I attended a talk in 2024 by Guy Davidson where he talked about memory management in C++ and file I/O.
A particular topic he talked extensively about was filemapping in C++ and how it improved file I/O susbstantially.
Curious, I decided to make my own filemapping library to see how far I could push the performance of this method.
File mapping is useful for many things, but it's most common case is to quickly read and write large amounts of data to files.
Performance was tested by writing / reading 100 MiB worth of data to / from a file using STL and using my own library (rapidio) averaged over 100 iterations.
The results are as follows:
As you can see, my own implementation using filemapping is far faster than STL's built-in solution.
The entire code can be found here at GitHub