Scribd
Upload
136 views41 pages

Dama Temporal Tables Coffing 201110

1. Teradata uses parallel processing across multiple AMPs to distribute data from tables across disks for fast retrieval. A parsing engine creates plans to retrieve data from AMPs. 2. Temporal tables allow querying of past and current data by using valid and transaction time periods to track changes over time without deleting rows. This allows querying of past states. 3. Valid time temporal tables track effective and expiration dates for rows, logically deleting without physically removing rows. Transaction time tables track when data is transactionally current.

Uploaded by

ypswamy3631
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
136 views41 pages

Dama Temporal Tables Coffing 201110

1. Teradata uses parallel processing across multiple AMPs to distribute data from tables across disks for fast retrieval. A parsing engine creates plans to retrieve data from AMPs. 2. Temporal tables allow querying of past and current data by using valid and transaction time periods to track changes over time without deleting rows. This allows querying of past states. 3. Valid time temporal tables track effective and expiration dates for rows, logically deleting without physically removing rows. Transaction time tables track when data is transactionally current.

Uploaded by

ypswamy3631
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 41

Teradata and Temporal Tables for DAMA

Tom Coffing (Tera-Tom) CEO, Coffing Data Warehousing [email protected] Direct: 513 300-0341 Website: www.CoffingDW.com

Relational Databases store their data in Tables

Parsing Engine

PE

SQL
AMP Access
Module Processors

BYNET 0 BYNET 1
AMP AMP AMP

When users login to Teradata a Parsing Engine will take each SQL request and come up with a Plan for the AMPs to retrieve the data from their disks.

Each time AMPs are added to the Teradata system the rows of each table are redistributed.

Parsing PLAN Engine PE

1 AMPs Open your Order Table

2 Bring back all 2011 Orders


3 ORDER BY Order_Date

AMP

AMP

AMP

AMP

AMP

AMP

AMP

Order Customer Item

Order Customer Item

Order Customer Item

Order Customer Item

Order Customer Item

Order Customer Item

Order Customer Item

Logical View of Teradata

All tables are spread across all AMPs in Teradata to take advantage of the parallel processing.

Every Teradata Table has one and only one Primary index which does 3 things.

1 2 3

Distributes the rows to the proper AMP


Fastest way to retrieve a row(s)

Incredibly important for Joins

There are two types of Primary Indexes:

UPI Unique Primary Index NUPI NON-Unique Primary Index

When is the Primary Index Created?


CREATE Table Employee_Table ( Emp_No Integer Not Null ,Dept_No Integer ,First_Name Varchar(12) ,Last_Name Char(20) ,Salary Decimal (10,2) ) Unique Primary Index (Emp_No) ;

The Primary Index is created at table create time.

UPI
Emp_No 2 4 6 8 10 12 14 16 Dept_no 100 200 300 400 400 400 100 200 First_Name Rakish Vu Mary Leona Sandy Matt Javier Shelby Last_Name Ratel Vatish Mason Lacy Stewart Mason Jones Stewart Salary 50000.00 45000.00 98000.00 15500.00 84000.00 65000.00 47000.00 52000.00

SELECT * FROM Employee_Table WHERE Emp_No = 16 ;


If users use the Primary Index in the WHERE Clause of their SQL it is a 1-AMP operation every time.

NUPI
Emp_No 2 4 6 8 10 12 14 16 Dept_no 100 200 300 400 400 400 100 200 First_Name Rakish Vu Mary Leona Sandy Matt Javier Shelby Last_Name Ratel Vatish Mason Lacy Stewart Mason Jones Stewart Salary 50000.00 45000.00 98000.00 15500.00 84000.00 65000.00 47000.00 52000.00

SELECT * FROM Employee_Table WHERE Last_Name = Mason;


If users use the Primary Index in the WHERE Clause of their SQL it is a 1-AMP operation every time.

* * *

UPI
Dept_No 100 Row
Take UPI value 100

Department
Department_Name Sales
Hash with secret Math Formula

Mgr_No 2000000

Budget 500000.00
AMP Numbers

Value 100 has a 32-bit Row Hash of 000000110

=6

1 2 3 4 Count 6 1 2 3 4 steps In 1 2 3 4 Hash Map 1 2 3 4 AMP 2 will hold this row

AMP
Department

AMP
Department

AMP
Department

AMP
Department

The PE takes the Primary Index of each row and runs it through a math formula and then looks at the hash map to determine which AMP will hold the row.

