Browsing by Author "Totounferoush, Amin"

Now showing 1 - 2 of 2

Open Access
Data-integrated methods for performance improvement of massively parallel coupled simulations
(2022) Totounferoush, Amin; Schulte, Miriam (Prof. Dr.)
This thesis presents data-integrated methods to improve the computational performance of partitioned multi-physics simulations, particularly on highly parallel systems. Partitioned methods allow using available single-physic solvers and well-validated numerical methods for multi-physics simulations by decomposing the domain into smaller sub-domains. Each sub-domain is solved by a separate solver and an external library is incorporated to couple the solvers. This significantly reduces the software development cost and enhances flexibility, while it introduces new challenges that must be addressed carefully. These challenges include but are not limited to, efficient data communication between sub-domains, data mapping between not-matching meshes, inter-solver load balancing, and equation coupling. In the current work, inter-solver communication is improved by introducing a two-level communication initialization scheme to the coupling library preCICE. The new method significantly speed-ups the initialization and removes memory bottlenecks of the previous implementation. In addition, a data-driven inter-solver load balancing method is developed to efficiently distribute available computational resources between coupled single-physic solvers. This method employs both regressions and deep neural networks (DNN) for modeling the performance of the solvers and derives and solves an optimization problem to distribute the available CPU and GPU cores among solvers. To accelerate the equation coupling between strongly coupled solvers, a hybrid framework is developed that integrates DNNs and classical solvers. The DNN computes a solution estimation for each time step which is used by classical solvers as a first guess to compute the final solution. To preserve DNN's efficiency during the simulation, a dynamic re-training strategy is introduced that updates the DNN's weights on-the-fly. The cheap but accurate solution estimation by the DNN surrogate solver significantly reduces the number of subsequent classical iterations necessary for solution convergence. Finally, a highly scalable simulation environment is introduced for fluid-structure interaction problems. The environment consists of highly parallel numerical solvers and an efficient and scalable coupling library. This framework is able to efficiently exploit both CPU-only and hybrid CPU-GPU machines. Numerical performance investigations using a complex test case demonstrate a very high parallel efficiency on a large number of CPUs and a significant speed-up due to the GPU acceleration.
Open Access
Efficient and scalable initialization of partitioned coupled simulations with preCICE
(2021) Totounferoush, Amin; Simonis, Frédéric; Uekermann, Benjamin; Schulte, Miriam
preCICE is an open-source library, that provides comprehensive functionality to couple independent parallelized solver codes to establish a partitioned multi-physics multi-code simulation environment. For data communication between the respective executables at runtime, it implements a peer-to-peer concept, which renders the computational cost of the coupling per time step negligible compared to the typical run time of the coupled codes. To initialize the peer-to-peer coupling, the mesh partitions of the respective solvers need to be compared to determine the point-to-point communication channels between the processes of both codes. This initialization effort can become a limiting factor, if we either reach memory limits or if we have to re-initialize communication relations in every time step. In this contribution, we remove two remaining bottlenecks: (i) We base the neighborhood search between mesh entities of two solvers on a tree data structure to avoid quadratic complexity, and (ii) we replace the sequential gather-scatter comparison of both mesh partitions by a two-level approach that first compares bounding boxes around mesh partitions in a sequential manner, subsequently establishes pairwise communication between processes of the two solvers, and finally compares mesh partitions between connected processes in parallel. We show, that the two-level initialization method is fives times faster than the old one-level scheme on 24,567 CPU-cores using a mesh with 628,898 vertices. In addition, the two-level scheme is able to handle much larger computational meshes, since the central mesh communication of the one-level scheme is replaced with a fully point-to-point mesh communication scheme.