Oracle Tables

for DBAs and Developers

Daniel A. Morgan
email: [email protected]
mobile: +1 206-669-2949
skype: damorgan11g
Thursday: 18 May, 2017 twitter: @meta7solutions
1
Introduction

2
Unsafe Harbor
 This room is an unsafe harbor
 You can rely on the information in this presentation to help you protect your
data, your databases, your organization, and your career
 No one from Oracle has previewed this presentation
 No one from Oracle knows
what I'm going to say
 No one from Oracle has
supplied any of my materials
 Everything we will discuss is
existing, proven, functionality

3
Daniel Morgan
 Oracle ACE Director Alumni
 Oracle Educator
 Curriculum author and primary program instructor at University of Washington
 Consultant: Harvard University
 University Guest Lecturers
 APAC: University of Canterbury (NZ)
 EMEA: University of Oslo (Norway)
 Latin America: Universidad Cenfotec, Universidad Latina de Panama, Technologico de Costa Rica
 IT Professional
 First computer: IBM 360/40 in 1969: Fortran IV
 Oracle Database since 1988-9 and Oracle Beta tester
 The Morgan behind www.morganslibrary.org
 Member Oracle Data Integration Solutions Partner Advisory Council
 Vice President Twin Cities Oracle Users Group (Minneapolis-St. Paul)
 Co-Founder International GoldenGate Oracle Users Group System/370-145 system console

 Principal Adviser: Forsythe Meta7

email: [email protected]
Twitter: @damorgan12c 4
My Websites: Morgan's Library

www.morganslibrary.org`

5
Forsythe (1:2)
 In business 46 years
 $1.2B in 2016
 Partner with more
than 200 technology
OEMs
 Dan Wilson
Account Manager

6
Forsythe (2:2)
 In business 46 years
 $1.2B in 2016
 Partner with more
than 200 technology
OEMs
 Dan Wilson
Account Manager

 Focusing on solutions to business problems ... not products

7
What Meta7 Brings To The Party
 Oracle only division of Forsythe
 Platinum Partner
 Focuses on the entire Oracle technology stack
 The entire line of Oracle infrastructure from x86
through the full stack of engineered systems and storage
 Oracle Database
 Design and Deployment
 Stability
 Security
 Scalability
 Data Integration (GoldenGate and ODI)
 Oracle Cloud
 DevOps
 Infrastructure as Code
 Jason Peterson, Solutions Architect and Maggie Mellon, Account Manager
 Focusing on solutions to business problems ... not products

8
Stability: IT Fire Fighting

9
Oracle Stack Security

10
Scalability: VLDBs
and Partitioning

11
Database Performance

12
Zero Downtime Migration

13
Just In Time IT Procurement

14
Content Density Warning

Take Notes ... Ask Questions

15
Why Am I Focusing On Oracle Tables?
 Because no one else is
 Because Oracle University doesn't teach this material
 Because there are 119 pages in the 12cR2 docs under CREATE TABLE
 Because no one knows the full syntax for basic DDL statements
 Because we have now spent more than 30 years talking about performance
tuning and yet the number one conference and training topic remains tuning
which proves that we need to stop focusing on edge cases and focus, instead,
on the basics
 Because explain plans, AWR Reports, and trace files will never fix a problem if
you don't know the full range of options available
 Because the best way to achieve high performance is to choose techniques
that reduce resource utilization

16
How Well Do You Know CREATE TABLE?

17
Legacy Architecture Tables

18
Legacy Data Dictionary: DBA_, ALL_, USER_
VIEW_NAME VIEW_NAME VIEW_NAME
------------------------------ ------------------------------ ------------------------------
 DBA_ALL_TABLES  DBA_SUMMARY_DETAIL_TABLES  DBA_UPDATABLE_COLUMNS
 DBA_APPLY_TABLE_COLUMNS  DBA_SYNC_CAPTURE_PREPARED_TABS  DBA_WM_VERSIONED_TABLES
 DBA_BASE_TABLE_MVIEWS  DBA_SYNC_CAPTURE_TABLES  DBA_XML_NESTED_TABLES
 DBA_CAPTURE_PREPARED_TABLES  DBA_TABLES  DBA_XML_OUT_OF_LINE_TABLES
 DBA_CLUSTERING_TABLES  DBA_TAB_COLS  DBA_XML_TABLES
 DBA_EVALUATION_CONTEXT_TABLES  DBA_TAB_COLS_V$  DBA_XML_TAB_COLS
 DBA_EXTERNAL_TABLES  DBA_TAB_COLUMNS
 DBA_FILE_GROUP_TABLES  DBA_TAB_COL_STATISTICS
 DBA_FILE_GROUP_TABLESPACES  DBA_TAB_COMMENTS
 DBA_FLASHBACK_ARCHIVE_TABLES  DBA_TAB_HISTGRM_PENDING_STATS
 DBA_HIST_DATABASE_INSTANCE  DBA_TAB_HISTOGRAMS
 DBA_LBAC_TABLE_POLICIES  DBA_TAB_IDENTITY_COLS
 DBA_MINING_MODEL_TABLES  DBA_TAB_MODIFICATIONS
 DBA_NESTED_TABLES  DBA_TAB_PARTITIONS
 DBA_NESTED_TABLE_COLS  DBA_TAB_PENDING_STATS
 DBA_OBJECT_TABLES  DBA_TAB_PRIVS
 DBA_PARTIAL_DROP_TABS  DBA_TAB_STATISTICS
 DBA_PART_TABLES  DBA_TAB_STATS_HISTORY
 DBA_PENDING_CONV_TABLES  DBA_TAB_STAT_PREFS
 DBA_QUEUE_TABLES  DBA_TAB_SUBPARTITIONS
 DBA_SECUREFILE_LOG_TABLES  DBA_TSTZ_TABLES
 DBA_SOURCE_TABLES  DBA_TSTZ_TAB_COLS
 DBA_SUBSCRIBED_TABLES  DBA_UNUSED_COL_TABS

19
Heap Tables (1:2)
 Almost every Oracle table ever built by everyone except Oracle Corp. is a
heap table stored in a tablespaces ... a description of "vanilla" as good as any
 Heap tables are optimized for persistent storage of data when you have little
idea of what you are putting in and less of an idea of how that data, after
storage will be used
 There isn't one good reason for making every table a heap table: Not one

 By default all Oracle tables are heap tables

 A heap table is a single segment consisting of one or more extents: each extent consisting
of one or more blocks
 A heap table stores data as an unorganized pile of rows ... unordered
 With a heap table each "8K" block can contain information from only a single table

20
Heap Tables (2:2)
SQL> CREATE TABLE state (state_abbrev VARCHAR2(2));

Table created.

SQL> desc state

Name Null? Type
------------------------------ -------- ----------------
STATE_ABBREV VARCHAR2(2)

SQL> SELECT dbms_metadata.get_ddl('TABLE', 'STATE')

2 FROM dual;

DBMS_METADATA.GET_DDL('TABLE','STATE')
---------------------------------------------------------
CREATE TABLE "UWCLASS"."STATE"
( "STATE_ABBREV" VARCHAR2(2)
) SEGMENT CREATION DEFERRED
PCTFREE 10 PCTUSED 40 INITRANS 1 MAXTRANS 255
NOCOMPRESS LOGGING
TABLESPACE "UWDATA"

21
External Tables (1:3)
 An external tables is a physical file stored in a file system on disk
 Columns can be defined as fixed length or delimited
 External tables can be created by the database
 DML is not allowed on external tables
 SELECT statements interact with external tables precisely the same way they interact with
internal heap tables
 External tables are a replacement for the legacy SQL*Loader tool
SQL> conn sys@pdbdev as sysdba
Enter password:
Connected.

SQL> CREATE OR REPLACE DIRECTORY ext AS 'c:\external';

Directory created.

SQL>
SQL> GRANT read, write ON DIRECTORY ext TO uwclass;

Grant succeeded.

22
External Tables (2:3)
-- create a file named
7369,KYTE,SME,20
7499,MILLSAP,SALESMAN,30
7521,NORGAARD,SALESMAN,30
7566,KOLK,MANAGER,20
7654,LEWIS,ANALYST,30

-- create a file named

1111,MORGAN,DIRECTOR,10
2222,HARDIE,MANAGER,30
3333,HAVEMEYER,CTO,10
4444,LOFSTROM,DEVELOPER,10
5555,TOWNSEND,MANAGER,30

-- create a file on the hard disk named cost.txt with the follow 4 lines:
YEAR PID CPU GROSS REVENUE
2003 def 2.00 123.4567890
2004 ABC 1.00 39.7288841651344
2005 xyz 1.99 1107.5458517352

A comma delimited flat file is export from a UNIX database and transferred for loading into a Windows based Oracle database. The lines are terminated with 0x0A
and not the 0x0A 0x0D pair for DOS/Windows. The fields are randomly double quoted, and some contain commas internal to the field's data like the following:

100001,"7123 HIGHLAND DR","03/28/1965"," Mercer Island, WA 98040","Morgan, Dan A", "","","63034630752",

14,1,"F","T","06/28/2009","",0,"","01/01/2010","N", "01/01/1999",""
100020,"5432 SOUTH 28TH ST","01/01/1951"," Mercer Island, WA 98040", "Burger,Tom",
"","2566400","ZPW345070938",64,1,"M","B","02/23/2000","",0,"", "12/31/1799","P","12/31/1979",""
-- save external file as c:\external\preprocess.dat.gz (if UNIX or LINUX use your home directory)

23
External Tables (3:3)
 Note the key words "ORGANIZATION EXTERNAL"
conn uwclass/uwclass@pdbdev

CREATE TABLE ext_tab1 (

CREATE TABLE <table_name> (
empno CHAR(4),
<column_definitions>)
ename CHAR(20),
ORGANIZATION EXTERNAL
job CHAR(20),
(TYPE oracle_loader
deptno CHAR(3))
DEFAULT DIRECTORY <oracle_directory_object_name>
ORGANIZATION EXTERNAL (
ACCESS PARAMETERS (
TYPE oracle_loader
RECORDS DELIMITED BY newline
DEFAULT DIRECTORY ext
BADFILE <file_name>
ACCESS PARAMETERS (
DISCARDFILE <file_name>
RECORDS DELIMITED BY NEWLINE
LOGFILE <file_name>
BADFILE ext:'bad_%a_%p.bad'
[READSIZE <bytes>]
LOGFILE ext:'log_%a_%p.log'
[SKIP <number_of_rows>]
FIELDS TERMINATED BY ','
FIELDS TERMINATED BY '<terminator>'
OPTIONALLY ENCLOSED BY '"'
OPTIONALLY ENCLOSED BY '<character>'
MISSING FIELD VALUES ARE NULL
MISSING FIELD VALUES ARE <VALUE | NULL>
REJECT ROWS WITH ALL NULL FIELDS
REJECT ROWS WITH ALL NULL FIELDS
(empno, ename, job, deptno))
(<column_name_list>))\
LOCATION ('demo1.dat'))
LOCATION ('<file_name>'))
PARALLEL
[PARALLEL]
REJECT LIMIT 0
REJECT LIMIT <UNLIMITED | integer>
NOMONITORING;
MONITORING | NOMONITORING];
SELECT * FROM ext_tab;

24
Global Temporary Tables (GTT) (1:2)
 One of the most expensive things you can do in an Oracle database is "DROP
TABLE"
 Global temporary tables address this issue by creating a permanent table
where the data is temporary
 GTTs can have
 Constraints
 Indexes
 Optimizer Statistics
 Synonyms
 Views
 GTTs are built in the temporary tablespace so the same METADATA is shared
by all users while the data remains private

25
Global Temporary Tables (GTT) (2:2)
 There are two types of GTTs
 ON COMMIT DELETE ROWS
 Data is deleted when a COMMIT is issued
 Use these primarily to preserve a temporary working data set or an intermediate result set
 ON COMMIT PRESERVE ROWS
 Data is deleted when the session ends
 Use these primarily to preserve a working data set across multiple transactions or as a result
table for a report writing/business intelligence tool
CREATE GLOBAL TEMPORARY TABLE gtt_ocdr (
zip_code VARCHAR2(5),
by_user VARCHAR2(30),
entry_date DATE)
ON COMMIT DELETE ROWS;

CREATE GLOBAL TEMPORARY TABLE gtt_ocpr (

zip_code VARCHAR2(5),
by_user VARCHAR2(30),
entry_date DATE)
ON COMMIT PRESERVE ROWS;

26
Index Organized Tables (IOT) (1:2)
 IOTs are wholly underutilized by developers and DBAs
 An IOT is a table stored as a B*Tree index rather than as an unorganized
heap of data
 Use IOTs where the size of the primary key index is a substantial fraction, or
larger, than the size of the table
 Use IOTs when some of the columns are accessed frequently and others may
be rarely or never accessed
 Look up tables, the children of foreign key referential constraints are often
perfect candidates for storage as an IOT

27
Index Organized Tables (IOT) (2:2)
CREATE TABLE labor_hour (
WORK_DATE DATE,
EMPLOYEE_NO VARCHAR2(8),
CONSTRAINT pk_labor_hour
PRIMARY KEY (work_date, employee_no))
ORGANIZATION INDEX;

CREATE TABLE compressed_iot (owner, object_type, object_name,

CONSTRAINT pk_compressed_iot
PRIMARY KEY(owner, object_type, object_name))
ORGANIZATION INDEX
COMPRESS 2 AS
SELECT owner, object_type, object_name
FROM all_objects;

CREATE TABLE labor_hour (

WORK_DATE DATE,
EMPLOYEE_NO VARCHAR2(8),
SUMMIT_WORK_ORDER_NO VARCHAR2(7),
DASH VARCHAR2(2),
CLASS_CODE VARCHAR2(6),
PAYCODE VARCHAR2(2),
ASSIGNED_CREW_NUMBER VARCHAR2(5),
TRANSFER_CREW_NUMBER VARCHAR2(5),
REFERENCE_TYPE VARCHAR2(1),
REFERENCE_NUMBER VARCHAR2(10),
OVERTIME_CODE VARCHAR2(1),
SHIFT_DIFFERENTIAL VARCHAR2(1) NOT NULL,
HOURS NUMBER(4,2) NOT NULL,
MOD_USER_ID VARCHAR2(30) DEFAULT USER,
MOD_USER_DATE DATE DEFAULT SYSDATE,
CONSTRAINT pk_labor_hour
PRIMARY KEY (work_date, employee_no, summit_work_order_no, dash, class_code, paycode, assigned_crew_number,
transfer_crew_number, reference_type, reference_number, overtime_code, shift_differential))
ORGANIZATION INDEX
INCLUDING hours
PCTTHRESHOLD 10
OVERFLOW TABLESPACE uwdata;

28
Partitioned Tables (1:7)
 Tables are a logical, not a physical, construct
 Partitions and Subpartition are maps
 Just as a tablespace maps many datafile
 Just as a regular heap table maps many blocks
 Partitioned tables map segments so that they appear to be a single object
 Partitions may optionally map many subpartitions for increased granularity
 In a partitioned table the segment is no longer the table it is the segment

29
Partitioned Tables (2:7)
 Multiple types of partitioning
 HASH (a database internally generated hashing algorithm) - least valuable and overused
 LIST (for example North, South, East West)
 RANGE (for example 2015, 2016, 2017, 2018) - most valuable and most commonly used
 Multiple types of RANGE partitioning
 Basic
 Interval (date or numeric)
 Reference
 SYSTEM
 Data stored in partitions can be
 Encrypted to protect sensitive information (for example credit cards)
 Compressed to save on storage costs (for example older data)
 Relocated to slower, less expensive, disk when it is less frequently accessed
 Made READ ONLY so that it cannot be accidentally changed
 Easily exported to flat files, spreadsheets, and other databases
 Exported into spreadsheets with lessened system impact

30
Partitioned Tables (3:7)
 System Partitioning
 Partitioned tables created by the
Oracle Database and owned by
one of the database's built-in
schemas
 User Partitioning
 Partitioned tables created by users
or applications installed by the
user
 Only available for Enterprise
Edition

31
Partitioned Tables (4:7)
 Segments exist within a tablespace
 Segments are a collection of extents
 Extents are a collection of data blocks
 Data blocks are mapped to disk blocks

Basic Heap Table

Object = Segment Extents DB Blocks O/S Blocks

32
Partitioned Tables (5:7)

Basic Heap Table

Segment=Object Extents DB Blocks O/S Blocks

Partitioned Table

Object <> Segment Extents DB Blocks O/S Blocks

33
Partitioned Tables (6:7)
 Speed with Very Large Databases
 Partition Pruning (Developer)

Oracle optimizes SQL statements to mark the partitions or subpartitions that need to be
accessed and eliminates (prunes) unnecessary partitions or subpartitions from access.
Partition pruning is the skipping of unnecessary index and data partitions or subpartitions
by a query

 Partition Elimination (DBA)

 Ease of Management
 The following built-in objects support partition tables
 DATAOBJ_TO_PARTITION function
 DBMS_PART package
 DBMS_PCLXUTIL package
 Partitions can be based on normal, hidden, and virtual columns
 It is always preferable, whenever possible, to use local indexes

34
Partitioned Tables (7:7)
 The following are examples of special use cases for table partitioning
CREATE TABLE orders OF XMLType
XMLTYPE STORE AS BINARY XML
VIRTUAL COLUMNS (site_id AS (XMLCast(XMLQuery('/Order/@SiteId' PASSING OBJECT_VALUE RETURNING CONTENT) AS NUMBER)))
PARTITION BY RANGE (site_id) (
PARTITION p1 VALUES LESS THAN (10),
PARTITION p2 VALUES LESS THAN (20),
PARTITION pm VALUES LESS THAN (MAXVALUE));

CREATE TABLE json_orders (

tx_id NUMBER(5),
tx_date DATE,
jsondata VARCHAR2(4000),
site_id AS (JSON_VALUE(jsondata, '$.siteId' RETURNING NUMBER)))
PARTITION BY RANGE (site_id) (
PARTITION p1 VALUES LESS THAN (10),
PARTITION p2 VALUES LESS THAN (20),
PARTITION pm VALUES LESS THAN (MAXVALUE));

CREATE TABLE ref_child (

table_name VARCHAR2(30) NOT NULL,
index_name VARCHAR2(30) NOT NULL,
CONSTRAINT fk_ref_child_parent
FOREIGN KEY(table_name) REFERENCES ref_parent(table_name))
PARTITION BY REFERENCE(fk_ref_child_parent);

35
XML Tables
 Use the optimized storage of XML tables when persisting storage of XML
documents

CREATE TABLE xml_lob_tab OF XMLTYPE;

CREATE TABLE uwclass$schema OF SYS.XMLTYPE

XMLSCHEMA "http://xmlns.oracle.com/xdb/XDBSchema.xsd"
ELEMENT "schema" ID 81
TABLESPACE uwdata;

CREATE TABLE orders OF XMLType

XMLTYPE STORE AS BINARY XML;

36
Object-Relational Tables

37
Object/Nested Tables with a NESTED TABLE
 The best advice I can provide you is don't build these ... you will sacrifice data
integrity, space, and performance
 That said ... here's how you do it
CREATE OR REPLACE NONEDITIONABLE TYPE CourseList AS TABLE OF VARCHAR2(64);
/

CREATE TABLE department (

name VARCHAR2(20),
director VARCHAR2(20),
office VARCHAR2(20),
courses CourseList)
NESTED TABLE courses STORE AS courses_tab;

INSERT INTO department

(name, director, office, courses)
VALUES
('English', 'Lynn Saunders', 'Breakstone Hall 205', CourseList(
'Expository Writing',
'Film and Literature',
'Modern Science Fiction',
'Discursive Writing',
'Modern English Grammar',
'Introduction to Shakespeare',
'Modern Drama',
'The Short Story',
'The American Novel'));

38
Object/Nested Table Based with a VARRAY
 The best advice I can provide you is don't build these ... you will sacrifice data
integrity, space, and performance
 That said ... here's how you do it
CREATE OR REPLACE TYPE Project AUTHID DEFINER AS OBJECT (
project_no NUMBER(2),
title VARCHAR2(35),
cost NUMBER(7,2));
/

CREATE OR REPLACE TYPE ProjectList AS VARRAY(50) OF Project;

/

CREATE TABLE department (

dept_id NUMBER(2),
dname VARCHAR2(15),
budget NUMBER(11,2),
projects ProjectList);

INSERT INTO department

(dept_id, dname, budget, projects)
VALUES
(42, 'Quantum Mechanics', '$424242.42',
ProjectList(Project(1, 'Standard Model', 2121'),
Project(2, 'Uncertainty Principal', 21.21)));

39
12c Container Architecture Tables

40
Common Tables (1:2)
 As of Oracle Database version 12.2 tables in an Application Root container
can be optionally created with shared metadata or shared data
 When metadata, the DDL, is shared into PDBs as pointers from the
Application Root data dictionary

SQL> ALTER PLUGGABLE DATABASE APPLICATION uw_app

2 BEGIN UPGRADE '1.0' TO '2.0'
3 COMMENT 'Adding New Table With Sharing';

SQL> CREATE TABLE serv_inst

2 SHARING=METADATA (
3 siid NUMBER(10),
4 si_status VARCHAR2(15),
5 type VARCHAR2(5),
6 installstatus VARCHAR2(1),
7 location_code NUMBER(10),
8 custacct_id VARCHAR2(10),
9 srvr_id NUMBER(10),
10* ws_id NUMBER(10));

Table created.

SQL> ALTER PLUGGABLE DATABASE APPLICATION uw_app END UPGRADE;

41
Common Tables (2:2)
 As of Oracle Database version 12.2 tables in an Application Root container
can be optionally created with shared metadata or shared data
 When data is shared the DDL and the table rows are shared into PDBs under
the application root

SQL> ALTER PLUGGABLE DATABASE APPLICATION uw_app

2 BEGIN UPGRADE '1.0' TO '2.0'
3 COMMENT 'Adding New Table With Sharing';

SQL> CREATE TABLE serv_inst

2 SHARING=DATA (
3 siid NUMBER(10),
4 si_status VARCHAR2(15),
5 type VARCHAR2(5),
6 installstatus VARCHAR2(1),
7 location_code NUMBER(10),
8 custacct_id VARCHAR2(10),
9 srvr_id NUMBER(10),
10* ws_id NUMBER(10));

Table created.

SQL> ALTER PLUGGABLE DATABASE APPLICATION uw_app END UPGRADE;

42
New 12cR2 Container Database Architecture

Seed Dev 1
PDB$SEED PDB

Application PDB PDB PDB PDBDEV PDBQA PDBUAT

Seed Customer1 Customer2 Customer3

43
Table Options

44
Tables in the 12.2 Data Dictionary (1:2)
 By default all Oracle tables are heap tables
SQL> desc dba_tables SAMPLE_SIZE NUMBER
Name Null? Type LAST_ANALYZED DATE
--------------------------- -------- --------------- PARTITIONED VARCHAR2(3)
OWNER NOT NULL VARCHAR2(128) IOT_TYPE VARCHAR2(12)
TABLE_NAME NOT NULL VARCHAR2(128) TEMPORARY VARCHAR2(1)
TABLESPACE_NAME VARCHAR2(30) SECONDARY VARCHAR2(1)
CLUSTER_NAME VARCHAR2(128) NESTED VARCHAR2(3)
IOT_NAME VARCHAR2(128) BUFFER_POOL VARCHAR2(7)
STATUS VARCHAR2(8) FLASH_CACHE VARCHAR2(7)
PCT_FREE NUMBER CELL_FLASH_CACHE VARCHAR2(7)
PCT_USED NUMBER ROW_MOVEMENT VARCHAR2(8)
INI_TRANS NUMBER GLOBAL_STATS VARCHAR2(3)
MAX_TRANS NUMBER USER_STATS VARCHAR2(3)
INITIAL_EXTENT NUMBER DURATION VARCHAR2(15)
NEXT_EXTENT NUMBER SKIP_CORRUPT VARCHAR2(8)
MIN_EXTENTS NUMBER MONITORING VARCHAR2(3)
MAX_EXTENTS NUMBER CLUSTER_OWNER VARCHAR2(128)
PCT_INCREASE NUMBER DEPENDENCIES VARCHAR2(8)
FREELISTS NUMBER COMPRESSION VARCHAR2(8)
FREELIST_GROUPS NUMBER COMPRESS_FOR VARCHAR2(30)
LOGGING VARCHAR2(3) DROPPED VARCHAR2(3)
BACKED_UP VARCHAR2(1) READ_ONLY VARCHAR2(3)
NUM_ROWS NUMBER SEGMENT_CREATED VARCHAR2(3)
BLOCKS NUMBER RESULT_CACHE VARCHAR2(7)
EMPTY_BLOCKS NUMBER CLUSTERING VARCHAR2(3)
AVG_SPACE NUMBER ACTIVITY_TRACKING VARCHAR2(23)
CHAIN_CNT NUMBER DML_TIMESTAMP VARCHAR2(25)
AVG_ROW_LEN NUMBER HAS_IDENTITY VARCHAR2(3)
AVG_SPACE_FREELIST_BLOCKS NUMBER CONTAINER_DATA VARCHAR2(3)
NUM_FREELIST_BLOCKS NUMBER INMEMORY VARCHAR2(8)
DEGREE VARCHAR2(10) INMEMORY_PRIORITY VARCHAR2(8)
INSTANCES VARCHAR2(10) INMEMORY_DISTRIBUTE VARCHAR2(15)
CACHE VARCHAR2(5) INMEMORY_COMPRESSION VARCHAR2(17)
TABLE_LOCK VARCHAR2(8) INMEMORY_DUPLICATE VARCHAR2(13)

45
Tables in the 12.2 Data Dictionary (2:2)
 By default all Oracle tables are heap tables
SQL> desc dba_tables SAMPLE_SIZE NUMBER
Name Null? Type LAST_ANALYZED DATE
--------------------------- -------- --------------- PARTITIONED VARCHAR2(3)
OWNER NOT NULL VARCHAR2(128) IOT_TYPE VARCHAR2(12)
TABLE_NAME NOT NULL VARCHAR2(128) TEMPORARY VARCHAR2(1)
TABLESPACE_NAME VARCHAR2(30) SECONDARY VARCHAR2(1)
CLUSTER_NAME VARCHAR2(128) NESTED VARCHAR2(3)
IOT_NAME VARCHAR2(128) BUFFER_POOL VARCHAR2(7)
STATUS VARCHAR2(8) FLASH_CACHE VARCHAR2(7)
PCT_FREE NUMBER CELL_FLASH_CACHE VARCHAR2(7)
PCT_USED NUMBER ROW_MOVEMENT VARCHAR2(8)
INI_TRANS NUMBER GLOBAL_STATS VARCHAR2(3)
MAX_TRANS NUMBER USER_STATS VARCHAR2(3)
INITIAL_EXTENT NUMBER DURATION VARCHAR2(15)
NEXT_EXTENT NUMBER SKIP_CORRUPT VARCHAR2(8)
MIN_EXTENTS NUMBER MONITORING VARCHAR2(3)
MAX_EXTENTS NUMBER CLUSTER_OWNER VARCHAR2(128)
PCT_INCREASE NUMBER DEPENDENCIES VARCHAR2(8)
FREELISTS NUMBER COMPRESSION VARCHAR2(8)
FREELIST_GROUPS NUMBER COMPRESS_FOR VARCHAR2(30)
LOGGING VARCHAR2(3) DROPPED VARCHAR2(3)
BACKED_UP VARCHAR2(1) READ_ONLY VARCHAR2(3)
NUM_ROWS NUMBER SEGMENT_CREATED VARCHAR2(3)
BLOCKS NUMBER RESULT_CACHE VARCHAR2(7)
EMPTY_BLOCKS NUMBER CLUSTERING VARCHAR2(3)
AVG_SPACE NUMBER ACTIVITY_TRACKING VARCHAR2(23)
CHAIN_CNT NUMBER DML_TIMESTAMP VARCHAR2(25)
AVG_ROW_LEN NUMBER HAS_IDENTITY VARCHAR2(3)
AVG_SPACE_FREELIST_BLOCKS NUMBER CONTAINER_DATA VARCHAR2(3)
NUM_FREELIST_BLOCKS NUMBER INMEMORY VARCHAR2(8)
DEGREE VARCHAR2(10) INMEMORY_PRIORITY VARCHAR2(8)
INSTANCES VARCHAR2(10) INMEMORY_DISTRIBUTE VARCHAR2(15)
CACHE VARCHAR2(5) INMEMORY_COMPRESSION VARCHAR2(17)
TABLE_LOCK VARCHAR2(8) INMEMORY_DUPLICATE VARCHAR2(13)

46
Table Storage Types (1:2)
 Tablespace
 A tablespace is the default table storage definition and is a logical mapping to space in one
or more physical datafiles
 Tablespaces are limited to storing one segment per block
 Tablespaces can be read-write or read-only
 The default Oracle tablespace definition wastes 10% of most tables
 Clusters
 A cluster is a storage definition that can be mapped onto a tablespace that suspends the
default rules and allows for capabilities that can be used to greatly enhance performance
 The Oracle data dictionary's incredible performance speed is based in large part on the
use of Clusters
 SecureFiles
 Like Clusters a SecureFile is a mapped portion of a tablespace that suspends the default
rules and enhances the ability to store LOBs (they have nothing to do with security)
 File System
 Used by external tables

47
Table Storage Types (2:2)
 Here's how Oracle makes use of Clusters to enhance data dictionary
performance
SQL> SELECT cluster_name, tablespace_name TBS_NAME, SQL> SELECT table_name
cluster_type CTYPE 2 FROM dba_tables
2 FROM dba_clusters 3 WHERE cluster_name = 'C_OBJ#'
3 ORDER BY 1; 4* ORDER BY 1;

CLUSTER_NAME TBS_NAME CTYPE TABLE_NAME

---------------------- --------- ----- ------------------------------
C_COBJ# SYSTEM INDEX ASSEMBLY$ -- Assemblies
C_FILE#_BLOCK# SYSTEM INDEX ATTRCOL$ -- Column attributes
C_MLOG# SYSTEM INDEX CLU$ -- Clusters
C_OBJ# SYSTEM INDEX COL$ -- Columns
C_OBJ#_INTCOL# SYSTEM INDEX COLTYPE$ -- Column types
C_RG# SYSTEM INDEX ICOL$ -- Index columns
C_TOID_VERSION# SYSTEM INDEX ICOLDEP$ -- Index column dependents
C_TS# SYSTEM INDEX IND$ -- Indexes
C_USER# SYSTEM INDEX LIBRARY$ -- Libraries
SMON_SCN_TO_TIME_AUX SYSAUX INDEX LOB$ -- LOBs
NTAB$ -- Nested tables
OPQTYPE$ -- Opaque Types
 C_OBJ# stores information from 17 REFCON$ -- Referential constraints
SUBCOLTYPE$ -- Subcolumn types
separate but related tables within a TAB$ -- Tables
TYPE_MISC$ -- Audits
single block minimizing I/O and join VIEWTRCOL$ -- View Column Attributes
overhead

48
Cluster Types
 Single Table Cluster By Hash
 Multi-Table Cluster By Hash CREATE CLUSTER sc_srvr_id (
srvr_id NUMBER(10))
 Multi-Table Cluster By Index SIZE 1024;

 Sorted Hash Clusters CREATE INDEX idx_sc_srvr_id ON CLUSTER sc_srvr_id;

CREATE TABLE cservers (

 Can reduce the overhead of an srvr_id NUMBER(10),
ORDER BY clause to zero network_id NUMBER(10),
status VARCHAR2(1),
CREATE CLUSTER sorted_hc ( latitude FLOAT(20),
program_id NUMBER(3), longitude FLOAT(20),
line_id NUMBER(10) SORT, netaddress VARCHAR2(15))
delivery_dt DATE SORT) CLUSTER sc_srvr_id (srvr_id);
TABLESPACE uwdata
HASHKEYS 9 CREATE TABLE cserv_inst (
SIZE 750 siid NUMBER(10),
HASH IS program_id; si_status VARCHAR2(15),
type VARCHAR2(5),
CREATE TABLE shc_airplane ( installstatus VARCHAR2(1),
program_id NUMBER(3), location_code NUMBER(10),
line_id NUMBER(10) SORT, custacct_id VARCHAR2(10),
delivery_dt DATE SORT, srvr_id NUMBER(10),
customer_id VARCHAR2(3), ws_id NUMBER(10))
order_dt DATE) CLUSTER sc_srvr_id (srvr_id);
CLUSTER sorted_hc (program_id, line_id, delivery_dt);

49
SecureFiles
 SecureFiles LOB storage is one of two storage types used with Oracle
Database 12c the other type is BasicFiles LOB storage
 The SECUREFILE LOB parameter enables advanced features, including
compression and deduplication (part of the Advanced Compression Option),
and encryption (part of the Advanced Security Option)
 Starting with Oracle Database 12c, SecureFiles is the default storage
mechanism for LOBs

CREATE TABLE sec_tab_dd (

rid NUMBER(5),
bcol BLOB)
LOB (bcol)
STORE AS SECUREFILE bcol2 (
TABLESPACE securefiletbs
RETENTION MIN 3600
COMPRESS ENCRYPT CACHE READS)
TABLESPACE uwdata;

50
ASM
 Not all ASM DiskGroups should be created equal
 For diskgroups not on SSD you can get more I/Os from the outside of the
platter than the inside
ALTER DISKGROUP data
ADD TEMPLATE datafile_hot ATTRIBUTE (HOT MIRRORHOT);

ALTER DISKGROUP archives

ADD TEMPLATE datafile_cold ATTRIBUTE (COLD MIRRORCOLD);

ALTER DISKGROUP data

MODIFY FILE '+data/orcl/datafile/users.259.679156903' ATTRIBUTE (HOT MIRRORHOT);

51
Column Data Types (1:2)
 Consider your options before choosing a data type
CREATE TABLE tstring ( CREATE TABLE tnumeric (
charcol CHAR(2000), deccol DEC(38),
charvaryingcol CHAR VARYING(4000), decimalcol DECIMAL(38),
charactercol CHARACTER(2000), doubleprecisioncol DOUBLE PRECISION,
charactervaryingcol CHARACTER VARYING(4000), floatcol FLOAT(126),
nationalcharvarying NATIONAL CHAR VARYING(2000), intcol INT,
nationalcharactervaryingcol NATIONAL CHARACTER VARYING(2000), integercol INTEGER,
ncharcol NCHAR(1000), numbercol NUMBER(38),
ncharvaryingcol NCHAR VARYING(2000), numberfcol NUMBER,
nvarchar2col NVARCHAR2(2000), numericcol NUMERIC(38),
varcharcol VARCHAR(4000), numericfcol NUMERIC,
varchar2col VARCHAR2(4000)); realcol REAL,
smallintcol SMALLINT);
SQL> desc tstring
Name Type SQL> desc tnumeric
--------------------------- -------------- Name Type
CHARCOL CHAR(2000) ------------------- --------------
CHARVARYINGCOL VARCHAR2(4000) DECCOL NUMBER(38)
CHARACTERCOL CHAR(2000) DECIMALCOL NUMBER(38)
CHARACTERVARYINGCOL VARCHAR2(4000) DOUBLEPRECISIONCOL FLOAT(126)
NATIONALCHARVARYING NVARCHAR2(2000) FLOATCOL FLOAT(126)
NATIONALCHARACTERVARYINGCOL NVARCHAR2(2000) INTCOL NUMBER(38)
NCHARCOL NCHAR(1000) INTEGERCOL NUMBER(38)
NCHARVARYINGCOL NVARCHAR2(2000) NUMBERCOL NUMBER(38)
NVARCHAR2COL NVARCHAR2(2000) NUMBERFCOL NUMBER
VARCHARCOL VARCHAR2(4000) NUMERICCOL NUMBER(38)
VARCHAR2COL VARCHAR2(4000) NUMERICFCOL NUMBER(38)
REALCOL FLOAT(63)
SMALLINTCOL NUMBER(38)

52
Column Data Types (2:2)
 Consider your options before choosing a data type
CREATE TABLE ttemporal ( CREATE TABLE tmisc (
date_col DATE, rowid_col ROWID,
int_d2s_col INTERVAL DAY TO SECOND, urowid_col UROWID,
int_y2m_col INTERVAL YEAR TO MONTH, anytype_col ANYTYPE,
ts_col TIMESTAMP, anydata_col ANYDATA,
tswtz_col TIMESTAMP WITH TIME ZONE, anydataset_col ANYDATASET);
tswltz_col TIMESTAMP WITH LOCAL TIME ZONE);
SQL> desc tmisc
SQL> desc ttemporal Name Type
Name Type ------------------- --------------
------------------- -------------- ROWID_COL ROWID
DATE_COL DATE UROWID_COL ROWID
INT_D2S_COL INTERVAL DAY(2) TO SECOND(6) ANYTYPE_COL ANYTYPE
INT_Y2M_COL INTERVAL YEAR(2) TO MONTH ANYDATA_COL ANYDATA
TS_COL TIMESTAMP(6) ANYDATASET_COL ANYDATASET
TSWTZ_COL TIMESTAMP(6) WITH TIME ZONE
TSWLTZ_COL TIMESTAMP(6) WITH LOCAL TIME
ZONE

53
Column Defaults
 Provides a default value during DML if the SQL
SQL> desc dba_tab_cols
Name Null? Type
----------------------- -------- --------------
statement would leave the column NULL OWNER
TABLE_NAME
NOT NULL
NOT NULL
VARCHAR2(128)
VARCHAR2(128)

 In 12c this can include a sequence generated COLUMN_NAME

DATA_TYPE
NOT NULL VARCHAR2(128)
VARCHAR2(128)
DATA_TYPE_MOD VARCHAR2(3)
surrogate key known as an identity column DATA_TYPE_OWNER
DATA_LENGTH
VARCHAR2(128)
NOT NULL NUMBER
DATA_PRECISION NUMBER
DATA_SCALE NUMBER
NULLABLE VARCHAR2(1)
CREATE TABLE customers (
COLUMN_ID NUMBER
cust_id NUMBER GENERATED ALWAYS AS IDENTITY DEFAULT_LENGTH NUMBER
INCREMENT BY 2 DATA_DEFAULT LONG
START WITH 100 NUM_DISTINCT NUMBER
MAXVALUE 110 LOW_VALUE RAW(2000)
HIGH_VALUE RAW(2000)
MINVALUE 100
DENSITY NUMBER
CYCLE NUM_NULLS NUMBER
CACHE 5 NUM_BUCKETS NUMBER
NOORDER, ...
first_name VARCHAR2(25) HIDDEN_COLUMN VARCHAR2(3)
VIRTUAL_COLUMN VARCHAR2(3)
last_name VARCHAR2(25));
...
COLLATION VARCHAR2(100)
COLLATED_COLUMN_ID NUMBER

