Measuring Ad Blocker Impact on Site Performance

1
MEASURING AD BLOCKER IMPACT ON SITE PERFORMANCE
2
Last week, a couple of blog posts came out
from Catchpoint and Rigor discussing the
performance implications of ad blockers, with a
focus on user experience. These posts were
timely as it happened, I presented on ad
blockers impact on site performance at the
Velocity Conference in Santa Clara, CA last week.
This post summarizes the findings I discussed in
my talk (and the aforementioned blog posts) for
those who were unable to attend the
conference. Measuring web performance can be
like peeling back the many layers of an onion due
to various interdependent components, such as web
server stack, network stack, characteristics of
the device and browser, and the application
itself. All these components MUST work together
to deliver a great user experience to the end
user. To drive higher business metrics, web
publishers want their pages to render fast and be
ready to be consumed quickly by their users. Thus
far most of the performance focus has been on
network-level optimizations like HTTP/2, and
application-level optimizations such as
optimizing the critical rendering path and image
compression. As applications are becoming richer,
more personalized and more interactive,
performance bottlenecks are tending more and more
toward user endpoints. Thus it has become really
important to look at factors that might impact
application performance in the browser. One
factor that we looked into are browser
extensions. BY FAR, ad blockers are the most
popular of the browser extensions. In May 2016,
one such ad blocker, Adblock Plus, announced more
than 100M active users. In researching tools
that could provide visibility into the impact ad
blockers have on web sites, apart from a little
anecdotal evidence and a few independent studies,
we couldnt find anything that would allow us to
measure the impact of extensions, particularly ad
blockers, in a reliable and standard way.
3

• What became clear was the dearth of tools
  available to measure the impact of ad blockers on
  a sites performance, so I added support for
  browser-based extensions on our private instance
  of WebPagetest. (Ive published a separate blog
  post with detailed instructions on setting this
  up in your own private WebPagetest instance.)
• With this tooling in place, we ran multiple
  experiments across different industry verticals
  such as eCommerce (2016 Internet retailer
  150) and Media (2016 Top 150 media sites). Here
  are some of our observations
• eCommerce sites tend to have fewer ads and
  trackers that ad blockers were able to block. On
  average, we observed ad blockers block two to
  three times more content for media sites than
  eCommerce ones.
• Ad blockers spend CPU cycles determining what
  content should be blocked. Below is a waterfall
  showing this particular scenario. The area
  highlighted in red shows that during a delay in
  the download, no bandwidth is being consumed,
  while the CPU is pegging constantly at 20 to 30

4

• Adblockers have to look up and evaluate a given
  resource against a huge list of patterns and
  access controls (a.k.a. filter lists). Take a
  look at these filter lists.
• Blocked list 580 KB (1.9 MB uncompressed)
• Allowed list 422 KB (1.9 MB uncompressed)
• There is CPU contention between browsers trying
  to render the page and ad blockers trying to
  block ad content, so end user experience is
  pretty likely to be worse.
• Performance varies on a site-by-site basis. While
  ad blockers might improve performance for
  ad-heavy media sites, eCommerce sites with no ads
  may see worse performance.
• The results of our tests of these sites with and
  without an ad blocker demonstrated the impact of
  ad blockers on site type.
• For moviemistakes.com, a media site with multiple
  ads and 1,102 HTTP requests, we witnessed both a
  better Speed Index and a faster Start Render with
  ad blocking, as expected. For Parents.fr, we also
  saw an improved Speed Index but a slower Start
  Render.
• With Apple.com, a site with no advertisements, we
  see something that we may not have anticipated
  if there is nothing to block, ad blockers have a
  negative impact on performance. With Apple.com we
  saw that Speed Index was worse and Start Render
  slower. Etsy.com, another site with no
  third-party advertisements, showed the same
  thing.
• For all sites, there was either a longer time to
  first byte (TTFB) with AdBlocker Plus, or no
  statistically significant difference between the
  tests. Ad blockers starts to kick in as soon as
  the request is made. For media sites like
  moviemistakes.com, the increase in TTFB is offset
  by the performance improvement of ad blocking.
  For eCommerce sites in which there was an
  increase in TTFB but no content to block, we
  obviously saw worse performance.

5

• The table below shows the TTFB, Speed Index and
  Start Render results for a single run of the
  sites tested and mentioned above. As noted, in
  one instance both metrics improved, in two both
  were worse, and for Parents.fr we saw one metric
  improve and one worsen.
• Given these observations, what does it mean to
  you?
• Ad blockers definitely have overhead, including
  consumption of your systems memory and CPU.
• Never assume ad blockers will improve performance
  across the board it depends on the site.
• Ad blockers do show a benefit when the tradeoffs
  of blocking content are greater than the overhead
  of ad blocking.
• Invest in tooling to measure the impact of ad
  blocking for users coming to your site. Extending
  your private WebPagetest instanceis a step in
  that direction.

Domain(click to side-by-side WPT result) TTFB (ms) TTFB (ms) Speed Index Speed Index Start Render (ms) Start Render (ms)
Domain(click to side-by-side WPT result) Without ad blockers With AdBlock Plus Without ad blockers With AdBlock Plus Without ad blockers With AdBlock Plus
Moviemistakes.com 320 ms 494 ms 4255 2830 3197 ms 2300 ms
Parents.fr 105 ms 107 ms 5981 3523 1297 ms 2105 ms
Apple.com 104 ms 136 ms 1361 2360 497 ms 1550 ms
Etsy.com 188 ms 182 ms 902 2107 1355 ms 2299 ms
6
Instart Logic is the worlds first endpoint-aware
application delivery solution that makes websites
and applications fast, secure, and easy to
operate.
Interested in learning more? Preview our image
optimization capabilities in the Playground
Contact Sales