[
Lists Home |
Date Index |
Thread Index
]
At 02:30 PM 8/25/2003, pop3 wrote:
>I agree that mappings among models are useful.
>
>But I wish to note that the farther one gets away from the hardware the
>less robust, slower and harder to optimize the resulting system becomes.
>Specifically, if the application design for a terabyte system works out to
>be a mapping between an rdbms and an XML tool like TeraText, and then a
>separate, different mapping between the XML tool set back to a rdbms, then
>my point is that the resulting system will have more failure points, be
>harder to tune, and generally less robust than a "native" system.
You seem to like broad generalizations. Adding layers can improve
performance - for instance, adding a cache or indexed access involves
adding another layer of software, and either one does make the system more
complex, but it also improves performance. Similarly, adding a query
optimizer makes the system significantly more complex, but it does not
really make it harder to use or hurt performance.
>Mappings, if you ask me, are bridges... like bridges they are an
>infrastructure that is high maintenance and which require above normal
>attention to support, and are subject to replacement periodically at high
>costs.
For our SQL/XML product, we have found that using the SQL/XML extensions
generally results in faster programs than building the XML "by hand" using
the DOM. The reason for this has a lot to do with the structure of the
mappings between XML and relational, and the number of database calls
required by the approaches people generally use when solving this problem.
Here's a paper that discusses this at a useful level of depth:
SQL/XML in JDBC Applications
http://www.datadirect-technologies.com/products/connectsqlxml/docs/sqlxml_whitep.pdf
I think this also applies to XQuery. It's pretty easy to imagine what it
would look like to add a section on XQuery to the above paper.
Jonathan
|
