Skip Headers
Oracle® OLAP User's Guide
11g Release 1 (11.1)

Part Number B28124-01
Go to Documentation Home
Home
Go to Book List
Book List
Go to Table of Contents
Contents
Go to Index
Index
Go to Master Index
Master Index
Go to Feedback page
Contact Us

Go to previous page
Previous
Go to next page
Next
View PDF

7 Administering Oracle OLAP

Because Oracle OLAP is contained in the database and its resources are managed using the same tools, the management tasks of Oracle OLAP and the database converge. Nonetheless, you should address tasks such as database tuning in the specific context of data warehousing.

This chapter contains the following topics:

Setting Database Initialization Parameters

Table 7-1 identifies the parameters that affect the performance of Oracle OLAP. Alter your server parameter file or init.ora file to these values, then restart your database instance. You can monitor the effectiveness of these settings and adjust them as necessary.

See Also:

Table 7-1 Initial Settings for Database Parameters

Parameter Default Value Recommended Setting Description

JOB_QUEUE_PROCESSES

0

Number of CPUs, plus one additional process for every three CPUs; in a multi-core CPU, each core counts as a CPU

For example, JOB_QUEUE_PROCESSES=5 for a four-processor computer

Controls the degree of parallelism in OLAP builds, as described in "Parallelism"

SESSIONS

Derived

2.5 * maximum number of simultaneous OLAP users

Provides sufficient background processes for each user

UNDO_MANAGEMENT

AUTO

(MANUAL in 10g)

AUTO

Specifies use of an undo tablespace

UNDO_TABLESPACE

Derived

Name of the undo tablespace, which must already be defined

Identifies the undo tablespace defined for OLAP use, as shown in "Creating an Undo Tablespace"


To set the system parameters:

  1. Open the init.ora initialization file in a text editor.

  2. Add or change the settings in the file, as described in Table 7-1.

  3. Stop and restart the database.

    On Windows, use the Services utility to stop and restart OracleService.

    On Linux, use commands like the following. Be sure to identify the initialization file in the STARTUP command.

    SQLPLUS '/ AS SYSDBA'
    SHUTDOWN IMMEDIATE
    STARTUP pfile=$ORACLE_BASE/admin/orcl/pfile/init.ora.724200516420
    

Storage Management

Analytic workspaces are stored in the owner's default tablespace, unless the owner specifies otherwise. All tablespaces for OLAP use should specify EXTENT MANAGEMENT LOCAL. Tablespaces created using default parameters may use resources inefficiently. You should create undo, permanent, and temporary tablespaces that are appropriate for storing analytic workspaces.

Creating an Undo Tablespace

Create an undo tablespace with the EXTENT MANAGEMENT LOCAL clause, as shown in this example:

CREATE UNDO TABLESPACE olapundo DATAFILE '$ORACLE_HOME/oradata/undo.dbf'
    SIZE 64M REUSE AUTOEXTEND ON NEXT 8M
    MAXSIZE UNLIMITED EXTENT MANAGEMENT LOCAL;

After creating the undo tablespace, change your system parameter file to include the following settings, then restart the database as described in "Setting Database Initialization Parameters".

UNDO_TABLESPACE=tablespace
UNDO_MANAGEMENT=AUTO

Creating Permanent Tablespaces for OLAP Use

Each dimensional object occupies at least one extent. A fixed extent size may waste most of the allocated space. For example, if an object is 64K and the extents are set to a uniform size of 1M (the default), then only a small portion of the extent is used.

Create permanent tablespaces with the EXTENT MANAGEMENT LOCAL and SEGMENT SPACE MANAGEMENT AUTO clauses, as shown in this example:

CREATE TABLESPACE glo DATAFILE '$ORACLE_HOME/oradata/glo.dbf'
   SIZE 64M REUSE AUTOEXTEND ON NEXT 8M MAXSIZE UNLIMITED
   EXTENT MANAGEMENT LOCAL SEGMENT SPACE MANAGEMENT AUTO;

Creating Temporary Tablespaces for OLAP Use

Oracle OLAP uses the temporary tablespace to store all changes to the data in a cube, whether the changes are the result of a data load or data analysis. Saving the cube moves the changes into the permanent tablespace and clears the temporary tablespace.

This usage creates numerous extents within the tablespace. A temporary tablespace suitable for use by Oracle OLAP should specify the EXTENT MANAGEMENT LOCAL clause and a UNIFORM SIZE clause with a small size, as shown in this example:

CREATE TEMPORARY TABLESPACE glotmp TEMPFILE '$ORACLE_HOME/oradata/glotmp.tmp'
   SIZE 50M REUSE AUTOEXTEND ON NEXT 5M MAXSIZE UNLIMITED
   EXTENT MANAGEMENT LOCAL UNIFORM SIZE 256K;

Spreading Data Across Storage Resources

Oracle Database provides excellent storage management tools to simplify routine tasks. Automatic Storage Management (ASM) provides a simple storage management interface that virtualizes database storage into disk groups. You can manage a small set of disk groups, and ASM automates the placement of the database files within those disk groups.

