A small benchmark: Bazaar, Git, Mercurial
I just performed some benchmark on Bazaar, Git and Mercurial. This is a very small benchmark, I may publish a more detailed one in a few weeks.
Some notes:
- The unit in the first chart is seconds
- I haven’t run git-gc before taking the directory sizes. Running git-gc saves about 1083 kilobytes
- This PDF file is licensed under the MIT License, see http://opensource.org/licenses/mit-license.php
Interesting figures. Please keep on comparing the VCSes.
Martin "mhb" Böhm
January 13, 2008 at 21:07
Regarding: “I haven’t run git-gc before taking the directory sizes”; is this also true about the timing numbers?
Having ran git-gc on the repository would have made the git times even better.
bartman
January 13, 2008 at 21:35
Good stuff. Can’t wait to see more details on times of uploading and downloading. I know Bazaar will lose there and Git will typically win it.
nixternal
January 13, 2008 at 22:28
Great info. Did you the comparison Eric S. Raymond is working on?
http://lists.gnu.org/archive/html/emacs-devel/2008-01/msg00092.html
Eddy Pronk
January 13, 2008 at 23:51
How did you perform the disk space usage benchmark?
There is another comparison where bzr wins in: http://bazaar-vcs.org/Benchmarks/SpaceEfficiency
Martin Albisetti
January 14, 2008 at 20:19
Looks like a nice set of results.
A few questions:
* Warm cache or cold? How much memory on the system? Several of these systems take heavy advantage of disk cache and dentry cache, and will run much faster with a warm cache.
* Why did you not run git-gc?
* When you go to create more results, please consider benchmarking patch application, branch creation, merging, and rebasing.
Anonymous
January 15, 2008 at 00:09
I would argue proper use of git includes using git-gc. It affects timing as well as storage size, as already noted. (Historical: the pack format was introduced as an afterthought to be economical on network transfers, and the pretty large speedup of heavy operations came as a surprise even to Linus)
ulrik
January 16, 2008 at 23:38
I look forward to the following set of benchmarks in Plain Old PNG format
(HTML, etc)
Thanks for these. As Lennart Poettering said, you can use Subversion or Git, or you can make your users learn Yet Another revision control system (which is a barrier to entry).
Ted Percival
January 18, 2008 at 15:31
Can you provide a little bit more details wrt the settings used, etc? What file format was used for Bazaar, for example?
Jelmer Vernooij
January 20, 2008 at 15:13