Interesting performance issue with String functions

[Former Member]

Hi All,

I was working on a performance issues on a 7.02 system and discovered that a very small change in the code made a dramatic difference to the performance. I have created a very simple program to illustrate this. The program is attached.

Number of records	Version one	Version two	% Change
100	239	161	33%
1,000	7626	1590	79%
10,000	945,304	16,783	98%
100,000	169,005,062	860,739	99%

I have sent the program to two of the SAP ABAP Mentors, Graham Robinson and Thomas Jung. Graham has run the program on a 7.31 system and had similar results. Thomas is fortunate to a 7.40 system and his results show the opposite result but the difference were very small.

I am interested to hear if anybody has an explanation as to why there is such a dramatic difference in performance.

Regards

Brent Talbot

[Former Member]

The version 1 code is:

  LOOP AT lt_data REFERENCE INTO lr_data.

    lv_html_data =

      lv_html_data &&

      |<OPTION from="unalloc" VALUE="{ lr_data->anumber }" birthid="">{ lr_data->anumber }|.

  ENDLOOP.

The Version 2 code which runs significantly faster is

LOOP AT lt_data REFERENCE INTO lr_data.

    lv_html_data2 =

      |<OPTION from="unalloc" VALUE="{ lr_data->anumber }" birthid="">{ lr_data->anumber }|.

    lv_html_data3 = lv_html_data3 && lv_html_data2.

  ENDLOOP.

[volker_borowski2]

Hmm,

I think in version one to make the "&&" operator work, you need to build an intermediate expression result for the part to the right of "&&". It has to be stored somwhere, so that "&&" can be processed afterwards, and as soon as the expression is calculated, this interimspace has to be deallocated to avoid a memory leak. Sounds pretty expensive to me, esp. on unicode systems.

In Version 2, the compiler probably recognizes, that the storage will be in lv_html_data2 (left of =), so no need for dynamic allocation and de-allocation of the interim result for concatenation. The target lv_html_data2 stays static.

The memorymanagement for the target, which gets longer in each loop is the same in both cases.

But it is just guessing, I have no trace for proof.

As far as the compared systems are involved, any unicode / non-unicode derivations?

Volker

[former_member183770]

In 7.40, the ABAP runtime goes to greater effort to determine that there is no right-hand-side occurrence of the target variable and hence it can be appended directly. (If there was a rhs occurrence, direct append would be invalid.) In 7.20, this recognition only works for the simple form in version 2, and version 1 has quadratic run time, as speculated by Volker.