Adblock Plus and (a little) more

New round of AMO's performance tests (2011-04-16) · 2011-04-17 17:42 by Wladimir Palant

I uploaded the overview for the new round of AMO‘s performance tests. The bug that skewed most Windows results so far has been fixed, so the results are more usable now. This week we have:

  • 2 add-ons (SimilarWeb and Personas Plus) have extremely high Windows 7 numbers (more than 200% slowdown) for no apparent reason. The same add-ons score around 25% on Windows XP. Update (2011-04-18): Both measurements appear to be correct, these add-ons really cause very significant slowdowns on Windows 7 (bug 650709). Why the same slowdowns supposedly don’t happen on other platforms is a good question, there might be bugs in Firefox itself involved. Update2 (2011-04-18): As Ted Mielczarek guessed correctly in bug 650709, the real issue here is IPv6 DNS resolution. Adding a pref to the test environment makes the delays introduced by these two extensions “normal” again.
  • 9 add-ons that weren’t tested because of being platform-dependent (bug 648225)
  • 2 add-ons (Google Toolbar and Memory Fox) weren’t tested because they have no stable version for download on addons.mozilla.org
  • 1 add-on (Tab Mix Plus) wasn’t tested because Talos couldn’t understand its install.rdf file (bug 648978)
  • 2 add-ons (Adblock Plus and Better Privacy) weren’t tested because the browser couldn’t be closed due to modal dialogs (bug 648222)
  • 1 add-on (NoScript) wasn’t tested because it prevented the test page from working correctly
  • 6 add-ons that were tested disabled because of not being compatible with Firefox (typically Thunderbird add-ons)
  • 29 add-ons that were tested disabled because of not being explicitly compatible with Firefox 4 (bug 648229)
  • 16 add-ons with test results probably considerably skewed by bug 648734 (standard deviation above 10% of the total Firefox start-up time). A lot more results were marked with a warning because of standard deviation above 2% of the total Firefox start-up time but I won’t count those.

Tags:

Comment [8]

  1. Haploid · 2011-04-17 18:25 · #

    So by using modal dialogs, you can ensure your addons scores well on the slowdown test. Congratulations on moving from 12th to 96th on the charts! </sarcasm>

    Reply from Wladimir Palant:

    To be fair, fixing this issue seems non-trivial.

  2. Benoit Jacob · 2011-04-17 18:27 · #

    Thanks for your very constructive approach to this whole performance tests story, improving the situation one bug at a time.

  3. Mike Ratcliffe · 2011-04-17 20:43 · #

    An impressive rundown of the statistics. It is a shame that Mozilla still aren’t adding the time in milliseconds to the results but your chart is much more informative.

    Thanks for your hard work.

  4. Alfred Kayser · 2011-04-18 10:36 · #

    How about testing the impact of themes? Some themes, such as Walnut2, will certainly impact the performance of FF in a negative way, but some may improve performance (LittleFox). (both are mine, so I am not blaming other people).

    Reply from Wladimir Palant:

    You are asking the wrong guy – I am not the one doing these performance measurements. You probably want to file a bug blocking https://bugzilla.mozilla.org/show_bug.cgi?id=599169. However, I don’t see any themes in the Top100 right now so it won’t be relevant in the short term.

  5. Anonymous · 2011-04-18 23:35 · #

    Personally, I’d greatly appreciate it if the onbeforeunload went away in Adblock Plus. Have you considered simply showing the ABP icon as “disabled” if the user has no filters?

    On a related note, would you consider not showing the first-run page at all if the user already has ABP filters?

    Reply from Wladimir Palant:

    Users who have filters shouldn’t see the first-run page. As to onbeforeunload, I plan to get rid of it in Adblock Plus 1.4 – Adblock Plus will simply subscribe you automatically and the first-run page will be about undoing this decision if necessary. Which should be a better course of action for the overwhelming majority of Adblock Plus users.

  6. HG2H · 2011-04-25 04:06 · #

    Hello,

    Great work, thanks for your efforts! I noticed you tested TorButton by Mike Perry and you noted TorButton v1.2.5 isn’t compatible with Firefox 4; you are correct. However, there is a TorButton alpha version (v1.3.2-alpha) that is compatible with Firefox 4. I posted links to the alpha version below[1,2].

    For what it’s worth, Mike and the Tor Project will be removing TorButton from AMO soon. This is due to slow AMO review process for updates and because TorButton will soon only be available via the Tor Project’s portable “Tor Browser Bundle” [https://www.torproject.org/projects/torbrowser.html.en] because toggling TorButton on and off is a flawed usage model [see: https://lists.torproject.org/pipermail/tor-talk/2011-April/020077.html]. Soon, Tor Browser Bundle will include a forked version of Firefox 4 by the Tor Project that includes a patch to fix the hardcoded SOCKS timeout bug in all Firefox versions [see bug #280661 at https://bugzilla.mozilla.org/show_bug.cgi?id=280661], as well as other bugs that Firefox devs are very slow to fix, even when provided with a patch by the Tor Project.

    [1] TorButton website: https://www.torproject.org/torbutton/index.html.en
    [2] TorButton current alpha XPI direct URL: https://www.torproject.org/dist/torbutton/torbutton-current-alpha.xpi

    I was going to run the test you ran with the ‘about startup’ add-on for TorButton, until I noticed you already did so :-).

    I wonder if you wouldn’t mind testing TorButton again, but use the FF 4 compatible TorButton alpha version this time? I ask because you can test on many more platforms than I (I am limited to Windows XP SP3). If you are willing to test the alpha TorButton I would be very happy to let Mike know the results, if you would not mind of course. I know he would be interested in the results too.

    Lastly, how did you test TorButton? I assume you installed TorButton, then started Firefox 4, then set the TorButton proxy settings to not use Tor. Or maybe I am assuming incorrectly and you tested TorButton with default proxy settings? If it’s the latter, did you have Tor running when you tested TorButton?

    I think the best test method for TorButton may be configuring TorButton to not use Tor, so the test result isn’t affected in some unseen way(s). That way you would only test TorButton load time (which should be) unaffected by issues such as using Tor with TorButton if TorButton is configured to use Tor and Tor is running, or not using Tor if TorButton is configured to use Tor but Tor is not running.

    If you add the following lines to user.js TorButton will start in the ‘on’ state but it will not try to load Firefox through Tor. The following settings will simply tell TorButton to not configure HTTP, HTTPS and SOCKS address:port in Firefox:

    user_pref(“extensions.torbutton.http_port”, 0);
    user_pref(“extensions.torbutton.http_proxy”, );
    user_pref(“extensions.torbutton.https_port”, 0);
    user_pref(“extensions.torbutton.https_proxy”, );
    user_pref(“extensions.torbutton.socks_host”, );
    user_pref(“extensions.torbutton.socks_port”, 0);
    user_pref(“extensions.torbutton.settings_method”, “custom”);

    Thanks so much!

    Reply from Wladimir Palant:

    I don’t test anything – I only provide a better version of AMO’s testing results because the official overview (https://addons.mozilla.org/en-US/firefox/performance/) has so many issues and provides so little real information.

  7. Danial · 2011-05-06 21:00 · #

    google rarely seems to update their Mozilla addons page, but the latest toolbar was updated in march.

    Reply from Wladimir Palant:

    https://addons.mozilla.org/en-US/firefox/addon/google-toolbar/ gives me 404 not found. So there is nothing to be tested.

  8. Danial · 2011-05-06 21:56 · #

    yeah, google hasn’t used mozilla addons for a long time

    the latest is usually found at http://www.google.com/toolbar/ff/index.html

Commenting is closed for this article.