bzr on Emacs performance suggestions

Óscar Fuentes ofv at wanadoo.es
Thu Nov 26 03:17:55 GMT 2009


"Stephen J. Turnbull" <stephen at xemacs.org> writes:

[snip]

> The bad news is reported elsewhere: performance is visibly poor
> compared to git.
>
> The worse news: git's standard infrastructure is in place even though
> it's at best going to be a mirror.  bzr offers only basic html and
> sftp access,

I'm more concerned about performance of everyday VC operations than of
branching.

[snipped some common-sense suggestions about the progress meter]

> In the bzr over http test, the download was clearly either stalling,
> or for long periods (in CPU cycles) in the background from the point
> of view of the rate measurement, because it would drop to near 0 about
> once a second, then bounce back up.  I can only speculate about cause
> if it was really stalling; maybe the httpd is heavily loaded compared
> to the git server?

I don't think so. Most likely bzr was having an elaborate conversation
with the http server (similar to what happens with `bzr branches'). The
cloning took 23 minutes for me: 8 minutes is required for transmitting
the raw data over the 6 Mb/s ADSL, 10 minutes of local CPU activity for
building the branch, this leaves 5 minutes for chit-chat. And sets a
minimum of 18 minutes for my combination of machine/connection. Unless
the smart server transmits the data on a form that relieves the local
client from a significant amount of work, the real bottleneck is at
home, where the branch is made from the stream of data.

It is impressive, though, that bzr's Python uses only 2.5 times more CPU
than git's C. AFAIK, bzr has no compiled parts, right?

-- 
Óscar




More information about the bazaar mailing list