Building a production-grade WMTS for satellite imagery sounds straightforward—until you hit cloud-native reality. This talk charts UP42’s journey from a 2024 hackathon to a live beta service, exposing the hard lessons learned while delivering STAC-catalogued imagery to professional GIS tools at scale.

The Architecture Battle
Why GeoServer? While commercial options were cost-prohibitive and emerging tools like TiTiler lacked maturity, GeoServer offered OGC-compliance and a proven REST API. However, "standard" setups quickly crumbled under B2B demands. We’ll dive into:

The FUSE Trap
How GCP Cloud Storage FUSE crippled tile-seeding performance, and why we pivoted to native GCP BlobStore plugins to unlock massive throughput.

The Scaling Wall
Why vanilla GeoServer’s clustering failed us, necessitating a strategic (and bumpy) migration to GeoServer Cloud’s microservices architecture.

The Integration Tax
Real-world troubleshooting of Helmfile fixes, CSI driver misconfigurations, and the "silent" bugs that haunt cloud-native geospatial stacks.

Hard-Earned Takeaways
We’re sharing our internal RFCs and benchmarks so you don’t have to learn the hard way. Learn why you must evaluate plugin maturity over hype, why FUSE is a bottleneck for tile writes, and why continuous load testing is the only way to survive the jump from "it works on my machine" to "it works for the world."

For any geospatial platform, the ability to serve imagery at scale is the ultimate "stress test." When our team at UP42 began building a production-grade WMTS service, we quickly realized that moving from a functional setup to a high-performance one requires more than just adding more hardware. This talk shares our iterative journey of migrating from a "Vanilla" GeoServer architecture to a microservices-based GeoServer Cloud environment, and the systematic load testing that guided every decision along the way.

We will walk through our "detective-style" approach to performance tuning. Using Apache JMeter to simulate heavy production loads, we treated our infrastructure as a series of integration points where bottlenecks could hide. Rather than a smooth transition, the move to a cloud-native architecture revealed a new landscape of challenges that required us to look deeper into the system than we had ever anticipated.

Throughout our testing phases, we uncovered a variety of hidden performance killers, including:
- Storage Hurdles: How standard cloud-mount solutions struggled with tile-writing workloads and why native cloud storage plugins became essential.
- Concurrency Caps: The realization that default configurations for thread limits and traffic control are often too conservative for modern cloud environments.
- The Proxy Trap: How internal communication between services can become a bottleneck even when individual components are performing well.
- Resource Optimization: The relationship between CPU/Memory allocation and the ability to handle parallel tasks like simultaneous seeding and streaming.

This presentation is a practical guide for anyone looking to push GeoServer beyond its default limits. We will share our "battle map" for isolating bottlenecks, bypassing load balancers for diagnostic testing, and the critical importance of keeping load testing continuous as your architecture evolves.

Key Takeaways
- The Migration Reality: GeoServer Cloud offers the foundation for horizontal scaling, but it requires a specialized tuning strategy compared to standalone instances.
- Systematic Isolation: Learn how to test individual microservices (GWC, Gateway, etc.) in isolation to pinpoint exactly where latency is introduced.
- Visibility Matters: The importance of combining log analysis, thread dumps, and performance metrics to solve "silent" performance issues.
- End-to-End Testing: Why you must test the full integration path early to find bottlenecks that only appear under high concurrency.