Get the latest tech news
Optimizing My Disk Usage Program
Increasing performance by reducing thread scheduling overhead and lock contention.
I also tested with some hash functions that avalanche sequential integers but the results were comparable to my original idea with just the modulo. APFS hands out IDs sequentially so the bottom bits toggle fastest, causing threads that are scanning directory trees in creation order to pile into the same shard when using inode % 128. This improves temporal locality and cuts down inter-shard contention in our multithreaded crawl even though the statistical distribution stays perfectly flat across the whole disk.
Or read this on Hacker News