Data ware house design
Download as PPTX, PDF•1 like•307 views
This document provides information on designing a data warehouse, including key concepts like star schemas, snowflake schemas, dimensions, facts, and granularity. It discusses why an OLTP database may not be suitable for reporting and emphasizes the importance of auditing in a data warehouse to track changes over time through lineage information. Dimension tables contain attributes and keys while fact tables contain measures and foreign keys to dimensions.
Data ware house design
- 1. SOFTWARE/WEB/MOBILE/DATABASE ARCHITECT, ENGINEER, AND DEVELOPER TORONTO, CANADA HTTP://SAYED.JUSTETC.NET HTTP://WWW.JUSTETC.NET Sayed Ahmed Logical Design of a Data Warehouse
- 2. OUR SERVICES Free Training and Educational Services Training and Education in Bangla: Bangla.SaLearningSchool.com Training and Education in English: www.SaLearningSchool.com English.SaLearningSchool.com Ask a question and get answers: Ask.JustEtc.net
- 3. TOPICS - KEYWORDS Design a Data Warehouse Star Schema Snow Flake Schema Dimension Tables Fact Tables Auditing Surrogate Keys Type 1, Type 2, Type 3, and Mixed solutions for slowly changing dimension data ( SCD management) Pivoting for Analysis To help with SSAS on data warehouse
- 4. TOPICS - KEYWORDS Design a Data Warehouse Additive measures Semi additive measures Hierarchies for dimensions Attributes in dimensions Attributes in lookup tables Long term data warehouse design Usually Star Schema Short term data warehouse design POC Usually snowflake schema
- 5. TOPICS - KEYWORDS Fact Tables measures foreign keys and possibly an additional primary key and lineage columns granularity of fact tables auditing and lineage needs Measures can be additive non-additive semi-additive
- 6. TOPICS - KEYWORDS dimension keys names attributes member properties translations and lineage
- 7. TOPICS - KEYWORDS attributes natural hierarchies many-to-many fact table relationships you can introduce an additional intermediate dimension
- 8. CONCLUSION Not much – right However, if you understand all the terms and can implement all these concepts in your data warehouse That will be great Not necessarily you will need to use all of these concepts; however, you may need to justify based on the situation, will all or any of these will help? What will help and what will not help Check our sub sequent videos and tutorials
- 9. THANK YOU Any Concerns? http://ask.justetc.net Or comment below...
- 10. TOOLS AND SOFTWARE REQUIREMENTS Download the Adventure Works databases OLTP database (LOB database) Data warehouse Database From http://msftdbprodsamples.codeplex.com/releases/view/55330 For this tutorial, you can just check our slides Though the following tools will help And probably check the details in the downloaded databases esp. The AdventureWorksDW2012 You will need help from SQL Server and SQL Server MGMT Studio Tools
- 11. REQUIRED TOOLS Useful/Required SQL Server Components Database Engine Services Documentation Components Management Tools - Basic Management Tools – Complete SQL Server Data Tools
- 12. DATA WAREHOUSE DESIGN – THE DETAILS Data Warehouse Logical Design Topics: Design and Implement a Data Warehouse Design and implement dimensions. Design and implement fact tables Design Auditing track the source and time for data coming into a DW through auditing i.e lineage information Why a Data Warehouse? It is hard to generate reports from OLTP/LOB/Transactional database To do Analysis on OLTP database data (some times) Get useful information/useful summarized and details data to be used to take business decisions
