Cross-sectional team Application Optimization
The cross-sectional team "Application Optimization" is responsible for the optimization and petascaling of applications in terms of performance, efficiency and parallel I/O. We provide users with knowledge and proper tools sets for their work on the Jülich HPC systems. We strongly interact with the Simulation Laboratories as well as the Cross-sectional teams "Methods and Algorithms" and "Performance Analysis".
Know-How and Focal Points
We are experienced in parallelisation libraries like MPI and OpenMP and are also investigation PGAS languages. We know about specialities of the local HPC systems and this makes it possible for us to guide the users in optimization strategies on the Jülich systems, e.g. for compiler and memory usage, task mapping, and I/O.
We collaborate with users in the need of on-demand support, but we also offer close long-term collaborations with user projects to optimize their applications on the Jülich HPC systems, e.g. to implement parallel I/O or a hybrid parallelisation.
We focus on assisting users to choose optimal parameters for their simulations, this leads to our interest in benchmarking:
- We monitor and benchmark standard simulation software packages.
- We perform evaluations and parameter studies of selected user applications.
- We benchmark new architectures and develop system benchmarks.
With this work we also assist the system administrators in evaluating the system behaviour.
We emphasize on equipping users with knowledge for their usage of HPC systems:
- We provide courses in MPI, OpenMP, and Parallel I/O.
- We organize the bi-yearly introductory course to the Jülich supercomputer systems.
- We collaborate in the Extreme Scaling Workshops as well as in PATC Training Events (e.g. PRACE Spring and Autumn School).
Research and Development
Our main research interest is the optimization of parallel I/O, this has led to the library SIONlib, which implements efficient parallel I/O of task-local data from massively parallel applications.
In addition, we have a strong interest in tool development with focus on application monitoring, optimization and benchmarking. We are currently working on the following projects:
- LLview in Eclipse PTP - Batch system and job monitoring
- JUBE - Benchmarking Environment
- Jumel - Juropa Memory Logger (more information can be found at Juropa via 'module help jumel')
Due to our benchmarking activities we also develop system benchmarks:
- Scalable MPI point-to-point benchmark Linktest
Collaborations with EU Projects
In the above outlined fields we also work in European projects:
- PRACE - Partnership for Advanced Computing in Europe
- LinkSCEEM - Establishment of a HPC eco-system in the Eastern Mediterranean region
- EIC - Exascale Innovation Centre
- DEEP-ER - Dynamic Exascale Entry Platform - Extended Reach