There is no English translation for this web page.
TaLPas: Task-based Load balancing and Auto-Tuning in Particle simulations
The knowledge of accurate thermophysical data is essential for understanding natural processes and the design of technical applications. Despite of longstanding efforts in experimental thermodynamics, there is only a scarce data base because laboratory work is time-consuming and costly, particularly when there are safety concerns. With suitable computational methods and high-performance computing (HPC), molecular modeling and simulation may substantially contribute to the thermodynamic data supply in a much more efficient manner.
The TaLPas project, which is funded by BMBF, deals with fundamental research on HPC software and is a cooperation of six academic institutions. Due to advanced many-core architectures, programming models for numerical simulation codes have significantly changed. More efficient hybrid approaches, that combine shared-memory with distributed-memory parallelization, are required. In order to reap the full benefits of HPC hardware for molecular simulations, an optimal node-level performance is intended. Furthermore, optimized workflow scheduling is vital to reach improved time-to-solution and thereby an efficient workload of supercomputers. In the TaLPas project, we are working on innovative auto-tuning based software solutions for optimal node-level performance of particle simulations that rest on massively scalable task-based sampling methods.
The chair of Thermodynamics and Process Engineering is working on a wide field of molecular dynamics (MD) and Monte-Carlo (MC) simulation techniques to predict accurate thermophysical data. Examples are atomistic simulations of homogeneous gas bubble formation, surface tension of classical fluids, multi-criteria optimization of molecular models, transport properties and the development of the molecular simulation codes ls1 mardyn and ms2.
Lehrstuhl für Wissenschaftliches Rechnen (Prof. Bungartz)
Technische Universität München
Wissenschaftliches Rechnen (Dr. Neumann)
Höchstleistungsrechenzentrum Stuttgart (Prof. Resch)
Visualisierungsinstitut (Prof. Ertl)
Lehrstuhl für Parallele Programmierung (Prof. Wolf)
Technische Universität Darmstadt
Fachgebiet Thermodynamik und Thermische Verfahrenstechnik (Prof. Vrabec)
Technische Universität Berlin
Lehrstuhl für Thermodynamik (Prof. Hasse)
Technische Universität Kaiserslautern
- © TU-Berlin
Equimolar quaternary liquid mixture tetrachlormethane + methanol + toluene + cyclohexance (MC simulation).