mmap is a good crutch when you 1. don't have busy polling / async IO API available and want to do some quick & dirty preloading tricks; 2. don't want to manage

mmap is a good crutch when you 1. don't have busy polling / async IO API available and want to do some quick & dirty preloading tricks; 2. don't want to manage the complexity of in-memory cache, especially cross-processes ones. Obviously if you have kernel-backed async IO APIs (io_uring) and willing to dig into the deeper end (for better managed cache), you can get better performance than mmap. But in many cases, mmap is "good-enough".