- 13. DATA WAREHOUSE DESIGN – THE DETAILS Why a Data Warehouse? Data in OLTP are heavily normalized. The goal was to keep one data only in one single place to reduce redundancy and consistency of data You may end up with many tables 100s, 1000s To generate reports you may need to join many tables – will be slow Historical data may not be there Data quality is also an issue For reporting or analyzing, you may need data from multiple databases across many departments
- 14. WHY A DATA WAREHOUSE? So you can create a Data Warehouse By cleaning data With historical data Combining data from multiple sources Denormalizing data Using specific design geared towards Data Warehouse design Some or many consider DW design is less complex than relational database design Though it also has some complex areas to address... (by those some or many)
- 15. SO WHAT DOES A DATA WAREHOUSE CONTAIN? Usually two schemas are used for a DW Star Schema-> looks like a star Snow Flake Schema Another one called Dimensional Model Includes both Star and Snow Flake in the same Data Warehouse Both Schemas has tables of two types Dimension Tables Fact Tables
- 16. SO WHAT DOES A DATA WAREHOUSE CONTAIN? Fact Tables are in the center A Fact table joins/combines all the data required for this reporting or for the business aspect of this reporting Usually combines the primary keys of different tables that contain data for this report/business aspect Dimension tables are all the other tables that contain actual data Dimension tables are the tables that contain data these can be the actual tables in the OLTP database without any modification (Snow Flake) Or Dimension tables can be newly created by denormalizing the existing OLTP databases (Star)
- 17. SO WHAT DOES A DATA WAREHOUSE CONTAIN? So, you know now what are dimension tables and what are fact tables Fact tables contain primary keys of all related tables (here they are foreign keys) Dimension tables contain data Usually, it’s better that you keep your data warehouse separate from your OLTP database So bring all the tables (dimension) here Or denormalize them and bring them here in the new database
- 18. SIMPLIFIED: WHAT ARE STAR AND SNOWFLAKE SCHEMAS If you just create Fact tables and take all the related tables from your OLTP/LOB databases You get a Snow Flake Schema Here all Dimension tables are still normalized (as you just took them from the actual database) This is easy – so good for short-term, quick, and experimental Data Warehouse One note, your reporting and analysis services queries (MDX, DMV) will be slow with Snow Flake Schemas
- 19. SO WHAT DOES A DATA WAREHOUSE CONTAIN? Now, when you denormalize the dimension tables You get the start schema The Fact tables remain the same for example Star Schema is kind of standard and used a lot Originally was developed in 1980’s
- 20. EXAMPLES: WHY REPORTING IN OLTP DATABASE IS NOT A GREAT IDEA Sales amount for internet sales by different countries and historical years
- 21. WHY REPORTING IN OLTP DATABASE IS NOT A GREAT IDEA issues that I did not mention before If your OLTP database was well designed (?) It may be hard to find the tables related to the reporting The table names and the column names can be tricky – do not follow any conventions – do not have meaning So it can be hard to find data for the reporting
- 22. WHY REPORTING IN OLTP DATABASE IS NOT A GREAT IDEA Note: Reality: The OLTP may not even be well designed (that makes reporting hard sometimes) even the relationships as well as normalization – here we assumed that OLTP is perfect In a long back project I had to re-write/verify/check/change/optimize/had to deal with (whatever you say) 100s (not really 100s, can be close to 100) of queries for a reporting system Had to change the interface from one button for one report (easy to get lost) Into a drop down list of reports The relations among data were arbitrary – actually had only in the mind of the designer – did not follow any standards – No ER – no standard concepts--- So it was a hard job.. Anyway..