AMP

AMP

AMP

AMP

1000 1010

1001 1011

1002 1013

1005 1016

1150 1231 1579

1151 1232 1580

1157 1237 1583

1159 1241 1588

UPI on Emp_No will cause Perfect distribution

A Unique Primary Index (UPI) will spread the rows of a table evenly across the AMPs almost every time.

AMP

AMP

AMP

AMP
NUPI on Last_Name will cause uneven distribution

Smith Smith Jones Zao

* *

Wilson Wilson Patel Patel Patel

* * * * *

Davis Davis Davis

* * *

Conner Kelson Jetson Lyle Pate

A Non-Unique Primary Index (NUPI) will spread the rows of a table unevenly across the AMPs because duplicate values go to the same AMP every time.

AMP

AMP

AMP

AMP

Order_Table
1 21 33 41 5 13 25 45 9 17 29 37 14 26 34 42 6 18 38 46 2 10 22 30

Order_Table
3 11 19 39 15 23 31 43 7 27 35 47

Order_Table
12 20 28 48 4 24 32 40 8 16 36 44

Order_Table January Dates February Dates March Dates

SELECT * FROM ORDER_Table WHERE ORDER_DATE BETWEEN January Dates 2010-01-01 and 2010-01-31;

Teradata even has the ability to Partition Tables for range queries so a Full Table Scan is prevented.

Temporal Tables are tables based on time. There are three types of Temporal Tables. 1) Valid Time Temporal Tables 2) Transaction Time Temporal Tables 3) Bi-Temporal Tables which combine both: Valid Time

Transaction Time

Below is a normal Employee_Table Employee_Table


Emp_No 1 2 4 Dept_No 100 200 400 First_Name Mary Hitesh Chen Last_Name Mason Patel Wang Salary 50000.00 60000.00 60000.00

Imagine if someone called and said, Does Hans Svenson work here?
SELECT * FROM Employee_Table WHERE Last_Name = Svenson ;
No Rows Returned

What if they asked, Did Hans Svenson ever work here? How would I know Im not a fortune teller.

Normal Table Vs ValidTime Temporal Table Employee_Table_Normal


Emp_No Dept_No First_Name Last_Name Salary

1 2 4

100 200 400

Mary Hitesh Chen

Mason Patel Wang

50000.00 60000.00 60000.00

Employee_Table_Temporal
Emp_No Dept_No First_Name Last_Name Salary 1 2 100 Mary 200 Hitesh Mason Patel ValidStart ValidEnd Forever Forever 50000.00 2010-01-01 60000.00 2010-01-02

3
4 4

300 Hans
400 Chen 400 Chen

Svenson
Wang Wang

80000.00 2010-01-03 2010-12-31


50000.00 2010-01-04 2011-01-20 60000.00 2011-01-20 Forever

Emp_No Dept_No First_Name Last_Name Salary 1 2 3 4 4 100 Mary 200 Hitesh 300 Hans 400 Chen 400 Chen Mason Patel Svenson Wang Wang

ValidStart

ValidEnd Forever Forever

50000.00 2010-01-01 60000.00 2010-01-02

80000.00 2010-01-03 2010-12-31 50000.00 2010-01-04 2011-01-20 60000.00 2011-01-20 Forever

Does Hans Svenson work here?


SELECT * FROM Employee_Table_Temporal WHERE Last_Name = Svenson ;
No Rows Returned
Keywords placed in front of your standard SQL can Return closed rows.

NONSEQUENCED VALIDTIME SELECT * FROM Employee_Table_Temporal WHERE Last_Name = Svenson ;


Emp_No Dept_No First_Name Last_Name Salary 3 300 Hans Svenson

ValidStart

ValidEnd

80000.00 2010-01-03 2010-12-31

Emp_No Dept_No First_Name Last_Name Salary 1 2 3 4 4 100 Mary 200 Hitesh 300 Hans 400 Chen 400 Chen Mason Patel Svenson Wang Wang

ValidStart

ValidEnd Forever Forever

50000.00 2010-01-01 60000.00 2010-01-02

80000.00 2010-01-03 2010-12-31 50000.00 2010-01-04 2011-01-20 60000.00 2011-01-20 Forever

The two major concepts I want you to understand are: 1. No rows are physically deleted, so the table is APPEND ONLY. Row will be logically deleted. 2. Rows are deleted and updated by logically closing out the row. So rows will be open when data is inserted and then logically closed on a delete. Rows that are open are current rows and can be queried exactly like a normal table, but rows that are deleted logically will be invisible unless the user uses special keywords in their SQL.

