Optimizing Large-Scale Vector Data Rendering in osgEarth(3.7.2)

[KLEEN2000]

Hello osgEarth team,

I am currently working on optimizing the loading and rendering of a very large volume of vector data in osgEarth (version 3.7.2).

When using the native osgEarth APIs, I found that the performance differs significantly from my expectations. As the dataset size increases even moderately:

**FeatureModelLayer is no longer able to handle the scale of my data.

When using FeatureImageLayer, tile generation becomes noticeably slower.**

Because of this, I am considering implementing a lower-level optimization approach. However, osgEarth is a large and complex system, and I am not sure where the appropriate entry point in the source code would be.

Let me describe my requirements more clearly:

**The number of features is on the order of millions.

I need to separate my custom tile rendering system from osgEarth’s native tile system.

The native tile system should be responsible only for rendering base imagery and elevation.

My custom tile rendering system needs to overlay on top of the native terrain tiles.

Both systems should run in synchronization (LOD selection, tile lifecycle management, culling, etc.).**

I hope this description makes sense.

I would greatly appreciate your guidance on the following questions:

**From which part of the source code should I begin if I want to implement a custom tile rendering pipeline?

What is the recommended way to integrate a custom tile system into osgEarth so that it runs alongside and remains synchronized with the native terrain tile system?**

Thank you very much for your time and support.

[gwaldron]

Have you experimented with TiledFeatureModelLayer? This newer class replaces the older FeatureModelLayer with a much better and more controllable structure for loading huge datasets.

We typically use OpenStreetMap vectors as the test dataset and it certainly runs into the millions (or even billions depending on how you count) of features. TFML handles it well as long as you are thoughtful about your setup. In the repo there are a couple examples including feature_tile_level.earth and osm.earth.

[KLEEN2000]

I am currently using OgrFeatureSource to open a shapefile and then attaching it to a TiledFeatureModelLayer.

After that, I add the TFML to a map that already contains a GdalDEM layer. However, I noticed that the call to map->addLayer(tfml) takes a very long time to complete.

The shapefile I am using contains approximately 3,500 polygon features and has a total size of about 6 MB.

Given this relatively small dataset size, I did not expect such a long delay during layer initialization. Could you help me understand what might be causing this behavior and where I should begin investigating?

[gwaldron]

Make sure your shapefile has a spatial index; you can use the ogrinfo tool to create one:

ogrinfo -sql "CREATE SPATIAL INDEX ON myfile" myfile.shp

Beyond that I'd need to see your TFML setup to help further. It's usually best to start with an earth file and tweak the setup before moving to code.

I would also recommend grabbing the master branch if you can - there have been updates to TFML since 3.7.2.

We are planning a 3.7.4 release soon.