Authors:

Hsin-Yu Tsern

Date Created:

Not specified

Course Title:

Professor:

Not specified

About Paper:

Many EM-multiphysics simulation algorithms are parallel-friendly, and traditional parallel solvers utilize CPU-based MPI and OpenMP to accelerate simulations through distributed memory and multithreading. However, algorithms that rely on static resource partitioning or exhibit poor resource coordination often suffer from communication bottlenecks and suboptimal CPU or memory utilization. In addition, many parallelization approaches require substantial code modifications, making them difficult to integrate with existing solvers. Moreover, while GPU acceleration has become increasingly important, many existing parallel frameworks do not take advantage of heterogeneous computing resources. In this project, a general hybrid parallelization framework- combining MPI, OpenMP, and GPU acceleration-is proposed and designed to dynamically manage computational resources and optimize memory usage. The framework successfully minimizes idle resources and overlaps communication with computation, significantly improving overall efficiency. Designed with flexibility in mind, the framework abstracts parallelization strategies from solver implementation, requiring only minimal code changes to integrate with existing solvers and can be applied to a wide range of problems. It also seamlessly incorporates heterogeneous accelerators-such as GPUs-further enhancing performance across diverse platforms. Preliminary benchmarks demonstrate strong scalability, accuracy comparable to the sequential algorithm, and reduced runtime on cluster environments by several orders of magnitude. Future extensions will integrate AI-driven algorithms with other emerging methods, enabling efficient solutions for increasingly complex real-world EM-multiphysics challenges. Keywords: Hybrid Parallelization; Em-Multiphysics Simulation; Dynamic Resource Management; Heterogeneous Computing; GPU

Source:

Purdue University / 2025

Topics:

No topics listed

Co-authors:

Hsin-Yu Tsern