TDWI World Conference Spring 2005

Transcription

1 TDWI World Conference Spring 5 Baltimore 18 May 5 Multidimensional Data Model of the SAusiness Information Warehouse How to build good performing data models with SAW Dr. Michael Hahne Dr. Michael Hahne 5 1 Agenda Architecture of the SAusiness Information Warehouse Extended Star Schema of the SAG Variants for modeling hierarchical dimension structures Temporal aspects and time stamping Modelling guidelines Graphical model representation with Visio Dr. Michael Hahne 5, TDWI WC Spring Blt 05 2

2 Agenda Architecture of the SAusiness Information Warehouse Extended Star Schema of the SAG Variants for modeling hierarchical dimension structures Temporal aspects and time stamping Modelling guidelines Graphical model representation with Visio Dr. Michael Hahne 5, TDWI WC Spring Blt 05 3 Architecture of SAP Business Information Warehouse Metadata exchange Third-Party- Tools Business Explorer Open Hub Service XML BAPI ODBO XML Administrator Workbench BW-Server OLAP-Processor Info-Cubes Administration Scheduling Metadata- Manager Data-Manager ODS Monitoring Staging-Area PSA XML FILE BAPI Service API DB Connect XML-Data Flat Files Non-SAP- Applications SAP R/2 SAP R/3 SAW DB Dr. Michael Hahne 5, TDWI WC Spring Blt 05 4

3 Data flow and integration architecture Info-Cubes Business-Rules Sales data update rules aggregation business consolidation ODS Transformation update rules Sales actual Sales budget transfer- and update rules Sales market data consolidation of different data sources PSA Extraction Sales data EMEA Sales data USA Sales budget Source systems Sales market data nontransformed source data Dr. Michael Hahne 5, TDWI WC Spring Blt 05 5 Types of Info-Cubes Form of the InfoCube, which contains a part of data related to a closed business area and is physically stored. Info-Cubes in the BW-Server Basis-Cubes Abfrage-Schicht ETL-layer Front-End periodical data acquisition Source systems Multi-Cube Remote-Cube drill through Superordinate Cube, which places the data from several cubes into a common context. It contains even no data! Transaction data aren t administered in the BW in this case, they are administered separate externally. In the BW is defined only the structure of the remote cubes. For reporting, the data is transferred via a BAPI into the BW. Dr. Michael Hahne 5, TDWI WC Spring Blt 05 6

4 Agenda Architecture of the SAusiness Information Warehouse Extended Star Schema of the SAG Variants for modeling hierarchical dimension structures Temporal aspects and time stamping Modelling guidelines Graphical model representation with Visio Dr. Michael Hahne 5, TDWI WC Spring Blt 05 7 Data targets in SAP BW Data targets are objects, in which transactional data is stored for the purpose of reporting and analysis. The most important are: Info-Cubes ODS-Objects Additionally there are further data targets in BW,which enable for example direct master data reporting (Info-Sets, Info-Objects) Dr. Michael Hahne 5, TDWI WC Spring Blt 05 8

5 Definition Info-Cube The InfoCube is acentral data storage, on which are based reports and analyses in SAP BW. It contains a delimited data volume for example of a specific well-defined business area or business unit. InfoCubes contain two data types: measures and characteristics. The term InfoCube designates atable structure, in which some relational tables are linked in the sense of the so-called Star Schema. (multidimensional data storage) Star Schema: Dimension tables are grouped star shaped around a central fact table. Dr. Michael Hahne 5, TDWI WC Spring Blt 05 9 Star Schema The Star Schema is the most frequent kind of representing multidimensional data structures in relational data bases. In the Star Schema facts are stored in a seperate fact table, whereas the characteristics are grouped in Dimension tables. The dimension tables are joined to the fact table with foreign key and primary key relationships (DIM ID). In this way all data records from the fact table are marked uniquely by a value combination of these foreign keys from the dimension tables. Dr. Michael Hahne 5, TDWI WC Spring Blt 05 10

