@WMDE-leszek wrote https://gerrit.wikimedia.org/r/#/c/345332 for T161592: Account for foreign repositories in RDF mapping. Since RDF generation is a performance hotspot, we should compare profiling data before and after applying the patch, to see what impact it has on performance.
Profiling should be done against a sample of live data, e.g. the first 50k items and properties. The benchmark should be repeated, to make sure it's not skewed by external factors.
Current state of the benchmark collected in the table below (as measured 2017-11-14)
I did a table of running it on master, the old patch and the alternative one.
Name | First run | Second run | Third run | Average | Total run time ratio: master/patched |
Master | 19.90 | 21.091 | 21.096 | 20.696 | -- |
https://gerrit.wikimedia.org/r/#/c/345332/26/ | 23.301 | 23.156 | 22.931 | 23.129 | 89.48% |
https://gerrit.wikimedia.org/r/#/c/390996/1/ | 20.965 | 20.716 | 21.337 | 21.006 | 98.52% |