By Martin Grob


In reality dimensions in an InfoCube are often designed by business terms (like material, customer etc.) This often leads to the impression that InfoCube dimensions should be designed based on business constraints. This although should not be the leading criteria and shouldn't drive the decision. 
Aside from the datavolume which depends on the granularity of the data in the InfoCube, performance is very much depending on how the InfoObject are arranged in the dimensions. Although this has no impact on the size of the fact table it certainly has one on the size of the dimensions.

How is a dimension then designed?

The main goal distributing the InfoObjects in their dimensions must be to keep the dimensions as small as possible. The decision on how many dimension and what InfoObjects go where is purely technical driven. In some cases this matches the organisational view but this would only be a conicidence and not the goal.

There is a few guidelines that should be considered assigning InfoObjects to dimensions:
  • Use as many dimensions as necessary but it's more important to minimize dimension size rather than the number of dimensions.
  • Within the dimension only characteristics that have a 1:n relation should be added (e.g. material and product hierarchy)
  • Within a dimension there shouldn't be n:m relations. (e.g. product hierarchy and customer)
  • Document level InfoObjects or big characteristics should be designed as Line-Item dimensions. Line item dimensions are not a true dimensions they have a direct link between the fact table and the SID table. 
  • The most selective characteristics should be at the top of the dimension table
  • Don't mix characteristics with values that change frequently causing large dimension tables. (e.g. material and promotions)
  • Consider also to combine unrelated characteristics it can improve performance by reducing the number of table joins. (you only have 13 dimensions so combine the small ones)

As a help the report (SE38) SAP_INFOCUBE_DESIGNS can be used.
This yellow marked dimension should be converted into a line item dimension if it contains a document level characteristic or it is simply bad design.

The maximum number of entries  a dimension potentially can have is calculated through the cartesian product of all SID's. (e.g. 10'000 customer and 1'000 product hierarchies lead to 10'000'000 possible combinations in the dimension table. It's unlikely that this is going to happen and while designing the dimension this should also be considered - analyzing the possibilities of all customers buying all products in this case.
In cases where there is an m:m relationship it usually means there is a missing entity between those two and therefore they should be stored in different dimensions.
Once data is loaded into the InfoCube a check on the actual number of records loaded into the dimension table vs. the number of record in the fact table should be done. As a rule of thumb the ratio should be between 1:10 and 1:20.

Degenerated Dimensions

If a large dimension table reaches almost the size of the fact table when measured the number of rows in the tables it's a degenerated dimension. The OLAP processer has to join two big tables which is bad for the query perfromance. Such dimensions can be marked as Line Item Dimensions causing the database not to create an actual dimension table. Checking the table /BIC/F<INFOCUBE> will then show that instead of the DMID dimension key the SID of the degenerated dimension table is placed in the fact table. (Field name RSSID). With this a join of the two tables is eliminated. Those dimensions can only hold one InfoObject as a 1:1 relationship must exist between the SID value and the DIMID.
Dimensions with a lot of unique values can be set to High Cardinality which changes the method of indexing dimensions. (ORA DB only) This results in a switch from a bitmap index to a B-Tree index.
Defining a dimension as Line Item Dimension / High Cardinality


Finding the optimal model and balancing the size and the number of dimensions is a delicate excercise.
Dimensions in MultiProvider do not have to follow the underlying InfoCubes definitions. Those can be focused on the end users need and be structrured by the organizations meaning. This does not affect the performance as the MultiProvider does not have a physically existing datamodel on the database.    
Designing the dimension in an InfoCube correctly can have a significant improvement on performance!

By Mohanavel at

Objective:    Introducing time delay in the process chain with help of standard SAP provided program RSWAITSEC.

Background:  With the use of interrupt process type in the process chain we can give the fixed delay to the subsequent process types when the interrupt step is reached.  But when we use the standard interrupt process type we have to mention the date and time or event name.
In many cases interrupt step might not help, if suppose an interrupt step is introduce to delay the subsequent processes by a definite period of time, and if all the steps above to the interrupt gets completed early then instead of passing the trigger to the subsequent step after the desired wait time, the interrupt will force the chain to wait till the conditions in the interrupt are satisfied.
In order to achieve the delay in the trigger flow from one process type to another in a process chain without any condition for the fixed time limit or event raise we can use this RSWAITSEC Program.

Scenario:   In our project one of the master data chain is getting scheduled at 23.00IST. This load supplies data to the report which is based on 0CALWEEK. The data load and an abap program in the process chain make use of SY-DATUM, so a load that starts on sunday 23:00 if doesn't complete by 23:59:59 hours (1 hour duration) then the entire data gets wrongly mapped to the next week. .This will cause discrepancy of data.
So it was required to schedule chain at 23:00 IST everyday except sunday, and at 22:45:00IST (15mins earlier) on Sundays.

Different Ways to Achieve the above Situation:
1.        Creating two different process chains and scheduling the 1st process chain at 23.00 IST for Monday to Saturday (Using factory calendar), scheduling the 2nd at 22.45 IST only for Sunday.
Disadvantage of 1st method:
Unnecessarily creating two chains for the same loads, this makes way to have multiple chains in the system.

   2.   Scheduling the same chain at 22.45IST and adding the decision step to find and give interruption of 15mins for Monday to Saturday.  So on Sunday it will  get start at 22.45.

Process Chain with Interrupt Process Type:

Description: PC with Interrpt Step(1).jpg
Disadvantage of 2nd method:
If suppose you want execute this chain with immediate in other time, then my interruption step will wait until 23.00 IST to get start the load for Monday to Saturday loads.

