JuliaComputing questions

[orso82]

General points of discussion for the meeting

1. High level, what we are doing
   1. GA theory, GA stewards largest fusion experiment in US, working of FPP design
   2. few IR&Ds, GH org with 58 julia repos, all private
2. Main IR&D named FUSE (400k lines of code over ~20 repos)
   1. FUSE covers all systems of a FPP (plasma physics, engineering, costing, bop, etc...)
   2. FUSE has 4 core devs with few contribs
   3. No time dependence, but to come, as well with integration with control systems (MATLAB/Symulink)
   4. Use ML (simple problems)
3. Looking for help/questions on technical
   1. using parametric types for dd.jl
   2. performance tuning
   3. lower compile time
   4. sys-images do not particularly lower compile time
   5. build julia package out of fortran code (e.g. CHEASE, TGLF) https://github.com/JuliaPackaging/BinaryBuilder.jl
   6. Package installation from private git repo and registry broken (SSH/GIT library issue)
   7. Cloud computing
4. can we buy a fixed price (eg. $10k) of their consulting services this year and have them do those in 2023
   1. how is billing done (per hour I assume) and how many hours does $10k buy us
   2. NDA

[jguterl]

• Support for GPUs: full portability of kinsol (Sundials.jl) : porting nvector on GPU
• Support for GPUs: kernelization of large heterogeneous functions (e.g network of binary reactions): methodology + best practice + debugging + profiling
• Julia GUI: best approach?, embedded multiple Makie windows ?
• remote interactive job submission + interactivity: best practice, IJulia, data transfer between julia session

[orso82]

Let's sort these tasks based on priority

• [ ] using parametric types for dd.jl
• [ ] performance tuning
• [ ] lower compile time
• [ ] sys-images do not particularly lower compile time
• [ ] build julia package out of fortran code (e.g. CHEASE, TGLF) https://github.com/JuliaPackaging/BinaryBuilder.jl
• [ ] Package installation from private git repo and registry broken (SSH/GIT library issue)
• [ ] Cloud computing

[orso82]

NOTES: For compile times, I don't know if you saw: https://sciml.ai/news/2022/09/21/compile_time/

Documentation on JuliaSim Controls: https://help.juliahub.com/juliasimcontrols/stable/

Overloading this is optimal for doing lazy fields: https://docs.julialang.org/en/v1/base/base/#Base.getproperty