Many Web developers have gotten the message: a faster site means people buy more, read more, interact more, and return more.
But apparently the message hasn't sunk in far enough, because the top 2,000 retail Web sites still take 10 seconds to load on average, according to a study by Strangeloop Networks released yesterday. The list of the top retail sites comes from Amazon's Alexa list of top sites.
Strangeloop is in the business of helping customers speed up their Web sites, so it has an agenda to push, but that shouldn't deter developers from taking a look at the study's findings.
One big issue in particular jumped out at me after a chat with Strangeloop President Joshua Bixby: Web developers should test their pages the way their readers see it, not the way they see it on their own machines.
Strangeloop's study uses a speed-testing tool called WebPagetest developed by Patrick Meenan, who started it at AOL but who now works for Google. That test adds delays called latency to round-trip communications to better simulate how ordinary people several steps removed from a Web site see it, Bixby said.
"Most tests are run out of a huge data center with absolutely no latency and with bandwidth that is outrageous," Bixby said. "These test machines in data centers are next to the content delivery machines. They're just sending bits across a cage." That typically hides problems that real-world users have.
Latency matters a lot, in particular because it's a problem that compounds as a Web browser has to make multiple network connections to Web servers to request new elements. There are lot of tricks to reduce the number of requests a browser must make, but the Strangeloop study shows the complexity of Web pages is increasing at the same time as economization measures and browser speed are improving.
"Pages continue to get bigger and continue to have more requests," Bixby said. "In some ways we're losing the battle--or maybe it's a stalemate. We're not getting much better."
Another step backward came in how long a repeat view of a Web page took. Returning to a Web site should go faster, since browsers cache resources on computers for faster retrieval later, but repeat views actually slowed. In last year's study, a repeat view took on average 5.10 seconds, but this year, it was 6.20 seconds.
One more finding concerned browsers. Here, Microsoft's IE9 edged out Google's Chrome, Firefox's Mozilla, and IE7 to win the speed crown. On average, IE9 took 7.12 seconds to load the pages compared to 7.15 seconds for Firefox 7, 7.5 seconds for Chrome, and 10 seconds for IE7.
"Microsoft has started to catch up," Bixby said. "More than catch up--IE9 is equal and in some tests surpass some of these other browsers."