WM(T)S-wrapper på COG

[norkart-alexanno]

• Undersøke hva finnes fra før)
• Hvordan støttes reprojeksjon i TiTiler (og de andre)? COG som 25833 - output som 3857? Hva er fordeler/ulemper?
• Hva med Mapproxy og evt custom plugins? Hva med OGC-støtte? Lage et eksempel?
• Lage en docker-demo TiTiler med 2-3 COGs (titiler demo scaffolding #76 )
• Oppsummere / reviewe oversiktstabellen til @espen080 over og dokumentere (kort)
• README.md med erfaringer

[espen080]

Vi har så smått startet å planlegge mot dette, det blir muligens aktuelt å gjøre en Proof of Concept med Norge i Bilder på COG + STAC + WM(T)S kompabilitet.

[espen080]

Basics

Hva er en WM(T)S server

OGC WMS
OGC WMTS

Relevante lenker

GeoServer COG

KI oppsummering

Summary Table

Solution	Plug & Play w/ COG	COG Structure Used (cloud/remote)	WMTS/TMS Output	True OGC WMS?	Notes
Titiler	✅	✅ (full)	✅	❌	Best for "COG-first" cloud work
COG-Explorer	✅	✅	✅	❌	Lightweight, usually demo/utility
pg_tileserv	❌	❌	✅	❌	Must ingest as rasters, not COG
MapServer	✅	❌ (local only)	✅	✅	Behaves like GeoServer for COGs
QGIS Server	✅	❌ (local only)	✅	✅	Standard GeoTIFF handling
GDAL+vsicurl	🟨 (dev)	✅	🟨	❌	For scripting or custom pipelines

Faktisk oppsummering

COG-Explorer

https://geotiffjs.github.io/cog-explorer/
https://github.com/geotiffjs/cog-explorer

Har ikke gått veldig dypt i denne, men virker som en web applikasjon for å overlaye COG, uten at det er noe kompabilitetslag for WMS/WMTS api.

pg_tileserv

https://access.crunchydata.com/documentation/pg_tileserv/1.0.11/
https://github.com/CrunchyData/pg_tileserv

Genererer tiles on the fly fra Postgresql, har såvidt jeg kan se ingen kompabilitetslag for WMS/WMTS

MapServer

https://mapserver.org/input/raster.html
https://mapserver.org/input/virtual-file.html

Ser ut til å støtte remote COG input ved å lene seg på GDAL + Virtuelt filsystem. Krever dog full MapServer instans.

Qgis Server

https://docs.qgis.org/3.40/en/docs/server_manual/introduction.html

Har ikke verifisert at denne kan lese filer remote, men den ser og ut som den lener seg tungt på GDAL og antas å ha lignende funksjonalitet som de andre server rammeverkene (Map/Geo)

GDAL

Jeg vurderer dette som custom implementasjon, teknisk fullt mulig, men antakeligvis svært ressurskrevende..

[espen080]

TiTiler

Denne virker lovende

Docs

Tester her COG + TiTiler

WMS Extension

https://developmentseed.org/titiler/api/titiler/extensions/wms/

[alexanno]

Notater fra drøfting i møte:

GeoServer støtter COGs som storage på "remote buckets". Men kan "WMS"-laget være "tynnere"? (og hva er tynt?)
Mapserver har også en form for COG-støtte - men kanskje ikke "remote buckets".
TiTiler er nyttig å undersøke
Mapproxy er lovende - men ikke sett på nøyere.

[alexanno]

SigTill har en demo på TiTiler - lyst til å teste.

• Legg inn som egen branch
• Bruke frigitte COG-data

[alexanno]

@alexanno laster opp frigitte geovekst-data til blob-storage

[espen080]

Mapproxy

https://mapproxy.github.io/mapproxy/latest/plugins.html#

COG Plugin? 🤔

[ben-worsley]

Oppsummering av min diskusjon med KI:

COG → WMTS/WMS: Cloud-Native Architecture Reference

The Core Challenge

Serve high-performance, OGC-compliant WMTS/WMS from COGs in S3 while maintaining backward compatibility. This is not about ideology—it's about sustainable, scalable service delivery over years of transition.

---

Two Proven Architectural Patterns

Pattern A: Dynamic Rendering

When: Medium concurrency, correctness prioritized over throughput.

S3 COGs
  ↓
MapServer / QGIS Server (via GDAL + /vsis3/)
  ↓
WMS/WMTS clients

Trade-offs:

• ✅ No tile precomputation, minimal redundancy
• ✅ Always reflects latest COG data
• ❌ Performance tuning non-trivial
• ❌ S3 latency leaks into response times

---

Pattern B: Hybrid (Recommended for scale)

When: National mapping, high concurrency, long-term service.