54
Hidden Columns (1:2)
 Hidden, aka Invisible, columns are columns that
SQL> desc dba_tab_cols
Name Null? Type
------------------------ -------- --------------
are invisible unless specifically named in a OWNER
TABLE_NAME
NOT NULL
NOT NULL
VARCHAR2(128)
VARCHAR2(128)
command or statement COLUMN_NAME
DATA_TYPE
NOT NULL VARCHAR2(128)
VARCHAR2(128)

 The following are examples where a hidden

DATA_TYPE_MOD VARCHAR2(3)
DATA_TYPE_OWNER VARCHAR2(128)
DATA_LENGTH NOT NULL NUMBER
column will not be visible DATA_PRECISION
DATA_SCALE
NUMBER
NUMBER
 SQL*Plus describe command NULLABLE
COLUMN_ID
VARCHAR2(1)
NUMBER
DEFAULT_LENGTH NUMBER
 SELECT * DATA_DEFAULT LONG
NUM_DISTINCT NUMBER
LOW_VALUE RAW(2000)
HIGH_VALUE RAW(2000)
DENSITY NUMBER
NUM_NULLS NUMBER
NUM_BUCKETS NUMBER
...
HIDDEN_COLUMN VARCHAR2(3)
VIRTUAL_COLUMN VARCHAR2(3)
...
COLLATION VARCHAR2(100)
COLLATED_COLUMN_ID NUMBER

55
Hidden Columns (2:2)

SQL> CREATE TABLE invis ( SQL> SELECT * FROM invis;

rid NUMBER, RID ACCT_NAME
acct_active NUMBER(1) INVISIBLE, ---------- --------------------
acct_name VARCHAR2(20)); 1 Morgan

Table created. SQL> SELECT rid, acct_active, acct_name

FROM invis;
SQL> desc invis 2
Name Type RID ACCT_ACTIVE ACCT_NAME
------------ -------------- ---------- ----------- --------------------
RID NUMBER 1 0 Morgan
ACCT_NAME VARCHAR2(20)

SQL> INSERT INTO invis

2 (rid, acct_active, acct_name)
3 VALUES
4 (1, 0, 'Morgan');

1 row created.

56
Virtual Columns
 Virtual columns are columns that do not
SQL> desc dba_tab_cols
Name Null? Type
----------------------- -------- --------------
physically exist from the standpoint of providing OWNER
TABLE_NAME
NOT NULL
NOT NULL
VARCHAR2(128)
VARCHAR2(128)
persistent data storage but rather virtualize the COLUMN_NAME
DATA_TYPE
NOT NULL VARCHAR2(128)
VARCHAR2(128)
result of a function or expression DATA_TYPE_MOD
DATA_TYPE_OWNER
VARCHAR2(3)
VARCHAR2(128)
DATA_LENGTH NOT NULL NUMBER
 Virtual columns can be used for constraints, DATA_PRECISION
DATA_SCALE
NUMBER
NUMBER
indexing, and partitioning NULLABLE
COLUMN_ID
VARCHAR2(1)
NUMBER
DEFAULT_LENGTH NUMBER
CREATE TABLE vcol ( DATA_DEFAULT LONG
NUM_DISTINCT NUMBER
salary NUMBER(8), LOW_VALUE RAW(2000)
bonus NUMBER(3), HIGH_VALUE RAW(2000)
total_comp NUMBER(10) AS (salary+bonus)); DENSITY NUMBER
NUM_NULLS NUMBER
desc vcol NUM_BUCKETS NUMBER
...
HIDDEN_COLUMN VARCHAR2(3)
SELECT column_id, column_name, virtual_column VIRTUAL_COLUMN VARCHAR2(3)
FROM user_tab_cols ...
WHERE table_name = 'VCOL' COLLATION VARCHAR2(100)
COLLATED_COLUMN_ID NUMBER
INSERT INTO vcol
(salary, bonus)
VALUES
(1,2);

57
Performance Tuning Tables

58
Table Tuning Rules
 Choose the optimal type of table and table storage options based on the way
the table will be accessed
 It is almost always better to accept a small penalty during INSERTs and
UPDATEs to gain an advantage for SELECT
 When designing tables ... thinking vertically ... not horizontally ... generally, the
more columns the more issues
 Order columns in the table by the probability they will be accessed in a filter or
join operation
 Do not make primary key columns NOT NULL
 Do not use the LONG or LONG RAW data types

59
Too Many Columns
 Oracle claims that a table can contain up to 1,000 columns: It is not true. No
database can do 1,000 columns no matter what their marketing claims may be
 The maximum number of real table columns is 255
 Break the 255 barrier and optimizations such as advanced and hybrid
columnar compression no longer work
 A 1,000 column table is actually four segments joined together behind the
scenes just as a partitioned table appears to be a single segment but isn't
 Be suspicious of any table with more than 50 columns. At 100 columns it is
time to take a break and re-read the Codd-Date rules on normalization
 Think vertically not horizontally
 Be very suspicious of any table with column names in the form "SPARE1",
"SPARE2", "..."
 The more columns a table has the more cpu is required when accessing
columns to the right (as the table is displayed in a SELECT * query ... or at the bottom if the table is
displayed by a DESCribe)

60
Column Ordering
 Computers are not humans and tables are not paper forms
 CBO's column retrieval cost
 Oracle stores columns in variable length format
 Each row is parsed in order to retrieve one or more columns
 Each subsequently parsed column introduces a cost of 20 cpu cycles regardless of
whether it is of value or not
 These tables will be accessed by person_id or state: No one will ever put the
address2 column into the WHERE clause as a filter ... they won't filter on
middle initial either
Common Design Optimized Design
CREATE TABLE customers ( CPU CREATE TABLE customers (
person_id NUMBER, 20 person_id NUMBER,
first_name VARCHAR2(30) NOT NULL, 40 last_name VARCHAR2(30) NOT NULL,
middle_init VARCHAR2(2), 60 state VARCHAR2(2) NOT NULL,
last_name VARCHAR2(30) NOT NULL, 80 city VARCHAR2(30) NOT NULL,
address1 VARCHAR2(30), 100 first_name VARCHAR2(30) NOT NULL,
address2 VARCHAR2(30), 120 address1 VARCHAR2(30),
city VARCHAR2(30), 140 address2 VARCHAR2(30),
state VARCHAR2(2)); 160 middle_init VARCHAR2(2));

61
Proof Column Ordering Matters in 12.1.0.2

CREATE TABLE read_test AS

SELECT *
FROM apex_040200.wwv_flow_page_plugs
WHERE rownum = 1;

SQL> explain plan for

2 select * from read_test;

PLAN_TABLE_OUTPUT
-------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
-------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 214K| 2 (0)| 00:00:01 |
| 1 | TABLE ACCESS FULL| READ_TEST | 1 | 214K| 2 (0)| 00:00:01 |
-------------------------------------------------------------------------------

-- fetch value from column 1

Final cost for query block SEL$1 (#0) - All Rows Plan:
Best join order: 1
Cost: 2.0002 Degree: 1 Card: 1.0000 Bytes: 13
Resc: 2.0002 Resc_io: 2.0000 Resc_cpu: 7271
Resp: 2.0002 Resp_io: 2.0000 Resc_cpu: 7271

