Performance issue in CIF integration models

[Piyush]

Hi Gurus,

We are implementing SAP SCM GATP with SAP IS Retail. The typical characteristics of retail industry like

a.     No Manufacturing activity
b.     Very high no. of SKUs
c.     Very high no. of Plants [DC and Stores]
d.     More no. of MARC table entries
e.     More no. of sales orders as well as purchase orders cycles

Currently we are doing performance testing. So applying the full data load in the system

[Nearly 30000 SKUs and 126 Plants; DCs + Stores]

During Integration Models creation, I had observed that system is taking too much time. In some cases it is taking more than 40000 seconds [More than 10 hours] for CFM1 transaction. I had scheduled the backorder jobs for the same as it is giving dump in the foreground. Whereas CMF2 is not taking much time and completes within few seconds to few minutes.

1. Customization                                - No performance issue

2. Plant                                               - No performance issue

3. Plant - Product                                - No performance issue

4. Plant - Product - ATP                      - No performance issue

5. Customer                                        - No performance issue

6. Plant - Product - Stock                    - Performance issue

7. Plant - Product - Sales order           - Performance issue

8. Plant - Product - Purchase Order    - Performance issue

In this process I had deactivated and deleted some these integration models. Then I had recreated these IMs. During recreation I had noticed that some of the IMs had taken only couple of seconds. Initially these IMs had taken > 10000 seconds in the background processing. While remaining IMs had taken the similar time for recreations.

I am unable to predict the behaviour of the system. I am having couple of questions -

1. Is SAP SCM APO is capable enough to handle such a large volume of data?

2. Why there is abrupt change in the processing time for CFM1 for same integration models?

3. Once the initial load is performed via CFM1 and CFM2, do I need to repeat the same process? In other words, what

    should be ideal frequency to run CFM1 and CMF2? I have read somewhere these IMs are to be recreated everyday.

4. What about the new article creation?

5. How the performance could be improved further?

6. If any one has experience of GATP implementation with IS retail then requesting you to please share your experiences

    as well as issues.

Thanks to you all in advance.

Piyush

[Former Member · Accepted solution]

Hi Piyush,

Have you generated runtime version and did a consistency check via program RCIFIMAX

As i remember, we also performance issue in the past, what we did was that, we broke the materials based on Material range and add the major DCs seperately. just try running CFM1 after splitting in to a few IMs , may be you can see the difference.

These are the programs which you need to run over the weekend / month

RAPOKZFX - Program for Adjusting Field MARC-APOKZ With Integration Models

RCFUPDCR - Reorganization of superfluousCIF update counters           

RCFORDCH - Check for Inconsistencies inTable CIFORDMAP

RCIFIMAX - Generate and reconcile the runtime version of active models          

RCIFIMDL - Deletion of superfluous filter object keys in CIF_IM* tables          

RCIFMTDE - Program for Maintaining the Materials Deactivated in CIF          

RDELALOG  - Delete Application Log Entries

3. Once the initial load is performed via CFM1 and CFM2, do I need to repeat the same process? In other words, what

    should be ideal frequency to run CFM1 and CMF2? I have read somewhere these IMs are to be recreated everyday. - the frequency could be weekend, you need to handle the new material creation / material extension which is done every day by extracting from MARD table or you can create a program and schedule the jobs to handle delta.

[Former Member · Accepted solution]

please also check this thread

http://scn.sap.com/thread/1080954

http://scmgmt.blogspot.in/2010/01/note-563806-faq-apo-cif.html

[Piyush · Accepted solution]

Hi Rorie,

Many many thanks for the reply.

I had executed these programs in the system. It had helped me to improve the performance significantly.

Particularly RDELALOG  - Delete Application Log Entries, had contributed significantly.

I had analyzed further. Now I am not facing much issues in the IM creation [RIMODGEN].

I am facing issues in the IM activation [RIMODAC2].

For example, a batch job of Material IM with 370,296 records had taken 32 seconds to complete IM creation [RIMODGEN]. While it had taken 3,863 seconds to complete IM activation [RIMODAC2].

So my questions are,

1. Does the RIMODAC2 always take such a significant time compared to RIMODGEN?

2. Is there any cap or limit for maximum no. of records in the particular IM?

Thanks & Regards

Piyush Ranpura

[Former Member · Accepted solution]

Hi Piyush,

What are your settings in RIMODAC2? are you using parallelize selection and are you using any special CIF settings?

RIMODAC2 should not take much time. Initially it can.

i think you should break that 370,296 records in to 2 IMs.

[Piyush · Accepted solution]

Hi Rorie,

Thanks for your prompt reply.

I am already having more than 350 IMs which are active. There are 5 million + entries in the MATLOC table.

Anyway, thanks for your reply again.

One question-

I am having an issue with the new article creation. If I have to go for the new article IM then I have to go for following steps-

1. CFM1 [IM Creation]

2. CFM3 [IM activation]

For article, I am going to maintain article field 'blank'. I will be the maintaining filter for article type[MARA-MTART] and article status. [MARC-MMSTA]. I don't want to perform the CFM1 step.

Only deactivation and reactivation of the existing IMs should be done. By this means I want the APO indicator of the article [MARC-APOKZ] for the new article to be 3.

Is it possible to avoid the CFM1 step for new article creation?

Thanks & Regards

Piyush Ranpura

[Former Member · Accepted solution]

Hi Piyush,

without fetching the data in to CFM1 and Generating an IM , there is no way you could send the material to APO. you cannot avoid CFM1.

Regards,

[babu_kilari4 · Accepted solution]

Fully agree with Rorie here. You need to run RIMODGEN program at any cost to "Generate the model" to pick the new material and then we run RIMODAC2 to activate the model.

Babu Kilari