on 07-01-2007 5:39 PM
Hello,
We are on BW-BPS 3.5 system.
After converting a floating point format into decimal one, all exit functions on our planning sequence are taking a long time to run and we noticed a lot of swaps.
A performance issue was traced for loop at xth_data instruction.
We had just changed the format of floating key figures on our infocubes.
Can this change have a bad impact on the runtime execution and what could we do ?
Thanks a lot for any answer.
Valerie RENEE.
Many Thanks Marc for your replies and your advice.
Valerie RENEE.
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.
Hello Marc,
The performance issue occurs in an exit module function executed in batch mode.
LOOP AT xth_data INTO <ls_data>.
All elements of planning package are to be changed.
The trace gives a total of 230000 records (in debug) to be performed. Inside this first loop, there is another loop which deals with currencies.
When team asked for help they just said nothing has changed but floating point format on key figures (preproduction system). So I wanted to figure out any lack in customizing or indices generation. They also noticed the same degradation on production system therefore they never changed the initial format on production system.
Actually team said that they probably made a bad estimation of the total of records in this loop. They are looking forward. It was the first time they had such a volume.
Anyway thanks for any information on floating point issues and internal table xth_data with BPS .
It seems to be difficult to change this format.
Thanks a lot.
Valerie RENEE.
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.
Hi Valerie,
230000 is certainly a lot of data and your ABAP coding needs to be very efficient to process it quickly.
You write "they said" and "probably". Get facts. How fast was it before? How fast is it now? Is the test case still the same? Anyway, SE30 is the right tool to analyze the issue.
I see that field-symbols are used. These are inherrently slow. If the landscape (dev, qa, prod) is using the same client, then you can replace the field-symbols with the generated structures. For example use ls_data type /1sem/_ys_data_100infocube instead of <ls_data> type any. You can then remove all the assigns and refer to the characteristics and key figures directly (ls_data-s_chas-0comp_code), which will speed up the coding.
Regards,
Marc
SAP NetWeaver RIG
Hello Valerie,
this change could have an impact since BPS works internally with floating point numbers in some places. However, you said you did already a trace. Which program in the loop at xth_data in?
Regards,
Marc
SAP NetWeaver RIG
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.
User | Count |
---|---|
87 | |
23 | |
11 | |
9 | |
8 | |
5 | |
5 | |
5 | |
5 | |
4 |
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.