I tried to replace std::unordered_map with boost::concurrent_flat_map and saw a memory footprint I could not explain. It was easy to find this line:
using multimutex_type=multimutex<mutex_type,128>; // TODO: adapt 128 to the machine
Can this be fixed?