RE: Data Modeling Question
- From: "Powell, Mark D" <mark.powell@xxxxxxx>
- To: "oracle-l" <oracle-l@xxxxxxxxxxxxx>
- Date: Thu, 7 Feb 2008 09:22:53 -0500
Unless your application normally only needs to retrieve a single
attribute for an item_code per transaction just add a second attribute
table to you application to handle attributes 1001+
Doing this will almost certainly result in better query performance than
using a parameter/value attribute table. I say this because we have a
vendor provided statistical process system build on just such a scheme.
Inserts are great but query performance is not so great.
-- Mark D Powell --
Phone (313) 592-5148
-----Original Message-----
From: oracle-l-bounce@xxxxxxxxxxxxx
[mailto:oracle-l-bounce@xxxxxxxxxxxxx] On Behalf Of Ahbaid Gaffoor
Sent: Wednesday, February 06, 2008 10:52 PM
To: oracle-l
Subject: Data Modeling Question
We have a data model which looks like the following:
ITEMS
==========
ITEM_CODE VARCHAR2(100) (Primary Key col 1)
ITEM_TYPE VARCHAR2(100)
ATTRIBUTE_001 VARCHAR2(4000)
ATTRIBUTE_002 VARCHAR2(4000)
...
ATTRIBUTE_250 VARCHAR2(4000)
ITEM_TYPE_LU
============
ITEM_TYPE VARCHAR2(100) (Primary Key col 1)
ATTRIBUTE_NAME VARCHAR2(1000) (Primary Key col 2) ATTRIBUTE_COLNUM
NUMBER
It's a developer's dream of how we can store attributes for an item, the
item type is used to decide which column holds what attributes by
storing that metadata in the ITEM_TYPE_LU table. Initially, some 7 years
ago the developer only dreamed of no more than 10 attributes per item.
Today there are 205 attributes, some fields hold data that's a few
characters and some the full 4000 characters.
I informed the development team that we face a max number of columns
(1000 when I tested on 10.2.0.3)
Now we are undertaking a redesign to provide a data model that allows us
to store an unlimited number of attributes.
The ideas we have in mind so far are:
1) Switch to a row based schema, so the two tables become one, each
attribute and its value taking a row for an item.
ITEMS
=====
ITEM_CODE VARCHAR2(100) (Primary Key col 1)
ATTRIBUTE_NAME VARCHAR2(250)
ATTRIBUTE_VALUE VARCHAR2(4000)
So a single item would look as follows:
SELECT * from ITEMS where ITEM_CODE = 'shoe-001'; ITEM_CODE
ATTRIBUTE_NAME ATTRIBUTE_VALUE
---------------------------------
shoe-001 size 10
shoe-001 color brown
shoe-001 price 15.45
shoe-001 manufacturer clarks
2) Use a CLOB column and store an XML document in it with all
information:
ITEMS
======
ITEM_CODE VARCHAR2(100) (Primary Key col 1)
ITEM_XML CLOB
SELECT * from ITEMS where ITEM_CODE = 'shoe-001';
---------------------------------
ITEM_CODE ITEM_XML
---------- --------------------------------------------------
shoe-001 <item>
<item_code>shoe-001</item_code>
<size>10</size>
<color>brown</brown>
<price>15.45</price>
<manufacturer>clarks</manufacturer>
</item>
3) Use Oracle's XML data types (not sure how this works)
I am partial to the first approach since I would be able to easily parse
data when a business request needs it, also I can make use of Oracle
partitioning to improve performance.
One other point to consider is that single attributes can be updated for
an item, in the row based model this is simple, however in the CLOB
based solution I am concerned that we need to replace the entire CLOB
for a single attribute value change.
What I'd like is any advice pros / cons of the CLOB (approach 2) and the
XML type (approach 3) solutions.
For scale, we are looking at close to 200 million items, each item
having about 150 attributes each.
I am proceeding with approach 1 for now.
many thanks for your input
Ahbaid
--
//www.freelists.org/webpage/oracle-l
--
//www.freelists.org/webpage/oracle-l
Other related posts:
