Accelerating the FlowSimulator: Strong scaling analyses and improvements of simulation toolchains in HPC

Kurzfassung

Multidisciplinary simulation toolchains are crucial to aircraft design and certification, yet the overall performance is often overlooked relative to the individual solvers. As a result, unexpected bottlenecks may remain undetected, particularly when workflows are scaled to large numbers of processors. This work investigates the scalability of steady aeroelastic workflows, combining strong-scaling analyses, timing logs, and automatic Python-layer instrumentation with Score-P.

Two representative workflows implemented within the \flowsim framework and using the CFD solver CODA are examined. Profiling reveals that simulation components typically considered non-critical may dominate the runtime. Specifically, bottlenecks are identified in a result output writing operation and a logging utility function. After applying custom solutions targeted to alleviate the detected bottlenecks, the overall runtime of both workflows could be significantly reduced, up to 3x speed-up.

These results demonstrate that scalable multidisciplinary workflows require careful analyses to identify simulation bottlenecks and performance optimization of components considered critical (e.g. CFD) may not suffice.