ASM spreads data evenly across all available storage resources to optimize performance and utilization. After you add or drop disks, ASM automatically rebalances files across the disk group.

Because OLAP is part of Oracle Database, you can use ASM to manage both relational and dimensional data.

ASM is highly recommended for analytic workspaces. A system managed with ASM is faster than a file system and easier to manage than raw devices. ASM optimizes the performance of analytic workspaces both on systems with RAC and those without RAC.

However, you do not need ASM to use Oracle OLAP. You can still spread your data across multiple disks, just by defining the tablespaces like in this example:

CREATE TABLESPACE glo DATAFILE 
   'disk1/oradata/glo1.dbf' SIZE 64M REUSE AUTOEXTEND ON NEXT 8M MAXSIZE 1024M
   EXTENT MANAGEMENT LOCAL SEGMENT SPACE MANAGEMENT AUTO;

ALTER TABLESPACE glo ADD DATAFILE 
   'disk2/oradata/glo2.dbf' SIZE 64M REUSE AUTOEXTEND ON NEXT 8M MAXSIZE 1024M,
   'disk3/oradata/glo3.dbf' SIZE 64M REUSE AUTOEXTEND ON NEXT 8M MAXSIZE UNLIMITED;

Dictionary Views and System Tables

Oracle Database data dictionary views and system tables contain extensive information about analytic workspaces.

Static Data Dictionary Views

Among the static views of the database data dictionary are several that provide information about analytic workspaces. Table 7-2 provides brief descriptions of them. All data dictionary views have corresponding DBA and USER views.

Table 7-2 Static Data Dictionary Views for OLAP

View Description

ALL_AWS

Describes all analytic workspaces accessible to the current user.

ALL_AW_OBJ

Describes the current objects in all analytic workspaces accessible to the current user.

ALL_AW_PROP

Describes the properties defined in all analytic workspaces accessible to the current user.

ALL_AW_PS

Describes the page spaces currently in use by all analytic workspaces accessible to the current user.


See Also:

  • "Querying the Data Dictionary" for a list of data dictionary views that describe OLAP dimensional objects

  • Oracle Database Reference for full descriptions of all data dictionary views

System Tables

The SYS user owns several tables associated with analytic workspaces. Table 7-3 provides brief descriptions.

Important:

These tables are vital for the operation of Oracle OLAP. Do not delete them or attempt to modify them directly without being fully aware of the consequences.

Table 7-3 OLAP Tables Owned By SYS

Table Description

AW$

Maintains a record of all analytic workspaces in the database, recording its name, owner, and other information.

AW$AWCREATE

Stores the AWCREATE analytic workspace, which contains programs for using OLAP Catalog metadata in Oracle Database 10g Release 10.1.0.2 and earlier releases. It exists only for backward compatibility.

AW$AWCREATE10G

Stores the AWCREATE10G analytic workspace, which contains programs for using OLAP Catalog metadata in Oracle Database 10g Release 10.1.0.3. The OLAP Catalog is not used by later releases. It exists only for backward compatibility.

AW$AWMD

Stores the AWMD analytic workspace, which contains programs for creating metadata catalogs.

AW$AWREPORT

Stores the AWREPORT analytic workspace, which contains a program named AWREPORT for generating a summary space report.

AW$AWXML

Stores the AWXML analytic workspace, which contains programs for creating and managing analytic workspaces for Oracle Database 10g Release 10.1.0.4 and later.

AW$EXPRESS

Stores the EXPRESS analytic workspace. It contains objects and programs that support basic operations. EXPRESS is used any time a session is open.

AW_OBJ$

Describes the objects stored in analytic workspaces.

AW_PRG$

Stores program data. Not currently used.

AW_PROP$

Stores analytic workspace object properties.

AW_TRACK$

Stores tracking data about access to aggregate cells. Not currently used.

PS$

Maintains a history of all page spaces. A page space is an ordered series of bytes equivalent to a file. Oracle OLAP manages a cache of workspace pages. Pages are read from storage in a table and written into the cache in response to a query. The same page can be accessed by several sessions.

The information stored in PS$ enables Oracle OLAP to discard pages that are no longer in use, and to maintain a consistent view of the data for all users, even when the workspace is being modified during their sessions. When changes to a workspace are saved, unused pages are purged and the corresponding rows are deleted from PS$.


Analytic Workspace Tables

Analytic workspaces are stored in tables in the Oracle database. The names of these tables always begin with AW$.

For example, if the GLOBAL user creates two analytic workspaces, one named MARKETING and the other named FINANCIALS, then these tables will be created in the GLOBAL schema:

AW$FINANCIALS
AW$MARKETING

The tables store all of the object definitions and data.

Build Logs

The first time you load data into a cube or dimension using Analytic Workspace Manager, a table named CUBE_BUILD_LOG is created in your schema. This table stores information about the build. Information about each subsequent build is added to the table and is identified by its own build identifier. CUBE_BUILD_LOG is populated whenever a cube is refreshed, whether by Analytic Workspace Manager, the database materialized view refresh subsystem, or a PL/SQL procedure. It is updated in real time during the build, so that you can check its status at any time by querying the table in SQL.

