Partitioning Oracle Sources in PowerCenter

2012 Informatica Corporation. No part of this document may be reproduced or transmitted in any form, by any means

(electronic, photocopying, recording or otherwise) without prior consent of Informatica Corporation.

Abstract
You can partition Oracle database table reads to increase performance. This article explains some techniques for
partitioning Oracle source data.

Supported Versions
Informatica PowerCenter 9.x

Table of Contents
Overview. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Database Partitioning. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Key Range Partitioning. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Configuring Key Range Partitioning. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Pass-Through Partitioning. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
MOD Function. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Configuring the Filter Condition with the MOD Function. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Function-Based Index. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Filter on ROWID. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Filter on ROWID Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
CreatePartitionInfo Mapping. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
DimReadTest Mapping. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

Overview
When an Oracle source is the bottleneck for PowerCenter session performance, you can increase performance by
configuring source partitioning in the session properties. With source partitioning, the PowerCenter Integration Service reads
multiple Oracle rows in parallel.
The PowerCenter Integration Service creates a reader thread for each pipeline partition. For a relational database source,
each partition issues an SQL statement to access the source data. To optimize performance, the SQL statements should
create efficient and fairly equal-sized data sets.
You can use the following approaches for partitioning Oracle source data:
Database partitioning
Creates a pipeline for each physical table partition in the database.
Key range partitioning
Distributes source rows into partitions based on the values of a port or set of ports.
Pass-through partitioning
Passes rows into static partitions based on a filter condition for each partition.

Database Partitioning
You can optimize session performance by using the database partitioning partition type for Oracle sources. Use database
partitioning for Oracle sources whenever possible.
Database partitioning creates a pipeline for each physical table partition in the Oracle database. When you use database
partitioning, the PowerCenter Integration Service queries the database system for table partition information and fetches
data into the session partitions. You can use any number of session partitions and any number of database partitions. When
the pipeline partitions do not equal the database partitions, the PowerCenter Integration Service generates SQL queries for
each database partition and distributes the data among the session partitions equally. However, you can improve
performance when the number of pipeline partitions equals the number of database partitions.
For Oracle sources that use composite partitioning, you can increase performance when the number of pipeline partitions
equals the number of database subpartitions. For example, if an Oracle source contains three partitions and two
subpartitions for each partition, set the number of pipeline partitions at the source to six.
You can use other partitioning methods when the Oracle table is not partitioned or when the table is partitioned in a way that
is not useful for extracting data sets of equal size.

Key Range Partitioning

Configure key range partitioning to partition Oracle data based on the value of a port or set of ports. With key range
partitioning, the PowerCenter Integration Service distributes rows of source data based the ports that you define as partition
keys. The PowerCenter Integration Service compares the port value to the range values for each partition and sends rows to
the appropriate partition.
An advantage of key range partitioning is that you can use key range partitioning with PowerCenter dynamic partitioning.
With dynamic partitioning, the PowerCenter Integration Service scales the number of session partitions at run time based on
the source database partitions or the number of nodes in a grid.
Use key range partitioning for columns that have an even distribution of data values. Otherwise, the partitions might have
unequal size. For example, a column might have 10 rows between key values 1 and 1000 and the column might have 999
rows between key values 1001 and 2000.
A disadvantage of key range partitioning is that Oracle might perform a table scan for each partition. Multiple queries might
run slower than one query because the Oracle database runs multiple concurrent table scans.
Note: If you create an database index on the partition key, the Oracle database might perform a table scan anyway. The
index might not increase performance.
With key range partitioning, a query for one partition might return rows sooner than another partition. Or, one partition can
return rows while the other partitions are not returning rows. This situation occurs when the rows in the table are in a similar
order as the key range. One query might be reading and returning rows while the other queries are reading and filtering the
same rows.

Configuring Key Range Partitioning

Configure key range partitioning to partition Oracle data based on port values.
1.

In the Workflow Manager, double-click the session you want to partition.

2.

Select the Source Qualifier partition point on the Partitions view of theMapping tab.

3.

Click Add to add partitions.

4.

Choose Key Range Partitioning and click OK.

The Edit Partition Key dialog box appears.

5.

Choose at least one port to use as the key.

6.

Click Edit Keys to enter a range of values for each partition.

Pass-Through Partitioning
With pass-through partitioning, the PowerCenter Integration Service passes all rows from one partition point to the next
partition point without redistributing data across partitions. All rows in a partition stay in that partition after crossing a partition
point. Pass-through partitioning is the default partitioning method.
When the session has pass-through partitioning, you can configure a filter condition for each static partition. You can use the
following filters to partition data:
MOD function
Use a MOD function to filter data into different partitions based on a value of a numeric column.
ROWID
Partition data by ROWID. Oracle can perform direct reads on rows of data by ROWID.

