Hi Miklos, On Fri, 05 Apr 2019 at 09:17:01 +0200, Miklos Vajna wrote:
One more aspect: perhaps I'm just unlucky, but on average, I found cgit to load much faster than these git.libreoffice.org links.
Hmm that's interesting. That's not something I'm able to reproduce by
taking 100 random commits from the last 10000 non-merge commits in
master:
$ git log -n 10000 --no-merges --pretty="%H" master | shuf -n 100 >/tmp/commit_list
$ ./measure.sh "https://git.libreoffice.org/core/+/%s%%5E%%21/"; </tmp/commit_list
[…]
average stddev minimum maximum
------------------- ------- ------ ------- -------
size (byte) 41906 75197 4564 409087
start transfer time 0.44 0.42 0.29 2.95
total time 0.56 0.46 0.30 2.96
$ ./measure.sh "https://cgit.freedesktop.org/libreoffice/core/commit/?id=%s"; </tmp/commit_list
[…]
average stddev minimum maximum
------------------- ------- ------ ------- -------
size (byte) 70122 63175 39247 390793
start transfer time 0.37 0.19 0.28 2.06
total time 0.59 0.34 0.29 2.22
(I'm not running running this from TDF's premises, by the way; though it
wouldn't matter as latency isn't the bottleneck here.)
Repeating the measurement with the same 10 random commits:
https://git.libreoffice.org/core/
average stddev minimum maximum
------------------- ------- ------ ------- -------
size (byte) 19691 17007 4334 64345
start transfer time 0.80 0.68 0.27 2.13
total time 0.86 0.66 0.29 2.17
start transfer time 0.12 0.09 0.08 0.38
total time 0.16 0.09 0.09 0.39
start transfer time 0.12 0.09 0.09 0.38
total time 0.16 0.09 0.09 0.39
https://cgit.freedesktop.org/libreoffice/core/
average stddev minimum maximum
------------------- ------- ------ ------- -------
size (byte) 52903 15767 39077 94857
start transfer time 0.39 0.16 0.30 0.85
total time 0.73 0.30 0.31 1.17
start transfer time 0.23 0.15 0.17 0.67
total time 0.56 0.31 0.33 1.30
start transfer time 0.23 0.15 0.18 0.67
total time 0.56 0.37 0.18 1.48
So on first glance it appears that occasionally a gitiles page takes
longer to load (though I wasn't able to reproduce the 20s you measured,
more like 2-3s in my case). However it's AFAICT rare enough that it
doesn't impact the average load time. And caching makes subsequent hits
constantly faster. (I dunno what's the caching strategy here, something
internal to JGit probably.)
Cheers,
--
Guilhem.
Attachment:
measure.sh
Description: Bourne shell script
Attachment:
signature.asc
Description: PGP signature