The call for an audit of Facebook’s metrics comes a week after the social network acknowledged inflating its video metrics.
(Page 2 of 2)
Retailers often have several connections from their data centers to the web so they can keep operating if one fails and so they can use lower-cost connections whenever possible.
Nanette Lepore, a retailer of women’s fashion apparel, uses three such connections and employs the PowerLink 100 Pro device from Ecessa to route traffic to the pipe that offers the best performance at any time, says Jose Cruz, I.T. director at the retailer.
Besides maximizing performance, the device can prevent site failures. In one case, when a construction crew cut a landline connection, the device switched traffic to the retailer’s wireless connection, keeping the site up.
In another, a denial-of-service attack by hackers coming in on one Internet link drove up traffic and spilled over to the other two, slowing down the site. PowerLink enabled the retailer to shut down the link under attack, allowing the other two web connections to resume normal operation. The PowerLink 100 device is priced at $2,995.
4. Test the weakest link ...
Many retailers test their sites in advance of the holiday season by simulating the load that many individuals accessing the site simultaneously create. But concurrent users isn’t necessarily the key metric for every retailer.
One large retail chain is most concerned about the capacity of its order management system. So testing service Keynote Systems Inc. has been putting that system to the test, simulating the load of many transactions going through the entire checkout process, using nonworking credit card numbers, says Donald Foss, director of Keynote’s global testing service, who declines to name the retailer client.
It’s also important to test a site with a realistic mix of the actions consumers initiate at a site, such as browsing, searching and purchasing, says Rodrigue of Sears Canada. Rodrigue says he has seen in past jobs cases of surprising results-such as an unexpected failure of the site’s database layer-that weren’t predicted by the modeling site engineers had done in advance.
5. ... and test until it breaks
A site crash is usually a disaster, but some large retail chains intentionally crash their sites in the months leading up to the holiday season. They just do it in the middle of the night when few people shop.
The point is to understand the capacity of the site, and to make sure crucial features will stand up to heavy loads. While he would not mention the retailers by name, Foss of Keynote Systems says one large chain has been applying exceptionally heavy loads to its site every other Friday night and another every Wednesday night, both in the wee hours. While the sites don’t always crash, they sometimes do, and usually are brought back up within 10 minutes, Foss says.
It’s important to test any new feature in a production environment-and not just on a server off to the side-because that’s the only way to know how the new functionality will stand up when working alongside all the other components of the site.
“We will overload the amount of concurrent users or page views and see what breaks first,” he says. “Then we make an effort to understand why the failure occurred, fix it and break it again. When launching new functionality it’s important to understand where the weak links are.”
Finding weak links in the middle of a summer’s night is surely better for an e-retailer than discovering them the day after Thanksgiving.
Click Here for the Web Site Performance Products & Services Guide