Emp_No Dept_No First_Name Last_Name Salary 1 2 3 4 4 100 Mary 200 Hitesh 300 Hans 400 Chen 400 Chen Mason Patel Svenson Wang Wang

ValidStart

ValidEnd Forever Forever

50000.00 2010-01-01 60000.00 2010-01-02

80000.00 2010-01-03 2010-12-31 50000.00 2010-01-04 2011-01-20 60000.00 2011-01-20 Forever

1) How many employees are currently employed? ______ 2) When did Hans Svenson leave the company? _________________ 3) What is the only column ever updated for a row physically? ____________ 4) What keyword determines that a row is still open? _____________________ 5) What was the date that Chen Wang received a raise? _______________ When Chen Wang received a raise the SQL was written like this: UPDATE Employee_Table_Temporal SET Salary = 60000 WHERE Emp_No = 4; 6) What two things did Teradata automatically do behind the scenes?____ ____

Emp_No Dept_No First_Name Last_Name Salary 1 2 3 4 4 100 Mary 200 Hitesh 300 Hans 400 Chen 400 Chen Mason Patel Svenson Wang Wang

ValidStart

ValidEnd Forever Forever

50000.00 2010-01-01 60000.00 2010-01-02

80000.00 2010-01-03 2010-12-31 50000.00 2010-01-04 2011-01-20 60000.00 2011-01-20 Forever

1) How many employees are currently employed? 3 2) When did Hans Svenson leave the company? December 31, 2010 3) What is the only column ever updated for a row physically? ValidEnd 4) What keyword determines that a row is still open? Forever 5) What was the date that Chen Wang received a raise? January 20, 2011 When Chen Wang received a raise the SQL was written like this: UPDATE Employee_Table_Temporal SET Salary = 60000 WHERE Emp_No = 4;

6) What two things did Teradata automatically do behind the scenes? Updated the ValidEnd column and inserted a new row to reflect the salary change.

Normal Table Vs TransactionTime Temporal Table Policy_Table_Normal


Cust_No Policy_No Amount

1 2 3

10 20 30

500,000.00 600,000.00 1,000,000.00

Policy_Table_Temporal
Cust_No Policy_No
1 2 10 20

Amount

TransactionTime

500,000.00 2011-01-01 , 9999-12-31 600,000.00 2011-01-02 , 9999-12-31

30

1,000,000.00 2011-01-03 , 9999-12-31

Cust_No Policy_No

Amount 500,000.00 600,000.00 1,000,000.00

Policy_Table_Normal

1 2 3

10 20 30

UPDATE Policy_Table_Normal SET Insured_Amount = 2000000.00 WHERE Customer_No = 1 ;

UPDATE Policy_Table_Temporal SET Insured_Amount = 2000000.00 WHERE Customer_No = 1 ;

Policy_Table_Temporal
Cust_No Policy_No 1 2 3 10 20 30 Amount TransactionTime 500,000.00 2011-01-01 , 9999-12-31 600,000.00 2011-01-02 , 9999-12-31 1,000,000.00 2011-01-03 , 9999-12-31

How will both tables look like after the UPDATE statement done on December 1, 2011?

Here are both tables after customer 1 was updated to an amount of 2,000,000 on December 1, 2011.
Cust_No Policy_No Amount 2,000,000.00 600,000.00 1,000,000.00

Policy_Table_Normal

1 2 3

10 20 30

Policy_Table_Temporal
Cust_No Policy_No Amount TransactionTime

1 2 3 1

10 20 30 10

500,000.00 600,000.00 1,000,000.00 2,000,000.00

2011-01-01 , 2011-12-01 2011-01-02 , 9999-12-31 2011-01-03 , 9999-12-31 2011-12-31 , 9999-12-31

Policy_Table_Temporal
Cust_No Policy_No 1 2 3 1 10 20 30 10 Amount TransactionTime 500,000.00 2011-01-01 2011-12-01 600,000.00 2011-01-02 9999-12-31 1,000,000.00 2011-01-03 9999-12-31 2,000,000.00 2011-12-31 9999-12-31