The Maintenance Wizard in Analytic Workspace Manager displays the relevant rows from this table at the end of every build on the Log page.

Partitioned Cubes and Parallelism

Cubes are often partitioned to improve build and maintenance times. For information about creating a partitioned cube, refer to "Choosing a Partitioning Strategy".

Querying Metadata for Cube Partitioning

To discover the current partitioning, query the ALL_CUBES data dictionary view. The PARTITION_DIMENSION_NAME, PARTITION_HIERARCHY_NAME, and PARTITION_LEVEL_NAME columns display partitioning information. For example, the following query shows that the Units Cube is partitioned on the Time dimension, the Calendar hierarchy, and the Calendar Year level.

SELECT partition_dimension_name, partition_hierarchy_name, 
     partition_level_name FROM all_cubes
     WHERE owner='GLOBAL' AND cube_name='UNITS_CUBE';

PARTITION_DIMENSION_NAME  PARTITION_HIERARCHY_NAME  PARTITION_LEVEL_NAME
------------------------- ------------------------- --------------------
TIME                      CALENDAR                  CALENDAR_YEAR

Creating and Dropping Partitions

The OLAP engine automatically creates and drops partitions as part of data maintenance, as members are added and deleted from the partitioning dimension.

For example, assume that in the sample Global analytic workspace, the Units cube is partitioned on the Time dimension, using the Calendar hierarchy, and at the Calendar Quarter level. The OLAP engine creates a partition for each Calendar Quarter and its children. The default top partition contains Calendar Years and all members of the Fiscal hierarchy. If Global has three years of data, then the Units cube has 13 partitions: Four bottom partitions for each Calendar Year, plus the top partition.

A data refresh typically creates new time periods and deletes old ones. Whenever a Calendar Quarter value is loaded into the Time dimension, a corresponding new partition is added to the cube. Whenever a Calendar Quarter value is deleted from the Time dimension, the corresponding empty partition is deleted from the cube.

Parallelism

You can improve the performance of data maintenance by enabling parallel processing. There are two levels of parallelism:

  • Parallel job execution: Loading and aggregating the data using multiple processes.

  • Parallel update: Moving the data from temporary to permanent tablespaces using multiple processes.

This number of parallel processes is controlled by these factors:

  • The number of objects that can be aggregated in parallel. Each cube and each partition (including the top partition) can use a separate process.

    You can control the number of partitions in a cube on the Partitioning tab of the cube property sheet in Analytic Workspace Manager.

  • The number of simultaneous database processes the user is authorized to run.

    This number is controlled by the JOB_QUEUE_PROCESSES parameter. The setting for this parameter is based on the number of processors, as described in "Setting Database Initialization Parameters". You can obtain the current parameter setting with the following SQL command:

    SHOW PARAMETER JOB_QUEUE_PROCESSES
    
    
  • For parallel update, the number of processes you allocate to the job. You can specify the number of processes in the Maintenance Wizard of Analytic Workspace Manager when specifying the task processing options, or on the Materialized View tab of the cube.

Suppose that a cube is partitioned on the Quarter level of Time, and the cube contains three years of data. The cube has 3*4=12 bottom partitions, JOB_QUEUE_PROCESSES is set to 8, and you set the parallelism option to 4 for the build. Oracle Database will process the cube in this way:

  1. Load and build the dimensions of the cube serially using a single process.

  2. Load and build the 12 bottom partitions in parallel using 4 processes. As soon as one process finishes, another begins until all 12 are complete.

    This cube could use the 8 processes allowed by JOB_QUEUE_PROCESSES, but it is limited to 4 by the build setting.

  3. Load and build the top partition.

The following are excerpts from CUBE_BUILD_LOG for a completed build that used two parallel processes.

SQL> SELECT slave_number, status, command, build_object, partition FROM cube_build_log 
      WHERE build_id='6';
 
SLAVE_NUMBER STATUS     COMMAND              BUILD_OBJECT                             PARTITION
------------ ---------- -------------------- ---------------------------------------- ----------
             STARTED    BUILD                BUILD GLOBAL.UNITS_CUBE USING (CLEAR VAL
                                             UES, LOAD, SOLVE)
 
             STARTED    UPDATE               UNITS_CUBE
             COMPLETED  UPDATE               UNITS_CUBE
             STARTED    COMMIT               UNITS_CUBE
             COMPLETED  COMMIT               UNITS_CUBE
             STARTED    SCHEDULE DBMS JOB    BEGIN DBMS_CUBE.BUILD_SLAVE('GLOBAL.UNIT
                                             S_CUBE USING (CLEAR VALUES, LOAD , SOLVE
                                              )', 'P10', 'JOB$_41', 6, 1);  END;
 
             COMPLETED  SCHEDULE DBMS JOB    BEGIN DBMS_CUBE.BUILD_SLAVE('GLOBAL.UNIT
                                             S_CUBE USING (CLEAR VALUES, LOAD , SOLVE
                                              )', 'P10', 'JOB$_41', 6, 1);  END;
 
             STARTED    SCHEDULE DBMS JOB    BEGIN DBMS_CUBE.BUILD_SLAVE('GLOBAL.UNIT
                                             S_CUBE USING (CLEAR VALUES, LOAD , SOLVE
                                              )', 'P9', 'JOB$_42', 6, 2);  END;
 
