Some unscientific timing results (on the Python source tree)

[David Cournapeau]

Jelmer Vernooij wrote:
> The smart server isn't significantly faster than plain http as far as I
> know. Pauls timings match what I have seen on repositories with tens of
> thousands of revisions in history. It's a lot better than what it used
> to be, but still behind on some other systems.

Does the size of the repository matter for pulling recent changes ? E.g 
will pulling 100 changes from a repository with 200 revisions be much 
faster than pulling the same 100 changes assuming 10000 revisions ? 
Intuitively, it should not, but I know nothing about how bzr works 
internally (or any DVCS for that matter).

I imported from bzr-svn one project with ~2000 revisions, put the branch 
on launchpad. Using bzr+ssh:// protocol, I get:

   - 4.30 minutes to branch the project from launchpad
   - 30 seconds to pull ~150 revisions (I branch the project at HEAD 
minus 150 revisions, and pull from the internet)
   - 1:30 minutes to pull ~650 revisions (same method: branch from HEAD 
minus 650 rev)

Now, using http:// instead, I get:

   - 8:50 minutes to branch the project from launchpad
   - 6:08 minutes to pull ~150 revisions
   - 4:52 minutes to pull ~650 revisions

That's a pretty significant difference. 30 seconds to pull recent 
changes is not great, but so-so (that's what I would get using svn, I 
think, for this case). 5 minutes is pretty bad.

cheers,

David