[ic] Performance issues with i18n

Stefan Hornburg racke at linuxia.de
Sat Dec 6 19:52:51 EST 2003


On Thu, 4 Dec 2003 19:17:22 +0100
Thomas Weiss <pater.noster at gmx.net> wrote:

> * On Wed, Dec 03 panos wrote:
> > I'm building an internationalized catalog with Interchange 4.9.9 and the
> > locale based database mappings as found in the foundation catalog. My
> > problem is that as soon as the mappings are used (by using a locale
> > other than the default) everything gets terribly slow. The reason is the
> > way Vend::Table::Shadow applies the mappings when querying the database.
> > 
> > For example when using the foundation catalog's data and database mapping
> > settings it takes about 80 sql queries to build the category menu (with
> > the category_vertical component) when using the default locale (no
> > mappings in use, no Vend::Table::Shadow), but with a non-default locale
> > in use it takes about 950 sql queries to do the same, which is IMHO way
> > too much to be efficient.
> 
> I've been hacking around a little bit and have been able to speed up the
> database mapping for this case by more than factor 3. The benchmarks
> have been done with the foundation catalog's process page with
> category_vertical being the only enabled component. Always a non-default
> locale setting was used. Interchange was running on my workstation, a
> PC with a P4 1.5GHz CPU and 512MB and the database server (PostgreSQL)
> was running on a machine with a PIII 550GHz CPU with 768MB RAM.
> 
> The number of database queries required to display the page has dropped
> from about 950 down to about 120 which raises the number of pages
> Interchange can server in a minute from 28 up to 95. When using single
> table mappings (i.e. the mapped column is in the same table as the
> original column) it served 106 pages in 1 minute.
> 
> The optimizations made were 1) omitting an unnessessary call to
> record_exists() in Vend::Table::Shadow and 2) letting the database
> server do the hard work applying the mappings instead of doing it
> manually.
> 
> Below are patches against Vend::Table::DBI (CVS Version 2.57) and
> Vend::Table::Shadow (CVS Version 1.43) which implement these
> optimizations. The patches are meant as a proof-of-concept kind of thing
> and there is of course room for improvement. The most important thing
> to do is support optimization even if the fallback option is set for
> the mapping. Currently optimization is done only if fallback is off
> but it can relatively easy be implemented (see patch no. 2 for details).
> 
> I'm looking forward to hear any opinions.

If I remember correctly the call to record_exists() has been introduced
because field() fails for GDBM databases on non-existant records.

Many thanks for your analysis/patches, I'll look in detail at it
after the release of 5.0.

Bye
	Racke


-- 
LinuXia Systems => http://www.linuxia.de/
Expert Interchange Consulting and System Administration
ICDEVGROUP => http://www.icdevgroup.org/
Interchange Development Team



More information about the interchange-users mailing list