             COMPLETED  SCHEDULE DBMS JOB    BEGIN DBMS_CUBE.BUILD_SLAVE('GLOBAL.UNIT
                                             S_CUBE USING (CLEAR VALUES, LOAD , SOLVE
                                              )', 'P9', 'JOB$_42', 6, 2);  END;
 
           1 STARTED    BUILD SLAVE          BUILD GLOBAL.UNITS_CUBE USING (CLEAR VAL P10
                                             UES, LOAD , SOLVE )
 
           1 STARTED    CLEAR VALUES         UNITS_CUBE                               P10
           2 STARTED    BUILD SLAVE          BUILD GLOBAL.UNITS_CUBE USING (CLEAR VAL P9
                                             UES, LOAD , SOLVE )
 
           1 COMPLETED  CLEAR VALUES         UNITS_CUBE                               P10
           1 STARTED    LOAD                 UNITS_CUBE                               P10
           2 STARTED    CLEAR VALUES         UNITS_CUBE                               P9
           1 COMPLETED  LOAD                 UNITS_CUBE                               P10
           1 STARTED    SOLVE                UNITS_CUBE                               P10
                  .
                  .
                  .
          10 STARTED    CLEAR VALUES         UNITS_CUBE                               P1
          10 COMPLETED  CLEAR VALUES         UNITS_CUBE                               P1
          10 STARTED    LOAD                 UNITS_CUBE                               P1
          10 COMPLETED  LOAD                 UNITS_CUBE                               P1
          10 STARTED    SOLVE                UNITS_CUBE                               P1
          10 COMPLETED  SOLVE                UNITS_CUBE                               P1
          10 STARTED    UPDATE               UNITS_CUBE                               P1
          10 COMPLETED  UPDATE               UNITS_CUBE                               P1
          10 STARTED    COMMIT               UNITS_CUBE                               P1
          10 COMPLETED  COMMIT               UNITS_CUBE                               P1
          10 COMPLETED  BUILD SLAVE          BUILD GLOBAL.UNITS_CUBE USING (CLEAR VAL P1
                                             UES, LOAD , SOLVE )
 
             STARTED    SCHEDULE DBMS JOB    BEGIN DBMS_CUBE.BUILD_SLAVE('GLOBAL.UNIT
                                             S_CUBE USING (CLEAR VALUES, SOLVE )', 'P
                                             0', 'JOB$_51', 6, 11);  END;
 
             COMPLETED  SCHEDULE DBMS JOB    BEGIN DBMS_CUBE.BUILD_SLAVE('GLOBAL.UNIT
                                             S_CUBE USING (CLEAR VALUES, SOLVE )', 'P
                                             0', 'JOB$_51', 6, 11);  END;
 
          11 STARTED    BUILD SLAVE          BUILD GLOBAL.UNITS_CUBE USING (CLEAR VAL P0
                                             UES, SOLVE )
 
          11 STARTED    CLEAR VALUES         UNITS_CUBE                               P0
          11 COMPLETED  CLEAR VALUES         UNITS_CUBE                               P0
          11 STARTED    SOLVE                UNITS_CUBE                               P0
          11 COMPLETED  SOLVE                UNITS_CUBE                               P0
          11 STARTED    UPDATE               UNITS_CUBE                               P0
          11 COMPLETED  UPDATE               UNITS_CUBE                               P0
          11 STARTED    COMMIT               UNITS_CUBE                               P0
          11 COMPLETED  COMMIT               UNITS_CUBE                               P0
          11 COMPLETED  BUILD SLAVE          BUILD GLOBAL.UNITS_CUBE USING (CLEAR VAL P0
                                             UES, SOLVE )
 
             COMPLETED  BUILD                BUILD GLOBAL.UNITS_CUBE USING (CLEAR VAL
                                             UES, LOAD, SOLVE)
 
156 rows selected.

Oracle Database allocates the specified number of processes regardless of whether all of them can be used simultaneously at any point in the job. For example, if your job can use up to three processes, but you specify five, then two of the processes allocated to your job cannot be used by it or by any other job.

If Oracle Database is installed with Real Application Clusters (RAC), then a script submitted to the job queue will be distributed across all nodes in the cluster. The performance gains can be significant. For example, a job running on four nodes in a cluster may run up to four times faster than the same job running on a single computer.

Analyzing Cubes and Dimensions

If your application executes queries directly against a single cube, you do not need to generate optimizer statistics for the cube. These queries are automatically optimized within the analytic workspace.

Optimizer statistics are used to create execution plans for queries that join two cube views or join a cube view to a table or a view of a table. They are also used for cost-based rewrite to cube materialized views. You need to generate the statistics only for these types of queries.

To generate optimizer statistics, use the DBMS_AW_STATS PL/SQL package. You can run this package in Analytic Workspace Manager as part of a cube script, in SQL*Plus, or in any other SQL interface. Generating the statistics does not have a significant performance cost.