6 Pros and cons of the general Star Schema Good performance with the analysis of data Very flexible when adding characteristics and measures Problems come along with - N: M relatonships and - unbalanced (unragged ) hierarchies because of the uniqueness of the primary keys in the dimension tables Therefor the SAG decided to extend the Star Schema. Master data is stored seperate and independent from InfoCubes in the so called Extended Star Schema. Dr. Michael Hahne 5, TDWI WC Spring Blt Extended Star Schema The Extended Star Schema enables access to: Master data tables and their corresponding attributes Text tables with extensiv multilingual captions External hierarchy tables for the structured data access Master data and hierarchy tables are joined to the fact table via the SID- Tables (pointer tables) and the dimension tables Dr. Michael Hahne 5, TDWI WC Spring Blt 05 12

7 Concept of master data Attributes Customer: Customer group Texts Customer : Language,caption Hierarchies Customer: Customer hierarchy Dimension Customer Info-Cube Dimension Time Attributes Time: Public holiday Fact table of the Info-Cube Dimension table Customer Dimension table Time Texts Time: Language,caption Dimension table Hierarchies Time: Calender hierarchy Attributes : Brand, Category Texts : Language,caption Hierarchies : hierarchy Dimension Dr. Michael Hahne 5, TDWI WC Spring Blt Connecting master data to cubes via SID-Tables C P T sold units revenue Fact table Info-Cube C SID-Cust SID-Group SID-Branch SID- Corp Dimension table Time Dimension table Dimension table Customer SID-Cust. Cust. SID-Region Cust. Text SID-Group Group Lampen Müller SID-Region Region 23 S5 SID table for attributes SID table for attributes Text tables for (language dependent) captions Region S5 Text Süd 5/Frankfurt SID table for attributes Master data Dr. Michael Hahne 5, TDWI WC Spring Blt 05 14

8 Different master data tables C SID-Cust. SID-Group SID-Branch SID-Corp Dimension table Customer S SID-Cust. Cust SID table BIC/SKunde Standard SID-Table P Q Cust. Cust. name Car wash Smith Cust DateFrom... DateTo... Master data tablebic/pkunde fornot time-dependent display attributes Region Master data tablebic/qkunde for time dependent Süd 5/Frankfurt display attributes X SID-Cust. Cust SID-Region 23 SID table BIC/XKunde for not time-dependent navigation attributes Y SID-Cust. Cust. DateFrom DateTo... SID-Cluster 12 SID table BIC/YKunde for time dependent navigation attributes Dr. Michael Hahne 5, TDWI WC Spring Blt Hierarchy tables Dim-ID SID-Cust 522 SID-Grp. 170 SID-Branch SID-Corp. Dimension table Customer S SID-Cust. 522 Customer SID table /BIC/SKunde Standard SID-Table I SID of Hierarchy 234 pred -21 succ 522 Hierarchy table /BIC/IKunde Parent-Child-Tuple of the hierarchies K SID of Hierarchy 234 node Cust.Group A SID -21 Hierarchy table /BIC/KKunde Text nodes of the hierarchies Dr. Michael Hahne 5, TDWI WC Spring Blt 05 16

9 Line-Item dimensions Fact table C P T Sold units revenue Fact table C P T Sold units revenue C Dimension table Customer SID-Cust. 522 Line-Item: Dimension table Is left out SID-Cust. Cust SID-Region 23 SID-Cust. Cust SID-Region 23 SID table for attributes SID table for attributes SID-Region Region 23 S5 SID table for attributes SID-Region Region 23 S5 SID table for attributes Dr. Michael Hahne 5, TDWI WC Spring Blt Complexity of the extended Star at a glance InfoCube (1) Fact-Table (2) Dimension tables (3) time independent SID time dependend SID conventional SID (4) SID Attributes Dr. Michael Hahne 5, TDWI WC Spring Blt 05 18

10 Agenda Architecture of the SAusiness Information Warehouse Extended Star Schema of the SAG Variants for modeling hierarchical dimension structures Temporal aspects and time stamping Modelling guidelines Graphical model representation with Visio Dr. Michael Hahne 5, TDWI WC Spring Blt Dimensions and balanced hierarchies dimension elements derived resp. aggregated elements level (of a consolidation tree) base elements resp. independent elements granularity Dr. Michael Hahne 5, TDWI WC Spring Blt 05 20

