Performance of Oracle xpath: two select scenarios

Question

I am using oracle 11g and have a table with an XMLType. There are many records in this table and each XML contains a list of 0 to n answers as follows:

<section>
  <answer id="100">
    <value>Something</value>
  </answer>
  ...
</section>

There are two ways to xpath for values.

Example 1:

extract(table.column, '//answer[@id=100]').getStringVal()

which returns

<answer id="100">
  <value>Something</value>
</answer>

Example 2:

extractvalue(table.column, '//answer[@id=100]/value/text()')

which returns

Something

There is a hit in performance extracting many text() values for a complex query, which has me asking the following question. Would I gain better performance if I utilized a library like xstream to parse out those values in java (example 1), or is it better to let oracle parse those values for me (example 2)?

I am trying to learn from other people's experiences or maybe some whitepapers before I commit time to rewrite a lot of daos and queries, so anything is much appreciated, thanks!

================== EDIT

Started looking into using Scala as an alternative to parsing xml in the JVM. Feel free to add Scala examples to the mix.

Answer 1

I finally got around to testing and here is what I found.

For a simple "reporting" jsp view using a table, I ran a query to populate 850 rows.

Given my codebase, I ran the query first as follows:

• Query contained 3 left joins on XMLTypes for each record.
• Each record extracted 15 "Example 2" values.
• Each value did not need XStream processing.

This took a total of:

8.27 seconds to run the query alone.

15 seconds total to go from start to a finished view

I then altered the query to do following:

• Query contained 3 left joins on XMLTypes for each record.
• Each record extracted 15 "Example 1" values.
• Each value also ran XStream processing to extract value(s).

This took a total of:

24.84 seconds to run the query alone.

60 seconds total to go from start to a finished view.

This is of course my codebase, but I think I can rule out the use of XStream as a mean of efficiency. But it seems that the XMLType and xpathing is not very efficient in the first place.