Static Dimension in Datawarehouse

Static Dimension are not extracted from the data source, but created in warehouse itself in the context of warehouse. These are usually loaded just once, manually or by the help of a procedure. The data of this dimension does not change or changed very rarely.

A few common examples of the static dimensions would be "DIM_SOURCE_SYSTEMS", "DIM_STATUS_CODE" or date time as well.

Please do not confuse it with the SCD Type 0. In SCD type 0, we insert data regularly but never update it. Also, they are loaded with the data extracted from source.

Click here to read about the other dimensions in datawarehousing.

OLTP vs OLAP explained

Both OLAP and OLTP are databases are designed for different purposes.

OLTP stands for Online Transaction Processing and is optimized for insert. update and delete operations. This is the reason these are used behind the application for regular business transactions.

OLAP stands for Online Analysis Processing and this databases are optimized for read operation and are widely used for analyzing, mining and reporting on large datasets.

It is important to note here that the database in which we store all the Facts and Dimensions is also OLTP. Post this we move the data to create a cube, which is stored in a OLAP database.

The image below explains the major differences of a OLTP and OLAP systems.

Now that we know the basic difference between OLTP and OLAP, lets now dig a little deeper.

• The core of any OLAP system is a Cube, which is just a three-dimensional spreadsheet, where as in OLTP data is stored in 2D tables.
• A cube consists of Numeric Facts called measures and are categorized by dimension data.
• The metadata of any cube is star schema or Snowflake Schema or Fact Constellation in relational(OLTP) database. Measures are derived from the records in the fact table and dimensions are derived from dimension tables.
•  While we can perform CRUD operation in OLTP, there are a few more operations which we could perform on a cube. These include Slicing, Dicing. Pivoting etc.

Different Operations on a OLAP cube in datawarehouse

The common operations on a cube are Slice, Dice, Drill-down, Roll-Up, pivot. These could be used individually or together to generate the desired picture of data.

Slice is an act of picking a subset of the cube by choosing a single value for one of its dimensions. This create a new cube with fewer dimension.
The picture below shows a slicing operation:   The sales figures of all categories and all regions is sliced for just year 2004.

Dicing allow the analysts to create and visualize a sub cube by selecting multiple values from multiple dimensions. The picture below shows dicing operation:

Drill down / up allows the analyst to navigate through the levels of data from most summarized to most detailed. The picture shows the drill down operation from summary category "Outdoor-Schutzausrüstung" to see the sales figures for the individual products.

Roll Up means summarizing the data along a dimension. The summarization  could be calculating, count, total across a dimension hierarchy or more complex formulas like profit from sales etc. The picture below shows the summation of sale for each categories.

Pivot allows the analysts to rotate the cube in space to the desired direction, which gives the data and visualizations a totally different perspective. The image below show that.

Fact Constellation schema in Data warehousing

Fact Constellation: Fact Constellation allows you to connect multiple facts to dimensions. It is combination of multiple stars connected with each other using dimension, hence called Galaxy or Fact Constellation schema.

As compared to Star and snowflake it is used for more sophisticated application. The dimension tables could be normalized or not as per the requirement.

If you do not understand the basic concepts of data warehousing, please visit my previous article.

If we consider the data warehousing scenario detailed here, below is how is galaxy or fact constellation schema would look like.

Characteristics of Fact Constellation schema:

• Complex design
• Different variants of aggregations
•  Suitable for more sophisticate application 

Snowflake schema

Snowflake Schema: When one fact is surrounded by a chain of more than one dimension, it is called snowflake. It is also called an extension of star schema by adding more dimensions.
In snowflake schema the table are normalized.

If you do not understand the basic concepts of data warehousing, please visit my previous article.

If we consider the data warehousing scenario detailed here, below is how a snowflake schema would look like.

Click here to compare the same design with star schema.

Characteristics of Snowflake schema:

• Less storage requirement because of normalized dimensions
• Less time to load as compared to star schema.
• Better slicing and dicing
• Join complexity increase
• Time to Query may increase a little depending on the scenario.

Click here to learn more about different types of schema.

Star Schema - The simplest of all

Star Schema : It is simplest data warehouse schema. It is called star because it resembles the star with points radiating out of the center. Fact tables is the center of the star and dimensions sit at the points.
Usually the fact table in the star schema is in 3rd normal form and dimensions are usually de-normalized. It is most simplest and most commonly used schema.

