Hi Christophe,

That's like a programatic plug-and-play indexes system.
I think it is not so easy for everyone to develop customized indexing class.
But "coupled with an appropriate index set definition" you saying is a very
good idea.
By developing these modules with xindices and experiencing those feature, we
can find out more good solutions.

By the way, I would like to know what kind of mechanism the other commercial
Native-XML-Database products have is implemented.
Anyone knows ?

Regards,
Takashi.


----- Original Message ----- 
From: "christophe reiss" <[EMAIL PROTECTED]>
To: <[email protected]>
Sent: Tuesday, May 11, 2004 6:33 PM
Subject: IndexManager.SAXHandler ... interface ?


> Hi,
>
> this is just an idea... sounds it realistic ?
>
> When tuning a base for an application, the model of repartition of
documents
> in collections and indexes can dramatically influence queries performance.
Of
> course, the model is lead by grammars of docs that are expected. However,
to
> reach optimum numbers, the designer has to keep in mind the way indexes
are
> used in queries, and, most of all, populated. By looking at the code, the
> masterpiece in document assignation thru indexes is the SAXHandler private
> class in IndexManager. Given the set of indexes, this class has a direct
> impact on the way indexes are filled, and thus by a side effect on queries
> performance (because the handler could introduced unbalanced indexes
> populations depending on the model). A different implementation of the
> SAXHandler, coupled with an appropriate index set definition, could
ultimately
> tune queries performances. Thus, my question is : would it make sense to
have
> a configurable SAXHandler, where default class would be the current one,
and,
> if specified in configuration, assigned class would be one defined by the
> user ? (SAXHandler implements org.xml.sax.ContentHandler and
> org.apache.xindice.xml.sax.CompressionHandler, so an SAXHandler interface
that
> extends both could be easily introduced).
>
> my 2 cts,
> Christophe.
>
>

Reply via email to