STANDARDwalkthrough
Vector Tiles vs Raster Tiles
A user rotates the map 45 degrees to orient it along their walking direction, then switches to dark mode because it is nighttime. With raster tiles, both actions are impossible without re-fetching entirely new tile sets from the server.
We have three options. Option one: raster tiles.
“The constraint: we need map tiles that the client can rotate, tilt into 3D, and re-style on the fly without additional network requests.”
The server pre-renders PNG images at each zoom level. Simple to serve, works on any device, but completely inflexible.
Rotation requires server-side re-rendering of every tile at every angle, which is absurd. Dark mode requires a second complete tile set, doubling storage to 1 PB.
Option two: vector tiles. The server sends raw geometry (road polylines, building polygons, labels) encoded as Protocol Buffers.
The client renders them locally using WebGL or GPU shaders. A vector tile at zoom 14 weighs 20 to 50 KB versus 100 to 300 KB for the equivalent raster tile, a 5 to 10x size reduction.
Option three: hybrid, where we use vector for urban areas and raster for satellite imagery. The winner for map rendering is vector tiles.
Google switched from raster to vector tiles in 2013, and the Mapbox Vector Tile (MVT) specification became the industry standard. Vector tiles support rotation, 3D tilting, custom styling, real-time label collision detection, and smooth zoom transitions, all client-side.
A full vector tile set for the planet at all zoom levels is roughly 50 to 80 TB, compared to 500 TB for raster. The trade-off is on the client side: vector tile rendering requires a GPU-capable device.
Older phones, embedded car displays, and low-end IoT devices may not have the graphics pipeline to render vector tiles at 60fps. For those, we fall back to raster.
Google Maps still serves raster tiles for its Static Maps API and for devices that cannot run WebGL. What if the interviewer asks: what about satellite imagery?
Satellite imagery is inherently raster (photographs from space). We serve satellite tiles as raster JPEGs and overlay vector data (roads, labels) on top.
The hybrid approach gives us the best of both worlds.
Related concepts