Dear SAP Community Member,
In order to fully benefit from what the SAP Community has to offer, please register at:
Thank you,
The SAP Community team.
Skip to end of metadata
Go to start of metadata

What tools can be used to help with performance tuning?

ST05 is the performance trace. It contain the SQL Trace plus RFC, enqueue and buffer trace. Mainly the SQL trace is is used to measure the performance of the select statements of the program.

SE30 is the Runtime Analysis transaction and can be used to measure the application performance. 

SAT transaction is the replacement of the pretty outdated SE30. Provides same functionality as SE30 plus some additional features.

ST12 transaction (part of ST-A/PI software component) is a combination of ST05 and SAT. Very powerful performance analysis tool used primarily by SAP Support.

One of the best tools for static performance analyzing is Code Inspector (SCI). There are many options for finding common mistakes and possible performance bottlenecks.

back to top

What are the steps to optimize the ABAP Code?

    1. Use WHERE clause in your SELECT statement to restrict the volume of data retrieved.  Very important !!
    2. Design your Query to Use as much index fields as possible in your WHERE statement
    3. Use INNER (or OUTER under some circumstances) JOIN in your SELECT statement to retrieve the matching records at one shot
    4. Avoid using nested SELECT statement and SELECT within LOOPs, better use JOINs or FOR ALL ENTRIES. Use FOR ALL ENTRIES when  the internal table is already there or the end of some processing. Try JOINs if the SELECT are right behind each other
    5. Avoid using INTO CORRESPONDING FIELDS OF TABLE during buffered access. Otherwise use the most appropriate for the program.
    6. Avoid using SELECT * and Select only the required fields from the table.
    7. Avoid using ORDER BY in SELECT statements if it differs from used index  (instead, sort the resulting internal table), because this may add additional work to the database system which is unique, while there may be many ABAP servers
    8. INDEX: Creation of Index for improving performance should not be taken without thought. Index speeds up the performance but at the same time adds two overheads namely; memory and insert/append performance. When INDEX is created, memory is used up for storing the index and index sizes can be quite big on large transaction tables! When inserting new entry in the table, all the index's are updated. More index more time. More the amount of data, bigger the indices, larger the time for updating all the indices
    9. Avoid Executing an identical Select (same SELECT, same parameter) multiple times in the program. Buffer in your abap code.
    10. Avoid using join statements if adequate standard views exist no performance impact
    1. Defining a table as buffered (SE11) can help in improving the performance but this has to be used with caution. Buffering of tables leads to data being read from the buffer rather than from table. Buffer sync with table happens periodically, only if something changes which is happen rarely. If this table is a transaction table chances are that the data is changing for a particular selection criteria, therefore application tables are usually not suited for table bufferung. Using table buffering in such cases is not recommended. Use Table Buffering for configuration data and sometimes for Master Data.. 
    2. Avoid using complex Selects on buffered tables-, because SAP may not be able to interpret this request, and may transmit the request to the database- The code inspector tells which commands bypass the buffer
  3. Internal tables
    1. Use HASHED tables where-ever possible. Otherwise SORTED tables. STANDARD tables should be the last choice.
    2. Use assign instead of into in LOOPs for table types with large work areas, if the data is being modified.
    3. When in doubt call transaction SE30 and check your code.
    4. If you must use a STANDARD table and you are using a READ, sort the table appropriately and use the addition BINARY SEARCH to speed up the search. 
  4. Miscellaneous
    1. PERFORM : When writing a subroutine, always provide type for all the parameters. This reduces the overhead which is present when system determines on it's own each type from the formal parameters that are passed. It also makes for more robust programming.

back to top

What is the difference between SELECT SINGLE and SELECT ... UP TO 1 ROWS?

  • SELECT SINGLE and SELECT UP TO n ROWS return the first matching row/rows for the given condition. It may not be unique, if there are more matching rows for the given condition.
  • With ORACLE database system, SELECT SINGLE is converted into SELECT ... UP TO 1 ROWS, thus they are exactly the same in that case. The only difference is the ABAP syntax prevents from using ORDER BY with SELECT SINGLE, but it is allowed with SELECT ... UP TO 1 ROWS. Thus, if several records may be returned and we want to get the highest record for example, SELECT SINGLE cannot be used, but SELECT ... UP TO 1 ROWS WHERE ... ORDER BY ... may be used.

back to top

Which is the better - JOINS or SELECT... FOR ALL ENTRIES...?

