Lieu : Bertrange, Luxembourg · Contrat : Stage · Rémunération : A négocier
At LuxProvide, our mission is to explore the frontiers of digital intelligence. By combining data science with high-performance computing, we deliver precise, secure, and sustainable solutions for a rapidly evolving digital world. Our goal is clear: to engineer a smarter and more sustainable future.
How?
We harness the potential of data for innovation through MeluXina, Luxembourg’s state-of-the-art supercomputer. Together, we develop advanced tools and technologies that go well beyond the conventional, driving real innovation.
Why?
We believe that innovation is a force for positive change. LuxProvide is more than a supercomputing facility, we are enablers of progress. Our role is to help shape a future where technology doesn’t just keep up, but genuinely improves the way we live, work, and solve society’s greatest challenges.
Project Context
Numerical simulations consist of the execution on a computer of a mathematical model that describes the behavior of real-world phenomena. They have applications in many areas such as life sciences, materials science, physics, climatology, as well as engineering, social sciences, economics.
Simulation software and computers evolve together. However, because computers and computer technology have a lifetime of a decade, simulation software has to be continuously adapted for their new execution platforms to guarantee correct results and achieve the best performance. This software maintenance, testing, and verification is an arduous task, in particular when numerical reproducibility cannot be guaranteed, for example, when porting a code to a new architecture (e.g., from CPU to GPU).
With this project, we would like to explore a new methodology, SimSem or Simulation Semantic, for the testing and verification of deterministic numerical simulations. The core idea is to annotate the simulation code to highlight the main characteristics of the simulation in terms of control flow and identify key variables and data items. Put together, these annotations will build a semantic representation of the simulation that is independent of the software implementation. The semantic insights given on the simulation data offer the possibility to compare independent executions, possibly from different implementations, while accounting for numerical variability inherent to parallel execution and heterogeneous architectures.
This approach can be applied to the testing, debugging, and verification of numerical simulation software in the long term. During this project, the SimSem methodology will be experimented with mini-apps and target use cases related to software engineering: result verification, regression testing and performance analysis.
Objectives
The project targets regression testing and high-level performance profiling, for the comparison of different implementations of the same problem. It aims to
- prototype the SimSem instrumentation library that can be used to annotate numerical simulation codes and allow collecting runtime traces of simulation executions;
- and design tools to compare executions, results, and performances of simulations, independently of their algorithms or execution platforms.
Tasks
- Define an initial set of semantic annotations
- Prototype the SimSem instrumentation library based on LTTng
- Annotate a selected set of mini-apps, offering various implementations
- Apply trace analysis to regression testing and performance profiling
- Experience with Linux, command line and control versioning systems (Git)
- Experience in using HPC platforms (e.g., Slurm)
- Knowledge in programming languages: C, C++ or Fortran
- Familiar with parallel programming models: MPI, OpenMPI, CUDA or similar
- Data processing: Python pandas/matplotlib or R tidyverse
- Fluent in English
Send your application including:
- An updated CV
- A short motivational letter (1⁄2 page)
- List of the courses followed during the Master studies