11 Unbalanced hierarchies derived resp. aggregated elements dimension elements level (of a consolidation tree) base elements resp. independent elements granularity Dr. Michael Hahne 5, TDWI WC Spring Blt Attributes Attributes of Dimension Dimension time Attributes of all dimension elements Attributes of level year year Attributes of elememts of level year Attributes of level month month Attributes of elements of level month Attributes of level day day Attributes of elements of level day Dr. Michael Hahne 5, TDWI WC Spring Blt 05 22

12 Hierarchies within a dimension via characteristics Country Region Customer SIDs Dimension table Characteristic Customer Characteristic Region Characteristic Country Dr. Michael Hahne 5, TDWI WC Spring Blt Hierarchies within a dimension via characteristics each level is represented by an InfoObject number of levels should be fixed generally faster than attributes and external hierarchies include the higher hierarchical levels to aggregates no predefined drill down paths Dr. Michael Hahne 5, TDWI WC Spring Blt 05 24

13 Navigation attributes as basis of hierarchical structures Country Region Customer SID Dimension table Characteristic Customer SIDs Master data Attributes Attribute Region Attribute Country Dr. Michael Hahne 5, TDWI WC Spring Blt 05 Navigation attributes as basis of hierarchical structures each level is represented by an InfoObject number of levels should be fixed include the higher hierarchical levels to aggregates no predefined drill down paths bad Performance without aggregates increased flexibility for reorganisation Dr. Michael Hahne 5, TDWI WC Spring Blt 05

14 External hierarchies in BW edges Customer Dimension table SID Characteristic Customer Child Parent Master data table Hierarchy (inclusion table) Text nodes Dr. Michael Hahne 5, TDWI WC Spring Blt 05 External hierarchies in BW Reasonable in the case of frequent changes of dimension structure Enables unbalanced structures Several hierarchies possible per Info-Object Poor performance similar to navigational attributes Problems with big hierarchies In the case of time dependency only time stamping the whole structure enables aggregates Dr. Michael Hahne 5, TDWI WC Spring Blt 05

15 Agenda Architecture of the SAusiness Information Warehouse Extended Star Schema of the SAG Variants for modeling hierarchical dimension structures Temporal aspects and time stamping Modelling guidelines Graphical model representation with Visio Dr. Michael Hahne 5, TDWI WC Spring Blt Time dependency: changes in consolidation trees PG 1 PG 2 change of structure PG 1 PG 2 P 1 P 2 P 3 P 4 P 5 P 4 deleted P 6 added P 3 changed P 1 P 2 P 3 P 5 P 6 Dr. Michael Hahne 5, TDWI WC Spring Blt 05 30

16 Example of Slowly Changing Dimensions dimension in Fact table P E dimension in (changed) (new) P E Period Dr. Michael Hahne 5, TDWI WC Spring Blt Reporting requirements -Szenarios Reporting scenario actual structure Rev. 300 Rev. 400 Reporting scenario old structure Rev. Rev. Reporting scenario historical truth Rev. Rev. 400 Reporting scenario comparable results Rev. Rev. Dr. Michael Hahne 5, TDWI WC Spring Blt 05 32

17 Scenario I : Report with actual structure dimension in Fact table Period P E (changed) (new) P E Dr. Michael Hahne 5, TDWI WC Spring Blt Query path actual structure with navigational attributes S-table of SID 4711 SID X-table of P E SID 4711 DIM ID 29 Fact table Period SID DIM ID Dimension Table Dr. Michael Hahne 5, TDWI WC Spring Blt 05 34

18 Query path actual structure with external hierarchy K-table of SID nodename Fact table -2-3 I-table of pred succ DIM ID 29 Period SID DIM ID group Dimensions-Tabelle Produkt Dr. Michael Hahne 5, TDWI WC Spring Blt Szenario II : Report with old structure Fact table dimension in Period P E Rev. Rev. Dr. Michael Hahne 5, TDWI WC Spring Blt 05 36

