[treerex]

I did this 20 years ago. My allocator had multiple regions (slabs) that were tuned to specific allocation patterns and lifetimes, with different algorithms for managing block size, fragmentation handling, and freeing. We could turn on detailed allocation/free logging, which included the file/line the action took place. From this we could put the system under load and then identify allocation hotspots and build models on block size and lifetimes and tune the allocators from that. It was pretty cool.