Digital maps are deeply integrated into modern life, underpinning navigation, location-based services, and logistics. Many widely used mapping applications rely on the openly available, planetscale OpenStreetMap (OSM) dataset. For managing, processing, and visualizing such geospatial data, the technology stack developed by Mapbox and its open-source forks maintained by the MapLibre organization are widely considered state-of-the-art, as evidenced by their adoption at major technology companies including Amazon, Meta, and Microsoft. However, core components of this stack were designed nearly a decade ago and no longer fully exploit the capabilities of contemporary hardware, modern APIs, or the demands posed by rapidly growing geospatial data volumes. This paper presents an overview of ongoing doctoral research that designs and partially implements a next-generation vector map rendering stack for the web. The research makes three primary contributions. First, it introduces COMTiles, a cloud-optimized tile container format that significantly reduces cloud access charges. Second, it presents MLT, a column-oriented vector tile format that employs lightweight compression to achieve high compression ratios while enabling fast decoding. Third, the work proposes the design of a novel GPU-driven web map renderer based on WebAssembly and WebGPU, detailing an execution model and optimization strategies that combine SIMD-based processing in WebAssembly with offloading of the vector tile processing pipeline to the GPU. Preliminary evaluations on planet-scale OSM data demonstrate cloud hosting cost reductions of approximately three orders of magnitude compared to software-as-a-service offerings, alongside encoded tile sizes up to six times smaller than Mapbox Vector Tiles with faster decoding. Together, these results indicate substantial potential for improved user experience, reduced latency, and more cost-effective delivery of large-scale vector basemaps.
Tremmel et al. (Thu,) studied this question.