Better Way of Achieving with RSWAITSEC program:

Scheduling the chain at 22.45 and adding the decision step to find whether its Sunday or other.  If Sunday then next step would be directly to the local chain, if the particular day is between Monday to Saturday then the next step would be with RSWAITSEC program(SAP std program).   In the program variant we have to mention the desired time delay in Secs(900 Secs).

     Compared to above two methods, this will be the  better way to achieve the desired output.  Even though if I run the process chain with start process as         immediate on other than Sunday’s my local chain will not wait until 23.00IST to reach, it will wait for 15mins and it will get triggered. 

       As this is the SAP provided one no need to move any TP for this, even in production we will be able to use directly.

Process chain with RSWAITSEC:
Description: PC with RSWAITSEC program(2).jpg

ABAP Process type with RSWAITSEC Program(which shown in the above PC):

Description: Program Variant for RSWAITSEC(3).jpg

Setting the Variant value (required time):

In the variant value we need to mention the desired limit of delay in Seconds.  My requirement is of with 15mins of delay, so I have given 900sec in the variant value.

Description: Variant Value Screen(4).jpg

So we can use this program in any stages of process chain to give the fixed period of delay.

Hope this will be helpful.

Dimension design: A different perspectiveObjective: The objective of this post is to simplify the understanding on dimension designs of an infocube and to decide upon the dimensions based on the repetition of the data held in the dimension tables.

Pre-requisites: An infocube is already created and active, and filled will data, which will be used for analysis of dimension tables.

Dimension to Fact Ratio Computation: This ratio is a percentage figure of the number of records that exists in the dimension table to the number of records in fact table or what percentage of fact table size is a dimension table. Mathematically putting it down, the equation would be as below:

          Ratio = No of rows in Dimension table X 100 / No of rows in Fact Table

Dimension Table Design Concept: We have been reading and hearing over and over again that the characteristics should be added into a dimension if there exists a 1:1 or 1:M relation and they should be in separate dimension if there exists a M:M relation. What is this 1:1 or 1: M? This is the relation which the characteristics share among each other.
For instance if one Plant can have only one Storage Location and one storage location can belong to only one plant at any given point of time, then the relation shared among them is 1:1.
If 1 Functional Location can have many equipment but one equipment can belong to only one functional location then the relation shared between the functional location and Equipment is 1:M.
If 1 sales order can have many materials and one material can exist in different sales orders then there absolutely is no dependence among these two and the relation between these two is many to many or M: M.

Challenges in understanding the relationship: Often we SAP BI consultants depend on the Functional consultants to help us out with the relationship shared between these characteristics / fields. Due to time constraint we generally cannot dedicate time to educate the functional consultants on the purpose of this exercise, and it takes a lot of time to understand this relationship thoroughly.

Scenario: An infocube ZPFANALYSIS had few dimensions which were way larger than the preferred 20% ratio. This had to be redesigned such that the performance was under 20% ratio.
This ratio could be either manually derived by checking the number of entries in the desired dimension table (/BIC/D<infocube name><dimension number>) to the fact table (/BIC/F<Infocube Name> or /BIC/E<Infocube name>) or a program SAP_INFOCUBE_DESIGNS can be executed in SE38 which reports this ratio for all the dimensions, for all the infocubes in the system.

We can find from the report that the total number of rows in the fact table is 643850. Dimension 2 (/BIC/DZPFANLSYS2) has around 640430 rows, which is 99% (99.49%)of the fact table rows and Dimension 4(/BIC/DZPFANLSYS4) has around 196250 rows, which is 30%  (30.48%)of the fact table rows.



Step 1: Analysis of the dimension table /BIC/DZPFANALSYS2 to plan on reducing the number of records.

Fact table:
Dimension table holds 1 record more than the fact table.
View the data in the table /BIC/DZPFANLSYS2 (Table related to Dimension 2) in SE12 and sort all the fields. This sorting will help us spot the rows which have repeated values for many columns, which will eventually lead to understanding the relationship between the characteristics (columns in dimension table).

Identifying the relationships:
Once the sorting is done we need to look out for the number of values that repeat across the columns. All the records which repeat could have been displayed in a single row with one dimension id assigned if all the columns had same data. The repetition is a result of one or more columns which contribute a unique value to each row. Such columns if removed from the table then the number of rows in the table will come down.

In the below screenshot I’ve highlighted the rows in green that were repeating themselves with new dimension IDs, as only 2 columns SID_ZABNUM and SID_0NPLDA have new values for every row. These two columns having new values for every row have resulted in rest of the columns repeating themselves and in turn increasing the data size in the dimension table. Hence it can be easily said that these two columns do not belong in this dimension tables, so the related characteristics (ZABNUM and 0NPLDA) need to be removed out of this dimension.
Few rows could be found which repeat themselves for most of the columns, but have a new value once in a while for some columns, as highlighted in yellow in the below screenshot. This indicates that these columns share a 1:M relation with the rest of the columns with repeated rows and these could be left in the same dimension.
Conclusion: The columns marked in green belong to this dimension tables and the columns marked in red needs to be in other dimension tables.
Step 2: Create a copy infocube C_ZPFAN and create new dimensions to accommodate ZABNUM and 0NPLDA.
ZABNUM was added to dimension C_ZPFAN8 and 0NPLDA was added to C_ZPFAN7. These were marked as line item dimensions as they have only one characteristic under them.
Analysed the issue with dimension 4 in the similar way and changed other dimensions to help the situation.

Post changes, loaded the data into the copy infocube C_ZPFAN and found the number of records in the dimension table /BIC/DC_ZPFAN2 to be 40286.

Ratio: 40286 / 657400 * 100 = 6.12 %


Dimension2 of the copy infocube: /BIC/DC_ZPFAN2
Even now there a few repeated rows and columns, but the ratio is within 20%. We can create up to 13 dimensions, but it is always better to keep a dimension or two free for future enhancements.

Hope this was helpful.