SAP Performance & Tuning

Bonus material for 201 ABAP Interview Questions

http://www.201interviewquestions.com/books/abap.htm

Material code : BM02

Author : P.Renjith Kumar
Genie Press

Architecture
The Architecture level to improve performance
The R/3 System has a three-layer client/server architecture,
With a Presentation layer, an Application layer and a Database layer.
The presentation layer and the application layer are scalable.
This means that if there is a hardware bottleneck, you can extend the system by adding
more

front ends and application servers.

The database layer, as the central data repository, is not scalable.

ABAP performance analysis areas of focus.

One goal is to reduce the run time of programs on the application server, thereby reducing the
CPU load.
Another goal is to reduce the database load. Reducing the database load is particularly important
since the database software is not scalable.
If the required data is in the R/3 buffers on the application server, accessing them requires
approximately 0.1 milliseconds for each data record.
If the data records are read from the database buffer, around 1 millisecond is required.
When the data is read from the disk, this requires approximately 10 milliseconds for each data
record.

Memory Allocation in SAP

An R/3 work process allocates around 5 MB of memory.

The R/3 table buffers allocate approximately 120 MB (40 MB for single record buffers, 80
MB for generic table buffers).

The data buffers of the database use around 500 MB of memory. The database on the disks
can reach a size of several terabytes.

The data transfer between front end and application server occurs in blocks of 2 KB.

The transfer between application server and database server occurs in blocks of up to 32
KB.

Work process
From presentation server

Application server
Dispatcher
Dispatcher

Queue

Work Process
WP

WP

WP

WP
Roll area

To DB server

General Definitions

Response time:
Time from the receipt of a user request to the sending of a response ( measured on the application
server; does not include network time between the presentation server and the application server).
Dispatcher wait time: Time spent by the user request in the dispatcher queue.
Roll-in: Time required to roll the user context in to the R/3 work process.
Load time: Time required for loading and generating R/3 Repository or ABAP Dictionary objects.
Processing time: = response time - dispatcher wait time - roll-in - roll-out - load time - database
time enqueue time - roll wait time
Enqueue time:
Time from sending an enqueue request to the R/3 enqueue server to the receipt of the results

Definitions
Database time: Time from sending an SQL statement to the receipt of the results (measured on the
application server; includes network time between the application server and the database server).

Roll wait time: Time in which the user context is rolled out of the work process pending response
to an RFC call.

Roll-out: Time required to roll the user context in to the roll buffer.

CPU time: Time spent by the CPU in processing the transaction step (measured by the operating
system; not an additive component of the response time).

Initial Checks to be done to measure Performance

Work Load Monitor

ST03 -> Performance Data base (Work Load Monitor )
Select Server , Time .

If DB response > 600 ms then there is fundamental problem with R3 / DB

ST 03

DB 01

SM 50

Snap Shot Analysis

SM50 ( Snap Shot analysis )

To identify performance critical objects .
To identify long running objects
How to identify long running process ?
Refresh the screen continuously , If a work process is there
for a long time then it is long running process
Important fields to know about the action to be done : Action / Table

General Flow to get into tuning process

Check R3 based WP overview from ( SM50 )

Find long running WP and note its Action / Table field
Find Action ( Sequential / Direct / Insert )
Call DB Monitor ( DB01)
Check if any lock is there . If so find the user who is responsible and Try
to remove the lock , Or delete the lock .
Then based on the analysis go for SQL Trace / ABAP Run time analysis

Initial Reporting consideration

Select Proper internal table types

Standard Tables :
Used generally . Can be sorted
Sorted Tables
Sorted automatically based on key
Hashed Tables
Used when I record is to be retreived . Good in performance . Work based on hash
Key

Debugger Finding Memory

Se38 ->ABAP Debugger -> Settings ->Memory Display On

SE30 (Runtime Analysis)

The runtime analysis is an additional development workbench tool that is quite useful for
analyzing performance of an ABAP / 4 Program or transaction. With this tool, the system can
display information about:
Executed instruction
Accessed execution time.
Tables and Types of access.
Chronological execution flow
ST05 (SQL Trace)
he SQL trace is a tool, which allows displaying and analyzing the contents for the database
calls, which are made by the reports and transactions written in ABAP/4. It monitors programs
and transactions on the database level. With the help of this facility for every open SQL
instructions, you can display, about which SQL Embedded (DECLARE, OPEN, FETCH)
Statement have been executed, besides analyzing the system performance.
SLIN or PROGRAM -> CHECK -> EXTENDED PROGRAM CHECK
ST03, ST02, ST04 are the tcode for workload, tuning and DB Performance Monitoring
codes.

SQL Trace

ST 05 ->SQL Trace -> Activate Trace

Go back to SE 38 .
Then Run The program from SE38 ,
Now again Come back to ST05 ,
Deactivate trace .
Now press Display Trace .

