Is the io_uring code faster than mmap because multiple worker threads? If we use mmap from multiple threads will we get almost the same result?