PHPSW: Profiling & Benchmarking, January 2016
13 Jan 2016
at MixRadio
Talk comments
The speaker failed to engage the audience...at least in part through a lack of facing us....It's hard to be enthralled when someone is talking to the screen rather than talking to the audience.
Although the subject matter was interesting it focused too much on the mechanics of PHPBench, rather than why we would want to be benchmarking stuff in the first place, or why PHP bench is better than using other tools. One clear example is that at one point during the talk, it was mentioned that the outlying results were discarded......why? Aren't these results also interesting?
Coming from somebody who's never done profiling before but had a general idea of how it works, this was a great talk and introduction to profiling and xhprof. It was really useful to see a real example of the process Michael went through to find bottlenecks and fix them over a number of iterations. Everything was explained really well.
For me, it would've been useful to see a bit more about how xhgui works with xhprof, but I guess there is only a certain amount of time and a lot of material to get through.
Great talk all in all.
It was interesting.....I probably wasn't quite the target audience for this as I've use xhprof very briefly before. I think the talk might (or might not, who am I to judge) be improved by:
* Explaining what the call graph diagrams mean a bit more clearly. The fact that the text on the diagram isn't easy to read probably isn't overcome-able, but explaining what the hierarchy of calls represents in general could be hammered home a bit more. And then for each of the scenarios tested, needs to be said more explicitly, as although that is clear when it's on a computer screen, it's not so clear when viewed through a projector.
* The 'hook' of the talk showing that naive measuring of optimizations might not be worth the time spent on it....instead of spending the time on that, it might be worth spending time on a more complex scenario (maybe something involving Doctrine) where there can be a considerable amount of time spent on code that isn't very visible to the programmer.
btw the description of the talk might be better focused - I enjoyed hearing about hearing about xhprof....but if people read the abstract and were expecting a "few options" rather than a nice focus on a particular tool, they might feel their expectations hadn't been met.