Whose Website is it
The performance impact of third-party content
It looks a lot like your website. It's your URL up there in the address bar, and there's your logo at the top of the page. But how much of what you see is really yours? And how much is happening in the background that you don't see and that's not yours?
The age of Web 2.0 is fully here. One of the most salient features it has brought is the aggregation of content and functionality from multiple sources, in real time, onto the pages of websites. It's a phenomenon that has developed steadily over the past several years to where today, externally sourced content is commonplace on most major websites. But because it grew relatively slowly (in Internet time), the need for external content strategies never seemed particularly urgent.
"It's sort of like the old metaphor of the frog in a warm pot of water," says Dave Karow, senior product manager for Keynote Systems. "You can turn up the water slowly and he'll just basically stew, but if you put him right into a hot pot of water he'll jump out. The technical capability for multiple sources of content has been there for a long, long time. And over time, more and more companies came to build their solutions as composite applications where components come from drastically different places.
"And then people woke up one day and thought, wow, I have control of what's behind my firewall, but two-thirds of my Web content is coming from somewhere else."
This multi-sourced content helps to create rich user experiences and to empower site owners with valuable analytical and tracking data. But it is also a pesky and elusive culprit in undermining performance and the very user experience it seeks to enrich. Understanding the performance ramifications of third-party content takes some detective work — but the financial consequences of ignoring it can be significant.
The High Cost of Poor Performance
Speed is the Web user's drug of choice. Study after study has documented the negative consequences of slow site performance.Internet performance guru Steve Souders rounded up some compelling examples culled from speakers at the annual Velocity conference that he co-chairs:
- A two-second slowdown on Bing takes a 4.3 percent chunk out of revenue per user.
- Google finds that a half-second increase in page load time — from 400 to 900 milliseconds — drops first result page searches by 25 percent.
- Shopzilla found that shaving 5 seconds — from 7 to 2 seconds — increased page views by 25 percent, revenues by 7 to 12 percent, and reduced hardware needs by 50 percent. 1O’Reilly Radar, “Velocity and the Bottom Line,” by Steve Souders, 7/1/09
Similarly, Aberdeen Research data shows that a one-second delay in Web application response time dings customer satisfaction by up to 16 percent. 2TechNewsWorld, “Web Site Performance: When Seconds Count,” by Andrew Borg, 12/17/09 And Joshua Bixby and Hooman Beheshti of Strangeloop, in their presentation at this year's Velocity, charted how conversion drops nearly in a straight line with each additional second it takes a landing page to load; by the time you reach 10 seconds, you've lost nearly half of your conversions, and if, in a nightmare scenario, 28 seconds have ticked by, nearly 80 percent of your conversions are gone. 3Web Performance Today, “The 16 Best Graphs of Velocity:Â A snapshot of the current web performance landscape, by Joshua Bixby, 7/1/10
Whether measured by revenue, conversion, page views, overall user experience, or any other metric, speed is a bottom-line factor for success. Speed can be engineered into a site to the extent that the site and its assets are controlled by the owner. But in the real-world Web, much of what the user ultimately sees — and much of what is loaded in the user's browser that the user doesn't see but still impacts load times — is out of the site owner's immediate control.
Where did all this stuff come from?
First, there's the obvious mash-up: network-delivered ads; social media feeds; headlines, photos, and other editorial from outside sources; externally hosted shopping cart applications and transaction processing; product information fed from vendors, and the like. Each and every one has its own performance profile, and cumulatively impacts overall page performance.
And then there's the not-so-obvious. Visibly, there is content fed from elsewhere in the site owner organization — another department, division, location. Invisibly, there are tags for tracking all manner of user behavior, from media campaign traffic to page views to any number of things for which the powers that be want data.
"We've seen pages where 80 percent of the content is tags," says Keynote Director of Internet Technologies Ben Rushlo. "So the question would be, what have things come to, when we're tagging and tracking much more than we are actually providing real content to the customer. There's a lot of misinformation in the industry. There's a sentiment that 'we have to tag' — to an extent that explodes the page size and complexity. The reality is, that's just not true.
"In this particular example," Rushlo continues, "Say the home page is made up of 80 requests, and 80 percent of those, more than 60 requests, are third-party tags — they're tracking tags, media buy tags, ad tags, and they're not providing any direct value to the consumer. They're providing some analytical value. But the most important point is, they're trusting these third parties to provide the same level of service that they as the site owner are providing, and the reality is, very few of them do."
This is an extreme example — but not too extreme. Google performance specialists Arvind Jain and Michael Kleber, speaking at Velocity this summer, postulated that the average Web page calls on 44 resources from 7 domains — much of it third-party content — and takes 4.9 seconds to load. 4Velocity presentation, “Don’t Let Third Parties Slow You Down,” Arvind Jain and Michael Kleber, Google, 6/10
The impact of all those calls is cumulative and can add up in a hurry. Google AdSense, for example, is responsible for 12.8 percent of page load time; Google Analytics, something less than 5 percent (potentially less if using Google's new asynchronous GA snippet); Doubleclick, 11.5 percent. The Digg widget has been measured on some sites to consume 9 to 14 percent of page load time; Facebook Connect, 12 to 17 percent. 5ibid Also at Velocity, Yahoo!'s Tony Ralph pointed out that single ads have been found to involve 11 hosts, 15 connections, and 6 domains. 6Velocity presentation, “Display Ads & Page Performance: A Brief Tour of the Ad Ecosystem,” by Tony Ralph, Yahoo!, 6/10
Given the morass of content and connections that is increasingly typical of significant websites, it's a wonder they perform with any consistency at all. And often, they don't.
