Executing in Parallel

Home | Previous Page | Next Page Design Decisions > Enhancing Performance >

Executing in Parallel

Parallelizable routines can execute in parallel across multiple processors.

To make a UDR parallelizable, apply the following rules:

Follow the guidelines for well-behaved user-defined routines.
Avoid any DataBlade API routine that involves query processing (mi_exec(), mi_exec_prepared_statement()), collections (mi_collection_*), row types, or save sets (mi_save_set_*).
Do not create rows that contain any complex types including another row type as one of the columns. Do not use the mi_row_create() or mi_value() functions with complex types or row types.
Avoid DataBlade API FastPath functions (mi_routine_*, mi_func_desc_by_typeid()) if the access method might pass them routine identifiers for nonparallelizable routines.
Specify the PARALLELIZABLE routine modifier in the CREATE FUNCTION or CREATE PROCEDURE statement for the UDR.

For more information about the following topics, refer to the IBM Informix: DataBlade API Programmer's Guide:

Guidelines for well-behaved user-defined routines
A complete list of nonparallelizable functions
FastPath function syntax, usage, and examples

For more information about the PARALLELIZABLE (and other) routine modifiers, refer to the Routine Modifier segment in the IBM Informix: Guide to SQL Syntax. For more information about parallelizable UDRs, refer to IBM Informix: User-Defined Routines and Data Types Developer's Guide.

To make an access method parallelizable

Create a basic set of parallelizable purpose functions.
The basic set, which enables a SELECT statement to execute in parallel, includes the following purpose functions: am_open, am_close, am_beginscan, am_endscan, am_getnext, and am_rescan.

An access method might not supply all of the purpose functions that define a basic parallelizable set. As long as you make all the basic purpose functions that you provide parallelizable, a SELECT statement that uses the access method can execute in parallel.
Add a parallelizable purpose function to the basic set for any of the following actions that you want the database server to execute in parallel.

Parallel SQL Statement

Parallelizable Purpose Function

INSERT (in a SELECT)

am_insert

SELECT INTO TEMP

am_insert

DELETE

am_delete

UPDATE

am_update

Important:

A parallelizable purpose function must call only routines that are also parallelizable. All the strategy and support functions for the operator class that the index uses must also be parallelizable.

The database server sets an am_parallel purpose value in the sysams system catalog table to indicate which access-method actions can occur in parallel. For more information, refer to Purpose Options.