Energy Efficiency At Node Level

The goal is to boost energy efficiency of molecular simulation software for process engineering applications at node level. Novel interfaces shall be implemented that allow the incorporation of energy and performance data into the molecular simulation software AutoPas. This will enable auto-tuning, i.e. automated shared-memory algorithm selection, at run time of the simulation to ensure energy-optimal solution procedures over the entire course of the simulation. In a similar way, visualization techniques are to be adapted to energy consumption of the underlying compute systems.

Model Accuracy/Approximate Computing and Energy Efficiency

The benefits and costs of HPC simulations put in relation such that good compromises can be found. In a first step, the costs of simulations are analyzed in a novel holistic approach in order to derive suitable means of improvement and then implement and test them. The aim of the work package is to develop methods to define a simulation task in such a way that a required (sufficient) accuracy is achieved taking into account the energy demand and environmental impact, instead of achieving the highest possible accuracy in principle. This novel analysis approach is applied to different simulation methods, e.g. molecular simulation and continuum simulations as well as hybrid simulation methods. Also, different hardware architectures will be included in the comparison.Model Accuracy/Approximate Computing and Energy Efficiency

Energy efficiency and node level

A goal of WindHPC is to combine for the first time existing standard and second life compute resources at a wind farm (WWIT) with a HPC center (HLRS) and study potential usage scenarios focusing on the sustainability of the run applications and workflows under various constraints. Therefore, WindHPC considers several energy metrics that will be evaluated for the different sites. Together with additional information about, e.g., resource availability these metrics will be used for the scheduling of the scientific workflows. To make this possible, WindHPC’s goal is to develop and implement interfaces for the information and data exchange between the sites and the various software components. The metrics will also be used to evaluate potential energy savings when applying distributed visualisation techniques for the analysis of simulation results

Integration of the components and results

This work package has the goal to integrate all different components and results from the other work packages into the combined WindHPC architecture. The key work here is the definition and implementation of interfaces for the various software parts, which allow to exchange the various information, e.g., the energy related metrics and resource availability necessary for the dynamic selection of algorithms, scheduling of workflows, and adaptation of the visualisation.

Digit twins, data management and data re-use

This work package has the goal to improve the handling and use of simulation data to increase the sustainability of obtained results, i.e., reducing the number of unnecessary or repeated simulations. Therefore, digital twins will be implemented for the selected process engineering applications in the form of workflows that can be used by other scientists. To make the obtained simulation data more useful with respect to the FAIR principles a set of onthologies will be created and used to define extended metadata schemes. These will then be applied to store simulation data into knowledge data bases for later access by other scientists. A second point that will be addressed in this work package is the characterisation of knowledge gain from the digital twins. This characterisation will then be combined with metrics around the simulation costs from other work packages to create a cost-benefit analysis allowing scientists to help with the selection of the most efficient method to answer their research questions under given constraints, e.g., required simulation accuracy and available hardware.