Evaluating the unavailability of interconnected power and communication networks with open-source tools on a petascale cluster

In reliability engineering, unavailability is defined as the probability that a system is not operational at a given point in time, typically due to failure or maintenance. A critical gap in reliability analysis by systematically evaluating the time-dependent unavailability of real interconnected power and communication networks in the Czech Republic is addressed in this work. These networks are modelled as acyclic graphs using open-source R packages. Unlike previous studies relying on commercial tools, the research presented here offers a novel, reproducible, and scalable framework. The main contribution lies in the innovative application and benchmarking of ftaproxim, an R package based on proxel simulation, which models ageing components during their entire life using various probabilistic distributions. This approach contrasts with traditional tools such as the FaultTree package, which are limited to asymptotic unavailability analysis. Here presented work evaluates both R packages on a real infrastructure model and compares their performance and computational efficiency on the Barbora supercomputer cluster against commercial software (Matlab). It is demonstrated how ftaproxim's tolerance and time-step parameters can be tuned for robust computational efficiency and accuracy, an aspect previously unexplored. The results of the presented study show that unavailability computations can be completed in approximately 5 h under optimal settings, with absolute errors ranging from 1.0 × 10 −4 to 9.6 × 10 −4 when compared to commercial solutions. This integrated approach, combining open-source tools, high-performance computing scalability, and interconnected infrastructure modelling, represents a significant advancement in the field. Open-source code, compatible with the petascale system Karolina, which is installed at the Czech IT4Innovations National Supercomputing Center in Ostrava, is provided to support reproducibility and future benchmarking.