HPPSS Panel @ SC23

2023-11-12

Panelists

How has distributed/parallel computation evolved?
- What role do you see Python playing in the future?
What are the key challenges non-expert and expert HPC users face when scaling Python workloads?
Users are often faced with some kind of rewrite when transitioning from laptop to supercomputer.
- How should users think about approaching this challenge?
- When should they look to
  - Cool-new-tools (Julia, Mojo, etc)
  - CPython (+ broader ecosystem)
  - Full rewrites into Fortran/C/C++ (and traditional HPC libraries)?

There are technical roadblocks to wider use of Python in HPC.
- Issues like running Python at scale from specific filesystems on supercomputers can cause issues and many best-practices not widely documented or understood.
- How should the HPC community organize knowledge to enable users of all backgrounds?
There are social roadblocks to the wider adoption of Python in HPC.
- For example, some computing centers do not like Python-based workloads running on their machines.
- What technical/communication/other solutions should we look to for changing hearts and minds and promoting accessibility of HPC through Python across a broad range of users?
How do you see or use Python to bring together traditional HPC applications, cloud, and the AI ecosystem?
- How should users think about constructing applications and workflows at scale?
  - while keeping informed about performance implication?

Panelists:
- Fernanda Foertter (Voltron Data)
- Torsten Hoefler (ETH Zurich)
- Siu Kwan Lam (Anaconda)
- Travis Oliphant (Quansight)
- Ben Zaitlen (NVIDIA)
- Richard Zamora (NVIDIA)