DBMS_AW_STATS has the following syntax:

DBMS_AW_STATS.ANALYZE
     (object       IN VARCHAR2);

The argument can be either a cube or a dimension. Example 7-1 shows a sample script for generating statistics on the Units cube and its dimensions.

Example 7-1 Generating Statistics for the Units Cube

BEGIN
     DBMS_AW_STATS.ANALYZE('units_cube');
     DBMS_AW_STATS.ANALYZE('time');
     DBMS_AW_STATS.ANALYZE('customer');
     DBMS_AW_STATS.ANALYZE('product');
     DBMS_AW_STATS.ANALYZE('channel');
END;
/

Although you cannot view the statistics directly, you can examine the execution plans, as described in "Viewing Execution Plans".

Monitoring Analytic Workspaces

Oracle Database provides various tools to help you diagnose performance problems. As an Oracle DBA, you will find these tools useful in tuning the database:

Oracle Database also provides system views to help you diagnose performance problems. The following topics identify views that are either specific to OLAP or provide database information that is pertinent to OLAP.

Dynamic Performance Views

Each Oracle Database instance maintains fixed tables that record current database activity. These tables collect data on internal disk structures and memory structures. Among them are tables that collect data on Oracle OLAP.

These tables are available to users through a set of dynamic performance views. By monitoring these views, you can detect usage trends and diagnose system bottlenecks. Table 7-4 provides a brief description of each view. Global dynamic performance views (GV$) are also provided.

See Also:

Oracle Database Reference for full descriptions of the OLAP dynamic performance views.

Table 7-4 OLAP Dynamic Performance Views

View Description

V$AW_AGGREGATE_OP

Lists the aggregation operators available in analytic workspaces.

V$AW_ALLOCATE_OP

Lists the allocation operators available in analytic workspaces.

V$AW_CALC

Collects information about the use of cache space and the status of dynamic aggregation.

V$AW_LONGOPS

Collects status information about SQL fetches.

V$AW_SESSION_INFO

Collects information about each active session.

V$AW_OLAP

Collects information about the status of active analytic workspaces.


Table 7-5 describes some other dynamic performance views that are not specific to OLAP, but which you may want to use when tuning your database for OLAP.

Table 7-5 Selected Database Performance Views

View Description

V$LOG

Displays log file information from the control file.

V$LOGFILE

Contains information about redo log files.

V$PGASTAT

Provides PGA memory usage statistics as well as statistics about the automatic PGA memory manager when PGA_AGGREGATE_TARGET is set.

V$ROWCACHE

Displays statistics for data dictionary activity. Each row contains statistics for one data dictionary cache.

V$SYSSTAT

Lists system statistics.


Basic Queries for Monitoring the OLAP Option

The following queries extract OLAP information from the data dictionary.

More complex queries are provided in a script that you can download from the Oracle OLAP web site on the Oracle Technology Network. For descriptions of these scripts and download instructions, refer to "OLAP DBA Scripts".

Is the OLAP Option Installed in the Database?

The OLAP option is provided with Oracle Database Enterprise Edition. To verify that the OLAP components have been installed, issue this SQL command:

SQL> SELECT comp_name, version, status FROM DBA_REGISTRY 
     WHERE comp_name LIKE '%OLAP%';

COMP_NAME                VERSION                        STATUS     
------------------------ ------------------------------ -----------
OLAP Analytic Workspace  11.1.0.6.0                     VALID      
Oracle OLAP API          11.1.0.6.0                     VALID      
OLAP Catalog             11.1.0.6.0                     VALID      

What Analytic Workspaces are in the Database?

The DBA_AWS view provides information about all analytic workspaces. Use the following SQL command to get a list of names, their owners, and the version:

SELECT owner, aw_name, aw_version FROM DBA_AWS;
 
OWNER      AW_NAME                        AW_VERSION
---------- ------------------------------ ----------
SYS        EXPRESS                        11.1
GLOBAL     GLOBAL                         11.1
SYS        AWCREATE                       11.1
SH         SH                             11.1
SYS        AWMD                           11.1
SYS        AWXML                          11.1
SYS        AWREPORT                       11.1
SYS        AWCREATE10G                    11.1

See Also:

"System Tables" for descriptions of the analytic workspaces owned by SYS.

How Big is the Analytic Workspace?

To find out the size in bytes of the tablespace extents for a particular analytic workspace, use the following SQL statements, replacing GLOBAL with the name of your analytic workspace.

SQL> SELECT extnum, SUM(dbms_lob.getlength(awlob)) bytes FROM aw$global 
     GROUP BY extnum;
 
    EXTNUM      BYTES
---------- ----------
         0  191776956

To see the size of the LOB table containing an analytic workspace, use a SQL command like the following, replacing GLOBAL.AW$GLOBAL with the qualified name of your analytic workspace.

SQL> SELECT ROUND(SUM(dbms_lob.getlength(awlob))/1024,0) kb 
     FROM global.aw$global;
 
        KB
----------
    187282

When Were the Analytic Workspaces Created?

The DBA_OBJECTS view provides the creation date of the objects in your database. The following SQL command generates an easily readable report for analytic workspaces.

