Hi CPU usage in ABAP coding

[peter_strauss]

Hello,

I have a custom program (programmed by somebody else) which is churning away all day long and consuming between 50% and 70% CPU as seen in the 'top' command. I'm on HP-UX with 8 CPUs installed. The CPU usage is consistently at this level. (I matched the PID from SM50 with the PID in 'top')

I ran abap traces in ST12 which found that between 60% and 70% of the time was spent in the following coding:

773 IF SY-SUBRC <> 0.

774 LOOP AT GIT_ZTABLEA TRANSPORTING NO FIELDS

775 WHERE KUNNR = PC_KUNNR "CUSTOMER CODEﾞ

776 AND ZZVCGRTNFG IN P_DRTNFG "FLAGﾞ(SELECTION CRITEREA)

777 AND ZZVCGSTPFG IN P_DSPDFG . "QUIT FLAGﾞ(SELECTION CRITEREA)

778 EXIT.

779 ENDLOOP .

780

781 *

782 PI_SUBRC = SY-SUBRC .

783 RETURN .

784

785 *

786 ELSE.

787 LOOP AT GIT_ZTABLEA TRANSPORTING NO FIELDS

788 WHERE ZZVCGGRPID = GFC_TABLEB-ZZVCGGRPID "GROUP ID

789 AND ZZVCGRTNFG IN P_DRTNFG "FLAGﾞ(SELECTION CRITEREA)

790 AND ZZVCGSTPFG IN P_DSPDFG. "STOP FLAGﾞ(SELECTION CRITEREA)

791 EXIT.

792 ENDLOOP .

I guess the idea here is to just find the number of rows in GIT_ZTABLEA for certain conditions.

Is there a simple and less intensive way of achieving this?

(apologies in advance for foreseen dumb questions. I'm very new to coding).

Peter

[Former Member]

use internal table of type sorted

for your loop that will reduce cpu usage

Regards,

Alpesh

[Former Member]

> LOOP AT GIT_ZTABLEA TRANSPORTING NO FIELDS

> 775 WHERE KUNNR = PC_KUNNR "CUSTOMER CODEﾞ

> 776 AND ZZVCGRTNFG IN P_DRTNFG "FLAGﾞ(SELECTION CRITEREA)

> 777 AND ZZVCGSTPFG IN P_DSPDFG . "QUIT FLAGﾞ(SELECTION CRITEREA)

sorted table ... is the usual answer but not in this case.

You should check the IN P_DRTNFG "FLAGﾞ(SELECTION CRITEREA) ?

What does it mean? I don`t undterstand the flag.

With the IN conditions I doubt that there is any optimziation possible, it will always scan.

If the two IN-flag are really flags then it should be possible to write equal condition

and get something

WHERE k = a AND z1 = b AND z2 = c.

The can be solved with sorted tables.

And from the coding you show with will exit after the first solution, so a READ would then also

be enough. The LOOP is only used because the WHERE is only possible with the LOOP.

but a LOOP is not really necessary because it check only the first solution.

So please revise the logic.

Then it will become really fast.

Siegfried

[Former Member]

Is this code called from within a loop or is it in a form called from within a loop?

Rob

[romanweise]

Hi,

i am a bit confused about the IN statements. The fields are describes as Flag, a flag is usually yes or no why using an in statement there?

Furthermore sorting the table could add to performance if the table is sorted by customer code as sorted tables look for the start dataset in binary search when the key or a key part is specified.

Rgds.

Roman

[Former Member]

that is the more general case of a often repeated recommendation.

sort GIT_ZTABLEA by ZZVCGGRPID.
....

read table GIT_ZTABLEA transporting no fields  
                  with key ZZVCGGRPID = GFC_TABLEB-ZZVCGGRPID.

IDX = SY-TABIX.

loop at GIT_ZTABLEA into WA from index IDX   

   if (        WA-ZZVCGGRPID = GFC_TABLEB-ZZVCGGRPID 
        and WA-ZZVCGRTNFG IN P_DRTNFG
        and  WA-ZZVCGSTPFG IN P_DSPDFG ). 
*  success  -> exit ???
   endif.

   if not ( WA-ZZVCGGRPID = GFC_TABLEB-ZZVCGGRPID.
      exit.
   endif.
endloop.

there is something missing, in your posting, you don't discriminate between the cae where something is found and nothing is found

Siegfried

[Former Member]

you can replace the above code like this but the thing is i dont see you keeping any record with any of these loops.

data: lv_index1 type  sy-index,
        lv_index2 type sy-index.

sort GIT_ZTABLEA by KUNNR ZZVCGRTNFG ZZVCGSTPFG

loop at GIT_ZTABLEA from lv_index1.

if GIT_ZTABLEA-KUNNR = PC_KUNNR or  ZZVCGRTNFG in GIT_ZTABLEA  or ZZVCGSTPFG in GIT_ZTABLEA.
lv_index1 = sy-index.
exit.
endif.

endloop.

Nafran.

[Former Member]

> you can replace the above code like this but

no, he can't oyur solution is simply wrong!

[Former Member]

ya both your and my code is wrong i guess coz even in yours the index is not updated and after looking at the main code again i think this will be the correct code

data: lv_index1 type  sy-index,
        lv_index2 type sy-index.
 
sort GIT_ZTABLEA by KUNNR ZZVCGRTNFG ZZVCGSTPFG

loop at GIT_ZTABLEA from lv_index1.
 
if ( GIT_ZTABLEA-KUNNR = PC_KUNNR and  ZZVCGRTNFG in GIT_ZTABLEA  and ZZVCGSTPFG in GIT_ZTABLEA. )
lv_index1 = sy-index.
exit.

endif.
 
endloop.

[Former Member]

> ya both your and my code is wrong i guess coz even in yours the index is not updated and

> after looking at the main code again i think this will be the correct code

nonsense, my code is correct, while yours is still wrong

You aussume that both tables are sorted in the same way and that both contain the same keys,

both anssumptions are wrong!

If a record is missing then you will never determine a new index!

The READ BINARY SEARCH determines the index, and that is the main point of the optimization.

Please think twice before you want to correct me.

Siegfried

[peter_strauss]

Thank you all. Even the "not so good" advice has been great.

It's good to know that the code does have room for improvement; it will take me some time to understand all the advice however I got exactly what I was looking for.