PHP performance in production

Measurements with the method attached, so you can repeat them. Including the numbers that are inconvenient for us.

• 22 September 2026

  Xdebug in production: we measured it

  40 ms became 1012 ms and sixty requests wrote 1.2 GB of cachegrind files. Five interleaved rounds, the method, and our own worst case in the same table.

• 20 September 2026

  What always-on tracing costs, and why sampling does not help

  We measured our own extension on a WordPress page. Recording every request could not be separated from the baseline; the function observer costs the same at any sample rate, and we explain why.

• 20 September 2026

  Running a PHP profiler on shared hosting without root

  .user.ini cannot load a PHP extension. Your own php.ini can. The difference, and a recipe verified as an unprivileged user.