19 Query path old structure with time-dep. nav. attributes S-table of productgroup SID 4711 SID Y-table of product P E SID Query key date DateFrom DateTo DIM ID 29 Fact table Period SID DIM ID Dimension Table Dr. Michael Hahne 5, TDWI WC Spring Blt Query path old structure with time-dependent hierarchy K-table of product SID -2-3 Version nodename I-table of product pred succ DIM ID 29 Fact table Period Version 1 2 DateFrom 0-01 Query key date DateTo SID DIM ID 29 Dimension table product Dr. Michael Hahne 5, TDWI WC Spring Blt 05 38

20 Szenario III : Report with historical truth dimension in P E dimension in (changed) (new) P E Fact table Period Rev. Rev, 400 Dr. Michael Hahne 5, TDWI WC Spring Blt Query path historical truth with characteristics S-table of productgroup Fact table SID 4711 SID SID Dimension table product DIM ID DIM ID Period Rev. Rev. 400 Dr. Michael Hahne 5, TDWI WC Spring Blt 05 40

21 Szenario IV : Report with comparable Results dimension in P E dimension in (changed) (new) P E Fact table Period Rev. Rev. Dr. Michael Hahne 5, TDWI WC Spring Blt Query path comparable results with time-dependent navigational attributes Dimension table product Prod. SID DIM ID S-Table of Prod.grp. SID 4711 UserFrom < UserTo > Query key date od. DIM ID 29 Fact table Period Prod. SID Prod. P E Prod.grp. SID UserFrom UserTo DateFrom Y-table of product DateTo Dr. Michael Hahne 5, TDWI WC Spring Blt 05 42

22 Agenda Architecture of the SAusiness Information Warehouse Extended Star Schema of the SAG Variants for modeling hierarchical dimension structures Temporal aspects and time stamping Modelling guidelines Graphical model representation with Visio Dr. Michael Hahne 5, TDWI WC Spring Blt Modelling Guidelines Modelling of dimensions Design of Info-Provider Dr. Michael Hahne 5, TDWI WC Spring Blt 05 44

23 Guidelines for conceptual modelling of dimensions Number of dimensions should be between four and ten (optimal between six and eight) Number of hierarchy levels (at most seven hierarchy levels) Number of elements per consolidation element (a maximum of fifteen to twenty elements is advisable) Determination of dimensions 1:1-relationship unsuitable (-> attributes) 1:N-relationship determine dimension hierarchy M:N-relationship rather two different dimensions) Dr. Michael Hahne 5, TDWI WC Spring Blt Guidelines for logical modelling of dimensions Model characteristics with high cardinality as a line item dimension Attributes that change frequently should be modelled as own dimension (use line item where possible!) Group characteristics with very low cardinality (e.g. scenario) in one dimension in order to reduce the number of dimensions and to fulfill the restriction of 16 dimensions at most Distribute characteristics of a hierarchy with high cardinality to seperate dimensions (parent characteristics in own dimension) Dr. Michael Hahne 5, TDWI WC Spring Blt 05 46

24 Criteria for the decision-making aid of the logical modelling of dimension structures in the BW Versioning Scope Performance Navigational paths Unbalanced dimension structures Leaves with multiple parent elements Structural changes and reorganisation Dr. Michael Hahne 5, TDWI WC Spring Blt Hierarchy-Guideline: Versioning External hierarchy Hierarchy within a dimension (characteristics) Hierarchy defined by navigational attributes Transactional view isn t possible (no as posted ) Different types of views are possible (hierarchy versions and time-dependent hierarchies) Only the transactional view ( as posted ) is possible Transactional view isn t possible (no as posted ) Time-dependent attributes enable different views Dr. Michael Hahne 5, TDWI WC Spring Blt 05 48

25 Hierarchy-Guideline: Scope External hierarchy Hierarchy within a dimension (characteristics) Hierarchy defined by navigational attributes Hierarchy is part of master data and valid for each Info-Cube in the system (where the underlying Info-Object is used) Only valid in the Info- Cube Hierarchy is part of master data and valid for each Info-Cube in the system (where the underlying Info-Object is used) Dr. Michael Hahne 5, TDWI WC Spring Blt Hierarchy-Guideline: Performance External hierarchy Hierarchy within a dimension (characteristics) Hierarchy defined by navigational attributes Aggregates should be used for good query performance Good performance (even without aggregates) Aggregates should be used for good query performance Dr. Michael Hahne 5, TDWI WC Spring Blt 05 50

