Tuning Simple Intermedia Queries

Question

I am comparing the performance of doing a text search with
standard sql and Oracle Intermedia.
I have a table with a text column NARRLONG, defined as below:
Name Null? Type
NARRPREFIX NOT NULL VARCHAR2(2)
NARRKEY NOT NULL VARCHAR2(240)
NARRLANG NOT NULL VARCHAR2(2)
NARRSEQ NOT NULL VARCHAR2(5)
NARRLSEQ VARCHAR2(5)
NARRCHADTE DATE
NARRFLAG CHAR(1)
NARRLONG VARCHAR2
(1260)
The table has a primary key on NARRPREFIX, NARRKEY, NARRLANG,
NARRSEQ and a text index on NARRLONG.
When I run the query:
select count(*) from maa240 where narrprefix = 'BA'
and narrkey like '000010000100001C%'
and (upper(narrlong) like ('%VDU%')
or upper(narrlong) like ('%INPUT%'))
I get 273 rows returned in approx. 16 secs.
When I run the query;
Oracle 8.1.7 > cat andy2.sql
select count(*) from maa240
where contains (narrlong, '(%VDU%) OR (%INPUT%)') > 0
and narrprefix = 'BA'
and narrkey like '000010000100001C%'
I get 273 rows returned in approx. 3 minutes.
Analyzing the table made no difference.
Can anyone point me in the right direction for tuning this
query ?
Thanks
Andy Lamberth
OCP DBA

Answer 1

I believe that the problem revolves around the way that
intermedia text now works. The contains search returns rowids
which are then accessed as necessary. In your example I suspect
that a very large number of rows match the criteria that you
have specified. Each row will then be read to check the other
(non-IM) criteria before giving you the results.
By contrast your non-IM query may well be just accessing an
ordinary index based on most/all of your criteria and therefore
reading comparitively few rows.
To confirm this either just run the IM query to get a count on
the number of rows it finds as matches or use tkprof to examine
trace files.
We use a large number of IM indexes and have found that queries
matching a large number of rows on an individual IM index
perform poorly. By contrast when there are (relatively) few
matches performance is certainly superior to ordinary indexes.