[TYPO3-Performance] General questions

[georg kuehnberger]

Matthes, let my give it a try:

Mathias Schreiber [TYPO3] wrote:
> Hi guys,
> 
> since most of you are interested in performance tuning of TYPO3 I guess, 
> I wanted to ask some quick questions in order to get this thing going.
> 
> 1) Bottleneck
>    What do you think is the bottleneck in your installations?
>    1.a) Apache Webserver + PHP
>    1.b) MySQL Server
neigther; but rather the useless ping-pong between php and mysql;


> 2) Tuning needed
>    2.a) Generating the pages
>    2.b) Fetching pages from DB cache
a !

> 3) Does your website use Logins?
>    3.a) no
>    3.b) yes, a little (stuff <10% of the pags goes here)
>    3.c) yes, quite a lot
>    3.d) our website requires a login for most of its content

This differs per project; the ones with more personalization suffer 
badly (performance-wise). Workaround for the really personalized stuff 
for me seems to use eID being loaded via ajax, which allows the rest 
nonpersonalized stuff to be cached by TYPO3;


> 4) Do you NEED USER_INT objects on your pages?
>    4.a) yes
>    4.b) yes, but I could handle this via AJAX and eID
>    4.c) no, but I have them because I was lazy at some point
>    4.d) no
>    4.e) USER_WHAT?

d) + c)


> 5) Workflow in general
> If I find a flaw, what do I do?
> Do I write a patch?
> If so, when will it be checked?
> If checked, when will it be applied?
> 
> THis list could of course be continued like forever but I gues if you 
> read the first few things you get the point...

In general I would vote for
a) focus on reducing the number of DB-calls, both in the core and also 
in the various "important" extensions.

b) working on more clever cache-concepts (like real object caching, 
cache-headers and more)

I get the point that the mentioned PHP-tricks (which function is faster 
than the other) are valid, however I dont feel like this would do any 
meaningful impact on performance compared to the other bottlenecks.

Recently I saw a T3-Site, which (on descent hardare) takes up to 10+ 
seconds to generate a non-cached FE-Page, simple for the reason of 
having 30+ Content-Elements / Plugins on the page, which results in 
endless ping-pong;

 From experience I know, that it's best to fix the slowest 20% and you 
will gain 80% performance-improovement.


As for the DBAL discussion & your question: yes we did one postgresql 
setup, however this was for demonstratingt the client that mysql is 
faster only (was ~ factor 8); so no we dont use it in production.
And meanwhile, as mysql has gained much more respect on the market 
(inkl. being bought by SUN), we're able to convince even Oracle-Clients 
  to buy into mysql.


hth regards georg