[TYPO3-core] RFC: #0011505: TYPO3 4.3 is slower than 4.2.8 -> re-implement native db caching

[Oliver Hader]

Hi Rupi,

thanks for the work you're putting into analysing and tracking down the
bottlenecks in TYPO3 4.3. However, I'd like to test and profile it on my
own and ask you not to commit it to Trunk until I put by vote to this
issue (even if you already got enough +1s). Thanks!

olly


Rupert Germann schrieb:
> Hi,
> 
> as one of the results of the performance meeting from last weekend here's a
> first RFC:
> 
> Type: Bugfix
> 
> Bugtracker references:
> http://bugs.typo3.org/view.php?id=11505
> 
> Branches: Trunk
> 
> Problem:
> the FLOW3 caching framework (F3CF) eats up a lot of performance.
> Despite there have been a bunch of performance related improvements lately
> in trunk, TYPO3 4.3 delivers cached pages still significantly slower than
> 4.2 (see numbers below).
> 
> Solution:
> re-implement the "traditional" way of caching additionally to the F3CF and
> make the caching engine configurable in the install tool.
> This gives all users that don't need another cache than the mysql database
> the chance to achieve the same performance with TYPO3 4.3 as they had in
> 4.2.x.
> Those users who benefit from the F3CF because they have the possibility to
> use other caching engines like memcached or APC will still have the
> possibility to set a checkmark in the install tool and use the caching
> framework.
> 
> Info:
> the sligtly decrease in troughput when comparing 4.3 patched with F3CF and
> 4.3 trunk is caused by the conditions which determine which caching engine
> is enabled. (I did already a benchmark to compare if(defined(const)) and
> if($GLOBALS['var']) - the latter one is faster)
> I also did some benchmarks without the condition and only the internal db
> caching enabled; the troughput actually reaches the values from 4.2.8
> without any decrease.
>  
> note: the bigger decrease that I measured earlier[1] between 4.3 without
> FCF3 and 4.2.8 seemed to be caused
> a) by new improvements in 4.3 (namely the xclass fix in the autoloader) and
> b) by differences in my test installations. This time I made shure that all
> installations use identical dbs and identical localconfs (except the db
> settings of course)
>  
> Since I do believe that the F3CF will be faster one day, consider this patch
> as preliminary - I'm mainly interested in the question if others can
> reproduce the results.
> 
> greets
> rupert
-- 
Oliver Hader
TYPO3 Release Manager 4.3