No matter how well a site is architected and quality tested, poor-performing outside content can slow it down – or in worst cases, break it. This summer, Twitter rolled out a new version of its official "tweet" button to anyone who wanted to add it to their site. Unfortunately, the button "essentially rendered Web browsers useless," according to tech blog Mashable, whose home page was jumbled by the new button — a button that gets viewed 750 million times every day. 7Mashable, “New Tweet Button Temporarily Hoses the Internet,” by Adam Ostrow, 8/13/10 With one little snippet of functionality, Twitter cast a fail whale across a large swath of the Web. Fortunately, most failures aren't so endemic and are typically confined to individual sites, as in these following examples.
"Now, the interdependencies are not typically large like that," Karow explains. "It's an unusual case of how effectively they broke that site, but the point is, you can be blind to what's going on. That's why you need to monitor it."
There's also the possibility that the outsourced content itself can fail.
"We've seen third-party tags on sites that are failing 100 percent of the time," Rushlo says. "And they're not failing elegantly, they're failing catastrophically — timing out, for example. And this has gone on for months and the site owners are not aware of it. In some cases, we see the page actually truncate loading if the third party fires off in the middle of that load, which is very common because people don't place their tags correctly at the bottom. So the user will get a malformed page. Or — we've all seen Internet Explorer spin and spin and never complete, or you try to scroll and it seems like the browser is locked. Most all of those problems are related to third-party tags."
How to Minimize Third-Party Problems
Some things you can control, some things you can measure, and some things you can manage.
Unfortunately, this is where many site owners stop. The site goes into production, and from behind the firewall, everything appears to be snappy. There's no reason users should be anything short of delighted. But unless the site is monitored out on the Web with all third-party content being fed, using a real browser just as a user would, there's no way to tell that everything is working and that pages are loading in an acceptable timeframe.
"People need a comprehensive way to watch how their site is performing," Karow says. "It has less to do with servers and their data center and more to do with how each of the separate providers on the page are delivering their value. Third parties by their nature act independently of your own quality assurance and functional testing teams. In a world where multiple providers are in your pages, you're increasingly less in control and more in need of rapid detection and response when something goes wrong."
Identifying and Dealing with Third-Party Crashers
Consistent, ongoing, external monitoring can provide data on how each page element is performing, and which elements are impacting performance, whether it's an ad, embedded tags, or an errant Twitter button. Keynote's Virtual Pages technology was designed specifically for this task, enabling quick, accurate performance measurement for any page component or combination of page components, internal or external.
Armed with that data, site owners can make decisions about how to resolve or manage third-party performance issues. If SLAs are in place, they can be triggered by the data. If tag vendors are coming up short, fixes can be negotiated. Or, if push comes to shove, costs and benefits can be evaluated to determine if it is really worth keeping the problematic content on the site at all.
John Masseria, manager of I/T engineering for Carnival Cruise lines, leverages independent monitoring data in managing his relationships with third-party vendors for Carnival's complex and multifaceted e-commerce site.
"I'm always trying to push the envelope in the area of, can I make it go faster?" Masseria says. "Can we make it more efficient? That's where Akamai fits in and Keynote fits in, because Keynote helps me to keep Akamai honest."
"It goes back to what Ronald Reagan said about the Soviets," Masseria concludes. 'Trust but verify.' And Keynote is our verification."
Taming the Tags
Business people love data, and as indicated earlier, the use of tracking tags on Web pages has simply exploded. Not surprisingly, more tags equals more performance management challenges, particularly since multiple vendors are usually involved, presenting multiple opportunities for glitches.
"As you add each of these vendors, what's the likelihood that they're all going to be good?" Rushlo asks. Not very high. Maybe you'll have several that are bad — and that is additive in some cases. If I've got problems with five of the tag vendors on my site, then I've got five problems and five ways that my site is being slowed down."
Rushlo offers up a three-step program to manage tags and minimize their impact on page performance.
"Our recommendation is, number one, put all the tags at the bottom of the page so you're shielding your user," he says. "So that as far as the user's concerned, the page is loaded even though more things might be happening behind the scenes with the tags.
"Number two is to track tag performance and understand the quality of all your tag vendors, so if they have a problem, you can work with them to clean it up.
"And number three, get rid of tags on the site that are no longer needed. Say you run a campaign for a week and now it's over, but your tags for it are still there. Make sure they're cleaned up where they're not needed. And of course, make sure you don't over-tag."
Making Everything Play Nicely Together
Multi-sourced content is here to stay — businesses need it, and users want the end result. Content can be tamed and made to perform well by consistently, continuously following these best practices:
- Be sure your site is architected properly so that third-party content will have minimal impact on page load times — four seconds is the magic number, beyond with users abandon your site in droves.
- Scrutinize every third-party component to be sure it's absolutely necessary; pare down the elements to only those needed to satisfy your site's business and revenue objectives.
- Monitor the performance of each page component continuously, from the field with real browsers just as users would experience your site; when performance issues come up, invoke your SLAs, negotiate a fix with the vendor, or lose the problem component.
- Practice good site hygiene — clean up unused tags on a regular basis.
Follow these steps, and you'll be well on your way to a site that quickly delivers a rich experience to your users.