynamic resource management, meaning the computer can decide on the fly how much power to use and how many steps to calculate in parallel. This helps the simulation run faster and use resources more efficiently because it adapts to how well the calculations are converging at different times.
What this means in practice
- •For weather simulation teams: Manage computing resources dynamically to improve speed and efficiency in weather forecasting that uses parallel-in-time integration methods.
- •For high performance computing teams: Optimize the use of supercomputing resources in climate and physics simulations by adapting the number of parallel timesteps and compute power during runtime.
Authors
Keerthi Gaddameedi, Dominik Huber, Martin Schreiber, Hans-Joachim Bungartz, Valentina Schüller, Tobias Neckel, Martin Schulz
Abstract
As computational resources continue to grow, the strong-scaling limitations of spatial parallelism motivate the pursuit of additional concurrency in the temporal dimension, particularly for applications with hard time constraints, such as weather and climate simulations. The Parallel Full Approximation Scheme in Space and Time (PFASST) is a parallel-in-time method based on Spectral Deferred Corrections (SDC). It computes multiple timesteps concurrently by coupling fine- and coarse-grid SDC sweeps using multigrid Full Approximation Scheme (FAS) corrections. However, PFASST's convergence is often problem-dependent, demanding a variable number of parallel timesteps and, hence, computing resources at different times throughout the simulation. Dynamic Resource Management (DRM) provides a remedy for this challenge by enabling the adaptive adjustment of computational resources and algorithmic parameters at runtime. In this work, we present our novel approach to extending PFASST with DRM, which enables (a) dynamic adaptation of computing resources, (b) adaptive selection of the number of PFASST iterations based on local convergence behavior, and (c) coupling of these two adaptations into a single resizing strategy. With this approach, we demonstrate for the first time that optimal configurations can be identified in real time for each application, rather than relying on static allocation. Furthermore, we show that convergence-informed tuning of PFASST improves resource utilization and convergence efficiency.