1. Data Warehousing

2. Databases support: Transaction Processing Systems
operational level decision
recording of transactions
Decision Support Systems
tactical and strategic decision making
analysis of historical records

3. Can one database support both?
RDBMS
DSS
TPS

4. Can one database support both?
RDBMS
DSS
TPS
low concurrency
large reads
significant aggregation
high concurrency
small transactions
limited aggregation
Yes… but at a cost in performance.

5. The Solution… TPS DSS Production Database (OLTP) Data Warehouse
The TPS doesn't need 10yrs of info; it can be kept lean.
You can also pre-compute certain common queries (adding up certain totals, etc)
You might *denormalize* data -- take two big tables that you know people
will be joining, and go ahead and join:
non-normalized, joined tables are only a problem when updating;
it's not a problem for read-only.
Extract,
Transport & Transformation
Load

6. OLTP vs DW Characteristics
OLTP Database
Data Warehouse
High Read/Write Concurrency
Primarily Read Only
Highly Normalized
Highly Denormalized
Limited Transaction History
Massive Transaction History
Very Detailed Data
Detailed and Summarized Data
"OLTP" -- on-line transaction processing.
"external data" might have interest rates, stock prices, competitor revenues, etc.
Limited External Data
Significant External Data

7. Data Marts (3-tier approach)
External Data
Sources
Data
Mart A
DSS
Data
Warehouse
Production
Database
(OLTP)
Data
Mart B
DSS
ETL
Data
Mart C
Transformation
& Limitation
DSS

8. Data Marts (bottom-up approach)
External Data
Sources
Data
Mart A
DSS
ETL
Production
Database
(OLTP)
External Data
Sources
Data
Mart B
ETL
DSS
ETL
Data
Mart C
DSS
External Data
Sources

9. Multi-dimensional (Sales) Data80
110 60 25
California 40 90 50 30
Utah 70 55 60 35
March 3
Arizona
March 2
March 1
Diet Soda
Lime Soda
Soda
Orange Soda

10. Cube Operations Cube (group by option)
Slice (implement in Oracle with where clause)
Dice (implement in Oracle with where clause)
Drill Down (implemented in report writers)
Roll-up (group by option)
Pivot (not implemented by Oracle (but by Access))

11. Cube Data Example Create table sales ( Item varchar2(20),
State varchar2(20),
Amount number(6),
Day date);
Insert into Sales
values('Soda','California',80,'01-Mar-2004');
values('Diet Soda','California',110,'01-Mar-2004'); …

12. Examine these queries Select * from sales;
Select Item, State, sum(amount)
from sales
group by Item, State;
group by Rollup(Item, State);
Select State, Item, sum(amount)
group by Rollup(State, Item);
group by Cube(State, Item);

13. Materialized Views
We looked at Materialized views earlier this year. Materialized views are one of the primary tools for Data Warehousing in Oracle.
Recall the materialized views are schema objects that can be used to summarize, precompute, replicate, and distribute data. Unlike, regular views, they are not constructed when requested (like a query) but are actually materialized in secondary storage making them much faster.
In data warehouses, materialized views are used to precompute and store aggregated data such as sums and averages. Materialized views in these environments are typically referred to as summaries because they store summarized data.

14. RDBMS Star Schema
A star schema presents a set of tables that are centered around an individual event of interest. This makes it easy for users of mining and reporting applications to analyze the event across many dimensions.
Item
Store
ItemID
StoreID
Name
Manager
Sales
UnitPrice
Street
SalesNO
Brand
City
SalesUnits
Category
Zip
SalesDollars
SalesCost
ItemID
Customer
Day
CustID
CustID
DayID
StoreID
Name
DayOfMonth
DayID
Phone
Month
Street
Year
City
DayOfWeek