Beruflich Dokumente
Kultur Dokumente
• For performance and data integration reasons, most mature OBIEE systems source
their data from one (or more) data warehouses
• For initial implementations though, data is often initially accessed directly
• OBIEE has some powerful features for creating integrated “virtual” logical models
‣ For bringing together data from disparate systems (“federation”)
‣ For integrating historic and real-time data (“fragmentation”)
‣ For creating calculations and derivations
‣ For handling multi-valued dimensions
• OBIEE also requires that any normalized data is converted into a “star schema”
• These data modeling techniques, coupled with more formal data warehouses, can
allow you to deploy very flexible, agile data models to support your reporting
• This presentation will take you through some of these data modeling techniques and will
discuss how these initial models are migrated to a more formal data warehouse
Metadata Interchange
Device Adaptive Content Session Management
Intelligent Request Generation
Services
Query Govern.
• Design goal for the BI Server metadata layer is to create a Star Schema
‣ Fact tables, Dimension tables
‣ Drill paths
• Source data can either be pre-integrated (data mart,
data warehouse) or integrated using
the BI Server (“federated”)
‣ Federated can be faster to deliver (report in place)
- Assumes data sources share common key values
‣ Data marts and data warehouses are simpler and can
be faster to query
• MDX (OLAP, Multi-dimensional) data is converted into
RDBMS row-sets
• In this presentation, we will look at how you can use the
data modeling features of OBIEE to create logical star schemas
• Global Company has a data warehouse, containing order and customer data
• Requirement is to take the existing warehouse (normalized)
and turn it into a star schema (denormalized)
• For now, OBIEE will be used to map the data
• Once you have your logical model and one or more physical models, you can map them
together
‣ The simpler the mapping, the better
• Drag and drop physical columns
on to the logical model
• Logical model will inherit
physical data types
• Calculations within the repository are defined using the Expression Editor
• Uses OBIEE SQL Syntax
• Reference functions,
expressions, logical
columns, constants,
variables
• Same SQL syntax across
all supported databases
• Ability to pass-through
function calls using
EVALUATE function
• OBIEE can combine and join data from multiple data sources
• Data is retrieved in separate SQL, MDX calls and joined using the BI Server
• Allows you to perform federated queries
• Watch out for performance though
‣ End goal should be to eventually move reporting data to a data warehouse
• Import the new schema into the physical model, ensure FK and PK
constraints are set
• Caused when one master table (i.e. a dimension) joins to two fact tables (sales, budget)
• You would expect the aggregations on each table to be correct, however regular
SQL queries can over-count totals due to cartesian joins forming
• Known in the Discoverer world as “fan trap” queries
SQL> select a.acc_name,
2 sum(sales),
3 sum(budget)
4 from fan_trap_accounts a, fan_trap_sales s, fan_trap_budget b
5 where a.acc_id = s.acc_id
6 and a.acc_id = b.acc_id
7 group by a.acc_name
8 order by a.acc_name
9 /
• The Oracle BI Server automatically detects fan trap queries, and generates two
separate SQL queries which it combines at the BI Server level
• Solves issue, functionally identical, but not quite as efficient as Discoverer
‣ Two SQL queries vs. one, and join done in BI Server memory
-------------------- Sending query to database named ora11g (id: <<2653>>):
+++Administrator:2a0000:2a0001:----2008/08/26 17:24:48
• In some cases, there might exist a many-to-many relationship between a fact table
and a dimension table
‣ Patient diagnoses, sales made by salespeople
• In entity relationship modeling, you would resolve
this many-to-many with and “intersection” table
• The intersection in dimensional modeling terms
is called a “bridge table”
• Marking a dimension table as a bridge table stops OBIEE thinking of it as a fact table
• Any dimensions that join to it will not now join to the main fact table
• To avoid metadata consistency issues, add the dimension table to the bridge LTS
and then add any dimension columns to the bridge logical table
‣ Remove the dimension table from the logical model, use the bridge table instead
‣ Dimension data can then be accessed as required
Gavin
‣ http://www.rittmanmead.com/2007/06/21/obiee-data-modeling-tips-3-ragged-hierarchies/
• The data modeling features of OBIEE are a useful way to create an “on-the-fly” logical
data model for your reporting
• It is possible to turn a 3NF model into a star schema
• Integrate data from multiple source (“federation”)
• Integrate historic and real-time data (“fragmentation”)
• Create calculations and report drill-paths
• This is often a useful first step when prototyping your system
• Over time, migrate the data you are reporting on to a formal data warehouse
• Use the modeling features of the OBIEE suite to bring in data from all your sources
• Contact me at mark.rittman@rittmanmead.com