High Performance Plsql

© 2009 Quest Software, Inc. ALL RIGHTS RESERVED

High Performance PL/SQL

Guy Harrison

Director of Development, Melbourne

www.guyharrison.net

2

Introductions

Buy Guy’s BookBuy Quest Products

3

http://www.motivatedphotos.com/?id=17760

4

Blue

Yellow

Red

0 10 20 30 40 50 60 70 80

Star trek shirt fatality analysis

Pct

5

Measuring PL/SQL performance• DBMS_PROFILER is the best way to find PL/SQL “hot

spots”:

6Scripts at www.guyharrison.net

7

Toad Profiler support

8

SQL*Navigator profiler support

9

11g Hierarchical profiler

$ plshprof -output hprof demo1.trc

10

Plshprof output

11

DBMS_HPROF tables

Scripts at www.guyharrison.net

12

When is PL/SQL faster?

• PL/SQL routines most massively outperform other languages when network round trips are significant.

13

Network traffic• Routines that process large numbers of rows and return simple

aggregates are also candidates for a stored procedure approach

14

Stored procedure alternative

15

Network traffic example

Stored Procedure

Java client

0 200 400 600 800 1,000 1,200 1,400 1,600 1,800

344

1703

297

313

Local Host

Remote Host

Elapsed time (ms)

16

Aspects of PL/SQL performance

Compilation and datatype Tweaks

PL/SQL code structure

PL/SQL Data handling

SQL Code

17

PLSQL_OPTIMIZE_LEVEL

0: No optimization

1: Minor (eliminate but not reorganize)

2: (default) Significantly reorganize code (array fetch, optimize loops)

3: 11g specific (in-lining, etc)

18

Compilation and datatype Tweaks

PL/SQL code structure

PL/SQL Data handling

SQL Code

19

It’s usually the SQL • Most PL/SQL routines spend most of their time executing

SELECT statements and DML• SQL tuning is a big topic but:

– Measure SQL overhead of PL/SQL routines first– Ensure best possible optimizer statistics – Consider adequacy of indexing– Learn how to use DBMS_XPLAN, SQL Trace, etc – Exploit 10g/11g tuning facilities (if licensed)– Don’t issue SQL when you don’t need to

20

SQL or PL/SQL?

Scripts at www.guyharrison.net

21

22

Compilation and datatype Tweaks

PL/SQL code structure

PL/SQL Data handling

SQL Code

23

Three ways of processing rows C

PU

& lo

gica

l rea

ds

•One at a time

•In Batches

•All at once .

Mem

ory Requirem

entsFour

24

One at a time

25

All at once

26

In batches

27

Array processing

1 10 100 1000 10000 100000 10000000

20

40

60

80

100

120

140

160

180

200

Bulk Collect Size

Ela

ps

ed

Tim

e

No bulk collect

Bulk collect without LIMIT

(Prior to 10g or PLSQL_OPTIMIZE_LEVEL <2)

28

Array processing

• PLSQL_OPTIMIZE_LEVEL>1 causes transparent BULK COLLECT LIMIT 100

1 10 100 1000 10000 100000 1000000 100000000

50

100

150

200

250

300

Array Size

Ela

sped

tim

e (s

)

10g or higher with PLSQL_OPTIMIZE_LEVEL >1

No bulk collect

Bulk collect without LIMIT

29

BULK COLLECT worst case scenario

30

Bind variables in Dynamic SQL • Using bind variables allows sharable SQL, reduces parse overhead and

minimizes latch contention • Unlike other languages, PL/SQL uses bind variables transparently

– EXCEPT when using Dynamic SQL:

31

Using bind variables

32

Bind variable performance • 10,000 calls

Bind variables

No Binds

0 1 2 3 4 5 6 7 8

3.42

7.84

Elasped Time (s)

33

NOCOPY• The NOCOPY clause causes a parameter to be passed

“by reference” rather than “by value”

34

NOCOPY performance gains• 4,000 row, 10 column “table”; 4000 lookups:

NOCOPY

NO NOCOPY

0 100 200 300 400 500 600 700 800 900

0.28

864.96

Elapsed time (s)

35

Associative arrays • Traditionally, sequential scans of PLSQL tables are used for caching

database table data:

36

Associative arrays• Associative arrays allow for faster and simpler lookups:

37

Associative array performance• 10,000 random customer lookups with 55,000 customers

Associative lookups

Sequential scan

0 5 10 15 20 25 30

0.04

29.79

Elapsed time (s)

38

Compilation and datatype Tweaks

PL/SQL code structure

PL/SQL Data handling

SQL Code

39

Reduce unnecessary Looping• Unnecessary loop iterations burn CPU

Well formed loop

Poorly formed loop

0 5 10 15 20 25 30 35

3.96

34.31

Elapsed time (s)

40

Remove loop Invariant terms• Any term in a loop that does not vary should be extracted

from the loop

PLSQL_OPTIMIZE_LEVEL>1 does this automatically

41

Loop invariant performance improvements

plsql_optimize_level=2

Optimized loop

Original loop

0 2 4 6 8 10 12

5.28

5.87

11.09

Elapsed time (s)

42

Recursion (see: recursion)• Recursive routines often offer elegant solutions*.• However, deep recursion is memory-intensive and

usually not scalable

* Known In Australia as “smart-ass solutions”

*

43

Recursion memory overhead

0 2000000 4000000 6000000 8000000 100000000

200

400

600

800

1000

1200

1400

Recursive

Non-recursive

Recursive Depth

PG

A m

emo

ry (

MB

)

44

Compilation and datatype Tweaks

PL/SQL code structure

PL/SQL Data handling

SQL Code

45

Number crunching (1)

SIMPLE_INTEGER

PLS_INTEGER

NUMBER

0 5 10 15 20 25

5.99

7.06

20.09

0.54

3.83

17.64

NATIVE

INTERPRETED

Elasped Time (s)

46

Number crunching (2)

Java stored procedure

PLSQL compiled, SIMPLE_INTEGER datatype

PLSQL interpreted, PLS_INTEGER data type

PLSQL interpreted, NUMBER data type

0 5 10 15 20 25 30 35 40 45 50

0.11

0.74

14.48

47.22

Elapsed time (s)

47

11g Function cache• Suits deterministic but expensive functions• Expensive table lookups on non-volatile tables

48

Function cache performance • 100 executions, random date ranges 1-30 days:

Function cache

No function cache

0 1 2 3 4 5 6

1.51

5.21

Elapsed time (s)

49

In-lining

Manual in-liningModular design

50

Automatic in-lining• PLSQL_OPTIMIZE_LEVEL = 3

• OR:

PRAGMA INLINE

Manual Inlining

No Inlining

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

2.54

2.56

4.95

Elapsed time (s)

51

Ran out of time for...

• Array insert using FORALL• Explicit vs. implicit cursors • RETURNING clause • Pipelined functions• Optimizing triggers• Short circuit evaluations• IF and CASE comparison ordering

52

Thank You – Q&A