GeoServer is mature, widely deployed, and well documented; yet production deployments routinely underperform significantly compared to what the hardware can actually deliver. The culprit is rarely the software itself: it is the gap between a working installation and a tuned one.

This talk benchmarks GeoServer against itself. No competing software, no vendor comparisons — just experiments isolating the cost of common operator decisions: data format choices, security configuration, logging verbosity, CRS handling, JVM settings, service limits and more. Each experiment is designed around a question a practitioner actually faces, and each answer comes with a number attached.

Topics will be drawn from: the real cost of leaving data in the wrong format for your query shape; how much security and logging configurations can silently degrade throughput; what reprojection actually costs at scale; how single-node and clustered deployments compare under sustained and spike load; and where vertical scaling stops being the answer, amongst others.