jackrabbit-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Alexander Klimetschek" <aklim...@day.com>
Subject Re: Scalability of using "reference" properties
Date Sun, 20 Jul 2008 18:06:16 GMT
Hi,

we generally recommend not to use references. Apart from the scaling
issues it has, there are also conceptual reasons. The content
modelling tips by David NĂ¼scheler, which you mentioned in your first
mail, can be found on the wiki - here is the link to the part about
references:

http://wiki.apache.org/jackrabbit/DavidsModel#head-ed794ec9f4f716b3e53548be6dd91b23e5dd3f3a

Regards,
Alex

On Sun, Jul 20, 2008 at 2:20 PM, sbarriba <sbarriba@yahoo.co.uk> wrote:
> Hi all,
>
>
>
> Are there plans to improve the efficiency of the implementation of
> references between nodes in JackRabbit or is it considered best practise NOT
> to use references at all?
>
>
>
> Based on our experiences any use of references will eventually hurt
> performance as the node set grows. It's a shame as references are really
> useful and the alternatives move the emphasis to the application to enforce
> node-to-node dependencies, which personally I think is wrong.
>
>
>
> Regards,
>
> Shaun
>
>
>
>
>
> From: sbarriba [mailto:sbarriba@yahoo.co.uk]
> Sent: 15 April 2008 14:14
> To: users@jackrabbit.apache.org
> Subject: Scalability of using "reference" properties
>
>
>
> Hi all,
>
> Last year David Nuescheler provided some very useful data modelling rules -
> 1 of which related to the use of "reference" properties.
>
>
>
> We had various, reasonable, use cases for which we felt references were
> appropriate e.g. to ensure referential integrity. While semantically they've
> worked well for us, as our repositories grow in size we're now seeing how
> expensive references can be where you have 1,000's of items referencing the
> same node. Monitoring our SQL logs (on MySql) we can see some pretty huge
> database operations which are getting incrementally slower as more and more
> items reference a node.
>
>
>
> Ignoring the data modelling semantic of using "reference" properties for the
> moment, is there anything that can be done in order to improve the
> performance of references or do you always have to design with this
> limitation in mind (hindsight being very useful)?
>
>
>
> We're facing a tricky remodelling/migration exercise to ensure further
> scalability.
>
>
>
> Regards,
>
> Shaun
>
>
>
>
>
>
>
>



-- 
Alexander Klimetschek
alexander.klimetschek@day.com

Mime
View raw message