Testing your site's vulnerability to third-party failure

7 August 2012  -   Tags: , ,

Poorly-optimized third-party scripts not only hinder usability, but can easily take down an entire site. In web performance, these culprits are commonly referred to as SPOFs – single points of failure.

Third-party SPOFs can easily nullify significant (and expensive) CDN performance gains, leaving users staring at soon-to-be-abandoned blank screens. But despite these clear performance drawbacks, measuring their impact is often an afterthought.

Strangeloop tested the top five ecommerce sites to see how SPOF interference affected performance. For the tests, all third-party domains were blackholed except the CDN domain. A series of side-by-side videos were generated for each site, with waterfalls included for quick identification of the major culprits.

Amazon.com

Waterfalls: normal vs. broken

Observations:

  • Site still performs reasonably well
  • Site looks mostly fine, except for empty ad spaces

Staples.com

Waterfalls: normal vs. broken


Observations:

  • When Omniture dies (see line 18 in waterfall) the entire site stalls for almost 30 seconds.
  • Site is entirely dependant on Omniture.

Apple.com

Waterfalls: normal vs. broken


Observations:

  • Very few third-party calls.
  • Nothing blocks. The entire internet could go down and this site would still work.

Dell.com

Waterfalls: normal vs. broken


Observations:

  • As with Staples, Omniture blocks the page request (see line 11 of the waterfall), this time for almost 20 seconds

OfficeDepot.com

Waterfalls: normal vs. broken


Observations:

  • This waterfall is the worst of the bunch.
  • The issue here seems to be that the file is an external JavaScript file very near the top of the page, which is why the effect is so bad. The page is white until it times out trying to connect to the broken domain.
  • The same behaviour happens when testing this page with HTTPWatch in IE9, Firefox 7, and Chrome 16.

How are top ecommerce sites using third-party scripts?

Strangeloop did an audit of the top 200 Internet Retailer sites to see how they were implementing third-party scripts. here are some of the results:

Average number of third-party scripts: 


Average # of 3rd-party scripts
Top 200 sites
6.7
Top 20 sites
3.5

*6.7 is significant, but some sites use many more than that…

Top sites, in terms of the number of third-party scripts used

Site # of 3rd-party scripts
Coastal Contacts
25
Express LLC
23
American Greetings
22
Urban Outfitters
21
The Sports Authority
20
Coldwater Creek
19
American Girl
19
RealNetworks/GameHouse 18
Chico’s FAS
18
Signature Styles/Spiegel
17
Boden USA
17

*When looking at the sheer volume of widgets and third-party tools available, these numbers aren't too surprising. 

Most-used third-party scripts

3rd-party script provider
Appearance in top 200 sites
Omniture
98
Google Analytics
97
DoubleClick Floodlight
49
DoubleClick
45
Google AdWords Conversion
45
Coremetrics
44
Right Media
40
Foresee 36
Microsoft Atlas
33
LeadBack
32
DoubleClick Spotlight
29
Turn
29
Facebook
27
Acerno
26
Rubicon
25
Channel Intelligence
24
Dotomi
22
Interclick
22
Traffic Marketplace
22
Adconion
19
Channel Advisor
19
Resonance
19

*Many people would guess Facebook because of its visibility, but this is a good reminder that “invisible” scripts are actually more prevalent than obvious content like social buttons.

Test conclusions

From this data, it's clear that site owners are implementing more and more third-party scripts, sometimes improperly. The sheer volume of scripts indicates that little analysis is being conducted as to how these scripts affect performance.  

Bottom line: it doesn't matter how well you optimize a site if a single script - or one errant line of JavaScript - can throw the whole operation.

How Strangeloop addresses third-party SPOFs 

To give site owners peace of mind as to how third-party scripts affect performance, both our Site Optimizer and Mobile Optimizer minimize the impact of third-party content by deferring and/or parallelizing third-party requests so they don't block key site content. To learn more, talk to a Strangeloop performance expert.


Post your comment

Posting comments has been disabled.

Comments

No one has commented on this page yet.

RSS feed for comments on this page | RSS feed for all comments