S3 COGs (authoritative)
  ↓
MapServer / QGIS Server (dynamic WMS rendering)
  ↓
MapCache or MapProxy (WMTS tile cache)
  ↓
CDN (optional but recommended)
  ↓
Clients (WMS / WMTS)

Trade-offs:

• ✅ WMTS: blazing fast, cached
• ✅ WMS: dynamic, correct
• ✅ COGs remain authoritative
• ✅ Clear separation of concerns
• ❌ Additional cache infrastructure

---

Solution Positioning

Solution	COG-Aware (Cloud)	WMTS	True OGC WMS	Best For
TiTiler	✅ Full (stateless)	✅ Dynamic	❌	Cloud-native streaming; XYZ/TMS; COG-first workflows
MapServer	✅ (via GDAL)	✅	✅	OGC-compliant rendering engine; proven at scale
QGIS Server	✅ (via GDAL)	✅	✅	Standards-heavy environments; easier config
MapCache	N/A (cache layer)	✅	✅	High-performance WMTS caching; paired with MapServer
MapProxy	N/A (cache layer)	✅	✅	Protocol adapter; multi-source federation; flexibility
MosaicJSON	✅ Full	🟨 Custom	🟨 Custom	Dynamic mosaic aggregation; cloud-native; scalable

---

High-Resolution Orthophotos (~18 cm)

Performance winner: MapCache
Flexibility winner: MapProxy

Both handle metatiling efficiently. Choose based on:

• MapCache → raw throughput, minimal overhead, tight MapServer coupling
• MapProxy → flexibility, protocol bridging, easier multi-source management

Critical: Neither reads COGs natively. GDAL (~S3 via /vsis3/) is the bridge. Cache efficiency matters far more than proxy overhead.

---

National Orthophoto Mosaic (Multi-Project)

Pattern: VRT or MosaicJSON aggregation

Approach	Storage	Scalability	Latency	Flexibility
GDAL VRT	None (virtual)	Medium	Depends on S3	Low (fixed)
MosaicJSON	Small metadata	High	Low (dynamic tiling)	High (add COGs dynamically)
Precomputed tiles	High (duplication)	Very High	Very Low	Low (batch process)

Recommendation:
VRT for conservative/proven; MosaicJSON for future-ready scalability.

Pair with:

• MapCache/MapProxy for OGC clients (WMS/WMTS)
• CDN for XYZ/TMS modern clients
• Optional TiTiler for high-volume dynamic API access

---

Technical Non-Negotiables

1. COGs must be perfect:

   • Internal tiling matches WMTS tile size (256–512 px)
   • Overviews exactly align with tile matrix
   • Compression for range-read efficiency (DEFLATE/ZSTD)

2. S3 access optimized:

   • Same region as servers
   • Persistent connections
   • Use /vsis3/ over presigned /vsicurl/

3. Cache is not optional at scale:

   • Dynamic WMTS from S3 alone will not survive load
   • MapCache, MapProxy, or CDN essential for WMTS

---

Strategic Forward Path

Phase 1 (Stable)

• COGs in S3
• MapServer + MapCache
• Full OGC compliance
• Zero architectural risk

Phase 2 (Modernize)

• Introduce TiTiler for internal APIs / XYZ access
• CDN edge caching
• Parallel OGC layer
• Gradual client migration

Phase 3 (Cloud-Native Core)

• OGC becomes compatibility façade
• TiTiler + MosaicJSON drives volume
• Thin WMS adapter for legacy clients
• Reduced operational complexity

---

The Honest Conclusion

You can have:

• Cloud-native COG storage
• High performance and scalability
• Long-term OGC compatibility

You cannot avoid:

• A caching/proxy layer (MapCache, MapProxy, or CDN)
• GDAL as the bridge between COG and rendering
• Some operational overhead during transition

The hybrid pattern (Pattern B) is not "old tech" —it is the proven, mature, enterprise-safe way to modernize storage while maintaining compatibility. It is genuinely cloud-native: object storage as source of truth, stateless compute, caching pushed outward, replaceable components.

---

Key Insight for Your Workshop

"Cloud-native is not 'no servers' or 'only TiTiler'—it means object storage as the source of truth, stateless compute where possible, and caching pushed outward. The hybrid OGC + cache model satisfies this definition while meeting real-world backward compatibility needs."