Performance analysis for two-level parallel transient stability constrained optimal power flow

Transient stability constrained optimal power flow (TSCOPF) is one the most computational-intensive applications in power systems. Research efforts were made to utilize high performance computing (HPC) technology to parallelize and accelerate TSCOPF solving process, but challenges still exist in performance analysis to identify bottlenecks and improve efficiency on practical HPC platforms such as Beowulf clusters. Based on first two authors' previous work [8, 12] - two-level parallel TSCOPF with reduced-space interior point method (RIPM), a systematic performance analysis approach is demonstrated in this paper. Comprehensive performance analysis procedures - wall time analysis, MPI/OpenMP profiling and tracing - is employed in order to discover performance bottlenecks. State-of-the-art performance analysis software are employed to generate and visualize performance data, providing guidelines for parallel performance enhancement. Numerical results on a 2746-bus system show effectiveness of the proposed approach and relative low overhead in solving large-scale multi-contingency TSCOPF on a Beowulf cluster.