Extensibility, deployment and distribution, process connections

[ERnsTL]

Problem:

• Multiple processes are slow:
  • full process switches (not just green thread or native thread switch)
  • and have to communicate via kernel = ring switch and syscall ... super slow
• but need better way to allow separate development of components and extensibility and the source-level built-in compilation is not the solution
  • also leads to LGPL license to apply to components
  • and thus to the custom-made and per-project components
  • not desired

Evaluate:

• microVMs:
  • unikraft / kraftkit is fastest according to microVM benchmarks:
  • Rust examples: https://github.com/unikraft/catalog/tree/main/examples
  • and they say "just communicate via TCP, it is so fast"
  • probably because of virtio or vmnet, which is basically shared memory between host and guest
• Running in UEFI
  • example: https://github.com/codyd51/uefirc
  • very bare metal, but not sure if tha TCP facilities on that level are made for performance
• loading of rlib's which can be found in the target/ directory in every Rust project using Cargo.
  • does that exist?

Connections:

• TCP (?)
• named pipes: flowd-go uses that
• shared memory: rtrb ringbuffer, currently in use