Performance & Bottlenecks

Understanding the performance characteristics of the Physics and Ecosystem engines is crucial for optimization.

Hotspots

Voxel Traversal

Iterating over voxels is the most computationally expensive operation.

• Location: GridBoxProcessor::processBox

• Impact: Even with OpenVDB’s efficient accessors, the sheer volume of voxels in a large world creates significant overhead.

• Mitigation: Use active voxel iterators where possible instead of dense iteration over the bounding box.

Entity Iteration

Linear scans of entities can become a bottleneck as the population grows.

• Location: PhysicsEngine::processPhysics

• Impact: Iterating over all entities with MovingComponent scales linearly with entity count.

• Mitigation: Use spatial partitioning or chunks to limit processing to active areas.

Lock Contention

TerrainStorage Lock

The TerrainStorage lock is a central contention point.

• Issue: If both engines and multiple worker threads try to lock the terrain grid simultaneously, performance degrades significantly due to thread blocking.

• Observation: High contention is observed during periods of intense water simulation (e.g., heavy rain) when many write operations are queued.

Memory Bandwidth

• Issue: The EcosystemEngine creates many small WaterFlow objects and pushes them to concurrent queues.

• Impact: High water activity can saturate memory bandwidth with allocation and queue operations, leading to cache thrashing.