The idea
This project examined wildfire point data at high spatial resolution. The scale of the density calculation made performance a central engineering challenge.
Approach
I developed a hybrid Python/C++ OpenMP pipeline to process 787,327 NASA FIRMS fire points across a 960,000-pixel grid, involving approximately 755 billion Haversine density calculations. Profiling exposed false sharing as a multi-core bottleneck; after diagnosing it, the reported speedup was 1.35× with four threads.
Reported outcomes
Tools & methods
More about the project
The work combined Python for data handling with C++ and OpenMP for computation. The important optimization finding was that shared-memory behavior created false sharing and limited multi-core performance.
What I took from it
This project made the cost of memory behavior visible: adding threads alone did not remove the bottleneck, so profiling and diagnosing false sharing mattered to improving throughput.