Currently we are working on the following funded projects.
The goal of this project is an automatic performance-modeling approach for cloud environments, capable of analyzing the performance behavior and scalability of individual cloud services as well as identifying their breaking point, the maximum load they can handle. An additional objective is to achieve an optimal trade-off between compliance with service level agreements (SLAs) and operational costs.
Deploying deep-learning models efficiently on edge devices often demands numerous iterations of tuning and code specialization. In this project, we focus on the computer system’s side of AI to automate the adaptation of AI-based vision applications to low-cost processors. Our portfolio of method ranges from scalable training to hardware-aware neural-network design and efficient model deployment, to enable affordable computer vision on a wide range of edge devices, such as autonomous vehicles, robots, and drones.
The flat storage hierarchies found in classic HPC architectures no longer satisfy the performance requirements of the growing share of data-processing applications. At the same time, the shift towards data-centric computing is accompanied by a disruptive change of the underlying storage technology towards multi-tier storage hierarchies with fast non-volatile memory with the potential to remove this bottleneck. The objective of this project is the development of an intelligent I/O stack for such architectures that helps exploit their full performance potential. Our focus is the proper balance between computational and I/O requirements.
This project will deliver the programming environment for future European exascale systems, adapting all levels of the software stack – including low-level drivers, computation and communication libraries, resource management, and programming abstractions with associated runtime systems and tools – to support highly heterogeneous compute and memory configurations and to allow code optimisation across existing and future architectures and systems. We contribute tools to map applications onto modular heterogeneous supercomputers.
A national high-performance computing center for computational engineering. In this project, RWTH Aachen and TU Darmstadt join forces to combine their existing strengths in HPC applications, algorithms and methods, and the efficient use of HPC hardware. NHR4CES aims to create an HPC ecosystem combining best practices of HPC and research data management. The focus of NHR4CES will be on engineering and materials science, and engineering-oriented physics, chemistry, and medicine. Our contribution lies in the area of parallelism and performance.
The key to understanding and ultimately improving the performance of HPC applications is performance measurement. Unfortunately, many HPC systems expose their jobs to substantial amounts of noise, leading to significant run-to-run variation. This makes performance measurements generally irreproducible, heavily complicating performance analysis and modeling. In this project, we develop methods and tools to make performance measurement and analysis more noise resilient.
Performance models help explore the design and configuration space of HPC applications. A performance model is a formula that expresses a performance metric such as execution time as a function of one or more execution parameters such as the size of the input problem or the number of processors. However, creating such models analytically is often too laborious. Empirical methods that learn such models form performance measurements present an attractive alternative. Unfortunately, the cost of the required experiments grows steeply with the number of parameters. In this project, we work on solutions to limit this cost, including choosing the right parameters in the first place.
The Human Brain Project is one of the three FET (Future and Emerging Technology) Flagship projects. More than 500 scientists and engineers at more than 140 universities, teaching hospitals, and research centers across Europe come together to address one of the most challenging research targets – the human brain. The project is building a research infrastructure called EBRAINS to help advance neuroscience, medicine, computing, and brain-inspired technologies. Our group helps make data-intensive computer programs running on EBRAINS more efficient.
The goal of this project is to enable legacy software systems to keep up with rapid technological advances of hardware and middleware platforms. Existing software must be retrofitted to fully exploit the potential provided by technological progress because re-development of the existing stock of production software too expensive. Our researh group focuses on the semi-automatic parallelization of sequential software that has not been designed with parallelism in mind.