If you do not understand the basic concepts of data warehousing, please visit my previous article.

If we consider the data warehousing scenario detailed here, below is how a star schema would look like.

Although, star schema require just one fact table at the center, but you could surely have multiple stars in your design.

Characteristics of  star schema

• Simplest design
• Better query performance.
• A little extra time to load the data into dimension because of de-normalization.
• More space because of redundancy in dimensions.

Click here to learn more about different types of schema.

Bridge Tables in datawarehouse

As the name suggests, bridge table is used to link tables when direct linking does not make sense. Usually used in a little complecated senarios when we have many to many relations between 2 tables.

Example: Consider we have multiple customer linked to a single bank account balance. Putting multiple customer keys in that fact would create obvious problems in the fact table. 

Bridge table is also used sometimes to manage variable hierarchies. For e.g. to maintain one primary customer and one or more secondary customers.

Bridge table come to rescue in such situation, but just like outrigger dimension don't use it too much. Remember, every query has to pay toll to navigate the bridge.

Although, it is not a dimension but I personally call it Bridge dimension because it has many properties of a dimension.

Click here to read about the other dimensions in datawarehousing.

What is Conformed Dimension and its advantages

People usually finds it difficult to understand the Conformed dimension, I am trying to explain it with an example. To start with, please do not confuse it confirmed / static dimension, because of the name.

Definition: A conformed is defined as a dimension, which has the same meaning to all the facts in a data warehouse, even if the facts belong to different data-marts.
It could one single table or multiple copies of same table across different marts. There is no hard rule  of the copies having the same table name. One copy could even be a sub set of another copy.

Let us now try to understand it with a example. Consider online movie tickets data warehouse scenario.
Lets assume we have divided our EDW into 2 data marts

• Marketing : Marketing team is more concerned about the no of tickets sold.
• Finance : Finance team is more concerned about the profit.

Both teams would require to know the stats for each cinema. Now, both teams might also have different terminologies. Marketing team might call it Cinems whereas finance team might call it Screen.
So in this scenario we could create 2 copies of same table with different names as shown below.

But why should we use it. Why don't we create 2 separate tables if we need. Below are a few disadvantages of not creating Conformed dimensions:

• There will more ETL work to generate the logic of different tables.
• More ETL will surely mean some extra maintenance effort.
• and possible difficulties in joining multiple data marts. 

Visit my post on Types of Dimensions to understand more about different types of dimensions.

List of dimensions in a Data Warehouse

Below is the list of dimensions types we should consider while designing any data warehouse. Click on the link below for a detailed definition and example.

• Basic dimension
• Junk dimension
• Conformed dimension
• Outrigger Dimension
• Bridge table
• Snowflaked dimension (Details article in progress)
• Role Playing dimension (Details article in progress)
• Degenerate Dimension (Details article in progress)
• Slowly Changing dimension (Details article in progress)
• Inferred dimension
• Shrunken Dimensions
• Static Dimensions

This list will keep on improving with time, so you can bookmark this page and  feel free to subscribe to the blog by entering your email

What is Junk Dimension and its purpose

Definition: By definition, it is a collections of random transnational code flags. It is usually created by joining multiple low cardinality dimensions into one.
As we merge different dimensions in to one, it will reduce the number of foreign keys in the fact table.

If we consider the Data warehouse scenario of Online Ticket Booking, The design should have following 3 tables, with fact having 2 foreign keys one for each dim.

• Fact_Tickets
• Dim Transaction Type
• Dim Cert Type

Both Dims are of very low cardinality(i.e. no of records in each dim would be very low)
Transaction Type : The values could be Net-banking, Card, Wallet. The chances of increasing this count is very low
Movie Certificate : The values could be U, A, U/A. Again the chances of increasing/Changing these is very rare.

The picture below describes the 2 possible designs of it, if  we keep these parameters in dimension table.

So with the help of junk dimension, we converted 2 forein keys in to one.
No of Records: Notice that, the no of records in the junk transaction is the multiplication of no of records in each dimension. In the example above there are 3 records for both Transaction and Certificate, so the total count in junk dimension is 9.
If we are merging 3 tables with 3 records each, the no of records in the junk dimension would be 27(3*3*3)

Ensure the dimensions we junk together should be of very low cardinality. I would suggest the no of records in the junk transaction should be less than thousand.

Please feel free to post your queries in the comments section and follow this blog.