It’s fall, which means it’s time for Radware’s quarterly State of the Union for Ecommerce Performance! If you’re not familiar with these reports, here’s a quick backgrounder:
Every three months, we analyze the performance of the top 500 retail websites, as ranked by Alexa. Using WebPagetest, a synthetic tool that simulates page load from a real user’s perspective across various browser types, we test the home page of each site and record a number of metrics, including size, composition, load time, start render time, and adoption of performance best practices.
The goal of this research is to gain ongoing visibility into the real-world performance of leading ecommerce sites – to learn how these sites perform for visitors sitting at home using the internet under normal browsing conditions – and to provide strategies and best practices that will help site owners serve their pages faster.
What I expected we’d find in our latest round of tests…
Prior to this latest round of tests, I made two big predictions. As it turns out, I’m only batting 0.500 on this one. Fortunately, I like surprises.
Assumption #1: Pages would be bigger and slower.
Since 2010, when we initiated these performance SOTUs (at Strangeloop, now part of Radware), we’ve consistently found that pages are getting bigger (in terms of number of objects and total bytes in) and slower (in terms of total load time). Assuming that this trend would continue in our latest report was kind of a no-brainer.
Assumption #2: Time to interact (TTI) would plateau or improve.
Our last report, which came out three months ago, marked the first time we included time to interact (TTI) as a performance metric. TTI is the moment that a page’s primary content — in the case of ecommerce sites, this is usually some sort of feature banner or carousel — renders in the browser and becomes interactive, i.e. has a usable call-to-action button or link.
Last summer, I was surprised (and borderline shocked) to learn that the median page took 4.9 seconds to become interactive. Ideally, pages should be interactive in 3 seconds or less. Like many folks in the performance community, I had assumed — wrongly, as it turns out — that most pages serve their key content within the first couple of seconds.
Given the fact that TTI surprised me once, you’d think that would make me somewhat jaded about my later assumptions. You would be wrong. I assumed that the slow TTI was an unfortunate performance blip, and that TTI would have improved, or at least plateaued, in our newest round of tests.
What we actually found…
Finding #1: Pages are bigger and slower in terms of total load time.
No surprise here: The median load time was 8.56 seconds, a 14% slowdown from the median of 7.48 seconds we recorded three months ago. This is due, in part, to the fact that the median page is now 1258 KB in size and contains 92 resources, meaning that pages today are 24% bigger than they were back in 2011.
Finding #2: Pages are slower in terms of time to interact.
Here’s the surprise: The median page took 5.3 seconds to load primary content, an 8% slowdown from the median of 4.9 seconds we recorded last quarter. The median number isn’t the only noteworthy finding here. We also discovered that only 18 of the top 100 sites had a TTI of 3 seconds or less, while 26 of the top 100 took 8 seconds or longer — with some nearing the 20-second mark — to become interactive.
While the TTI slowdown is disheartening, there’s a bright side…
Time to interact (TTI) is taking hold as the performance metric to watch.
If you care about user experience, then time to interact is arguably the metric you need to care about the most. It’s been extremely heartening to see TTI — along with its kissing cousin, above-the-fold time (AFT) — starting to get attention inside and outside the performance community as the metric to watch.
The great thing about TTI: It pinpoints the most critical moment in a page load, from the end user’s perspective.
The not-so-great thing about TTI: Right now, there’s no reliable way to automatically measure it. For our research, we identified TTI the old-school way: human eyeballs. Using WebPagetest, we downloaded the filmstrip version of each site’s median load, then analyzed each filmstrip and pinpointed the moment at which the primary page element (almost always a feature banner/carousel) loaded and became interactive (the main CTA button loaded). If anyone develops a tool that can consistently, reliably measure TTI, let me know!
But wait! There’s more…
I’ll be using this space to discuss our other findings. If you can’t wait, you can download the report and infographics here:
DOWNLOAD: State of the Union: Ecommerce Page Speed & Web Performance [Fall 2013]
As a former senior researcher, writer, and solution evangelist for Radware, Tammy Everts spent years researching the technical, business, and human factor sides of web/application performance. Before joining Radware, Tammy shared her research findings through countless blog posts, presentations, case studies, whitepapers, articles, reports, and infographics for Strangeloop Networks.