ST05 Components

SQL Trace
Buffer Trace
Enquee Trace
RFC Trace

SQL Trace Internal Mechanism

SQL Trace Goals

The goal of using an SQL Performance Trace is to find SQL statements with a high
optimization potential. Use three user sessions.
One user session is for the trace list, one for the compressed summary, and one for
identical selects.
From the trace list you can access Explain SQL or the ABAP code.
An expensive SQL statement is indicated when a database operation takes longer
than 200,000 milliseconds, or when more than 10 FETCHes are required for a
database operation.
In addition, a series of SQL statements that are similar in structure usually indicate
nesting that can be optimized considerably. If the sum of SQL statements that are
similar in structure take more than 200,000 milliseconds, they can be regarded as
expensive.

Run Time analysis

SE30 -> Give Report name and Execute , Then click
Analyze ,You will get DB time and ABAP run time

Run Time Analysis ( SE 30 )

Runtime analysis ( Comparing more 2 programs )

Binary Search

Very effective measure to increase performance in Select queries

REPORT

ZPR_PER_T1

REPORT

ZPR_PER_T1

TABLES : VBAK .
TABLES : VBAK .
Data
data

Begin of itab occurs 0

Include structure vbak
end of ITAB.

Select * from vbak into table

itab .
Read table itab with key
vbeln = '400151' .

Data Begin of itab occurs 0 .

Include structure vbak .
data end of ITAB.
Select * from vbak into table
itab .
Read table itab with key
vbeln = '400151' BINARY
SEARCH.

Steps to Improve Performance

First see runtime analysis ( SE30 ), try to make 50% of ABAP time and Database time.
Next SQL Trace (ST05), check which table is taking more time and try to minimize it by
using full key or creating index for the where clause.
Next see logic in the program, try to avoid multiple reads on same table and try to
minimize unnecessary data
Next try to remove for all entries if it has large amount of data in the for all entries
internal table.
Next try to read Header table first than item table.
Next try to put joins
Next try to remove nested select's.

Steps to Improve Performance

Select field in sequence as defined in database

free intrenal table memory wnen table is not required for
further processing.

Comparison of select statements

Some changes to statements to improve performance

Not Recommended

Recommended

REFRESH TAB_DEST.
LOOP AT TAB_SRC INTO TAB_DEST.
APPEND TAB_DEST.
ENDLOOP

TAB_DEST[ ] = TAB_SRC[ ].

Aggregate Function

Use already provided aggregate functions instead of

Not Recommended
Recommended
manually
coding it in ABAP.
Maxnu = 0.
Select * from zflight where airln =
LF and cntry = IN.

Select max( fligh ) from zflight int

maxnu where airln = LF and cntry
= IN.

Check zflight-fligh > maxnu.

Maxnu = zflight-fligh.
Endselect.

Similary use MIN, AVG,COUNT,and

SUM as needed.

Avoid append statement

Not Recommended

Recommended

Refresh: itab_flight.

Refresh: intab_flight.

Select * from zflight into

intab_flight.

Select * from zflight into table

intab_flight

Append intab_flight.
Clear intab_flight.
Endselect.

Modifying lines of internal table

Not Recommended
Loop at int_fligh.
If int_fligh-flag is initial.
Int_fligh-flag = X.
Endif.
Modify int_fligh.
Endloop.

Recommended
Int_fligh-flag = X.
Modify int_fligh
transporting flag where
flag is initial.

How to improve Performance

1.

Avoid using SELECT...ENDSELECT... construct and use

SELECT ... INTO TABLE.

2.

Use WHERE clause in your SELECT statement to

restrict the volume of data retrieved.

3.

Design your Query to Use as much index fields as

possible from left to right in your WHERE statement

4.

Either enable buffering in your database table or

create Indexes to speed up the query.

5.

Use FOR ALL ENTRIES in your SELECT statement to

retrieve the matching records at one shot.

How to improve Performance

6.

Avoid using nested SELECT statement, SELECT within

LOOPs.

7.

Avoid using INTO CORRESPONDING FIELDS OF

TABLE. Instead use INTO TABLE.

8.

Avoid using SELECT * and Select only the required

fields from the table.

9.

Avoid nested loops when working with large internal

tables.

10. Use assign instead of into in LOOPs for table types

with large work areas

How to improve Performance

11.

When in doubt call transaction SE30 and use the examples

and check your code

12.

Whenever using READ TABLE use BINARY SEARCH addition

to speed up the search. Be sure to sort the internal table
before binary search.

13.

Use "CHECK" instead of IF/ENDIF whenever possible.

14.

Use "CASE" instead of IF/ENDIF whenever possible.

15.

Use "MOVE" with individual variable/field moves instead of

"MOVE-CORRESPONDING", creates more coding but is more
effcient.

Some useful Transactions related to performance analysis

SM51 App. Servers Overview

STAT Display Statistical Records
ST05 SQL Trace
SE30 Runtime Analysis
ST03 Analysis of Workload
DB02 Database Performance : Tables and Indexes.
DB05 Analysis of Table w.r.t. Indexed Fields
ST04 Database Performance Analysis : Oracle View
SM66 Global Work Process Overview (Over All App Servers)
SM50 Process Overview