Research tools for users - created by NHR4CES
and RWTH Aachen University
At NHR4CES, our Cross-Sectional Groups and Simulation & Data Labs (SDLs) are dedicated to providing users with comprehensive support for their scientific goals. Our mission is to empower researchers through high-performance computing (HPC), offering state of the art infrastructure such as CLAIX-2023 and Lichtenberg II. We assist users at every stage of their projects—from initial application to implementation and final reporting.
Beyond infrastructure, our SDLs provide scientific consulting, developing a range of tools designed to simplify workflows and make HPC more accessible. These resources add significant value to the research community. Explore our repositories and open-source tools any many more to find what’s helpful for you. If you need further assistance, don’t hesitate to contact us.
DEA
– a platform that enables intuitive and easy use of HPC in a patient-centered structure
Helpful for?
Researchers and clinicians working with medical (timeseries) data and any kind of compute intense application
And how?
“The Diagnostic Expert Advisor (DEA) is a platform for intuitive use of HPC in a patient-centric structure, enabling quick setup of machine learning models on medical data. Developed in the SDL Digital Patient, it supports research on the NHR4CES HPC infrastructure”, says Richard Polzin, researcher from our SDL Digital Patient.
#opensource #medicaldata #ML
FEATHERS
– a framework for automatic optimization of neural architectures and hyperparameters
Helpful for?
Users, who might not have access to data on all machines, making manual tuning of architectures and hyperparameters impractical or even impossible
And how?
“FEATHERS automatically optimizes neural architectures and hyperparameters in federated learning, where data is distributed across multiple owners. It uses gradient-based neural architecture search and bandit-based hyperparameter optimization to identify high-performing models”, explains Jonas Seng, member of our CSG Data Science and Machine Learning.
#opensource #ML #hyperparameter
DaVE
– a curated Database of Visualization Examples
Helpful for?
Users who want to find, explore, and run advanced visualization techniques suitable for their data
And how?
“DaVE is a centralized repository for discovering visualization examples through a simple search. It offers seamless integration into workflows with adaptable containers. Whether you’re exploring advanced visualizations or practical solutions for simulations, DaVE helps find useful resources. Users can also contribute by easily adding their own examples, regardless of experience level”, says Tim Gerrits, leader of our CSG Visualization.
#visualization #repository #database
SLURMminer
– a tool designed to analyze SLURM systems in High Performance Computing clusters
Helpful for?
Researchers and HPC administrators who need to analyze SLURM-managed HPC systems efficiently
And how?
“SLURMminer is a tool for analyzing SLURM systems in HPC clusters. It uses process mining techniques to generate event logs, extract process models, and visualize key metrics. By offering insights into workflow execution, system utilization, workload balancing, and anomaly detection, it helps researchers and administrators assess performance and analyze user behavior efficiently.”, states Zahra Sadeghibogar, member of our CSG Data Science and Machine Learning.
#ML #code #SLURM
DiscoPoP
– a framework for semi-automatic parallelization of sequential C/C++ codes
Helpful for?
Researchers and developers interested in parallelizing their compute-intensive sequential code
And how?
“DiscoPoP was developed by the CSG Parallelism and Performance. By utilizing static and dynamic program analyses, it is able to identify potential for parallelization and offer guidance during the parallelization process”, says Lukas Rothenberger, researcher in the CSG Parallelism and Performance.
#opensource #parallelization #codeanalysis
Extra-P
– a new hyperparameter optimization algorithm
Helpful for?
Researchers and developers optimizing the performance/scalability of their compute/communication intensive application
And how?
“Extra-P is an automatic tool for performance modeling that helps identifying scalability bugs by describing how performance metrics like execution time scale with input size or processor count. Developed in the CSG Parallelism and Performance, it supports performance optimization on the NHR4CES HPC infrastructure”, explains Gustavo de Morais, researcher in the CSG Parallelism and Performance.
#opensource #scalabilityanalysis #performancemodeling
Coscine
– a research data management platform for your research project
Helpful for?
Researchers from participating universities or with an ORCID, who need a research data management platform for their research project
And how?
At RWTH Aachen University, we develop and maintain Coscine, a research data management (RDM) platform. It allows project participants to manage data and metadata through various services (so-called resources) in line with the FAIR principles. Particularly useful in the context of HPC, Coscine supports automation and metadata extraction, helping researchers manage large datasets and improve data reusability.”, explains Marcel Nellesen, member of our CSG Data Management.
#Coscine #Datamanagemt #Data
IBO-HPC
– a new hyperparameter optimization (HPO) algorithm that allows users to provide feedback at any time during runtime
Helpful for?
Users, who want more control over their optimization process and to boost the performance of hyperparameter optimization
And how?
“Finding optimal hyperparameters is essential for high-performing machine learning models. Traditional hyperparameter optimization (HPO) approaches are either manual, which is time-consuming, or fully automated, which doesn’t utilize user knowledge. IBO-HPC is a new HPO algorithm that addresses this by allowing users to provide feedback during optimization, combining automation with expert input to accelerate the process. This hybrid approach retains the efficiency of automatic methods while leveraging valuable insights from users to speed up optimization”, says Jonas Seng, member of our CSG Data Science and Machine Learning.
#ML #code #HPO
BoSSS
– an open-source framework
Helpful for?
Researchers, whose main field of application is multiphase flows with moving interfaces
And how?
“BoSSS (Bounded Support Spectral Solver) provides a general foundation for the scientific development, application and evaluation of higher-order discretization schemes based on the Discontinuous Galerkin (DG) and extended DG (XDG) method. It aims to overcome the gap between prototype codes with limited performance and generality on the one hand, and highly-optimized single-purpose research codes on the other hand. “, says Martin Smuda, member of our SDL Fluids.
#CFD #Solver #Opensource