RE: fast search or scan for words in selected field

Even better with this method, you conceivably have only one index file.
Consider the following file structure:

SoundexKey char(4)
CompanyKey char(10)

For the soundex below, you would have these entries, supposing "My Test
Company" had a company ID of 1:

C515, 1
M000, 1
T230, 1

Then the search SQL is extremely simple

Select distinct CompanyKey from searchfile
Where SoundexKey in (soundex(:search1), soundex(:search2),
soundex(:search3), ...)

Expand this generically by adding a "Key type"

SoundexKey, KeyValue, KeyType
M000, 1, Company

Then do a

Select distinct KeyValue, KeyType from searchfile where soundexkey in
(soundex(:search1)) and KeyType = "Company"

Then you have a generic search file for a lot more data than just companies:
vendors, customers, names, product descriptions, etc. A single file like
this can accommodate a variety of searching needs.


Loyd Goodbar
Senior programmer/analyst
BorgWarner
E/TS Water Valley
662-473-5713

-----Original Message-----
From: Vern Hamberg [mailto:vhamberg@xxxxxxxxxxxxxxxxxxxxxxxxx] 
Sent: Friday, September 10, 2004 13:03
To: Midrange Systems Technical Discussion
Subject: RE: fast search or scan for words in selected field

You're right - the LIKE predicate has to look through the entire field if 
it has the wildcard at the beginning.

The suggestion to use SOUNDEX to reduce disk space usage sounds (heh) good 
to me. There'd be a max length of 4, and SQL has a SOUNDEX() function. The 
words in "MY TEST COMPANY" would have the following soundex values:

MY    TEST  COMPANY
M000  T230   C515

There's not much point in the shorter words, since ME and MA and MI and MY 
would all get the value M000. And case is ignored.

Searches could still use

WHERE indexval = soundex(:search)

or

WHERE soundex(:search) = indexval

BTW, indexes might not be used because of the derived expression, according 
to the Indexing Strategies paper from IBM.

Vern

At 12:31 PM 9/10/2004, you wrote:
>I'm with Joe on this one.  I can't see how indices would help with
>substring searches.  Then again, just because I can't comprehend it
>doesn't mean it wouldn't.
>
>Rob Berendt
>--
>Group Dekko Services, LLC
>Dept 01.073
>PO Box 2000
>Dock 108
>6928N 400E
>Kendallville, IN 46755
>http://www.dekko.com