Hi, On Tue, 8 May 2007, Theodore Tso wrote: > On Tue, May 08, 2007 at 11:38:46AM -0400, Nicolas Pitre wrote: > > On Tue, 8 May 2007, Theodore Ts'o wrote: > > > > > Signed-off-by: "Theodore Ts'o" <tytso@xxxxxxx> > > > > I'd prefer if tests were performed on the performance impact before > > changing the default depth. If done separately from this patch then the > > commit log could contain those results as well. > > The following results are on a recent git repository, using time to > record the real, user, and sys times on the two commands: "git-gc > --no-reuse-delta --window=X --depth=Y" and "git log --pretty=oneline > -S'object' > /dev/null". All of these tests were done with a hot > cache, so disk speed didn't enter into the calculations. > > git-gc git log -S'object' pack size > w=10,d=10 27.1s/25.2s/0.3s 20.8s/20.4s/0.1s 15292k > w=10,d=30 23.8s/22.3s/0.2s 21.2s/20.9s/0.1s 12996k > w=10,d=50 24.8s/22.4s/0.4s 21.8s/21.2s/0.1s 12340k > w=100,d=100 24.1s/22.8s/0.3s 22.4s/21.8s/0.2s 11772k > > w=30,d=10 45.0s/43.1s/0.4s 20.8s/20.5s/0.1s 14388k > w=30,d=30 35.8s/34.1s/0.3s 21.6s/21.1s/0.1s 11800k > w=30,d=50 34.6s/33.0s/0.3s 22.1s/21.4s/0.1s 11376k > w=30,d=100 34.0s/32.2s/0.3s 22.2s/21.6s/0.1s 11012k > > w=50,d=10 56.1s/54.3s/0.4s 21.3s/20.5s/0.1s 14224k > w=50,d=30 47.2s/45.4s/0.4s 21.6s/21.0s/0.1s 11496k > w=50,d=50 44.5s/43.0s/0.3s 21.7s/21.2s/0.1s 11108k > w=50,d=100 44.3s/42.7s/0.4s 22.4s/21.7s/0.1s 10824k > > So a couple of things immediately become evident. First of all, as > Junio predicted, changing --depth makes no difference to the git-gc or > git log times. The latter is thanks to the delta chaching. Secondly, > changing --depth does make a signficiant difference to the pack size. > > Finally, --window does help somewhat in reducing the pack size, but it > _significantly_ increases the time to calculate the pack. > > My conclusion given this quick benchmark is that it seems to me that > changing the defaults of --depth to 50, and keeping --window at 10, is > a reasonable thing to do. I'd be happier if that test was done on _at least_ the kernel repo, if not something larger, _plus_ having the numbers on page faults. Swapping can kill performance substantially... git.git is small. Ciao, Dscho - 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
