First workshop at ISC 2025
Schedule
▶️ with links to the slides
9:00 - 11:00: Slot 1 (2hr)
- Introduction and Welcome Tom Vander Aa (imec)
- application people war stories (25' each)*
- Cloudy with a chance of NaNs, Ioan Hadade (ECMWF)
- Parables from the Pursuit of Performance Portability, Andrey Alekseenko (KTH)
- Welcome to the jungle: understanding communications in HPC clusters, Laura Bellentani (CINECA)
- 1 programming model talk
11:00 am – 11:30 am Coffee break
11:30 - 13:00 : Slot 2 (1.5hr)
- 2 programming models / tools talks (25' each)
- No Magic! To expose complexity helps users and developers, Mirko Rahn (Fraunhofer ITWM)
- Parallel Performance Measurement and Analysis made easy?!, Bernd Mohr (JSC)
- Panel/Discussion/Closing
- Everyone!!
Overall Conclusion
The Good
- CI/CD, packaging, containers
- large effort in software engineering, but important
- we need more €€€ for software
- giving back to the community, open source, eg EESSI initiative
- Tools: strace, tracing,
- Many tools exist
- But are these the tools we need? Do they scale? Do they support XXX?
- Interaction between domain experts and HPC experts
- Worth the investment of time/effort?
The Ugly
- Performance = f(kernel, tool, library, slurm, HW versions, environment variables, bugs)
- there are too many versions
- How to package SW and manage SW version
- Some versions solve bugs but introduce others
- Observability (ECMWF vs 'open' centres)?
The Bad
- HW complexity
- Should we hide the complexity?
- Vendor dependency
- Proprietary software stacks and vendor lock in
- Open source is better?
- Do users know how good their code performs?