-- fetch value from column 193

Final cost for query block SEL$1 (#0) - All Rows Plan:
Best join order: 1
Cost: 2.0003 Degree: 1 Card: 1.0000 Bytes: 2002
Resc: 2.0003 Resc_io: 2.0000 Resc_cpu: 11111
Resp: 2.0003 Resp_io: 2.0000 Resc_cpu: 11111

62
It Still Matters in 12.2.0.1
 Run the query while performing a 10053 Level 1 Trace
ALTER SESSION SET tracefile_identifier = 'test_plan2';

ALTER SESSION SET EVENTS '10053 trace name context forever, level 1';

SELECT * FROM audsys.aud$unified WHERE inst_id = 2 AND ROWNUM = 1;

SELECT * FROM audsys.aud$unified WHERE con_id = 9 AND ROWNUM = 1;

ALTER SESSION SET EVENTS '10053 trace name context OFF';

Column 101
Column 1
SINGLE TABLE ACCESS PATH SINGLE TABLE ACCESS PATH
Single Table Cardinality Estimation for AUD$UNIFIED[AUD$UNIFIED] Single Table Cardinality Estimation for AUD$UNIFIED[AUD$UNIFIED]
SPD: Return code in qosdDSDirSetup: NOCTX, estType = TABLE SPD: Return code in qosdDSDirSetup: NOCTX, estType = TABLE

kkecdn: Single Table Predicate:"AUD$UNIFIED"."INST_ID"=2 kkecdn: Single Table Predicate:"AUD$UNIFIED"."CON_ID"=9

Column (#1): INST_ID(NUMBER) Column (#101): CON_ID(NUMBER)
AvgLen: 22 NDV: 0 Nulls: 81 Density: 0.000000 AvgLen: 22 NDV: 0 Nulls: 81 Density: 0.000000
Estimated selectivity: 0.010000 , col: #1 Estimated selectivity: 0.010000 , col: #101
Table: AUD$UNIFIED Alias: AUD$UNIFIED Table: AUD$UNIFIED Alias: AUD$UNIFIED
Card: Original: 81.000000 Rounded: 1 Computed: 0.000000 Non Adjusted: 0.000000 Card: Original: 81.000000 Rounded: 1 Computed: 0.000000 Non Adjusted: 0.000000
Scan IO Cost (Disk) = 4.000000 Scan IO Cost (Disk) = 4.000000
Scan CPU Cost (Disk) = 85364.400000 Scan CPU Cost (Disk) = 245364.400000
Cost of predicates: Cost of predicates:
io = NOCOST, cpu = 50.000000, sel = 0.010000 flag = 2048 ("AUD$UNIFIED"."INST_ID"=2) io = NOCOST, cpu = 50.000000, sel = 0.010000 flag = 2048 ("AUD$UNIFIED"."CON_ID"=9)
Total Scan IO Cost = 4.000000 (scan (Disk)) Total Scan IO Cost = 4.000000 (scan (Disk))
+ 0.000000 (io filter eval) (= 0.000000 (per row) * 81.000000 (#rows)) + 0.000000 (io filter eval) (= 0.000000 (per row) * 81.000000 (#rows))
= 4.000000 = 4.000000
Total Scan CPU Cost = 85364.400000 (scan (Disk)) Total Scan CPU Cost = 245364.400000 (scan (Disk))
+ 4050.000000 (cpu filter eval) (= 50.000000 (per row) * 81.000000 (#rows)) + 4050.000000 (cpu filter eval) (= 50.000000 (per row) * 81.000000 (#rows))
= 89414.400000 = 249414.400000
Access Path: TableScan Access Path: TableScan
Cost: 4.002375 Resp: 4.002375 Degree: 0 Cost: 4.006624 Resp: 4.006624 Degree: 0
Cost_io: 4.000000 Cost_cpu: 89414 Cost_io: 4.000000 Cost_cpu: 249414
Resp_io: 4.000000 Resp_cpu: 89414 Resp_io: 4.000000 Resp_cpu: 249414
Best:: AccessPath: TableScan Best:: AccessPath: TableScan
Cost: 4.002375 Degree: 1 Resp: 4.002375 Card: 0.000000 Bytes: 0.000000 Cost: 4.006624 Degree: 1 Resp: 4.006624 Card: 0.000000 Bytes: 0.000000

How much are you willing to pay for accessing column 101?

63
Stop Making Primary Keys NOT NULL
 the only value in creating a needless, and meaningless, check constraint?
SQL> CREATE TABLE t (
2 recid NUMBER);

Table created.

SQL> desc t
Name Null? Type
--------------------------- -------- ----------------
RECID NUMBER

SQL> ALTER TABLE t

2 ADD CONSTRAINT pk_t
3 PRIMARY KEY (recid);

Table altered.

SQL> desc t
Name Null? Type
--------------------------- -------- ----------------
RECID NOT NULL NUMBER

 Is that you get to waste a little more CPU than everyone else

64
Compressed Tables
 Table compression comes in three separate flavors
 Basic ... no additional cost and including in the database license
 Advanced ... requires the Advanced Compression Option license
 Hybrid Columnar ... requires Exadata, ZFS, or Pillar Data Systems storage
 Compression does far more than save disk space ... it can greatly enhance
performance by minimizing I/O
 Compression and decompression require cpu so carefully calibrate usage on
systems that are cpu bound

65
DBMS_SPACE
 This package provides segment space information not currently available
through the standard views
 Use the CREATE_TABLE_COST and CREATE_INDEX_COST to determine
the sizes of tables and indexes prior to their creation
 Use FREEBLOCKS and SPACE_USAGE to determine the free blocks
consumed by tables, partitions, LOBs, clusters, and indexes
 Use OBJECT_SPACE_USAGE, SPACE_USAGE, and UNUSED_SPACE to
learn how efficiently space allocated to an object is used
 Use VERIFY_SHRINK_CANDIDATE to determine whether an object can be
DECLARE
shrunk to drop the high water mark, recover space, and improve performance
ub NUMBER;
ab NUMBER;
dbms_space.create_table_cost(
BEGIN
tablespace_name IN VARCHAR2,
dbms_space.create_table_cost('UWDATA',28,250000,0,ub,ab);
avg_row_size IN NUMBER,
row_count IN NUMBER,
dbms_output.put_line('Used Bytes: ' || TO_CHAR(ub));
pct_free IN NUMBER,
dbms_output.put_line('Alloc Bytes: ' || TO_CHAR(ab));
used_bytes OUT NUMBER,
END;
alloc_bytes OUT NUMBER);
/
Used Bytes: 7880704
Alloc Bytes: 8126464

66
In-Memory Clause (1:2)
 To greatly increase access
speed for joins and filters
consider the In-Memory option
which covers
 Full Database Caching
 In-Memory Aggregation
 In-Memory Column Store

67
In-Memory Clause (2:2)

68
Automated Data Optimization (ADO) (1:2)
 ADO is implemented with two Oracle capabilities ... heat maps utilizing the
DBMS_HEATMAP package and Integrated Lifecycle Management (ILM)
utilizing the DBMS_ILM and DBMS_ILM_ADMIN packages
 ILM can be used to apply policies to table data that automatically apply
management policies
 Compression
 Data Tiering
 In-Memory Option
CREATE TABLE rowmove_test (
testcol VARCHAR2(20))
ENABLE ROW MOVEMENT;

SELECT table_name, row_movement

FROM user_tables;

ALTER TABLE rowmove_test DISABLE ROW MOVEMENT;

SELECT table_name, row_movement

FROM user_tables;

69
Automated Data Optimization (ADO) (2:2)
CREATE TABLESPACE tier1_ts DATAFILE '/u01/app/oracle/oradata/orcl12c/orcl/tier1.dbf' SIZE 10M AUTOEXTEND ON NEXT 5M;
CREATE TABLESPACE tier2_ts DATAFILE '/u01/app/oracle/oradata/orcl12c/orcl/tier2.dbf' SIZE 10M AUTOEXTEND ON NEXT 5M;
CREATE TABLESPACE tier3_ts DATAFILE '/u01/app/oracle/oradata/orcl12c/orcl/tier3.dbf' SIZE 50M AUTOEXTEND ON NEXT 5M;