SQL> SELECT owner, object_name, created, status FROM dba_objects 
      WHERE object_name LIKE 'AW$%' AND object_name!='AW$' 
      GROUP BY owner, object_name, created, status 
      ORDER BY owner, object_name;
 
OWNER      OBJECT_NAME     CREATED   STATUS
---------- --------------- --------- -------
GLOBAL     AW$GLOBAL       05-JUL-07 VALID
SYS        AW$AWCREATE     30-JUN-07 VALID
SYS        AW$AWCREATE10G  30-JUN-07 VALID
SYS        AW$AWMD         30-JUN-07 VALID
SYS        AW$AWREPORT     30-JUN-07 VALID
SYS        AW$AWXML        30-JUN-07 VALID
SYS        AW$EXPRESS      30-JUN-07 VALID
 
7 rows selected.

OLAP DBA Scripts

You can download a file that contains several SQL scripts from the Oracle OLAP web site on the Oracle Technology Network. These scripts typically extract information from two or more system views and generate a report that may be useful in monitoring and tuning a database. To download the file, use this URL:

http://www.oracle.com/technology/products/bi/olap/olap_dba_scripts.zip

Table 7-6 describes these scripts. For more information, refer to the README file provided with the scripts.

Table 7-6 OLAP DBA Scripts

SQL Script Description

aw_objects_in_cache

Identifies the objects in the buffer cache that are related to analytic workspaces.

aw_reads_writes

Tallies the reads from temporary and permanent tablespaces, the writes to cache, and the rows processed in analytic workspaces.

aw_segment_size

Calculates the size of analytic workspace segments in tablespaces on disk.

aw_size

Displays the amount of disk space used by each analytic workspace.

aw_tablespaces

Provides extensive information about the tablespaces used by analytic workspaces.

aw_total_size

Tallies the sizes of all analytic workspaces accessible to the current user.

aw_users

Identifies the users of analytic workspaces.

aw_wait_events

Describes the wait events experienced by users of analytic workspaces over the previous hour.

buffer_cache_hits

Calculates the buffer cache hit ratio.

cursor_parameters

Indicates whether the database parameters that limit the number of open cursors are set too low.

olap_hit_ratio

Identifies the PGA, OLAP page pool, and OLAP hit/miss ratio for every user of analytic workspaces in the database.

olap_pga_performance

Determines how much PGA is in use, the size of the OLAP page pool, and the hit/miss ratio for OLAP pages for each user.

olap_pga_use

Determines how much PGA is consumed by the OLAP page pool to perform operations on analytic workspaces.

session_resources

Identifies the use of cursors, PGA, and UGA for each open session.

shared_pool_hits

Calculates the shared pool hit ratio.


Scripts for Monitoring Performance

Several of the scripts listed in "OLAP DBA Scripts" provide detailed information about the use of memory and other database resources by OLAP sessions. You can use these scripts as is, or you can use them as the starting point for developing your own scripts.

Example 7-2 shows the information returned by the session_resources script. It lists the use of resources such as cursors, PGA, and UGA.

Example 7-2 Querying Session Resources

SQL> @session_resources
 
USERNAME             NAME                                VALUE
-------------------- ------------------------------ ----------
GLOBAL:95            opened cursors cumulative             101
                     opened cursors current                  3
                     session cursor cache count             31
                     session cursor cache hits              68
                     session pga memory                1219292
                     session pga memory max            1219292
                     session stored procedure space          0
                     session uga memory                 432700
                     session uga memory max             432700
 
 
9 rows selected.

Monitoring Disk Space

Several of the scripts listed in "OLAP DBA Scripts" provide detailed information about the use of disk space by analytic workspaces. Example 7-3 shows the information returned by the aw_size script. It lists all of the analytic workspaces in the database, the disk space they consume, and the tablespaces in which they are stored.

Example 7-3 Querying the Use of Disk Space By Analytic Workspaces

SQL> @aw_size
 
Analytic Workspace                            On Disk MB Tablespace
---------------------------------------- --------------- --------------------
GLOBAL.GLOBAL                                     239.38 GLOBAL
SYS.AWCREATE                                        9.81 SYSAUX
SYS.AWCREATE10G                                     1.38 SYSAUX
SYS.AWMD                                            7.00 SYSAUX
SYS.AWREPORT                                        1.50 SYSAUX
SYS.AWXML                                          12.00 SYSAUX
SYS.EXPRESS                                         2.69 SYSAUX
                                         ---------------
Total Disk:                                       273.75
 
7 rows selected.

Backup and Recovery

You can backup and recover analytic workspaces using the same tools and procedures as the rest of your database.

Oracle Recovery Manager (RMAN) is a powerful tool that simplifies, automates, and improves the performance of backup and recovery operations. RMAN enables one time backup configuration, automatic management of backups, and archived logs based on a user-specified recovery window, restartable backups and restores, and test restore/recovery.

RMAN implements a recovery window to control when backups expire. This lets you establish a period of time during which it is possible to discover logical errors and fix the affected objects by doing a database or tablespace point-in-time recovery. RMAN also automatically expires backups that are no longer required to restore the database to a point-in-time within the recovery window. Control file auto backup also allows for restoring or recovering a database, even when an RMAN repository is not available.