- 23. WHY REPORTING IN OLTP DATABASE IS NOT A GREAT IDEA In such cases Tools such as SQL Profiler might help you could create a test environment, try to insert some data through an LOB application have SQL Profiler identify where the data was inserted Another, issue with this particular example No lookup for dates and years You need to extract The tables may not contain even historical data No date field So no historical data
- 24. WHY REPORTING IN OLTP DATABASE IS NOT A GREAT IDEA If sales data reside in multiple databases even by multiple departments How do you merge Identify and match Customer data can be in different database with no common identification Data quality can be low Data missing Partial data Inconsistent data in multiple databases Data can be represented differenlt in different database M or F for gender 1 or 0 for gender
- 26. TOTAL DW: MULTIPLE STAR SCHEMAS You saw one Star Schema for Internet Sales You can see another for Offline Sales Another for Accounting Your DW has many such Star Schemas And these start schemas need to be connected/related They will be connected when you use the same dimensions for them i.e. If two star schemas have the same dimension they can share that dimension Called: shared or conformed dimensions For SSAS, you can use shared dimensions only There is a concept of private dimension Not a great idea in practical and real life applications You cannot connect/compare/verify the data over the shared dimension
- 29. SNOW FLAKES WILL BE MORE AND MORE NORMALIZED Everything can be normalized Or the first level can be normalized others are not
- 30. NORMALIZED PRODUCT DIMENSION In the Star Schema, you could use these normalized product table to get snow flake schema (partially.) Could use all normalized dimensions to get full snow flake
- 31. SNOW FLAKE In Snow flake, you may see partial than full snow flakes in reality Though, in reality, better to go for star schema Queries will be faster
- 33. GRANULARITY The number of Dimension Tables connected to a fact table Dimension of a star schema Cube = 3 dimension SSAS operates/analyzes on Cube
- 34. AUDITING AND LINEAGE I will be very short on this In data warehouse, you may want some auditing tables For every update, you should audit who made the update, when it was made, and how many rows were transferred to each dimension and fact table in your DW
- 35. AUDITING AND LINEAGE You will need additional fields/columns in your dimension and fact tables to track When, and who, and from where the row data was/were updated Your ETL process needs to be updated If you used SSIS for the ETL Modify SSIS packages so that you can record these information
- 36. THANK YOU Any Concerns? http://ask.justetc.net Or comment below...
- 37. DESIGNING DIMENSIONS Keys . Used to identify entities Name columns . Used for human names of entities Attributes . Used for pivoting in analyses Member properties . Used for labels in a report Lineage columns . Used for auditing, and never exposed to end users
- 38. For analysis Pivot Table Pivot Graph For Dimensions The fields used as for pivoting are called Attributes Not all columns are attributes Attributes: based on what analysis are done In previous, slide you saw the different types of columns
- 39. Attributes For pivoting, discrete attributes with a small number of distinct values is most appropriate Should not be continuous Keys are not good candidates for pivoting and analysis To make continous column for pivoting Concert/utilize it as a small set of discrete values
- 40. SSAS can discretize continuous attributes. Not always great – need business perspecyive as well Age and Income are not good candidates for auto discretize Naming columns to identify the entity Not good for pivoting or keys Address such as Columns used in reports as labels only, not for pivoting, are called member properties. Can include translations
- 41. Lineage and auditing columns Used for auditing data Never exposed to the users
- 43. Possible Attributes BirthDate (after calculating age and discretizing the age) MaritalStatus Gender YearlyIncome (after discretizing) TotalChildren NumberChildrenAtHome EnglishEducation (other education columns are for translations) EnglishOccupation (other occupation columns are for translations) HouseOwnerFlag NumberCarsOwned CommuteDistance
- 45. FullDateAlternateKey (denotes a date in date format) EnglishMonthName CalendarQuarter CalendarSemester CalendarYear Drill Down attributes CalendarYear →CalendarSemester → CalendarQu arter → EnglishMonthName → FullDateAlternateKey .
- 46. why dimension columns used in reports for labels are called member properties. In a Snowflake schema, lookup tables show you levels of hierarchies. In a Star schema, you need to extract natural hierarchies from the names and content of columns. Nevertheless, because drilling down through natural hierarchies is so useful and welcomed by end users, you should use them as much as possible.
- 47. SLOWLY CHANGING DIMENSIONS Type 1 History lost Type 2 Keeps all history Type 3 Keeps partial history You can use a combination For some columns type1 for others type 2
- 52. DESIGNING FACT TABLES Fact tables include measures, foreign keys, and possibly an additional primary key and lineage columns. Measures can be additive, non-additive, or semi-additive. For many-to-many relationships, you can introduce an additional intermediate dimension.
- 53. Fact tables Collection of measurements on a specific aspects of business Measure columns sales amount, order quantity, and discount amount.