SELECT * FROM Policy_Table_Temporal ;


Cust_No Policy_No 1 2 3 10 20 30 Amount 2,000,000.00 600,000.00 1,000,000.00

Policy_Table_Temporal
Cust_No Policy_No 1 2 3 1 10 20 30 10 Amount TransactionTime 500,000.00 2011-01-01 2011-12-01 600,000.00 2011-01-02 9999-12-31 1,000,000.00 2011-01-03 9999-12-31 2,000,000.00 2011-12-31 9999-12-31

TRANSACTIONTIME as of 2011-06-30 SELECT * FROM Policy_Table_Temporal ;


Cust_No Policy_No
1 2 3 10 20 30

Amount
500,000.00 600,000.00 1,000,000.00

Temporal Tables are tables based on time. There are Four types of time concepts. 1) Current Time uses normal SQL and the answer set reflects the current rows that are open. 2) AS of Time uses an exact point in time and the answer set reflects the current rows open at that particular point in time. 3) Sequenced Time uses Period of Applicability and the answer set reflects the current rows open during that period of time. 4) Non-Sequenced Time ignores the ValidTime or TransactionTime and the answer set reflects the Rows by ignoring the temporal time.

Lets CREATE a Bi-Temporal Table


CREATE MULTISET TABLE Property_Owners ( Cust_No INTEGER ,Prop_No INTEGER ,Prop_Val_Time PERIOD (DATE) NOT NULL as VALIDTIME ,Prop_Tran_Time PERIOD (TIMESTAMP(6) with TIME ZONE) NOT NULL as TRANSACTIONTIME ) PRIMARY INDEX(Prop_No) ;

What is it about this Bi-Temporal table CREATE statement that makes Teradata know the table is Bi-Temporal?

CREATE MULTISET TABLE Property_Owners ( Cust_No INTEGER ,Prop_No INTEGER ,Prop_Val_Time PERIOD (DATE) NOT NULL as VALIDTIME ,Prop_Tran_Time PERIOD (TIMESTAMP(6) with TIME ZONE) NOT NULL as TRANSACTIONTIME ) PRIMARY INDEX(Prop_No) ;

What is it about this Bi-Temporal table CREATE statement that makes Teradata know the table is Bi-Temporal? The Keywords VALIDTIME and TRANSACTTIME

Lets CREATE a Bi-Temporal Table


CREATE MULTISET TABLE Property_Owners ( Cust_No INTEGER ,Prop_No INTEGER ,Prop_Val_Time PERIOD (DATE) NOT NULL as VALIDTIME ,Prop_Tran_Time PERIOD (TIMESTAMP(6) with TIME ZONE) NOT NULL as TRANSACTIONTIME ) PRIMARY INDEX(Prop_No) ;

What does a PERIOD Data Type mean?

CREATE MULTISET TABLE Property_Owners ( Cust_No INTEGER ,Prop_No INTEGER ,Prop_Val_Time PERIOD (DATE) NOT NULL as VALIDTIME ,Prop_Tran_Time PERIOD (TIMESTAMP(6) with TIME ZONE) NOT NULL as TRANSACTIONTIME ) PRIMARY INDEX(Prop_No) ;

A Period Data Type means a beginning and ending date or Timestamp such as:

Timestamp

2011-01-01 , 9999-12-31 Or 2011-01-01 , 2012-06-30 Or

Forever Closed Timestamp Forever

2011-01-01 08:09:290000-05:00, 9999-12-31 23:59:59.999999+00:00

Lets CREATE a Bi-Temporal Table


CREATE MULTISET TABLE Property_Owners ( Cust_No INTEGER ,Prop_No INTEGER ,Prop_Val_Time PERIOD (DATE) NOT NULL as VALIDTIME ,Prop_Tran_Time PERIOD (TIMESTAMP(6) with TIME ZONE) NOT NULL as TRANSACTIONTIME ) PRIMARY INDEX(Prop_No) ;

What type of PERIOD Data Types are required for ValidTime and TransactionTime?