Export and Import

You can copy analytic workspaces in several different ways, either to replicate them on another computer or to back them up.

The owner of an analytic workspace can create an XML template, or export the schema to a dump file. Only users with the EXP_FULL_DATABASE privilege or a privileged user (such as SYS or a user with the DBA role) can export the full database or create a transportable tablespace.

See Also:

Cube Materialized Views

A cube materialized view is an Oracle OLAP cube that has been enhanced with the capabilities of a materialized view at build time.

Acquiring Information From the Data Dictionary

The data dictionary contains numerous static views that provide information about materialized views. They list cube materialized views along with all other materialized views. The ALL_MVIEW_DETAIL_SUBPARTITION view is the only exception, with no information about cube materialized views.

See Also:

Oracle Database Reference for complete descriptions of the data dictionary views

Identifying Cube Materialized Views

USER_MVIEWS contains a row for each materialized view owned by the current user. The following query lists the materialized views owned by the GLOBAL user. The CB$ prefix identifies a cube materialized view.

SQL> SELECT mview_name, refresh_mode "MODE", refresh_method "METHOD", 
     last_refresh_date "DATE", staleness FROM user_mviews;
 
MVIEW_NAME                     MODE   METHOD   DATE      STALENESS
------------------------------ ------ -------- --------- -------------------
CB$TIME_CALENDAR               DEMAND FORCE    10-JUL-07 FRESH
CB$TIME_FISCAL                 DEMAND FORCE    10-JUL-07 FRESH
CB$PRODUCT_PRIMARY             DEMAND FORCE    10-JUL-07 FRESH
CB$CUSTOMER_SHIPMENTS          DEMAND FORCE    10-JUL-07 FRESH
CB$CUSTOMER_SEGMENT            DEMAND FORCE    10-JUL-07 FRESH
CB$CHANNEL_PRIMARY             DEMAND FORCE    10-JUL-07 FRESH
CB$UNITS_CUBE                  DEMAND COMPLETE 10-JUL-07 FRESH
 
7 rows selected.

The example shows the cube materialized views defined by Analytic Workspace Manager: One for each dimension hierarchy and one for each cube.

Identifying the Refresh Logs

Oracle Database can maintain a set of logs on the master tables for the cube materialized views. These logs support incremental (fast) refresh of the cube. The script generated by the Materialized View Advisor creates a log for each fact and dimension table to record any changes to the data. The following query lists the materialized view logs owned by the GLOBAL user:

SQL> SELECT master, log_table FROM user_mview_logs;
 
MASTER                         LOG_TABLE
------------------------------ ------------------------------
CHANNEL_DIM                    MLOG$_CHANNEL_DIM
CUSTOMER_DIM                   MLOG$_CUSTOMER_DIM
PRODUCT_DIM                    MLOG$_PRODUCT_DIM
TIME_DIM                       MLOG$_TIME_DIM
UNITS_FACT                     MLOG$_UNITS_FACT

Initiating a Data Refresh

You can initiate a data refresh of a cube materialized view in several different ways using Analytic Workspace Manager or a PL/SQL package:

  • Automatic Refresh: On the Materialized View tab for a cube, you can create a regular schedule for the materialized view refresh subsystem, as described in "Adding Materialized View Capability to a Cube".

  • Maintenance Wizard: The Maintenance Wizard is available for refreshing all cubes and dimensions, including cube materialized views.

  • DBMS_CUBE: The DBMS_CUBE PL/SQL package is available for refreshing all cubes and dimensions.

  • DBMS_MVIEW: The DBMS_MVIEW PL/SQL package contains several procedures for use with cube materialized views.

Using DBMS_CUBE

DBMS_CUBE can be used to create and populate an analytic workspace. You can use it to maintain any cube, including cube materialized views.

The following command initiates a complete refresh of UNITS_CUBE, which is enabled as a cube materialized view. It automatically refreshes any stale dimensions before refreshing the cube.

SQL> EXECUTE dbms_cube.build('GLOBAL.UNITS_CUBE');
 
PL/SQL procedure successfully completed.

You can determine the refresh method from USER_MVIEWS, as shown in "Identifying Cube Materialized Views".

Using DBMS_MVIEW

DBMS_MVIEW can be used to refresh all types of materialized views. These refresh procedures can be used with cube materialized views:

  • REFRESH refreshes a list of one or more materialized views.

  • REFRESH_ALL_MVIEWS refreshes all materialized views that meet certain criteria.

  • REFRESH_DEPENDENT refreshes all materialized views that depend on a particular master table and meet certain criteria.

Dimensions must be refreshed before the cube. An error is raised during refresh of a cube materialized view if any of its associated dimension materialized views are stale. The procedures in DBMS_MVIEW can refresh multiple materialized views in one call, but they do not guarantee the appropriate refresh order. You must be sure to list all the dimension materialized views before the cube materialized views when using this package.

The following command initiates a refresh of the materialized view for the CHANNEL_PRIMARY hierarchy. Only the Complete refresh type is valid for dimensions.

SQL> EXECUTE dbms_mview.refresh('CB$CHANNEL_PRIMARY', 'C');
 
PL/SQL procedure successfully completed.

