Performance of REGEXP_LIKE vs LIKE?
zstrin...@gmail.com
I'm curious if anyone has done any performance testing of REGEXP_LIKE
vs LIKE. There are certain situations where the syntax of REGEXP_LIKE
is cleaner and shorter than the comparable LIKE expression. A common
search request for us involves doing a case-insensitive, wildcard
search of a 3 million record table, for a series of text strings.
For example:
SELECT *
FROM big_table bt
WHERE UPPER(TRIM(bt.txt)) LIKE '%ABRASION%'
OR UPPER(TRIM(bt.txt)) LIKE '%DERMATOLOGICAL%'
OR UPPER(TRIM(bt.txt)) LIKE '%PSORIASIS%'
The equivalent regular expression search is:
SELECT *
FROM big_table bt
WHERE REGEXP_LIKE(UPPER(TRIM(bt.txt)),
'(ABRASION|DERMATOLOGICAL|PSORIASIS)')
The regular expression syntax is cleaner, especially when you have a
lot of strings to search for! However, the LIKE expression runs in 20
seconds, while the REGEXP_LIKE one runs in 60 seconds. Has anyone
else noticed this? Any way to speed it up?
BTW there's some kind of limit to the string you can pass to
REGEXP_LIKE. You'll get the error message: ORA-12733: regular
expression too long
Thanks,
Zstringer
sybr...@hccnet.nl
<zstrin...@gmail.com> wrote:
Regular expressions are implemented using Java. You force Oracle to
call Java, an interpreted language. The delay would be caused by the
architecture.
--
Sybrand Bakker
Senior Oracle DBA
Thomas Kellerer
> The regular expression syntax is cleaner, especially when you have a
> lot of strings to search for! However, the LIKE expression runs in 20
> seconds, while the REGEXP_LIKE one runs in 60 seconds. Has anyone
> else noticed this? Any way to speed it up?
>
I thought you should be able to create a function based index to support
the regex match but I'm not sure (could have been Postgres where I did
that the last time ;) )
Thomas
keithj...@hotmail.co.uk
You write that "Regular expressions are implemented using Java. You
force Oracle to call Java, an interpreted language. The delay would be
caused by the architecture."
I don't think this is always correct. I am using Oracle XE at home and
it doesn't install Java and you can't write Java stored procedures.
You can still use the "regexp_like" function however. I think
therefore that in Oracle XE the regular expressions are not written in
Java. Someone else can confirm this is the same in Enterprise and
Standard Oracle versions.
Please correct me if I am wrong - I am still learning Oracle.
Thank you
John
zstrin...@gmail.com
I forgot to mention that I'm using Oracle 10g. I've seen
implementations of regular expressions in earlier versions of Oracle
that were done in Java. However, in 10G I think that they're
implemented using PL/SQL. Can anyone confirm?
Peter
Vladimir M. Zakharychev
wrote:
Actually, they are implemented in C and built into the database kernel
in all editions. As for the origins of the implementation - hard to
tell, but very unlikely that it's Perl. :)
Regards,
Vladimir M. Zakharychev
N-Networks, makers of Dynamic PSP(tm)
http://www.dynamicpsp.com
Vladimir M. Zakharychev
wrote:
> Hello fellow netters,
>
> I'm curious if anyone has done any performance testing of REGEXP_LIKE
> vs LIKE. There are certain situations where the syntax of REGEXP_LIKE
> is cleaner and shorter than the comparable LIKE expression. A common
> search request for us involves doing a case-insensitive, wildcard
> search of a 3 million record table, for a series of text strings.
>
> For example:
>
> SELECT *
(> FROM big_table bt
> WHERE UPPER(TRIM(bt.txt)) LIKE '%ABRASION%'
> OR UPPER(TRIM(bt.txt)) LIKE '%DERMATOLOGICAL%'
> OR UPPER(TRIM(bt.txt)) LIKE '%PSORIASIS%'
>
> The equivalent regular expression search is:
>
> SELECT *
> FROM big_table bt
> WHERE REGEXP_LIKE(UPPER(TRIM(bt.txt)),
> '(ABRASION|DERMATOLOGICAL|PSORIASIS)')
>
> The regular expression syntax is cleaner, especially when you have a
> lot of strings to search for! However, the LIKE expression runs in 20
> seconds, while the REGEXP_LIKE one runs in 60 seconds. Has anyone
> else noticed this? Any way to speed it up?
>
> BTW there's some kind of limit to the string you can pass to
> REGEXP_LIKE. You'll get the error message: ORA-12733: regular
> expression too long
>
> Thanks,
> Zstringer
How about Oracle Text? It's as simple as
CREATE INDEX IX$CTX$BIG_TABLE#TXT ON BIG_TABLE (TEXT) INDEXTYPE IS
CTXSYS.CONTEXT PARAMETERS ('FILTER NULL_FILTER SYNC ON COMMIT')
/
(be aware that SYNC ON COMMIT option is unsafe in 10.1 up to 10.1.0.5
- there are a couple of bugs with it that can corrupt the index on
subsequent optimization, fixed in 10.1.0.5 and, I believe, 10.2.0.2)
and then you query like this:
SELECT * FROM BIG_TABLE
WHERE CONTAINS(TXT,'abrasion OR dermatological OR psoriasis') > 0
or
... CONTAINS(TXT,'about(abrasion) or about(psoriasis)') > 0
or ... well, I won't go into copying the Text Reference here, check it
out yourself... :) Using Text should reduce your query response times
dramatically while being much more flexible with regard to the way you
search for the right information in unstructured data in and outside
of the database. Just open the Text Reference and discover a whole new
world of powerful and efficient full text indexing and searching
supplied by Oracle out of the box. And it comes with all Oracle
editions for no extra cost. :)
Hth,
Rene Nyffenegger
You can't, at least on 10g XE:
SQL> create table foo_bar_t (a number, b varchar2(20));
Table created.
SQL> create index foo_bar_i1 on foo_bar_t (upper(b));
Index created.
SQL> create index foo_bar_i2 on foo_bar_t(regexp_like(b, '^[[:digit:]]+$'));
create index foo_bar_i2 on foo_bar_t(regexp_like(b, '^[[:digit:]]+$'))
*
ERROR at line 1:
ORA-00904: "REGEXP_LIKE": invalid identifier
Maybe, I am missing something, then please let me know.
Cheers, RenÚ
--
Rene Nyffenegger
http://www.adp-gmbh.ch
Thomas Kellerer
On 20.04.2007 08:52 Rene Nyffenegger wrote:
> On 2007-04-19, Thomas Kellerer <JUAXQO...@spammotel.com> wrote:
>> On 18.04.2007 22:52 zstrin...@gmail.com wrote:
>>> The regular expression syntax is cleaner, especially when you have a
>>> lot of strings to search for! However, the LIKE expression runs in 20
>>> seconds, while the REGEXP_LIKE one runs in 60 seconds. Has anyone
>>> else noticed this? Any way to speed it up?
>>>
>> I thought you should be able to create a function based index to support
>> the regex match but I'm not sure (could have been Postgres where I did
>> that the last time ;) )
>
> You can't, at least on 10g XE:
OK, then it was indeed with PostgreSQL ;)
Thanks for the clarification
Regards
Thomas
Maxim Demenko
Because regexp_like is not a function but condition.
Here (on 10.2.0.3):
SQL> create index foo_bar_i2 on foo_bar_t(regexp_replace(b,
'^([[:digit:]]+)$','\1'));
Index created.
Best regards
Maxim