Cube, Rollup and Materialized Views:

Mining Oracle Gold

John King
King Training Resources
6341 South Williams Street
Littleton, CO 80121-2627 USA
www.kingtraining.com
800.252.0652 or 303.798.5727

Copyright @ 2000, John Jay King, All rights reserved

 Objectives
Become aware of Oracle8i Release 2 (8.1.6)
Analytic Functions at a high level

Learn about the Cube and Rollup enhancements

to GROUP BY

Be aware of Materialized Views and how they may

be used to engineer more-useful and faster systems

Know how to use the Cube, Rollup, and Materialized

views to enhance systems
Oracle 8.1.6 Aggregates
AVG REGR_R2
CORR REGR_SLOPE
COUNT REGR_SXX
COVAR_POP REGR_SYY
REGR_SXY
COVAR_SAMP
STDDEV
GROUPING
STDDEV_POP
MAX
STDDEV_SAMP
MIN
SUM
REGR_AVGX
VAR_POP
REGR_AVGY
REGR_COUNT VAR_SAMP
REGR_INTERCEPT VARIANCE
Oracle8i Version 2 (8.1.6)
Analytic Functions
Oracle 8.1.6 includes a new set of functions designed to provide
expanded support for data mining operations -
(this topic is too rich to fully cover in the context of this paper)
The analytic functions are divided into four "families"
Lag/Lead Compares values of rows to other rows in same table:
LAG, LEAD
Ranking Supports "top n" queries: CUME_DIST, DENSE_RANK,
NTILE, PERCENT_RANK, RANK, ROW_NUMBER
Reporting Aggregate Compares aggregates to non-aggregates
(pct of total):
RATIO_TO_REPORT
Window Aggregate Moving average type queries:
FIRST_VALUE, LAST_VALUE
The analytic functions allow users to divide query result sets into
ordered groups of rows called partitions
(not the same as database partitions)
Oracle8i Version 2 (8.1.6)
Analytic Function Clauses
Along with the new functions came new clauses
(again, too rich to cover completely here):

analytic_function ( ) OVER (analytic clause)

Analytic clause
Query_partition_clause-Order_by clause-Windowing clause
Query partition clause
PARTITION BY list,of,cols
Windowing clause
RANGE or ROWS ...
Order by clause
ORDER BY col,list
Analytic Function: RANK (8.1.6)
RANK provides rankings of values with gaps where sets
of rows have equal values
(DENSE_RANK removes gaps)

SQL> run
1 select deptno,ename,sal,
RANK() OVER (PARTITION BY DEPTNO ORDER BY SAL DESC) salrank
2 from emp where deptno in (10,30)
DEPTNO ENAME SAL SALRANK
----------- ----------------- ---------- --------------
10 KING 5000 1
10 CLARK 2450 2
10 MILLER 1300 3
30 BLAKE 2850 1
30 ALLEN 1600 2
30 TURNER 1500 3
30 MARTIN 1250 4
30 WARD 1250 4
30 JAMES 950 6
Old Aggregate, New Usage (8.1.6)
Analytic function clauses may be used with many existing
aggregates

SQL> run
1 select deptno, ename, sal,
2 ,round(avg(sal) OVER (PARTITION BY deptno) , 0) avg_sal
3 from emp
4* where deptno in (10,20)
DEPTNO ENAME SAL AVG_SAL
----------- ---------------- ---------- ----------
10 KING 5000 2917
10 CLARK 2450 2917
10 MILLER 1300 2917
20 JONES 2975 2175
20 FORD 3000 2175
20 SMITH 800 2175
20 SCOTT 3000 2175
20 ADAMS 1100 2175
 ROW_NUMBER (8.1.6)
ROW_NUMBER allows ranking by criteria

SQL> run
1 select deptno, ename, sal,
2 ROW_NUMBER() OVER (PARTITION BY deptno ORDER BY by sal desc)
3 sal_rank
4 from emp
5* where deptno in (10,20)
DEPTNO ENAME SAL SAL_RANK
----------- ----------------- ---------- ---------------
10 KING 5000 1
10 CLARK 2450 2
10 MILLER 1300 3
20 FORD 3000 1
20 SCOTT 3000 2
20 JONES 2975 3
20 ADAMS 1100 4
20 SMITH 800 5
 Cube and Rollup
K CUBE and ROLLUP extend GROUP BY
K ROLLUP builds subtotal aggregates at any level,
including grand total
K CUBE extends ROLLUP to calculate all possible
combinations of subtotals for a GROUP BY
K Cross-tabulation reports are easy with CUBE
K Oracle8i Release 2 (Oracle version 8.1.6) began release
in February 2000, its new Analytic functions include:
ranking, moving aggregates, period comparisons,
ratio of total, and cumulative aggregates
Normal GROUP BY Functionality
Normally, GROUP BY allows aggregates
(sub-totals) by specific column or set of columns
Before Oracle8i SQL required JOIN or UNION to combine
subtotal information and grand totals in a single SQL
query
ROLLUP creates subtotals and grand totals in the same
query along with intermediate subtotals
CUBE adds cross-tabulation information based upon the
GROUP BY columns
GROUP BY (without CUBE or ROLLUP)

