SDP94: Performance Tuning Benchmarks

[Former Member]

Dear APO Admin/DP Experts,

I have a generic question related to loading and drill down time in interactive demand planning in SCM7.0

I do appreciate there are several parameters responsible for the response time, I wish to know from your personal experience, what is the possible response time for the actions below.

Assume

Key Figures in Data view: 12

Characteristics in Planning book: 8

CVC's: 100K, with product being one of the char

Storage buckets: 48 months

Time buckets: 36 past months and 12 future months.

Brand new SCM7.0, Ehp1

Sufficient hardware

Sufficient work processes.

Sufficient licenses.

5 users are logged in SDP94

Action1: I load shuffler with all Products .1000 products.

What should be the ideal response time for loading these products to the grid

Action2: I now drill down product in the header.. multiple.. all 1000

What should be the ideal response time ?

You can give any other example. I wish to know is there something wrong with our environment or everyone has had complaints wrt response time esp. in Production environment. In Action 2, I note a response time of 28 minutes. This is pathetic and impractical. I had to increase the dialog response time to do actually complete this.

Thanks

Guru

[Former Member · Accepted solution]

Hi,

to my knowledge, Action2 is impractical in APO (in the same way than one avoids checking visually 1000 rows in an Excel sheet plus the performance issue). How does a user navigate through 1000 products in the same screen? In this case, one either:

1) breaks down the products in families and check by family (if you select 100 products and then drill-down, it should be much faster)

2) use DB alerts to check only the interesting products via /SAPAPO/AMON1 (plan by exception). Examples: forecasting error, anomalies (negative forecast...), this kind of stuff.

Even in a background job your drill-down action would consume a lot o ftime. If you really need to drill-down all your products simultaneously, you may try to save some time by hiding key figures you won't be using, hiding graphics or creating an aggregate in your MPOS at the product level (http://help.sap.com/saphelp_SCM700_ehp02/helpdata/en/59/a3453ff168e07fe10000000a114084/frameset.htm, check OSS 503363), so liveCache will read directly the product-level numbers instead of aggregating through the whole database.

thanks,

J.

[Former Member · Accepted solution]

Thanks James for your prompt response. Its another story, how miserable does demand planner make his/her life with lots of products and key figures in SDP94. There could be a functional purpose for doing so with some or all key figures. esp. when in a 3 hour meeting, planners need to decide the demand plan for 4400 products (thats the number of products being planned actually) for next n periods. A 7,50,000 row excel sheet I made this morning took less then 10 seconds to execute with 2 dozen formulas in it. So Excel is clearly a superior user friendly tool for demand planners, performance of DP not withstanding.

My question was on what should be the "ideal" response time for actions above in SDP94 because I need to test the worst possible use case of demand planning platform by the planners. They could be super efficient and smart enough to scroll through 1000 products or they could be dumb but as a "consultant" I would hate to give them reasons for slow performance after go-live esp. when they are anyway going to download everything to excel and work peacefully and then copy past the results back to SDP94. To download data first SDP94 needs to be loaded and then drilled to whatever level users fancy.

Thanks

Guru

[Former Member · Accepted solution]

Hi Gulab,

I wouldn't say it is clearly superior (at least if you want dissagregation, fixing, or if we compare standard APO DP with standar forecasting functionality in Excel - no models, no outlier detection, sharing data among different people without emailing excels, etc...), but I absolutely get your point.

Background processing of your 2 dozen formulas on your 5000 products would be a matter of seconds if parallelized (may be 20, but close to your 10), what you can't get is good performance for seeing everything at the same time in SDP94, it is absolutely true it is not designed for that.

If your users are going to upload only forecast, did you create a planning book with a single key figure (forecast, or forecast and history), to speed up the process? Or do they really need the 12 keyfigures? Naively one would expect the system to go 10 times faster with a single keyfigure, I guess  this would be the most important parameter for you. Also for downloading, if you don´t need the future, you could create a planning book with only your 36 months of past history (1/4 time reduction).

And for your numbers:

- action 1 took me 5 seconds for 1700 products. 

- action 2: I never try to drill-down more than 400-500 products, above this I usually experience huge response times.1 key figure, 12 months, 15 seconds. Extrapolation to 48 months gives 1 minute, for 12 keyfigures roughly 15 minutes. So your performance seems similar.

regards,

J.

[Former Member · Accepted solution]

Guru,

My question was on what should be the "ideal" response time for actions above in SDP94

I don't know (and can't know) what it would be for your implementation. 

I believe that Action1, with proper design of the PA, should be possible in a time frame that is acceptable to most users, providing there are no default macros. 

I agree with expert James where he considers that Action2 is generally impractical from a business user's perspective.  For Action2, depending upon the number of macros that  have been implemented, I have seen similar levels of performance to yours in productive systems.  I have also seen better performance as well. 

It is fruitless for you to ask us such questions, because in the end it has no impact upon whether or not your client is satisfied.  If you need a detailed analysis of the performance of a DP solution in the client's system, and which steps you can take to satisfy the client's requirements, you must commission an expert to analyze and optimize your system and your solution.  If such an expert does not exist on your team, then SAP (and some consulting companies) will perform this analysis for you on a consulting basis.

A 7,50,000 row excel sheet I made this morning took less than 10 seconds to execute

SAP DP is not Excel! Do not expect that because DP displays are 'grid based' like Excel that DP should act like Excel.  Such comparisons are irrelevant, and it would be a disservice to the client for a consultant to raise them. In the background behind DP, you will find a collection of relational databases, not a single table as you would find in a simple Excel sheet.

FYI I personally have created an Excel book, containing multiple lookups against multiple sheets (similar to DP functionality), that takes an hour to perform a single calc, and with less rows than your example.  That doesn't mean that Excel is slow, it only means that I created a design that was impractical for Excel.

...when they are anyway going to download everything to excel and work peacefully and then copy paste the results back to SDP94.

If this is your expectation, then you should consider creating a standalone download/upload functionality, rather than asking the users to execute //SDP94; drill down; and export/import.

Best Regards,

DB49

[Former Member · Accepted solution]

Thanks James for providing some reassuring numbers. I was only keen on order of magnitude response times as I have not been a part of any other productive DP before.

Thanks DB49 for providing some clues. The reason I provided asssumptions above was to limit further assumptions. Surely performance can differ from system to system. My customer cant afford another tuning consultant at this moment so I will make do with what most "functional" consultants do best...offer some lip service and contain the design.. limited key figures per view, only required initialization of storage buckets, parallelize jobs and process chains and such.. which is all in place.

Yes by now, with whatever few screens I dwelled on, I did realize DP is not Excel but Excel can be a superior and efficient DP tool if used well.

Thanks for your responses. I am sinking it all in

Regards

Guru