[TYPO3-core] About table types (was: Re: RFC: core speedup)

Dmitry Dulepov dima at spamcop.net
Tue Mar 28 20:26:57 CEST 2006


-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hi!

The major problem is that we cannot just set table types to innodb
because many users do not use innodb. Also not all tables should be in
InnoDb. Generally innodb is better when concurrent
select/insert/update/delete is required. MyISAM will lock table in such
case, thus only one request (of may be 100!) is processed, than next one
and so on. InnoDb will just do all in parallel. At the same time InnoDb
is worse than MyISAM for tables, where data is primarily _selected_
(like pages and tt_content tables) or "count(*)" is used (InnoDb always
does a full scan and counts rows while MyISAM always knows exact count
or can quickly find it). Thus we could greatly optimize system if we
could use different table types for different tables with care.

I did not think yet how DBAL can handle it :)

Kasper, what do you think about small "optimization" subproject for 4.5?

Dmitry.

Kasper SkЕrhЬj wrote:
> Now we are talking!
> 
> I always wondered what you would use your database skills for with  
> TYPO3 and now I know!
> 
> - kasper
> 
> 
> On Mar 28, 2006, at 19:22 , Dmitry Dulepov wrote:
> 
> Hi!
> 
> Follow up. I am installing new server. Without major tuning it gives:
> ~ 1000 rq/s for test.txt
> ~ 830 rq/s for <? phpinfo(); ?>
> ~ 350 rq/s for a script that lists all databases and tables (so, no  
> any
> mysql caching)
> ~ 9 rq/s with typo3
> 
> eAccelerator is installed.
> 
> I added a couple of indexes and moved cache_pages, cache_pagesection,
> fe_sessions and fe_session_data to InnoDb to avoid locking when  
> records
> are simultaneously selected/inserted/updated/deleted in these tables.
> Now I got around 19 rq/s. Two times more with just a primitive db
> optimization...
> 
> One more note: "grep -R mysql_close /srv/www/web1/web" did not reveal
> anything. So, if typo3 has no_pconnect=0, it starts loosing  
> connections
> very soon. Setting no_pconnect to 1 solves this problem but we would
> better close connection properly.
> 
> Command line for testing was:
> 	/opt/lampp/bin/ab -t 1000 -c 50 "http://testsite/"
> "ab" runs on a separate PC connected to the server through 100mbit hub
> (no other computers in the hub).
> 
> This is tested on RC1 system.
> 
> Dmitry.
> 
> Dmitry Dulepov wrote:
>>>> Many slow pieces slow down the whole system. Sometimes significantly.
_______________________________________________
TYPO3-team-core mailing list
TYPO3-team-core at lists.netfielders.de
http://lists.netfielders.de/cgi-bin/mailman/listinfo/typo3-team-core

> - kasper

> "A contribution a day keeps the fork away"
> -------------------------------
> kasper2006 at typo3.com | +45 20 999 115 | skype: kasperskaarhoej |  
> gizmo: kasper_typo3


> _______________________________________________
> TYPO3-team-core mailing list
> TYPO3-team-core at lists.netfielders.de
> http://lists.netfielders.de/cgi-bin/mailman/listinfo/typo3-team-core

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.2 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iD8DBQFEKX/xRVcYnKJ8N6kRAmmsAKDBv0C3fOfbLBlS/Fg994v+1xZ/XQCfbxYZ
XRenfinvFru2ypNAfaiutxI=
=wZx9
-----END PGP SIGNATURE-----



More information about the TYPO3-team-core mailing list