code_aster and HPC computing
Master parallel computing with code_aster: from HPC fundamentals to diagnosing and optimizing the performance of your industrial computations on a cluster.
Course summary
Do you recognize yourself?
- You're running increasingly large computations but don't know when or how to parallelize, or where the actual computational cost lies.
- You're unsure whether to use shared memory, distributed memory, or MPI/OpenMP hybridization, and can't gauge the impact of that choice on your resources and turnaround time.
- You've tried enabling parallelism in code_aster but aren't sure which keywords to use or how to structure your input data accordingly.
- You're getting disappointing speed-up on a cluster and don't know how to diagnose whether the issue lies with the solver, the model partitioning, or the sizing of the requested resources.
Why our training in particular?
From fundamentals to industrial computation
You start from the basics of parallel computing and work up to an industrial case run and measured on a cluster. Every theoretical concept is immediately connected to hands-on practice in code_aster.
Two hands-on sessions to make it stick
You don't just flip on some options: you parallelize a computation and measure its speed-up, then you diagnose and optimize the performance of a computation that scales poorly.
Hands-on trainers, right there with you
Our engineers and PhDs use code_aster daily on industrial projects at Simvia. On site, they share best practices and pitfalls in real time — the kind you won't find in any documentation.
Close, hands-on supervision
Small group of trainees, with a trainer on hand to unblock anyone stuck during the hands-on work.
A reusable diagnostic method
You leave with a simple framework for identifying, on your own computations, whether the bottleneck comes from the solver, the partitioning, or the resource sizing.
Audience
- Industry: computation engineers running code_aster on large-scale models who want to reduce turnaround time.
- Academia: researchers and PhD students running intensive computations that need to scale.
Prerequisites
Basic practice of code_aster (setting up input data, running a computation). General notions of parallel computing are helpful but not required. No prior HPC experience needed.
Detailed programme
Morning: parallel computing fundamentals and code_aster's HPC architecture
- HPC and FE computing: why parallelize? Where does the cost actually sit in a finite element computation?
- Parallel computing fundamentals: shared memory, distributed memory, MPI/OpenMP hybridization.
- Partitioning and solvers: domain decomposition and parallel solving strategies.
- HPC in code_aster: code architecture, available parallelization options and possibilities.
Afternoon: implementation, hands-on practice and performance diagnosis
- Practical implementation: keywords, parallelization options, structuring input data for a parallel computation.
- Hands-on: parallelizing a code_aster computation: running a computation across multiple cores/processes, measuring speed-up and parallel efficiency.
- Performance and best practices: scaling, choosing resources, diagnostic indicators to watch.
- Hands-on: diagnosing and optimizing an industrial computation: analyzing a computation that scales poorly, identifying the bottleneck (solver, partitioning, resources) and adjusting it.
Practical details
Available for online training only
Dates, place, language, prices and registration are given on each session page.
Accessibility
We are committed to making our training accessible to people with disabilities. Every situation is unique, so please contact us before registering so that we can study together the educational, technical or organisational adaptations that can be put in place.