26 Hierarchy-Guideline: Navigational paths External hierarchy Hierarchy within a dimension (characteristics) Hierarchy defined by navigational attributes Drill-down path is predefined by the structure of the consolidation tree Levels can be skipped because there isn t a predefined drill-down path (all characteristics in a dimension are equal) Levels can be skipped because there isn t a predefined drill-down path (all navigational attributes of a characteristic are equal) Dr. Michael Hahne 5, TDWI WC Spring Blt Hierarchy-Guideline: Unbalanced dimension structure External hierarchy Hierarchy within a dimension (characteristics) Hierarchy defined by navigational attributes Unbalanced hierarchies are possible Each characteristic corresponds to a certain level of the hierarchy, therefore only balanced structures are possible Each characteristic corresponds to a certain level of the hierarchy, therefore only balanced structures are possible Dr. Michael Hahne 5, TDWI WC Spring Blt 05 52

27 Hierarchy-Guideline: Leaves with multiple parent elements External hierarchy Hierarchy within a dimension (characteristics) Hierarchy defined by navigational attributes Many-many relationships between the levels of the hierarchy are possible and consolidated correctly Many-many relationships between hierarchy levels are only possible in the way they are defined by the transactions ( as posted view) Many-many relationships between levels are impossible Dr. Michael Hahne 5, TDWI WC Spring Blt Hierarchy-Guideline: Structural changes and reorganisation External hierarchy Hierarchy within a dimension (characteristics) Hierarchy defined by navigational attributes Quick change and reorganisation possible Reloading cube(s) is required for reorganisation Reorganisation is possible (additional attributes and/or changes of master data) Dr. Michael Hahne 5, TDWI WC Spring Blt 05 54

28 2-Layer concept of Cube-Modelling Multi-Cube user oriented Basis-Cube Basis-Cube Basis-Cube physical optimized Dr. Michael Hahne 5, TDWI WC Spring Blt Agenda Architecture of the SAusiness Information Warehouse Extended Star Schema of the SAG Variants for modeling hierarchical dimension structures Temporal aspects and time stamping Modelling guidelines Graphical model representation with Visio Dr. Michael Hahne 5, TDWI WC Spring Blt 05 56

29 Dimension with one characteristic Dimension { } Characteristic T Short, Medium, Long Dr. Michael Hahne 5, TDWI WC Spring Blt Line-Item-Dimension LI Dimension { } Characteristic T LD Short, Medium, Long Dr. Michael Hahne 5, TDWI WC Spring Blt 05 58

30 Display attributes of a characteristic Dimension { } Characteristic T Short, Medium, Long D Display-Attrib. D Display-Attrib. Dr. Michael Hahne 5, TDWI WC Spring Blt Navigational attributes of a characteristic Dimension { } Characteristic N Nav-Attribute N TD Nav-Attribute Dr. Michael Hahne 5, TDWI WC Spring Blt 05 60

31 Hierarchy within a dimension with characteristics Dimension { } Characteristic Base level { } Characteristic Middle level { } Characteristic Top level Dr. Michael Hahne 5, TDWI WC Spring Blt Hierarchical structure with navigational attributes Dimension { } Characteristic Base level N Nav-Attribute middle level N Nav-Attribute top level Dr. Michael Hahne 5, TDWI WC Spring Blt 05 62

32 External hierarchy Dimension { } Characteristic Hierarchy (external) T Hierarchy (external) T T T Hierarchy (external) Dr. Michael Hahne 5, TDWI WC Spring Blt Modelling of Basis-Cubes Dr. Michael Hahne 5, TDWI WC Spring Blt 05 64

33 Contact Dr. Michael Hahne Freiherr-vom-Stein-Str. 13a Bretzenheim Germany URL / fon 0049 (671) fax 0049 (671) Dr. Michael Hahne 5, TDWI WC Spring Blt 05 65