Re: [Plone-i18n] New LinguaPlone releases with large performance improvements

[Maurits v. R. <m.v...@ze...>]

Op 28-09-10 11:25, Hanno Schlichting schreef:
> Hi.
>
> I spent some time in the last days to speedup LinguaPlone again.
> Testing this both locally and over the weekend on a large production
> site looks promising. I have seen rendering times to be cut in half.
> Locally they went down from 370ms to 190ms. On the production site on
> average the rendering times went down from ~2 seconds to ~1 second.
> This only affects view performance and has no impact on editing
> performance.
>
> The changes are released in LinguaPlone 3.2a1 for Plone 3.3 and
> LinguaPlone 4.0a3 for Plone 4.
>
> If you have a large LinguaPlone based site, you might want to test
> this out. Note that this is very new code and marked as alpha, so be
> careful. The test site in question had 6 languages and ~190.000
> content objects in it. The changes I made should be more noticeable
> the higher each of those numbers are.
>
> I'd appreciate feedback from anyone who tests this out. I would expect
> improvements in the 30-50% range.

I tested LP 3.2a1 on Plone 3.3.4, not on a large site (about 300 MB, 
plus 100 MB blob storage), 8 languages.

I compared the requests per second (using 'ab' to get 30 requests) of on 
the one hand the current setup of LP 3.0.1 and the old LanguageIndexes 
and on the other hand LP 3.2a1 with FieldIndexes.  We are doing some 
non-standard stuff, so it does not make sense so go into details of what 
happens on each page (I tried 3), but the improvements were 2%, 16% and 
26%.  Not bad for such a still fairly small site.

I did not notice strange behaviour, but my testing was fairly light.

> Note: If you are still using a LanguageIndex and don't need it's
> functionality (fallback from content in sublanguage codes like "de-de"
> to main language codes like "de"), you should replace it with a normal
> FieldIndex. This is the default for new sites created with LinguaPlone
> 3.1+ and will give you very noticeable improvements on its own.

I also compared 3.2a1 with LanguageIndexes versus FieldIndexes in the 
portal_catalog and uid_catalog.  On two out of three pages it did not 
matter; on an anonymous page view that lists a few dozen content items, 
the difference was a bit more noticeable, with requests per second 
increased by 5%.  So for this fairly small size of Data.fs it does not 
matter much, though it is still nice.  I can imagine it becomes more and 
more interesting the bigger the site is.

Thanks for all your work, Hanno!

-- 
Maurits van Rees
Programmer, Zest Software