Rendering the World

Pre-rendering a street map of the world has never seemed reasonable: it would take over 50 days and 80 terabytes of storage to render the 20 billion tiles between zoom levels 0 and 17 assuming the fastest speeds and smallest file sizes. Instead, rendering tiles on demand from beefy servers and caching heavily has been the default method for serving tiles for global street maps.

I’ll explain how fast image compositing and sqlite tile storage enable an analysis of maps in terms of information and redundancy, greatly changing the "dreaded" quadratic curve of tiled maps. By splitting maps into distinct layers and rendering only the relevant information from each we can reduce render times to only a few days and storage costs from terabytes to gigabytes. The fully rendered final product makes it possible to offer a global base map to a wide audience while maintaining high performance and affordability.

Transcript