yet another dvcs shootout (this time python-devel)

[Brendan Cully]

On Wednesday, 05 November 2008 at 20:42, Dennis Brakhane wrote:
> I doubt that that was a fair comparison.
> 
> On Wed, Nov 05, 2008 at 10:19:55AM +0100, Arne Bab. wrote:
> > -----------------------------------------------------------------------------------
> > $ time bzr branch http://code.python.org/python/3.0
> > Branched 37407 revision(s).
> 
> 37407 revisions/changesets, vs.
> 
> > $ time hg clone http://code.python.org/hg/branches/py3k/
> > added 2532 changesets with 16692 changes to 4528 files
> 
> 2532 changesets
> 
> Seems like the hg branch either was created later or is outdated.

that looks unfair, I agree.

> Maybe somebody could convert the bzr-branch to hg and clone this. The cloning should
> probably also happen via the loopback interface, to keep network latencies from biasing
> the statistics.

But I don't agree with this. The source repositories are hosted on the
same site. Network latency is a large part of actual performance - you
want more pipelining, less round trips. If you want an interesting
synthetic test, you could use netem/tokenbucket to force a bandwidth
limit (say 1Mbps) and consistent latency (100ms?). Assuming infinite
bandwidth and zero latency is not going to give you a good idea about
real world performance.