Bazaar repository size benchmarks

[John Arbash Meinel]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Pieter de Bie wrote:
| Hello,
|
| I did some benchmarks on repository size for repositories with full
| history. I compared the size of Git, Bazaar as well as Mercurial
| repositories. The results of this experiment can be seen here:
| http://vcscompare.blogspot.com/2008/06/git-mercurial-bazaar-repository-size.html
|
| The short conclusion is that Bazaar comes out last in the test. It's
| repositories are on average 2.8 times as big as those of Git.
| Mercurial also does better.
|
| This should be put in contrast with
| http://bazaar-vcs.org/Benchmarks/SpaceEfficiency , where Bazaar comes
| out on top. The reason for this is that my benchmarks include full
| repository history and also pack the repository before measuring
| repository size. Furthermore, the SpaceEfficiency benchmark does not
| take hardlinks in account (which Git uses when cloning a repository).

Did you delete the .bzr/repository/obsolete-packs/* after doing the 'bzr pack'?

That will cut about 50% of the size out. When we repack, we always save the old
packs for a bit. So that OS scheduling issues don't destroy data that you cannot
recover.

If you are issuing "bzr pack", that copies the entire repository.

|
| I think I did a more realistic comparison of repository size. As such,
| is it perhaps necessary to update the benchmark page and the front
| page (which states that "Bazaar's default storage format is highly
| efficient, better than its main competitors in most cases")?
|
| - Pieter

I agree that it is more realistic, though I'm curious to make sure it is an
apples-to-apples comparison. I would also be interested in understanding where
we are lacking, as there are probably a few simple things that we could tweak.
(For example, we currently have a maximum delta chain of 200 for deltas. This
tends to have a larger impact on inventories. Hg has an unlimited length, and
Git always breaks the inventories down into a per-directory entry.)

John
=:->
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (Cygwin)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iEYEARECAAYFAkhCwz4ACgkQJdeBCYSNAAP+yACeKy5iEi18eRRJ3GHKzZdupfpv
W5AAoKAQv1OW6pnczCBC24Ha1BaOKego
=/zh+
-----END PGP SIGNATURE-----