Normally GROUP BY sorts on GROUP BY

columns, then calculates aggregates
SQL> select deptno Department
2 ,job
3 ,sum(sal) "Total SAL"
4 from emp
5 group by deptno,job
6 /
DEPARTMENT JOB Total SAL
---------- --------- ---------
10 CLERK 1300
10 MANAGER 2450
10 PRESIDENT 5000
20 ANALYST 6000
20 CLERK 1900
20 MANAGER 2975
30 CLERK 950
30 MANAGER 2850
30 SALESMAN 5600
 GROUP BY ROLLUP
K ROLLUP provides aggregates at each GROUP BY level
SQL> col Department format a20
SQL> break on Department
SQL> select nvl(to_char(deptno),'Whole Company') Department
2 ,nvl(job,'All Employees') job
3 ,sum(sal) "Total SAL"
4 from emp
5 group by rollup (deptno,job)
6 /
DEPARTMENT JOB Total SAL
-------------------- ------------- ---------
10 CLERK 1300
MANAGER 2450
PRESIDENT 5000
All Employees 8750
20 ANALYST 6000
CLERK 1900
MANAGER 2975
All Employees 10875
30 CLERK 950
MANAGER 2850
SALESMAN 5600
All Employees 9400
Whole Company All Employees 29025
 NULL Values in
CUBE/ROLLUP Rows
Subtotal and grand total lines generated by ROLLUP
substitute NULL for column values not present in the
manufactured output row
The example uses the NVL function to replace NULLS
Some columns might normally contain NULL values,
thus, normally occurring NULLS would be grouped with
rows manufactured by ROLLUP or CUBE
 GROUPING Function
To improve dealing with the NULL values present in the
rows created by ROLLUP (and CUBE discussed later),
Oracle provides the new GROUPING function
GROUPING returns a value of 1 if a row is a subtotal
created by ROLLUP or CUBE, and a 0 otherwise
The following example shows the same query used
previously, with DECODE used in conjunction with
GROUPING to more-elegantly deal with the null values
created by ROLLUP and CUBE

(Note: sample data contains no null values, the results

from this query and the previous query are the same).
 GROUPING Example
SQL> col Department format a20
SQL> break on Department
SQL> select decode(grouping(deptno),1,'Whole Company'
2 ,'Department ' || to_char(deptno)) Department
3 ,decode(grouping(job),1,'All Employees',job) job
4 ,sum(sal) "Total SAL"
5 from emp
6 GROUP BY ROLLUP (deptno,job)
/
DEPARTMENT JOB Total SAL
-------------------- ------------- ---------
Department 10 CLERK 1300
MANAGER 2450
PRESIDENT 5000
All Employees 8750
Department 20 ANALYST 6000
CLERK 1900
MANAGER 2975
All Employees 10875
Department 30 CLERK 950
MANAGER 2850
SALESMAN 5600
All Employees 9400
Whole Company All Employees 29025
 GROUP BY CUBE
CUBE automatically calculates all possible combinations of subtotals
SQL> select decode(grouping(deptno),1,'Whole Company','Department '
|| to_char(deptno)) Department
2 ,decode(grouping(job),1,'All Employees',job) job
3 ,sum(sal) "Total SAL"
4 from emp GROUP BY CUBE(deptno,job)
DEPARTMENT JOB Total SAL
-------------------- ------------- ---------
Department 10 CLERK 1300
MANAGER 2450
PRESIDENT 5000
All Employees 8750
Department 20 ANALYST 6000
CLERK 1900
MANAGER 2975
All Employees 10875
Department 30 CLERK 950
MANAGER 2850
All Employees 9400
Whole Company ANALYST 6000
CLERK 4150
MANAGER 8275
PRESIDENT 5000
SALESMAN 5600
All Employees 29025
GROUP BY/ROLLUP/CUBE
CUBE add subtotals for all combinations of categories
(total salary for each job type was added in the example)
If there were three GROUP BY columns (i.e. country,
customer_id, product):
GROUP BY would produce aggregates each unique combination
of the three columns showing the aggregate for each product
ordered by each customer within each country
ROLLUP would add aggregates showing the total products by
country and customer_id, total products by country, and a grand
total of all products sold
CUBE would add aggregates for each product regardless of
country or customer id, aggregates for each customer_id
regardless of country or products ordered, and aggregates of
each product by country regardless of customer id
 Materialized Views
Oracle's SNAPSHOT is a query result table created
periodically to facilitate distribution or replication of data
Materialized Views in Oracle8i use similar technology to
allow a view's results to be stored as materialized in the
database for use by subsequent SQL statements
View materializations are refreshed periodically based
upon time criteria (defined at creation) or upon demand
View data is "old" until the view is refreshed
Indexes may be defined for Materialized Views
Materialized views can improve performance of frequent
requests for aggregate data or complex data
 CUBE/ROLLUP & Analytic Functions (8.1.6)

