Site Performance with Keynote
As Ian discussed in his outstanding recent post on the business value of Web site Performance Monitoring, Google demonstrated that even a sub-second response time delay has a lingering effect on user engagement, even weeks soon after the site was sped up once again. A delay of this length might not be severe sufficient to warrant a vital alert, but it has a measurable impact on a number of metrics including revenue per user and user satisfaction.
About the identical time as Google demonstrated this result, a study conducted by Forrester Investigation and commissioned by Akamai found that most consumers anticipate a web page to load in under two seconds—compared to an expectation of under four seconds reported three years earlier. More than half of respondents said that speedy page loading was an critical factor for continued web site loyalty.
For factors such as these, I think it's critical not to undervalue medium-term and long-term performance monitoring.
Keynote has several capabilities in this region which includes long term trending, time history graphs, and service level reports. Let's take a look at every of these capabilities in far more detail.
Keynote retains aggregated web performance information and provides lengthy-term trend graphs for up to two years. For example, I can tell that the performance of the Google house page has been fairly consistent more than the last couple of years, though lately site performance appears to be hovering closer to .five seconds than the .3 second range exhibited in late 2008/early 2009.
For medium-term analysis, Keynote retains raw performance information for six weeks that users can analyze across a variety of dimensions making use of time history graphs.
In the graph below, I've broken out the response time for the Google home page over the last four weeks into times for DNS lookup, connection, initial byte download, and content download. As one would expect, of these components, the one that takes by far the most time is content download.