D4.5: Meso- and multi-scale modelling E-CAM modules IV

In this report for Deliverable 4.5 of E-CAM, nine software modules in meso– and multi–scale modelling are presented. The modules represented efforts integrated into larger software packages and thus extend their applicability. The first package is DL_MESO_DPD [1, 2] that provides a toolbox to do Dissipative Particle Dynamics (DPD) simulations, routinely used in an industrial context to find out the static and dynamic behaviour of soft-matter systems. The DL_MESO packages has previously (in [3]) been identified as one of the important codes for mesoscale simulations within E-CAM, and the single- and multi-GPU version of DL_MESO_DPD have already been introduced in deliverables D4.2[4] and D4.3[5], respectively, and recently tested up to 4096 GPUs with results presented in D4.4[6]. In the current deliverable, we present the following 4 modules that relate to DL_MESO_DPD’s extension, performance and optimisation onmulti-GPU architectures: • Surface boundary conditions • Many-body DPDmodel on single GPU • Long Integer on DL_MESO_DPD multi-GPU • Load balancing formulti-GPU DL_MESO. The second software package covered in this report is A Load-balancing Library (ALL), which is a dynamic load balancing library that is both efficient and widely applicable. Load-balancing is a complex topic, particularly when considering multi-scale approaches, due to heterogeneity in both computational approaches and target architectures. Therefore, it has has not yet been implemented in a number of important codes of the E-CAM multi-scale community, e.g. DL_MESO, DL_POLY, Espresso, Espresso++, to name a few. Other codes (e.g. LAMMPS) have implemented somewhat simpler schemes, which might however turn out to lack sufficient flexibility to accommodate all important cases. Therefore, the ALL library was created in the context of an Extended Software DevelopmentWorkshop (ESDW) within E-CAM, where code developers of CECAM community codes were invited together with E-CAM postdocs, to work on the implementation of load balancing strategies. The goal of this activity was to increase the scalability of these applications to a larger number of cores on HPC systems, for spatially inhomogeneous systems, and thus to reduce the time-to-solution of the applications. ALL aims to provide an easy way to include dynamic domain-based load balancing into particle based simulation codes. The library is developed in the Simulation Laboratory Molecular Systems of the Juelich Supercomputing Centre at Forschungszentrum Juelich. It includes several load-balancing schemes, with additional approaches currently being added. The following list gives an overview about the currently included schemes, which each represent a module presented in this report: • Tensor method • Staggered Grid method • UnstructuredMesh scheme • Voronoi Mesh scheme • Histogram method. A short description is written for each module, followed by a link to the respectiveMerge-Request on theGitLab service of E-CAM. These merge requests contain detailed information about the code development, testing and documentation of the modules