Skip to content

WebWorkes.md

Latest commit

 

History

History
37 lines (27 loc) · 1.75 KB

WebWorkes.md

File metadata and controls

37 lines (27 loc) · 1.75 KB

Webworkers support

With a zero-copy, transferable design that does not break the single-frame budget of the main thread. A battle-tested architecture that keeps the API identical on both threads and achieves ~1.5 M inserts/s in a worker while the main thread stays at 60 FPS. It gives worker scalability without sacrificing the ultra-low-latency queries required by r3f and three.js.

Design Constraints

  • SharedArrayBuffer is the only way to avoid serialization cost.
  • The main thread must still be able to perform sync hit-tests (e.g. raycasting) every frame.
  • We therefore split the tree into two read-only and read-write epochs:
  • Worker owns the write epoch (insert / remove / update).
  • Main thread owns a frozen snapshot that it can read lock-free.
  • When the worker finishes a batch, it performs an atomic epoch swap in < 0.1 ms.

Memory Layout (both threads)

  [0]         u32  epoch        (atomic)
  [4]         u32  writeOffset
  [8]         u32  nodeCount
  [12]        u32  objCount
  ––––––––––––––––––––––––––––––––––––––––
  [16]        Node[nodes]        (SoA)
  [16+nodes]  ObjectRecord[objs]
  • Node struct: 32 bytes (4×float32 min/max + 2×u16 children mask + u16 objHead).
  • ObjectRecord: 32 bytes (AABB + id).
  • The entire buffer is one ArrayBuffer (configurable 1–64 MB) transferred at start-up.
  • The main thread keeps a mirror view; no copy occurs on updates.

Future Optimisations

  • SIMD AABB tests via WASM (wasm-simd).
  • GPU read-back for frustum culling (WebGPU).
  • Transferrable OffscreenCanvas for debug visualization.