CREATE MULTISET TABLE Property_Owners ( Cust_No INTEGER ,Prop_No INTEGER ,Prop_Val_Time PERIOD (DATE) NOT NULL as VALIDTIME ,Prop_Tran_Time PERIOD (TIMESTAMP(6) with TIME ZONE) NOT NULL as TRANSACTIONTIME ) PRIMARY INDEX(Prop_No) ;

What type of PERIOD Data Types are required for ValidTime and TransactionTime? ValidTime can be either a date or a Timestamp TransactionTime must be a Timestamp written exactly as above!

On January 1, 2011 Tera-Tom buys property 100 which is beach front property! Tera-Tom is Customer_No 1 in your table and in your heart!
CREATE MULTISET TABLE Property_Owners ( Cust_No INTEGER ,Prop_No INTEGER ,Prop_Val_Time PERIOD (DATE) NOT NULL as VALIDTIME ,Prop_Tran_Time PERIOD (TIMESTAMP(6) with TIME ZONE) NOT NULL as TRANSACTIONTIME ) PRIMARY INDEX(Prop_No) ;

INSERT INTO PROPERTY_OWNERS (Cust_No, Prop_No) VALUES (1, 100) ;

INSERT INTO PROPERTY_OWNERS (Cust_No, Prop_No) VALUES (1, 100) ;

Below is what the table looks like internally after Tera-Tom bought the property on January 1, 2011 (2011-01-01)
Property_Owners
Cust_No Prop_No
1 100

ValidTime
2011-01-01 , 9999-12-31

TransactionTime
2011-01-01 , 9999-12-31

Note: The TransactionTime should have a Timestamp, but not shown here for space reasons.

On February 14, 2011 Tera-Tom sells property 100 to Socrates who is Customer_No 2! UPDATE Property_Owners SET Cust_No = 2 WHERE Prop_No = 100 ;

Property_Owners
Cust_No Prop_No
1 100

Table before the UPDATE

ValidTime

TransactionTime

2011-01-01 , 9999-12-31 2011-01-01 , 9999-12-31

What does the table look like after the UPDATE? How many rows are in the new table?

UPDATE Property_Owners SET Cust_No = 2 Socrates bought on WHERE Prop_No = 100 ; February 14, 2011 (2011-02-14) Property_Owners
Cust_No Prop_No
1 100

ValidTime
2011-01-01 , 9999-12-31

TransactionTime
2011-01-01 , 9999-12-31

Property_Owners
Cust_No Prop_No
1 1 100 100

ValidTime

TransactionTime

2011-01-01 , 9999-12-31 2011-01-01 , 2011-02-14 2011-01-01 , 2011-02-14 2011-02-14 , 9999-12-31

100

2011-02-14 , 9999-12-31 2011-02-14 , 9999-12-31

On April 1, 2011 Socrates sells property 100 but through a different mortgage company so our mortgage company no longer owns property 100! DELETE FROM Property_Owners WHERE Prop_No = 100 ;
Table before the DELETE

Property_Owners
Cust_No Prop_No
1 100

ValidTime

TransactionTime

2011-01-01 , 9999-12-31 2011-01-01 , 2011-02-14

1
2

100
100

2011-01-01 , 2011-02-14 2011-02-14 , 9999-12-31


2011-02-14 , 9999-12-31 2011-02-14 , 9999-12-31

What does the table look like after the DELETE? How many rows are in the table now?

DELETE FROM Property_Owners WHERE Prop_No = 100 ;

Property_Owners
Cust_No Prop_No
1
1 2

ValidTime

TransactionTime

100
100 100

2011-01-01 , 9999-12-31 2011-01-01 , 2011-02-14


2011-01-01 , 2011-02-14 2011-02-14 , 9999-12-31 2011-02-14 , 9999-12-31 2011-02-14 , 9999-12-31

What does the table look like after the DELETE that happened on April 1, 2011 (2011-04-01)?
Cust_No Prop_No
1 1 100 100

ValidTime

TransactionTime

2011-01-01 , 9999-12-31 2011-01-01 , 2011-02-14 2011-01-01 , 2011-02-14 2011-02-14 , 9999-12-31

2
2

100
100

2011-02-14 , 9999-12-31 2011-02-14 , 2011-04-01


2011-02-14 , 2011-04-01 2011-04-01 , 9999-12-31

The End Thank You

Teradata and Temporal Tables for DAMA

Tom Coffing (Tera-Tom) CEO, Coffing Data Warehousing [email protected] Direct: 513 300-0341 Website: www.CoffingDW.com

You might also like