Xmlagg is executing slow on unicode database

Question

The database is Oracle 11.2.0.3.0 running a exadata.
I am trying to create XML using the following:
select
xmlserialize(DOCUMENT
xmlroot(
xmlelement("InterchangeEducationOrgCalendar",
xmlattributes(xml_text3 as "xmlns:xsi", xml_text1 as "xsi:schemaLocation",xml_text2 as "xmlns" ),
(select xmlagg(
xmlelement("CalendarDate",
xmlelement("Date", to_char(caldate,t_date_format)),
xmlelement("CalendarEvent",'xxxxxxxxxxxxxx'),
xmlelement("EducationOrgReference",
xmlelement("EducationalOrgIdentity",
xmlelement("StateOrganizationId", a.educationorganizationid)))
from calendardate a
) -- CalendarDate
, version '1.0'
AS CLOB) XMLRESULT
from (select t_xsi_location xml_text1, default_namespace xml_text2, xsi_namespace xml_text3 from xml_ctl);
if I run the above SQL statement in my non-unicode database it runs in about 45 seconds but in the unicode database it takes about 2 minutes and 20 seconds. The databases(init.ora parameters) are the same except one in unicode and the other is not. The above table has the same stats, same indexes, same number of rows in both databases as well.
The above query is selecting 2,000,000 rows from calendardate.
Can anyone tell me why it is slower in the unicode database?
Thanks in advance.
Edited by: 972551 on Apr 6, 2013 5:36 PM

Answer 1

Kit.net wrote:
I would expect to see slower performance since unicode is twice the character size.Not necessarily twice, depends on the encoding scheme : UTF-8, UCS-2 or UTF-16.
In the latest versions of the database, Unicode is supported via the AL32UTF8 db character set. It's a variable-width multibyte character set that encodes characters using 1 to 4 bytes.
So for occidental languages mostly based on the ASCII range, there's little overhead in storage size.
I agree though that UTF-8 string processing ought to be slower than a fixed-width encoding.
Back to the specific issue with SQL/XML functions, one thing that could be tried too is the NOENTITYESCAPING option.
It disables the runtime scan of element and attribute values in order to escape invalid XML characters such as &, < or >.
If we're sure the value doesn't contain invalid chars then it could save time :
SQL> select xmlserialize(content
  2           xmlelement(noentityescaping "emp",
  3             xmlattributes(noentityescaping empno as "id")
  4           )
  5         )
  6  from scott.emp
  7  where empno = 7369 ;
XMLSERIALIZE(CONTENTXMLELEMENT
<emp id="7369"></emp>