Skip to content

plan: record the tiling-alternatives evaluation (SVG/raster/streaming-ABI) #120

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
AdaWorldAPI merged 1 commit into from
Aug 12, 2026
Merged

plan: record the tiling-alternatives evaluation (SVG/raster/streaming-ABI) #120

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
28 changes: 28 additions & 0 deletions claude-notes/plans/2026-08-12-osm-vector-perf.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -98,3 +98,31 @@ Same honesty note as the parent plan: no slab in this container, so
verification is synthetic-fixture through the real client code; the live
deploy is the real test. The wire-size ratio is measured on synthetic
shapes with realistic point counts.

## Alternatives evaluated (operator ask, 2026-08-12, post-#119)

Asked: compare the merged design against (a) SVG/server tiling and (b) a
further-compressed "streaming ABI as projection". Conclusion, with the axes
the operator named:

- **SVG tiling: dominated on every axis, rejected without measurement.**
Server pays serialization, wire pays XML bloat (~3-5× over binary), client
still parses + rasterizes SVG DOM. It is the raster option's server burden
plus the vector option's traffic burden plus DOM cost.
- **Server raster tiles: right only behind a CDN.** Raster wire is O(pixels)
(~15-40 KB/tile, density-immune) so it wins traffic at maximum city
density — but q2-cockpit is one Railway container, so every tile would be
rasterized per viewer ("render load is on server, bad" — operator). Also
buys a rasterizer dep + style engine. Re-enters only with a CDN, and then
as a hybrid (raster overview, vector city).
- **Merged design (#119): the right point to measure from.** Server work is
mask + fold + LE encode; wire measured 5.1× under JSON before gzip; warm
revisits are 304s; client is ~150 native draws/frame.
- **The escalation path if city-zoom traffic still bites is OSM1-v2, not
raster:** after pixel-grid simplification consecutive points are adjacent
pixels, so i16 tile-relative + delta + zigzag-varint (the MVT trick) cuts
another ~4-6× → the original 60 MB JSON view lands around ~1 MB gzipped.
Pair with the bake-side `.tiles` sidecar (pre-computed per-tile ranges,
SIMD batch projection at bake — the genuine ndarray home) and serving
becomes a file-range read. Both recorded above under "ndarray — where it
actually fits".