Full-system acceleration (~40× faster)

[mavdol]

Full-system acceleration

Improving every workload's speed while keeping vpod's three pillars: portability, two-layer isolation, and full-system completeness.

Goal: ~40× faster than the current version.

Phases (In progress) :

• Network / TLS offload: Builds on the terminating proxy to let guest tools skip TLS work entirely on their side while the host still does real, verified TLS to the origin. HTTPS is now about as fast as plain HTTP.
• Warm Python: A preforking interpreter server so every python3 (or pip, if included) call starts from a warm process instead of paying the cold start each time.

  Note: sbx.code.run already starts from a warm process; this is for tools that want to call python3 directly from the shell.

• Prebaked environments: Bake common venvs and dependencies into the snapshot at build time, turning repeated runtime installs into instant startup.
• AOT gates: A focused spike and profiling pass to test and confirm where the real wins are before committing to AOT.
• AOT full-system translation: Translate hot guest code (userspace and kernel) ahead of time to lift the core interpretation ceiling.
• Softmmu fast path: Speed up the per-access address translation that every memory operation goes through.
• Device / I/O acceleration: Faster virtio/filesystem paths, including host-FS passthrough, for I/O-heavy tools.