Why ByteHouse’s GIS Engine Beats Traditional Spatial Databases in Real‑World Analytics

Geospatial Analytics in the Digital Age

Geospatial analytics has become a key tool for enterprises to gain market insights, from precise ad targeting to e‑commerce logistics optimization. A typical example is a coffee‑shop chain that uses population density, traffic flow, and competitor location data to select new store sites that maximize profit.

ByteHouse GIS: Technical Overview

ByteHouse, the Volcano Engine’s enhanced version of ClickHouse, recently added high‑performance GIS capabilities. The engine combines OLAP processing with spatial analysis, supporting OGC‑standard geometry types (Point, LineString, Polygon, MultiPoint, MultiLineString, MultiPolygon, GeometryCollection) and more than 50 spatial functions.

Key implementation details include:

Use of Google S2 to convert latitude/longitude into a one‑dimensional key, followed by sorting.

Data is stored column‑wise; after S2 sorting, points are split into small chunks and each chunk is indexed with an R‑Tree.

The R‑Tree enables fast block‑level filtering for polygon or radius queries, dramatically reducing the amount of data read.

Vectorized execution and CPU‑level optimizations for 2‑D spatial functions improve end‑to‑end query performance.

Compatibility with common GIS file formats (WKT, WKB, GeoJSON, ShapeFile, Parquet, CSV, Arrow).

Benchmark Tests

Using the 21 GB NYC Taxi dataset (169 M rows), ByteHouse’s GIS functions were benchmarked against ClickHouse, StarRocks, PostGIS and DuckDB. Two representative functions were evaluated:

ST_Within – spatial containment query.

ST_DistanceSphere – spherical distance calculation.

Results show that:

ByteHouse consistently achieved sub‑second latency for ST_Within, outperforming PostGIS (>6 s for large zones) and DuckDB (no spatial index).

For ST_DistanceSphere, ByteHouse kept query time under 0.1 s, comparable to ClickHouse and StarRocks, while DuckDB lagged due to lack of indexing.

Throughput (queries per second) was highest for ByteHouse, thanks to reduced I/O and vectorized processing.

Conclusions

ByteHouse’s GIS engine delivers lower query latency and higher throughput than traditional spatial databases, while maintaining OGC compatibility and supporting a wide range of file formats. For projects where geospatial analysis is a core component, PostGIS remains a solid choice. However, when spatial queries are part of larger big‑data workloads and performance is critical, ByteHouse, ClickHouse, StarRocks or DuckDB are more suitable.