Combine Analytic Functions and Clauses with CUBE and ROLLUP

SQL> run
1 select decode(grouping(deptno),1,'Whole Company
2 ,'Department ' || to_char(deptno)) Department
3 ,decode(grouping(job),1,'All Employees',job) job
4 ,sum(sal) "Total SAL
5 ,ROW_NUMBER() OVER (PARTITION BY deptno ORDER BY sum(sal)) rownbr
6* from emp where deptno in (10,20) group by rollup (deptno,job)

DEPARTMENT JOB Total SAL ROWNBR

-------------------- ------------- ---------- ------------
Department 10 CLERK 1300 1
MANAGER 2450 2
PRESIDENT 5000 3
All Employees 8750 4
Department 20 CLERK 1900 1
MANAGER 2975 2
ANALYST 6000 3
All Employees 10875 4
Whole Company All Employees 19625 1
 CREATE MATERIALIZED VIEW

create materialized view dept_summary

refresh start with sysdate next sysdate + 1
as
select dept.deptno
,dname
,count(*) nbr_emps
,sum(nvl(sal,0)) tot_sal
from scott.emp emp
,scott.dept dept
where emp.deptno(+) = dept.deptno
group by dept.deptno,dname;
 Creation Caveats
ORACLE recommends names not exceed 19 characters,
so that generated names are 30 characters or less
STORAGE, TABLESPACE, LOB, CACHE, LOGGING,
CLUSTER, and partitioning are similar to CREATE TABLE
BUILD IMMEDIATE is default, can do BUILD DEFERRED
ON PREBUILT TABLE allows use of existing tables; the
Materialized View name and the Table name must match
REFRESH controls reloading rate, START WITH specifies
the first refresh, NEXT specifies subsequent refreshes
(see the Oracle8i Replication manual for specifics)
AS describes the query for the materialized view, just
about any query may be used with a few restrictions
Oracle8i Release 2 allows query to contain ORDER BY
 Using Pre-built Tables
Basing a materialized view upon an existing table (ON
PREBUILT TABLE) allows the use of existing tables and
indexes
Using ON PREBUILT TABLE requires that the underlying
table and the materialized view share the same name
and schema
WITH REDUCED PRECISION allows a refresh to work
properly even if some columns generate different
precision than originally defined
 Pre-Built Table: Example Table

create table dept_summary_tab

as
select dept.deptno
,dname
,count(*) nbr_emps
,sum(nvl(sal,0)) tot_sal
from scott.emp emp
,scott.dept dept
where emp.deptno(+) = dept.deptno
group by dept.deptno,dname;
 Pre-Built Table: Example MView

create materialized view dept_summary_tab

on prebuilt table
with reduced precision
refresh start with sysdate next sysdate + 1
as
select dept.deptno
,dname
,count(*) nbr_emps
,sum(nvl(sal,0)) tot_sal
from scott.emp emp
,scott.dept dept
where emp.deptno(+) = dept.deptno
group by dept.deptno,dname;
MView Refresh via PL/SQL
An Oracle-provided PL/SQL packaged
procedure DBMS_MVIEW.REFRESH may
be used to refresh upon demand
Careful! This procedure COMMITs
changes in the active transaction

begin
dbms_mview.refresh('dept summary tab');
end;
/
 Conclusion
CUBE and ROLLUP reduce work necessary to code and
create aggregates often requested by management to
provide competitive or summary information
CUBE and ROLLUP provide mechanisms for using a
single SQL statement to provide data that would have
required multiple SQL statements, programming, or
manual summarization in the past
Materialized Views reduce the impact of frequently
executed queries by storing results and refreshing them
on a periodic basis
These tools may be used to mine Oracle databases for
the golden information frequently in demand today
To contact the author:
John King
King Training Resources
6341 South Williams Street
Littleton, CO 80121-2627 USA
1.800.252.0652 - 1.303.798.5727
Email: john@kingtraining.com