crell felt that performance testing was necessary to in order to evaluate the effects of adding the new Drupal kernel. XHProf has been used by numerous users to run performance testing, get stats and numbers on function calls, etc.
Early in the issue, the question arises "can we swallow a 6-12% performance regression cost on the expectation that macro-level improvements later plus the DX, flexibility, and testability benefits will be worth it?"
Since the kernel has been adopted, and introduced this performance regression, this issue finishes with a discussion on what can be done on the micro-level in related features to regain some lost performance.
Continue investigation and pursue micro-optimizations. The most mentioned candidates include the Event Dispatcher and Auto Loader. A new broader issue was opened regarding this and other performance regressions:.
It looks as though this issue was left open by catch for informational purposes (#45).
Original report by crell
Per catch's request, we should run some performance analysis of the kernel branch to make sure we're not shooting ourselves in the foot too badly. :-) Some performance regression is acceptable at this stage, provided that it's something we have a plan to address in future work. We should also see if there's any quick-wins we can get performance-wise before the patch goes in.
I would suggest looking at the basics:
/node (with a bunch of nodes)
The changes in the kernel patch are primarily in the late bootstrap / page dispatch process, so I think they should be fairly consistent across different types of pages. Data can of course prove me wrong in that expectation.