Estimating the execution time of fully-online multiscale numerical simulations

Abstract

In this paper, we propose a methodology for estimating the execution time of simulations driven by multiscale numerical methods. The methodology explores the idiosyncrasies of multiscale simulators to reduce the uncertainty of predictions. We use the multiscale hybrid-mixed (MHM) finite element method to validate our methodology. We compare our proposed technique with prediction models automatically selected and calibrated by Auto-WEKA. We show that the models obtained with our technique are competitive when compared with the models coming from Auto-WEKA, being interpretable and with much less computational effort during the learning process.