YiiBoilerplate
YiiBoilerplate copied to clipboard
clevertech
low performance
My yii1 and yii2 fresh installs handle 150-220 request per second, but fresh install of boilerplate handles only 20-25. What is the reason?
Well, if you go as far as to profile newly-installed application templates, you can also look at the xdebug dump output with cachegrind and find the bottleneck yourself. :-| We did not do this, and your finding is interesting, by the way.
First idea that comes into my mind is that application templates which come with Yii 1.1 and Yii 2 assume your app is in production mode, and you need to enable debug mode manually. And in debug mode everything works slower. Try to re-run your profiling after putting an empty file named PRODUCTION_MODE in root of the codebase. Look at the check_prod_mode.php script for details. YiiBoilerplate assumes that you in debug mode by default, and you need to enable production mode manually. This is by design, because you have production mode on just a single deploy machine and almost always need the debug mode in all other cases.
So, profiling showed that bottleneck was in using CDbHttpSession for session, as you can see below
When I commented session part, performance increased for 2.5 times, that is 50 requests per second, here is profile
But it is still slower where fresh yii1 which handles 150 r. p.s. As you see in second profile bottleneck is in CMap::mergeArray, YiiBase::import and YiiBase::autoload. It is because YiiBoilerplate uses very expensive method mergeArray for complex config files.
Actually I did not think that using database for session degrades performance.
Whoa, such a slow updates. Actually, database can be the botteneck if establishing connection to it is costly or it runs on a slow disk. Probably enabling DB caching will help, but I am not sure in this case, because most of the time were spent in UPDATE query.
Can you please provide the same profile charts for the fresh Yii 1.1 installation?
I certainly could not anticipate that combining arrays in PHP can be so costly as to decrease performance threefold.
my test is simple
ab -n 10 -c 2 http://yii-boilerplate result: 90 req p.s.
ab -n 10 -c 2 http://yii-fresh result: 150 req p.s.