SeisSol Performance Analysis with Omni-Path Networking

This white paper presents an analysis of SeisSol's performance and scalability using Cornelis Networks Omni-Path on 3rd Generation Intel Xeon Scalable processors. It details the evaluation of an industry-standard benchmark case, TPV33, which simulates seismic wave behavior. The study decomposes the benchmark into three resolutions and conducts scaling studies on a cluster of up to 32 dual-socket Intel Xeon Platinum 8358 nodes. Performance metrics such as job throughput, effective CPU utilization, and MPI profiling are documented. The paper outlines the setup for simulations, including the use of hybrid OpenMP/MPI versions of SeisSol and the configuration of MPI communications. Results indicate varying scaling efficiencies across different mesh resolutions, with higher resolutions yielding better performance. The findings suggest that low-resolution simulations provide faster turnaround, while high-resolution simulations optimize resource utilization. The paper concludes with recommendations for performance tuning based on observed MPI communication patterns.