In most scenarios INNER JOIN performs better than FOR ALL ENTRIES, and should be used first. Only if there are performance issues should FOR ALL ENTRIES be considered, and careful measurements taken before and after to validate whether there really are performance gains. 

The effect of FOR ALL ENTRIES needs to be observed first by running a test program and analyzing SQL trace. Certain options set by BASIS can cause FOR ALL ENTRIES to execute as an 'OR' condition. This means if the table being used FOR ALL ENTRIES has 3 records, SQL Trace will show 3 SQL's getting executed. In such a case using FOR ALL ENTRIES is useless. However of the SQL Trace shows 1 SQL statement it's beneficial since in this case FOR ALL ENTRIES is actually getting executed as an IN List.

JOINS are recommended over FOR ALL ENTRIES. There is no real limit to the number of tables that can be joined; however greater complexity can make maintenance harder, and if there are problems with the join, make it harder to resolve them. If the JOIN is being made on fields which are key fields in both the tables, it reduced program overhead and increases performance. 

In some scenarios, you are presented with an internal table. In these situations, you may have no choice but to use FOR ALL ENTRIES.

Here is a code with join :

back to top

User Collect Statement to do Sum in the internal table.

Instead of using logic to do summation use collect statement. COLLECT is especially efficient with HASHED tables.

Avoid use of nested loops

For example: if there is a loop like this.  Condition added, otherwise there is no optimization:

in the production environment it may be possible that such a loop takes a lot of time and dumps.

Instead we can use ... BINARY SEARCH added, otherwise no improvement!!! Better still - use a HASHED or SORT TABLE.

If you have a sorted table - the internal table can be read like this:


  1. Guest

    Why is it necessary to use CHECK instead of IF/ENIF? It is more difficult to read.

  2. Guest

    I would not recommend usage of FOR ALL ENTRIES for large tables. If using FOR ALL ENTRIES than SAP BASIS should change application server default parameters of

    rsdb/max_blocking_factor = a
    rsdb/min_blocking_factor = b
    rsdb/max_in_blocking_factor = c

    See  SAP HINT 881083. If possible use joins instead!  FOR ALL ENTRIES waste network bandwidth, application server CPU and database server IO and CPU.

  3. Guest

    i have noticed in sap standard program using select and endselect which according to coding standards should not be used.

    can anyone tell me why SAP uses it?



  4. GURU, select ... endselect can be found in old programs because when SAP create new statements in the abap kernel, it would be dangerous and expensive to rewrite all the programs which were using the old statements they replace. You will find the same issue for many many statements (and technologies, like logical database for example)

  5. I just confirmed that SELECT COUNT(*) UP TO 1 ROWS...WHERE... is the best to check existance:

    SELECT SINGLE COUNT(*)...WHERE... measure:         318
    SELECT SINGLE key INTO...WHERE... measure:         177
    SELECT COUNT(*) UPTO 1 rows...WHERE... measure:         141
    SELECT...INTO...WHERE...ENDSELECT... measure:         170

  6. To be honest I find this page a bit poor. Too few information for such an important area.

  7. Guest

    I must say while using for all entries make sure of deleting duplicate entries from key table..

    it will effect the performance too much

  8. What about Like statement in select query ?

    -> In table MARA, we have a standard index 'MPN' for fields MFRPN and MFRNR.

    In most of the cases, we will have a select statement like SELECT ( FIELDS) FROM MARA WHERE MFRPN = LV_MFRPN.

    In this case if I use something like SELECT ( FIELDS) FROM MARA WHERE MFRPN = LV_MFRPN AND MFRNR LIKE '%'.

    Does this improve performance? Here by mentioning MFRNR in the select query we are matching the fields of Index. Will this work for performance ?

    -> Similarly SELECT ( FIELDS) FROM VBAP WHERE VBELN = LV_VBELN AND POSNR LIKE '%'. Will it do better than

  9. What is the Volume of data for which we need to go for Binary search and for the Hash table?