Code review comment for lp:~stub/launchpad/page-performance-report

We want to make slow pages faster. At the moment we do this by optimizing test pages, but we really have no idea if we have made other pages slower. When we optimize, we need to ensure the system improves overall rather than playing whack a mole, speeding up some pages while causing others to timeout.

This hasn't been a genuine problem so far, but we want to evaluate using memcache. memcache is unusual in that it will slow down the initial page renderings due to overhead in stuffing data into the caches, and speed up repeat renderings. This branch is to provide us with tools to confirm if our experiments succeeded or failed.

The reporting tool is basic, but meets our initial needs in the current state. An example report from edge data can be found at http://people.canonical.com/~stub/lpperf.html .

To get meaningful graphs I had to use a pretty severe logarithmic scale on the y-axis and the numerical stats (mean, median, standard deviation) are pretty useless with real Launchpad data, but it is a start. The issue here is that slow requests are actually vastly outnumbered by fast requests in most of Launchpad, which is surprising to developers who spend their time worrying about the minor percentage of pages that generate timeouts and create the bulk of complaints.