It seems hybrid database

Question

Hi all :-D,
I´ve designed a OLTP base with all normalized tables.
But there are tables that contains millions of rows.
The reports of the application were simple, but
now the customer needs complex reports (complex selects)
with many tables with millions of rows to analyze the result
in the web application, returning millions rows.
This base seems a hybrid database.
What´s the best way to design this?
I thought to create another schema with design only star schema
and to create materialized views to load data from oltp schema
to this schema.
Today, The application user has synonyms to oltp schema. So,
I thought to create other synonyms in this user linking to
tables of star schema.
What do you think about this design?
Is it valid? am I´m triping? :-)
In this case, what´s the best way?
thank U very much!!!!

Answer 1

Hmm nobody answered yet ;)
OK If you have big box and much available resources you can of course create reports on normalized data model. Reports will be fully up to date, (as you have necessary resources!) nobody complains and everything is fantastic!
Unfortunately most of the time users already complain about slowness of the system and additional reports will definitely take off the last CPU and IO unit :)
So that depends on your needs.
Of course if you are creating denormalized model for reports in the same database you'll anyway take away some resources from oltp users, unless you have overnight silence period and do both data load and report generation only overnight. Of course your aim is to make necessary resources for data load and report generation on denormalized model LESS than those necessary for report generation for normalized schema.
Doing both OLTP and reports on the same database also makes some other challenges like you can set only one optimizer mode at instance level, so you have to remember alter it either for OLTP app or report app. Most probably you have now first_rows(_n), but for reports you'll need all_rows. Also some other parameters like workarea_size_policy could be different for OLTP and Warehouse apps.
Speaking about app user that has synonyms to oltp schema. I suggest you also to create another app user for star schema, because oracle has built in features to manage load on the db - profiles and resource plans - but both of them work only in user level, so most probably you'll need to manage resources for oltp and reports app differently.
<shameless note>
You can look at my paper about searching in normalized schemas using materialized views http://www.gplivna.eu/papers/mat_views_search.htm
if you need it ;)
</shameless note>
Gints Plivna
http://www.gplivna.eu/