On Sat, 17 Mar 2007, Linus Torvalds wrote: > > So the cache doesn't need to be "complete". It will get good hit-rates > even from being very simple. I think I have a very simple and cunning > plan, I'll try it out asap. Ok, got distracted by guests coming to look at the new puppy, so it took longer than it should have, but the following is a simple two-patch series that improves path-following by a factor of almost 2.5 for me. The cache is *really* simple. It's just a 256-entry hashed cache of the last few base entries, and it brings down my test-case of git log drivers/usb/ > /dev/null from 2.5s to just over 1s. I have *not* tuned or tweaked this at all, and maybe there are better ways to do this, but this was simple as hell and obviously quite effective. It also speeds up "git blame", for all the same reasons. Before (best times out of a run of five): [torvalds@woody linux]$ time git blame drivers/char/Makefile > /dev/null real 0m1.585s user 0m1.576s sys 0m0.004s after: [torvalds@woody linux]$ time ~/git/git blame drivers/char/Makefile > /dev/null real 0m0.763s user 0m0.644s sys 0m0.120s so it's a factor of two there too (just a random file, I'm not at all going to guarantee that this is really consistent - it should get more testing etc). The first patch just does some obvious re-factoring and setting up (no real code changes). The second patch just uses the new functions to actually add a cache. Linus - To unsubscribe from this list: send the line "unsubscribe git" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html