Predicting the execution time of COSMO weather forecast models

This data set is the work of E. Di Giacomo Master Thesis at the University of Bologna.

Predicting the execution time of a numerical weather forecast model is a complex task. Generally, these models simulate the evolution of atmospheric weather and they are typically used for the production of weather forecasts, one or multiple times a day. Given their computational complexity, they require large computing capabilities, such as High Performance Computing systems. In these systems, job scheduling and resource allocation are carefully managed to optimize the usage of the finite and expensive hardware resources; in particular, several allocation and related pricing decisions are based on estimates of the duration of the application submitted, such as the execution time of weather forecast models.
A reliable prediction of execution time allows for a better management of the overall system, an improved planning of the model execution, as well as the identification of possible anomalies during the execution, thus providing great benefits to both system administrators and users.

This data set regards a particular weather forecast model, namely the COSMO model, the weather forecasting model used at the the Hydro-Meteo-Climate Structure of Arpae Emilia-Romagna. The data set contains many execution times of the COSMO meteorological model run under a variety of different scientific parameters and parallelization levels.