Virtual Institute — High Productivity Supercomputing


BMBF (2017 - 2019)

Task-based Load Balancing and Auto-tuning in Particle Simulations (TaLPas)

This project is part of the BMBF Program “Fundamental research for HPC software in high performance and leadership computing”. The official project website can be found here.


With the approaching exascale era, HPC architectures undergo radical changes. Due to the high-energy consumption of these huge systems optimal exploitation of supercomputers becomes more and more important. On the node-level, the increase in compute cores requires using “MPI+X” programming models. These and many other issues pose challenges for the programmers and the users of both software and hardware. In particular, standard tuning approaches become prohibitively complex and not always effective.

The main goal of TaLPas is to provide a solution for fast and robust simulation of many, potentially dependent particle systems in a distributed environment. These systems, for example, arise in studies of droplet formation at the molecular level, or in uncertainty quantification and parameter identification. Particle simulations are also used in other problems, such as fluid dynamics and astrophysics.

The TaLPas project pursues the following specific objectives:

  • The development of innovative, auto-tuning-based particle simulation software in form of an open-source library to leverage optimal node-level performance. This will guarantee an optimal time-to-solution for small- to mid-sized particle simulations.
  • The development of a scalable task scheduler to yield an optimal distribution of potentially dependent simulation tasks on available HPC resources.
  • The combination of both auto-tuning-based particle simulation and scalable task scheduler, augmented by an approach to resilience. This will guarantee robust (i.e., fault-tolerant) sampling evaluations on peta-scale and future exascale platforms.

Project partner TU Darmstadt, a member of VI-HPS, contributes Extra-P, a tool for automated performance modeling. In the project, TU Darmstadt will employ it to support the task scheduler in the process of finding optimal execution configurations for individual tasks. For this purpose, TU Darmstadt aims to develop a library version of Extra-P and introduce an ability to refine models on the fly as new measurements become available. Extra-P will be also used to model the runtime of particle simulations as a function of various parameters that influence the simulation. The goal is to identify bottlenecks and reduce the time-to-solution for large-sized particle simulations.



Promoted by the German Federal Ministry of Education and Research (BMBF)