Why is bzr pull via http soooooo sloooooow?

[Stefan Reichör]

John Arbash Meinel <john at arbash-meinel.com> writes:

> Do you have pycurl installed? (I'm assuming you have cElementTree already).
>
> If you don't have a caching DNS, we have some small issues where urllib
> doesn't cache the host addresses, so for every request we get another
> DNS request. We have the same issue with pycurl, but we have a couple
> proposed fixes for this. (Ultimately with pycurl we can do partial
> (range) requests, etc.)
>
> Also the number 14 is a little bit misleading. There are only 14
> mainline revisions, but it amounts to > 100 total merged revisions. (The
> encoding merge itself was > 80 revisions).

Thanks for your detailed answer.

Today I tried to pull with/without pycurl. Here are my results:

Without pycurl:
  % time bzr pull
  Using saved location: http://bazaar-vcs.org/bzr/bzr.dev/
  All changes applied successfully.
  2 revision(s) pulled.
  bzr pull  3.89s user 0.30s system 4% cpu 1:36.19 total


With pycurl:
  % time ./bzr pull
  Using saved location: http://bazaar-vcs.org/bzr/bzr.dev/
  All changes applied successfully.
  2 revision(s) pulled.
  ./bzr pull  3.65s user 0.17s system 19% cpu 19.905 total


So using pycurl was about 5 times faster in this case!


Stefan.