Impact & Valorization
Parallel-in-time for your HPC application?
Impact in applications
Want to know what time parallelization can offer for you? Interested to try it yourself?
Impact in science & industry
Multi-scale simulation. Multilevel, iterative solution strategies (typical of parallel-in-time methods) naturally match with the idea of most current multi-scale approaches. TIME-X explores such links explicitly.
Optimization and control. The scalability potential of parallel-in-time methods can be exploited even further in control and optimization problems, where forward-in-time simulation can act as a constraint. One then needs to design algorithms that integrate the parallel-in-time iterations with the iterations of the optimization algorithm. TIME-X has a transformative impact on computational practice in such contexts.
Uncertainty quantification. When the model contains some randomness, one needs to average over large numbers of simulations. TIME-X is readily combining parallel-in-time methods with methods for uncertainty quantification.
HPC software engineering. In TIME-X, we study software engineering techniques that can have great impact on how prototyping and testing of new algorithmic ideas can be efficiently done for a variety of hardware systems.
With Exascale computing, we have the potential to dramatically improve our capabilities in simulation and prediction, virtual experiment, rapid prototyping and real-time control. This will create a huge amount of new possibilities for industry and industry sectors. With the methods and software developed in TIME-X, we are able to exploit exascale power for:
- time-critical application, such as the examples from the Health and Medicine domains (see above),
- large-scale applications, such as the fusion or climate cases (see above).
The broad range of possible applications will enable new simulation-based services for, e.g., medical or engineering applications. These services will hide the complexity of exascale computing from the end-user, but they will deliver immediate response for time-critical applications.
Want to know what TIME-X can offer for you? Interested to try it yourself?
Underneath, we list a selection of software developed by the TIME-X consortium, mainly relating to the targeted application domains.
Adaptive time parallization for molecular dynamics simulations
An adaptive implementation of the parareal algorithm for molecular dynamics (MD). The dynamics of the system is simulated using the MD software LAMMPS, which can model 2D or 3D systems with sizes ranging from only a few particles to billions. Simulations can be sped up by factor of 20.
Time parallelization of electric machines at Bosch
An in-house implementation of parareal to optimize and analyze Bosch’s motors, in particular for e-mobility applications. It is tightly integrated in Bosch’s electric machine simulation code to quickly compute the steady-state behavior of electric machines. Led to 28 times faster simulation approach.
Parallel-in-time for quenching problems at CERN
A high-performance-computing implementation of parareal that accelerates the simulation of quenching of superconducting accelerator magnets. Those simulations are necessary to ensure operational safety at CERN. Various what-if scenarios can now be simulated in acceptabel time.
DynMPI: dynamic management of resources with MPI
The DynMPI publicly available software realizes dynamic resources with MPI Sessions and PMIx. It provides extensions of the MPI Sessions Interface for handling changes of the application’s resources during runtime. The software is based on the publicly available Open MPI, OpenPMIx and PRRTE implementations.
pySDC: Python implementation of spectral deferred correction
The pySDC software is intended for rapid prototyping and educational purposes. New ideas such as sweepers or predictors can be tested and first toy problems can be easily implemented. It includes the integration of adaptivity and first resilience test cases and first steps toward GPU usage are taken.
SWEET (climate & weather)
This software allows a fast exploration & prototyping of time discretization methods for PDEs. The main applications are certain dynamical cores for climate & weather simulations as used by, e.g., the European Centre for Medium-Range Weather Forecasts.
TEMF Parareal for electromagnitcs
TEMF Parareal is an open-source implementation that allows for local parallel execution and for parallel execution over a cluster of machines, possibly combined with local parallel execution at each machine of the cluster. Speed-up of factor 10 realized.
Time parallelization for cardiac electrophysiology
Implementation of explicit stabilized methods in a parallel-in-time context for solving stiff differential equations appearing for instance in monodomain models in cardiac electrophysiology. These methods are well suited for integration with PFASST.
Showcases & videos
- pySDC tutorials & computational experiments
- Showcasing dynamic resources with MPI applied to a 2D heat equation solver (video)
- Showcasing tools assisting in developing and debugging of dynamic MPI programs (videos)
- SWEET tutorials
- TEMF Parareal – Example of an industry-relevant machine model (“im_3kw”)
- Web applications for predicting the performance of a PinT algorithm and devising a good scheduling strategy
- Demonstrator software illustrating the use of machine learning to create coarse models for PinT algorithms
Short stories on our impact
Educational Website for first performance analysis of PinT algorithmsRead more…: Educational Website for first performance analysis of PinT algorithms
PinT 2023: 12th Workshop on Parallel-in-time IntegrationRead more…: PinT 2023: 12th Workshop on Parallel-in-time Integration
CERN investigates PinT for superconducting magnet simulationRead more…: CERN investigates PinT for superconducting magnet simulation
PinT multirate explicit stabilised method for cardiac electrophysiologyRead more…: PinT multirate explicit stabilised method for cardiac electrophysiology
PinT for Molecular DynamicsRead more…: PinT for Molecular Dynamics
Design of extensions to MPI standardRead more…: Design of extensions to MPI standard