Page MenuHomePhabricator

Provide an easy-to-run suite of high-level benchmarks
Open, LowPublic

Description

There should be an easy way to set up a basic, SQLite-backed MediaWiki instance, import a few representative articles from Wikipedia, and measure the backend response time to a set of representative user requests.

Details

Reference
bz68821

Event Timeline

bzimport raised the priority of this task from to Low.Nov 22 2014, 3:33 AM
bzimport set Reference to bz68821.
Krinkle updated the task description. (Show Details)
Krinkle moved this task from Inbox, needs triage to Radar on the Performance-Team board.
Krinkle renamed this task from Provide an easy-to-run suite of performance regression tests to Provide an easy-to-run suite of high-level branchmarks.Jun 6 2019, 8:39 AM
Krinkle added subscribers: aaron, Krinkle.

The idea here would be to have a benchmark script, like the ones we already have in MediaWiki's maintenance/benchmarks directory.

In addition to the few that are in there, it might be useful to adopt a practice of running these before major releases together with a re-run on the previous release for comparison (on the same hardware), and documenting the results.

@aaron also mentioned during our offsite that it might be interesting to go beyond just a simple maintenance script (which means installation, configuration and extensions has to be done manually, and may not be consistent between runs). And instead, to perhaps have it be a shell script that also performs these steps. For example, it could install MediaWiki into a temporary directory using SQLite, then install a number of extensions, then import a number of templates and articles, and then perform the various benchmarks.

For example, it could involve a macro benchmark for viewing a given article with object caches and parser cache cleared in-between. That might be enough for a starting point, additional scenarios could be added.

Krinkle renamed this task from Provide an easy-to-run suite of high-level branchmarks to Provide an easy-to-run suite of high-level benchmarks.Jun 9 2019, 4:35 PM
Krinkle moved this task from Watching to Limbo on the Performance-Team (Radar) board.