Comparison between dimensional modeling and paradigm modeling

One 、 Two modeling ideas

about Inmon and Kimball Two modeling methods can be described at length , But the theory is boring , Especially the obscure words , I don't think you'll get much after reading it , So I use popular language to extract the core concepts according to my own understanding .
Paradigm modeling

Paradigm modeling is the father of data warehouse Inmon Advocated by ,“ Data warehouse ” The word is what the master defined , This modeling method is consistent with the paradigm theory 3NF, there 3NF And OLTP Medium 3NF It's a little different ： In a relational database 3NF It is the abstraction of entity object relationship for specific business process , And the data warehouse 3NF It's a topic oriented abstraction from an enterprise perspective .
Inmon From the perspective of process, the model is top-down , Top down refers to the flow of data ,“ On ” That's upstream of the data ,“ Next ” That's downstream of the data , From distributed heterogeneous data sources -> Data warehouse -> The data mart . Data oriented , Then explore step by step to get the data that meets the expectation as much as possible , Because data sources are often heterogeneous , So more emphasis will be put on data cleaning , Extract data into entities - relational model , The concept of fact table and dimension table is not emphasized .
One 、 Two modeling ideas
about Inmon and Kimball Two modeling methods can be described at length , But the theory is boring , Especially the obscure words , I don't think you'll get much after reading it , So I use popular language to extract the core concepts according to my own understanding .

3、 ... and 、 Comparison of the two models
The characteristics of the two modeling methods

Paradigm modeling ： From the specific examples in the previous section, we can see the advantages of paradigm modeling ： Be able to integrate data model of business system , It is convenient to realize the model of data warehouse ; The same data is only stored in one place , No data redundancy , Data consistency is guaranteed ; Data decoupling , Convenient maintenance . But at the same time, it brings disadvantages ： There are many tables ; When querying, there are many associated tables, which reduces the query performance .
Dimensional modeling ： The structure of the model is simple , Analysis oriented , In order to improve query performance, data redundancy can be increased , De normalized design , Short development cycle , Ability to iterate quickly . The disadvantage is that there is a lot of redundancy in the data , The preprocessing phase is expensive , Late maintenance trouble ; Another problem is that the consistency of data caliber cannot be guaranteed , The reason is explained later .
Let's look at dimensional modeling ： Data will be extracted as facts - Dimension model , Dimension is the angle of view , Look at a problem from a different perspective , You'll come to a different conclusion , And to get this conclusion , You need the measurement in the fact table , What is measurement , Fields of numeric type in fact table .
example ： The sales situation of each commodity of a company in all parts of the country , Dimensions are cities and commodities across the country , Measurement is the unit price of goods , Calculate sales from different dimensions ： Check the sales of yoghurt in Beijing , Sales of pure milk in Shanghai , That's how different dimensions are combined . On the condition of dimension , Calculate the sum of the unit prices of the goods , That is to say sum measurements , We can get the result we want .

Dimensional modeling , It's modeling by dimensions , But if the dimension design is unreasonable , Will it bring problems ？

If we think of provinces as a separate dimension , City as a dimension , Count the population of the city . At this time, provinces and cities are separate dimensions , There is no relationship between them , The following will happen ：

Guangdong province, hangzhou 1500

Zhejiang Province guangzhou 1200

At this time, the data caliber is inconsistent , Finally, it leads to inaccurate data results . And paradigm modeling doesn't have this problem , Because entities are emphasized in paradigm modeling - relational model , So there must be a belonging relationship between provinces and cities , There will be no inconsistency between provinces and cities .

therefore , Paradigm modeling can guarantee Consistency of caliber , And dimensional modeling can't ！

Four 、 Two kinds of modeling mixed scenes

Through the above sections, we have understood the ideas of paradigm modeling and dimension modeling, as well as the similarities and differences between them , Advantages and disadvantages . So can we mix the two modeling methods , Let them give full play to their respective advantages . Now let's take a look at .

Paradigm modeling Must conform to the quasi three paradigm design specification , If hybrid modeling is used , Then the source table also needs to conform to the limitation of normal form modeling , That is, the source data must be the data of operational or transactional systems . adopt ETL Extract transform and load to data warehouse ODS layer ,ODS Layer data is consistent with source data , therefore ODS Layer data is also in line with the paradigm design specifications , adopt ODS The data of , Using the paradigm modeling method , Build a data warehouse for atomic data EDW, Then based on EDW, Using dimension modeling method to build data mart .

Combine the specifications of the two modeling methods , Hybrid modeling follows “ loose coupling 、 hierarchical ” To implement the basic principles of architecture . The hybrid data warehouse architecture method mainly includes the following Key steps ：
Business requirements are built step by step 、
Hierarchical storage of data 、
Integrate atomic data standards 、
Maintaining consistency dimensions, etc