CREATE TABLE order_hdr(

order_no NUMBER NOT NULL,
order_date DATE NOT NULL,
order_note VARCHAR2(500))
PARTITION BY RANGE (order_date)
(
PARTITION orders_2016_q3 VALUES LESS THAN (TO_DATE('01/10/2016', 'DD/MM/YYYY')) TABLESPACE tier3_ts,
PARTITION orders_2016_q4 VALUES LESS THAN (TO_DATE('01/01/2017', 'DD/MM/YYYY')) TABLESPACE tier3_ts,
PARTITION orders_2017_q1 VALUES LESS THAN (TO_DATE('01/04/2017', 'DD/MM/YYYY')) TABLESPACE tier2_ts
ILM ADD POLICY TIER TO tier3_ts READ ONLY SEGMENT AFTER 3 MONTHS OF NO ACCESS,
PARTITION orders_2017_q2 VALUES LESS THAN (TO_DATE('01/07/2017', 'DD/MM/YYYY')) TABLESPACE tier2_ts
ILM ADD POLICY TIER TO tier3_ts READ ONLY SEGMENT AFTER 3 MONTHS OF NO ACCESS,
PARTITION orders_2017_q3 VALUES LESS THAN (TO_DATE('01/10/2017', 'DD/MM/YYYY')) TABLESPACE tier1_ts
ILM ADD POLICY TIER TO tier2_ts READ ONLY SEGMENT AFTER 1 MONTHS OF NO ACCESS,
PARTITION orders_2017_q4 VALUES LESS THAN (TO_DATE('01/01/2018', 'DD/MM/YYYY')) TABLESPACE tier1_ts
ILM ADD POLICY TIER TO medium_storage_ts READ ONLY SEGMENT AFTER 2 MONTHS OF NO ACCESS)
ILM ADD POLICY ROW STORE COMPRESS BASIC SEGMENT AFTER 1 MONTHS OF NO ACCESS;

70
Temporal Validity
 Associates one or more valid time dimensions with a table and makes data
visible depending on its time-based validity, as determined by the start and
end dates or time stamps of the period for which a given record is considered
valid

71
More You Should Know About Tables

72
Enabling Row Movement
 After a rowid is assigned to a row piece, the rowid can change in special
 circumstances if row movement is enabled
 Partition key updates
 Flashback Table operations
 Shrink table operations
CREATE TABLE rowmove_test (
testcol VARCHAR2(20))
ENABLE ROW MOVEMENT;

SELECT table_name, row_movement

FROM user_tables;

ALTER TABLE rowmove_test DISABLE ROW MOVEMENT;

SELECT table_name, row_movement

FROM user_tables;

73
Row Dependencies
 ORA_ROWSCN pseudocolumn
CREATE TABLE scnprec (
 Returns, for each row, the conservative upper testcol VARCHAR2(20))
ROWDEPENDENCIES;
bound system change number (SCN) of the
most recent change to the row SELECT table_name, dependencies
FROM user_tables;
 This is useful for determining approximately
SELECT current_scn
when a row was last updated FROM v$database;
 It is not absolutely precise, because Oracle INSERT INTO scnprec VALUES ('ABC');
tracks SCNs by transaction committed for the COMMIT;
block in which the row resides INSERT INTO scnprec VALUES ('DEF');
COMMIT;

INSERT INTO scnprec VALUES ('GHI');

COMMIT;

SELECT ORA_ROWSCN, ROWID, testcol

FROM scnprec;

ORA_ROWSCN ROWID TESTCOL

---------- ------------------ --------
6022658 AAATuZAABAAAZxhAAA ABC
6022661 AAATuZAABAAAZxhAAB DEF
6022670 AAATuZAABAAAZxhAAC GHI

74
DBMS_COMPARISON
 Use this fully documented and supported built-in package to compare data in
tables, views and materialized views
 Can converge (fix) data discrepancies automatically

dbms_comparison.create_comparison( exec dbms_comparison.create_comparison(comparison_name=>'UWCompare',

comparison_name IN VARCHAR2, -- cannot contain spaces schema_name=>'SCOTT',
schema_name IN VARCHAR2, object_name=>'DEPT',
object_name IN VARCHAR2, dblink_name=>NULL,
dblink_name IN VARCHAR2, remote_schema_name=>'ABC',
index_schema_name IN VARCHAR2 DEFAULT NULL, remote_object_name=>'DEPT',
index_name IN VARCHAR2 DEFAULT NULL, scan_percent=>90);
remote_schema_name IN VARCHAR2 DEFAULT NULL,
remote_object_name IN VARCHAR2 DEFAULT NULL, DECLARE
comparison_mode IN VARCHAR2 DEFAULT CMP_COMPARE_MODE_OBJECT, ct dbms_comparison.comparison_type;
column_list IN VARCHAR2 DEFAULT '*', BEGIN
scan_mode IN VARCHAR2 DEFAULT CMP_SCAN_MODE_FULL, dbms_comparison.converge('UWCOMPARE', 2, ct,
scan_percent IN NUMBER DEFAULT NULL, dbms_comparison.CMP_CONVERGE_LOCAL_WINS, TRUE);
null_value IN VARCHAR2 DEFAULT CMP_NULL_VALUE_DEF,
local_converge_tag IN RAW DEFAULT NULL, dbms_output.put_line(ct.scan_id);
remote_converge_tag IN RAW DEFAULT NULL, dbms_output.put_line(ct.loc_rows_merged);
max_num_buckets IN NUMBER DEFAULT CMP_MAX_NUM_BUCKETS, dbms_output.put_line(ct.rmt_rows_merged);
min_rows_in_bucket IN NUMBER DEFAULT CMP_MIN_ROWS_IN_BUCKET); dbms_output.put_line(ct.loc_rows_deleted);
dbms_output.put_line(ct.rmt_rows_merged);
END;
/

75
DBMS_REDEFINITION
 Used to online redefine
 Table columns
 Table column names
 Table column data types
 Constraints
 Indexes
BEGIN
 Triggers dbms_redefinition.can_redef_table('PM', 'PRINT_ADS', dbms_redefinition.cons_use_pk);
dbms_redefinition.redef_table(
uname => 'PM',
tname => 'PRINT_ADS',
table_compression_type => 'ROW STORE COMPRESS ADVANCED',
table_part_tablespace => 'NEWTBS',
index_key_compression_type => 'COMPRESS 1',
index_tablespace => 'NEWIDXTBS',
lob_compression_type => 'COMPRESS HIGH',
lob_tablespace => 'NEWLOBTBS',
lob_store_as => 'SECUREFILE');
dbms_redefinition.sync_interim_table('PM', 'PRINT_ADS', 'INT_PRINT_ADS');
dbms_redefinition.finish_redef_table('PM', 'PRINT_ADS', 'INT_PRINT_ADS');
END;
/

76
Wrap Up

77
Conclusion
 How comfortable are you with your knowledge of CREATE TABLE?
 Probably feel pretty good about DROP TABLE
 But how about all of the ALTER TABLE syntax?
 The most important principle in creating optimal tables is the same as
everything else in Oracle is "do the least work"
 Minimize CPU utilization
 Minimize I/O
 Take the load off the storage array
 Off the HBA cards
 Off the SAN switch
 Minimize network utilization
 Bandwidth
 Round Trips
 Minimize the memory footprint
 Stop using 7.3.4 best practices ... and start using the best practices
for 11gR2 and above

78
 *
ERROR at line 1:
ORA-00028: your session has been killed

Thank You

Daniel A. Morgan
email: [email protected]
mobile: +1 206-669-2949
skype: damorgan11g
twitter: @meta7solutions 79