Performance & Debugging

This is the oldest thread on the site and the one that has changed least. The tools are unrecognisable; the method is not. Measure the thing, find where the cost actually is, and be suspicious of the answer you were hoping for.

The recent work is cutting worker memory in PHP with Judy arrays — I modernised php-judy for PHP 8.1–8.5, added a pure-PHP polyfill and a PSR-16 cache, and benchmarked it head-to-head. The older posts hold up better than their version numbers suggest: MySQL multiple column index misuse is still the clearest example I have of an index that looks right and is not, and Zend Platform performance results is a 2007 benchmark I keep around as a reminder of how much of this is cyclical.

On instrumentation and debugging: JETM for Spring applications, graphing JMX values with Ganglia and JPype, phpdbg, and the one I still enjoy most — fixing a Flash security sandbox violation by reading the RTMP handshake in a packet capture, because the documentation did not mention the flag that mattered.

The legacy posts carry a warning banner and they mean it: versions have moved on, so treat the specifics as history and the reasoning as the point.

7 posts

2026 Cutting worker memory in PHP with Judy arrays I modernized php-judy for PHP 8.1–8.5 and added the two pieces it lacked: a pure-PHP polyfill and a PSR-16 cache with O(range) prefix invalidation.