Refresh Methods

In Analytic Workspace Manager, you can specify the COMPLETE, FAST, or FORCE methods for refreshing a cube. Two additional methods, FAST_PCT and FAST_SOLVE, are invoked by the materialized view subsystem. They are not separate choices.

Refresh Method Descriptions

Table 7-7 describes the refresh methods that are supported on cube materialized views.

Table 7-7 Refresh Methods For Cube Materialized Views

Refresh Method Description

COMPLETE

Deletes and recreates the cube.

This option supports arbitrarily complex mappings from the source tables to the cube.

FAST

Loads and re-aggregates only changed values, based on the materialized view logs.

The source for the refresh is the incremental differences that have been captured in the materialized view logs, rather than the original mapped sources. These differences are used to incrementally rebuild the cube. Only cells that are affected by the changed values are re-aggregated.

This option supports only simple mappings for cube materialized views, that is, where no expressions (other than table.column), views, or aggregations occur in the query defining the mapping.

The materialized view subsystem determines whether to perform a FAST or a FAST_PCT refresh.

FAST_PCT

Loads and re-aggregates data only from changed partitions.

The Partition Change Tracking method is not available for cube materialized views enabled for query rewrite. This method works best when the source table and the cube are partitioned on the same dimension.

FAST_PCT does not use change logs. It is always available, and it is always faster than a COMPLETE refresh. The materialized view subsystem determines whether to perform a FAST or a FAST_PCT refresh.

FAST_SOLVE

Loads and re-aggregates only changed values, based on the original mapped data source.

FAST_SOLVE is a new type of refresh only for cube materialized views. It incrementally re-aggregates the cube even when the refresh source is the original mapped source instead of the materialized view logs. The aggregation subsystem identifies the differences and then incrementally re-aggregates the cube.

This option is supported for arbitrarily complex mappings from the source tables to the cube. To discover whether a FAST_SOLVE refresh has occurred, review the CUBE_BUILD_LOG table as shown in "Fast Solve Refreshes".

FORCE

Loads and re-aggregates only changed values, using the best method possible.

The materialized view system first attempts a FAST refresh. If it cannot FAST refresh a cube materialized view, it performs a FAST_SOLVE refresh.


Fast Solve Refreshes

The build log lists the CLEAR LEAVES command when the FAST SOLVE method was used. Example 7-4 shows the rows of CUBE_BUILD_LOG concerned with building UNITS_CUBE.

See Also:

"Build Logs"

Example 7-4 Identifying a FAST SOLVE Refresh

SQL> SELECT build_object, status, command FROM cube_build_log
     WHERE build_object='UNITS_CUBE'
     AND build_id=8;
 
BUILD_OBJECT                   STATUS     COMMAND
------------------------------ ---------- --------------------
UNITS_CUBE                     STARTED    CLEAR LEAVES
UNITS_CUBE                     COMPLETED  CLEAR LEAVES
UNITS_CUBE                     STARTED    LOAD
UNITS_CUBE                     COMPLETED  LOAD
UNITS_CUBE                     STARTED    SOLVE
UNITS_CUBE                     COMPLETED  SOLVE
UNITS_CUBE                     STARTED    UPDATE
UNITS_CUBE                     COMPLETED  UPDATE
UNITS_CUBE                     STARTED    COMMIT
UNITS_CUBE                     COMPLETED  COMMIT
 
10 rows selected.

Using Query Rewrite

Query rewrite changes a query to select data from the materialized views instead of calculating the result set from the master tables. The transformation is fully transparent to the client, and requires no mention of the materialized views in the SQL statement. In the case of cube materialized views, the query is written against the tables or views of a star or snowflake schema, and it is transformed into a query against a cube materialized view. This transformation can result in significant improvements in runtime performance.

Query rewrite requires optimizer statistics on the cubes and dimensions. You can discover whether a query will be rewritten by generating and examining its execution plan.

Oracle Database uses two initialization parameters to control query rewrite:

  • QUERY_REWRITE_ENABLED: Enables or disables query rewrite globally for the database.

  • QUERY_REWRITE_INTEGRITY: Determines the degree to which query rewrite monitors the consistency of materialized views with the source data. The trusted or stale tolerated settings are recommended when using rewrite to cube materialized views.

See Also:

Acquiring Additional Information About Cube Materialized Views

Oracle Database has numerous PL/SQL packages for managing materialized views. Cube materialized views are already optimized to provide the best performance, so you have no need to use most of these packages. Few design decisions remain for you to make. For this reason, the TUNE_MVIEW procedure of DBMS_ADVISOR is disabled for cube materialized views.

However, there are a few packages that you may find useful, as shown in Table 7-8.

Table 7-8 PL/SQL Packages for Cube Materialized Views

Package Description

DBMS_METADATA

Returns the metadata for an object.

DBMS_MVIEW

Executes data refreshes. See "Initiating a Data Refresh".

You can use the EXPLAIN_REWRITE and EXPLAIN_MVIEW procedures to obtain information about cube materialized views.

DBMS_XPLAN

Displays an execution plan. See "Viewing Execution Plans".