SCALABILITY TESTS OF THE OPEN-SOURCE OPENFOAM CODE

Partner: HYDROGRAV

field: WATER MANAGEMENT

Hydrograv GmbH is a spin-off company from the Karlsruhe Institute of Technology – the Institute for Hydromechanics and the Technical University of Dresden – the Institute of Urban and Industrial Water Management (ISI, Prof. Dr. Peter Krebs). These institutions, which are international leaders in their disciplines, are the alpha and omega in the field of optimisation of wastewater treatment plants, and they both work intensively on the development of simulation tools. Hydrograv uniquely combines interdisciplinary advanced know-how in the field of advanced water management technologies.

Hydrograv develops innovative and efficient solutions for water management systems optimisation by means of Computational Fluid Dynamics (CFD) simulations considering multiphase flow. CFD simulations of multiphase flows are enormously time-consuming, and HPC infrastructure utilisation is the key factor for speeding up the whole process. Hydrograv uses the well-established open-source OpenFOAM code for their simulations.  Although several benchmarking tests are available for OpenFOAM, it is difficult to apply these results to other HPC clusters that differ in hardware architecture and software used.

Hydrograv together with the Czech National Competence Centre submitted an application to the EuroHPC Benchmark and Development Access call for proposals and were granted access to the Karolina supercomputer operated by IT4Innovations National Supercomputing Center at VSB - Technical University of Ostrava, in the Czech Republic. Using this supercomputing cluster, hydrograv performed scalability tests for the OpenFOAM code using one of their real-world cases. In order to gather comprehensive information, a total of four different setups with varying mesh sizes were used (1M cells, 6M cells, 11M cells and 17.5M cells). All of the setups were run using one, two, and three compute nodes of Karolina to evaluate the behaviour of the solver over several compute nodes. In cases where only one compute node was used, runs with 1, 16, 32, 64, and 128 compute cores were performed.

From the user’s point of view, it is important to find out what the real-time requirements for such simulations are, and what costs would have to be considered for further such simulations.