MOD Function
You can create a filter condition to partition data with the Oracle MOD function. The Oracle MOD function receives two
numeric input values and returns the remainder. For example MOD(4,2) = 0 and MOD(4,3)=1.
To use the MOD function to filter rows, define the Source Qualifier partition type as pass-through. The PowerCenter
Integration Service generates a WHERE clause that includes any filter condition you enter in the session properties.
Enter the filter condition for each partition on the Transformations view of the Mapping tab. The filter overrides any filter
condition that you set in the Designer when you configure the Source Qualifier transformation.
For example, if the session has two partitions you might configure the following function for the first partition:
MOD(columnName,2)=0

Configure the following function for the second partition:

MOD(columnName,2)=1

When the value of the column is an even number, the first partition receives the row. When the column value is odd, the
second partition receives the row.
When you configure the MOD function, choose a numeric column that has an even distribution of values. You can use a key
column. Do not use a column that has few values because the partitions will be unequal sizes. For example, if a column can
contain zero or one, you cannot partition the row into more than two partitions.

Configuring the Filter Condition with the MOD Function

Create a MOD statement for each partition.
The following example shows how to use the MOD function to partition Oracle source data in a session with four partitions.
1.

In the session properties, configure pass-through partitioning.

2.

Configure four partitions.

3.

On the Source Filter attribute, enter the following filter conditions for each partition:
Partition#1:
Partition#2:
Partition#3:
Partition#4:

MOD(InvoiceID,4)=0
MOD(InvoiceID,4)=1
MOD(InvoiceID,4)=2
MOD(InvoiceID,4)=3

The following figure shows where to configure the MOD functions in session properties:

Function-Based Index
You can define a function-based index on a MOD function to increase performance and eliminate full table scans. With a
function-based index, Oracle performs fast index range scans. Using a function-based index can affect all SQL statements
that have the matching predicate.
To create a function-based index use the following SQL syntax:
create index invoiceID-mod4-idx on MOD(InvoiceID,4)

If you change the number of partitions in the session, you need to build the function-based index to match the SELECT query.
Note: Without a function-based index, MOD partitioning performance is similar to range-based partitioning. Each query
typically requires a table scan.

Filter on ROWID
When a session reads all the rows in a table, you can configure the session partitions to read a table by the Oracle ROWID.
The ROWID is the physical address of a row in the table. Oracle performs direct reads on a row using ROWID.
To filter with ROWID, you need to determine the ROWID values in the database table. You can configure a SQL statement
that returns the ROWID for specific rows in a table. For example, the following SELECT statement returns the ROWID and
last name of each customer in department 20:
SELECT ROWID, last_name FROM Invoices WHERE Dept = 20

To partition the table read using ROWID, run a SQL query that returns a minimum and maximum ROWID for each partition
you plan to have in the session. After you determine the minimum and maximum ROWID, configure the Source Read
partition filters using the minimum and maximum ROWID for each partition.
For example, a session that reads the Invoices table has four partitions. Configure the following SQL statement to return the
minimum and maximum ROWID for each partition:
SELECT min(ROWID), max(ROWID), tile from (select ROWID, ntile(4) over (order by ROWID) as tile from Invoices) group
by tile order by 1

The query returns the following values:

MIN ROWID
AAATtYAA9AAAAALAAA
AAATtYAA9AAAK5kAAK
AAATtYAA+AAAAZzAAB
AAATtYAA+AAAK9DAAJ

MAX ROWID
AAATtYAA9AAAK5kAAJ
AAATtYAA+AAAAZzAAA
AAATtYAA+AAAK9DAAI
AAATtYAA+AAAlMIAAJ

TILE
1
2
3
4

In the session properties, configure the filter condition for each of the partitions. Configure the filter condtions with the
minimum and maximum ROWID values from the SQL query.
rowid
rowid
rowid
rowid

between
between
between
between

chartorowid('AAATtYAA9AAAAALAAA')
chartorowid('AAATtYAA9AAAK5kAAK')
chartorowid('AAATtYAA+AAAAZzAAB')
chartorowid('AAATtYAA+AAAK9DAAJ')

and
and
and
and

chartorowid('AAATtYAA9AAAK5kAAJ')
chartorowid('AAATtYAA+AAAAZzAAA')
chartorowid('AAATtYAA+AAAK9DAAI')
chartorowid('AAATtYAA+AAAlMIAAJ')

A disadvantage to partitioning with ROWID is that you must maintain the minimum and maximum values for the filter
conditions. When the table contains new rows, and the filter conditions do not contain the new ROWID values, the session
does not select the rows. You can automate the process to maintain the ROWID values in the filter conditions.

Filter on ROWID Example

The following example shows how to partition a session with the Oracle pseudocolumn ROWID. You can download the
workflows for this example from the following location: https://communities.informatica.com/docs/DOC-8126.
The session contains 16 partitions. The example has the following workflows:
CreatePartitionInfo
Runs an SQL query to determine the ROWIDs in a table. Generates a parameter file that contains a set of SQL
WHERE clauses based on the current ROWID values .
DimReadTest
Reads the parameter file to determine how to partition the source rows in the session. The DimReadTest workflow
performs a direct read by ROWID for each partition.

CreatePartitionInfo Mapping
Create a mapping that returns a parameter file containing the ROWID based WHERE clauses.

The mapping contains the following objects:

Partition_RowIDs
Oracle source that contains the data you want to read in a session.
SQ_Partition_RowIDs
Source Qualifier that contains the SQL query that retrieves the minimum and maximum ROWID values for each
partition you configure in DIMReadTest.
Exptrans
Expression transformation that returns a WHERE clause in the output port for each partition.
Partition_Where_Clauses
Flat file target that is the parameter file for DimReadTest.

SQ_Partition_RowIDs Source Qualifier

The mapping has two input parameters:
$$number_of_partitions
The number of partitions to include in the DimReadTest session. For this example, the value is 16.
$$source_table_name
The name of the source table to query. The table name is DIM_COM_ACCOUNT_TERM1.
Note: The session does not use an input parameter file. For this example, you can manually modify the values in the
Mapping Designer.
The SQL query in the Source Qualifier Properties tab contains the following text:
select min(rowid) AS MIN_ROWID, max(rowid) AS MAX_ROWID, PARTITION_NUM
from (select rowid, ntile($$number_of_partitions) over (order by rowid) as PARTITION_NUM from $
$source_table_name) group by PARTITION_NUM order by 1

Exptrans Expression Transformation

The Expression transformation receives the MIN and MAX ROWIDs for 16 partitions.
The Expression transformation has the following expression in the Where_Clause_ output port:
'$$where_p' || PARTITION_NUM || '=rowid between chartorowid(' || CHR(39) || MIN_ROWID || CHR(39) || ') and
chartorowid(' || CHR(39) || MAX_ROWID || CHR(39) || ')'

Target
Configure the target session properties and select the Use Header Command Output Header option. The PowerCenter
Integration Service adds a header to the target. It appends the contents of the partition_where_header.txt file.
Configure the Header Command field to generate a header row. The Header Command contains the following text:
cat /u01/app/infa_shared/presales/sdorcey/TgtFiles/partition_where_header.txt

The partition_where_header.txt file contains the following text:

[Global]

Any session include use the parameter file.

The following figure shows the Header Options and Header Command attributes for the target session properties:

DimReadTest Mapping
The DimTest mapping reads the parameter file from the CreatePartitionInfo mapping in order to determine how to partition
the source data and perform fast parallel reads. The DimTest mapping reads source rows in 16 partitions. The mapping
passes each row through a Filter transformation.
Note: The Filter transformation returns no rows in the target. You can change the example to include different transformations.

The DimReadTest mapping contains the following objects:

DIM_ COM_ACCOUNT_TERM1
Oracle source that contains the data you want to read in the session.
SQ_DIM_COM_ACCOUNT_TERM1
Source Qualifier that contains the SQL query that retrieves the minimum and maximum ROWID values for each
partition you configure in DIMReadTest.
FILTRANS
Filter transformation that returns no rows for this example. The expression is set to False.

10

DIM_COM_ACCOUNT_TERM
Target that receives no rows.
The following figure shows the parameter file path in the General Options section of the Properties tab:

DimReadTest Source Qualifier

Configure the Source Filter attribute for each partition on the Mapping tab of the session properties. To navigate to the
Source Filter attribute, select the Source Qualifier SQ_DIM_COM_ACCOUNT_TERM in the Navigation panel. Scroll to the
Source Filter attribute.
Enter an attribute for each of the 16 partitions. The naming convention for the attribute value is "$$where_pN" where "N" is
the partition number.
For example:
Attribute
Value
Partition#1 $$where_P1
Partition#2 $$where_P2
Partition#3 $$where_p3

The following figure shows the Mapping tab:

11

If you change the number of partitions in the DimReadTest session, change the $$number_of_partitions parameter in the
CreatePartitionInfo mapping to match the number of partitions in the session.

Authors
Ellen Chandler
Principal Technical Writer
Stan Dorcey
Sr. Product Specialist

12