Tuning Oracle Stored Procedures

Guy Harrison CTO, APM Business Unit, Quest Software

Agenda
Using PL/SQL in place of standard SQL Principles of code optimization applied to PL/SQL Using array processing in PL/SQL Using and optimizing packages and procedures Optimizing PL/SQL cursors Using the PL/SQL profiler A comparison of Java and PL/SQL performance Tuning Java stored procedures

Using PL/SQL in Place of Standard SQL

Exploit a procedural approach to data processing.
Overcome optimizer limitations

Reduce client server traffic Divide and conquer complex statements Be aware of the drawbacks
Coding/maintenance overhead Inability to return result sets in many development tools

Operations that are More Efficient in PL/SQL

Finding maximums and minimums
8.1.6 analytical functions are an alternative approach

Correlated updates
The WHEN clause contains a query that is equivalent to a query in the SET clause

Any SQL statement that repeats subqueries Denormalization

Using triggers

Code Optimization LOOP and IF Statements

Minimize the number of iterations
Use EXIT to cease looping where necessary

Remove any statements within a loop that could be processed outside the loop
Especially when you have nested loops

Specify the most probable condition first in a compound IF statement

Try to reduce the number of evaluations

Example of a Poorly Formed Loop

Effect of Loop Optimization

128.65

deoptimal loop

optimal loop

0.54

20

40

60

80

100

120

140

Elapsed time (s)

Code Optimization Recursion

Recursive routines invoke themselves Elegant from a coding perspective, but consume memory and are usually slower than an iterative alternative Almost all recursive algorithms have a non-recursive equivalent

Example of Recursive and Non-Recursive Solutions

Code Optimization - Recursion

44.8

Recursive

Non-recursive

1.23

10

20

30

40

50

Elapsed Time (s)

Array Processing
The BULK COLLECT statement allows data from a query to be loaded directly into a PL/SQL table FORALL allows data to be inserted into a table directly from a PL/SQL block For large operations, array processing is typically 10 times faster than non-array

Example of Array Processing

Array Processing
33.23 Insert 3.02

30.04 Select 4.73

12

18 Elapsed time (s)

24

30

36

Oracle8i PL/SQL array processing

Tradional (pre-8i) PL/SQL processing

WHERE CURRENT OF cursor

Inside a cursor loop, WHERE CURRENT OF allows the current row to be directly updated. The cursors SQL must be defined with the FOR UPDATE clause Saving the rowid as using that for the lookup can be a useful alternative
This can cause some overhead This is faster than looking up the row by the primary key

But you need to check that the row has not been altered since you opened the cursor

WHERE CURRENT OF cursor

Use primary key 33.9

Use WHERE CURRENT OF

16.2

Save ROWID

8.35

10

15

20 Elapsed time (s)

25

30

35

40

Example of WHERE CURRENT OF cursor

NOCOPY clause
If a subroutine takes a PL/SQL table as an argument, a copy of the table is created for the subroutine
This may be time consuming, especially if the subroutine is called repeatedly

NOCOPY allows the subroutine to work on the table directly

This is sometimes called passing by reference vs. passing by value

Example of NOCOPY

Effect of NOCOPY
115.18

Not using NOCOPY

Using NOCOPY

0.04

20

40

60

80

100

120

140

Elapsed time (s)

Avoiding Recompilation
Use packages rather than procedures.
Providing the package header remains constant, changes to subroutines will not cause recompilation of dependent procedures

Pin large or performance critical packages in the shared pool

Dbms_shared_pool.keep

Optimizing Triggers
Ensure that triggers fire only when required
Use WHEN clause to restrict trigger execution to specific logical conditions Use the UPDATE OF clause to restrict trigger execution to changes to specific rows

Example of a Well Formed Trigger

Explicit Cursors
Although you can embed SQL directly in a routine, explicitly creating a cursor is more efficient
An implicit cursor does a second fetch to ensure that only one row is returned This is especially important if the SQL performs a table scan

Example of Explicit and Implicit Cursors

Explicit Cursors
5827 table scan 143

Implicit cursor 4 Index lookup 4 Explicit cursor

1000

2000

3000 Block gets

4000

5000

6000

Using the PL/SQL Profiler

The DBMS_PROFILER package allows you to identify hot spots in your code Surround your PL/SQL call with DBMS_PROFILER.START_PROFILER and DBMS_PROFILER.END_PROFILER The output is stored in PLSQL_PROFILER_RUNS, PLSQL_PROFILER_UNITS and PLSQL_PROFILER_DATA Quests TOAD and SQL Navigator provide graphical support for the profiler

Java vs. PL/SQL Stored Procedures

Java stored procedures can be written in JDBC or SQLJ In general, Java stored procedures are more complex to write and maintain However, Java stored procedures can execute faster for computationally intensive routines

Performance Comparison Computation

19.39 Basic maths 4.11

Floating point arithmetic

0.83

148.33

15.82 Integer arithmetic 1.65

20

40

60

80

100

120

140

160

Execution Time (s) Java PL/SQL

Performance Comparison Database Intensive Routines

44.97 Batch processing 95.06 40.97 OLTP 66.38

20

40

60

80

100

Elapsed time (s) Java PL/SQL

Bind Variables in Java Stored Procedures

Unlike PL/SQL, bind variables are not transparently implemented in JDBC Although using bind variables requires extra coding, using literals will increase parse overhead and cause contention You should use Prepared statements in JDBC An alternative is to issue an ALTER SESSION SET CURSOR_SHARING=FORCE

Array Processing in JDBC

As in PL/SQL, you should use array processing where appropriate
Use the setFetchSize method for queries Use the addBatch and executeBatch methods for inserts

Thank you