Bazaar performance

[Matt Nordhoff]

I'm not a developer, but...

Guido Ostkamp wrote:
> Hello,
> 
> after finishing a conversion from Git to Bazaar with help of Ian
> Clatworthy (thanks, again, Ian!), I did some initial performance tests
> with Bazaar compared to Git and Mercurial on my large repository (~11000
> files, > 41000 changesets, repo size > ~500 MB).
> 
> These are the results (with bazaar 1.4):
> 
> Cloning from remote host via ssh/sftp:
> 
> Bazaar:    6:31 min   (bzr clone sftp://...)
> Mercurial: 3:39 min   (hg clone ssh://...)
> Git:       1:23 min   (git clone ssh://...)

Did you try bzr over bzr+ssh? (You need bzr installed on the remote side
for it to work.)

I doubt bzr will win, but with your current test it's at a disadvantage:
git and hg are using smart servers, but bzr is using dumb sftp.

<snip the other tests>

> So in any of the cases, bazaar is unfortunately the slowest system.
> Especially the poor performance on logging information queries appears
> strange.
> 
> I would be interested to understand the technical reasons.
> 
> As far as I know, the main differences between bazaar and the other two
> systems are that Bazaar is dealing with directory versioning, while the
> others are not and of course that Git is implemented in C instead of
> Python.
> 
> But does directory versioning really count in that much?
> 
> Do Git and/or Mercurial have data structures that Bazaar doesn't have
> which support faster access?

I would think directory versioning has just about nothing to do with it.

Robert Collins wrote about performance 2 months ago:

<http://thread.gmane.org/gmane.comp.version-control.bazaar-ng.general/39163/focus=39209>

Basically, when bzr was started, they didn't focus on performance, and
now that they are, there's a lot of work that needs to be done.

Also, I think the new pack format made partial-history operations much
faster, but whole-history operations slower, exacerbating some of the
current problems (like log).

> Regards
> 
> Guido
--