Parallel algorithm design and performance optimization in computational physics
Xin Hua Zhou · IET conference proceedings. · 2026
In the field of computational physics, large-scale numerical calculation and complex data processing tasks are highly parallel, and the computational efficiency can be significantly improved through parallel calculation. Firstly, this paper introduces the basic principles of parallel algorithm design, including load balancing, reducing communication overhead and ensuring data consistency, and expounds two design methods of data parallelism and task parallelism in detail. By adopting divide-and-conquer strategy and block strategy, combined with MPI non-blocking communication technology, the task is efficiently parallelized. In terms of performance optimization, this paper puts forward some strategies, such as communication optimization, computing kernel optimization and memory access optimization. Through Halo aggregation communication, computing-communication overlap, cyclic block, SIMD vectorization and other technical means, the efficiency of parallel computing is significantly improved. Taking the parallel optimization of two-dimensional unsteady heat conduction equation as an example, this paper shows the implementation process and effect of the optimization strategy in detail. The experimental results show that, after comprehensive application of various optimization strategies, the speedup ratio is 6.49 times in 256 processes, and the communication ratio is reduced to 8.7%. With the increase of the number of processes, the optimization strategy shows nearly linear scalability in large-scale simulation, which verifies its effectiveness. The research in this paper provides theoretical guidance and practical reference for parallel algorithm design and performance optimization in computational physics, and is of great significance for improving the efficiency of solving computational physics problems.