20767: Implementing a SQL Data Warehouse

$1,735.00
$1,735.00
$1,735.00
Start Date Time Days Price GTR Availability
July 17, 2017 09:00 (EST) 3 $1,735.00 Register
September 25, 2017 09:00 (EST) 3 $1,735.00 Register
November 27, 2017 09:00 (EST) 3 $1,735.00 Register

About this course

This three-day instructor-led course provides students with the knowledge and skills to provision a Microsoft SQL Server 2016 database. The course covers SQL Server 2016 provision both on-premise and in Azure, and covers installing from new and migrating from an existing install.

 

Audience profile

The primary audience for this course are database professionals who need to fulfil a Business Intelligence Developer role.  They will need to focus on hands-on work creating BI solutions including Data Warehouse implementation, ETL, and data cleansing. 

 

At course completion

After completing this course, students will be able to:

  • Provision a Database Server.
  • Upgrade SQL Server.
  • Configure SQL Server.
  • Manage Databases and Files (shared).

Prerequisites

In addition to their professional experience, students who attend this training should already have the following technical knowledge:

  • Basic knowledge of the Microsoft Windows operating system and its core functionality.
  • Working knowledge of relational databases.
  • Some experience with database design.

 

Course Outline

 

Module 1: Introduction to Data Warehousing

This module describes data warehouse concepts and architecture consideration.

 

Lessons

  • Overview of Data Warehousing
  • Considerations for a Data Warehouse Solution

 

Lab: Exploring a Data Warehouse Solution

  • Exploring data sources
  • Exploring an ETL process
  • Exploring a data warehouse

 

After completing this module, you will be able to:

  • Describe the key elements of a data warehousing solution
  • Describe the key considerations for a data warehousing solution

 

Module 2: Planning Data Warehouse Infrastructure

This module describes the main hardware considerations for building a data warehouse.

 

Lessons

  • Considerations for data warehouse infrastructure.
  • Planning data warehouse hardware.

 

Lab: Planning Data Warehouse Infrastructure

  • Planning data warehouse hardware

 

After completing this module, you will be able to:

  • Describe the main hardware considerations for building a data warehouse
  • Explain how to use reference architectures and data warehouse appliances to create a data warehouse

 

Module 3: Designing and Implementing a Data Warehouse

This module describes how you go about designing and implementing a schema for a data warehouse.

 

Lessons

  • Designing dimension tables
  • Designing fact tables
  • Physical Design for a Data Warehouse

 

Lab: Implementing a Data Warehouse Schema

  • Implementing a star schema
  • Implementing a snowflake schema
  • Implementing a time dimension table

 

After completing this module, you will be able to:

  • Implement a logical design for a data warehouse
  • Implement a physical design for a data warehouse

 

Module 4: Columnstore Indexes

This module introduces Columnstore Indexes.

 

Lessons

  • Introduction to Columnstore Indexes
  • Creating Columnstore Indexes
  • Working with Columnstore Indexes

 

Lab: Using Columnstore Indexes

  • Create a Columnstore index on the FactProductInventory table
  • Create a Columnstore index on the FactInternetSales table
  • Create a memory optimized Columnstore table

 

After completing this module, you will be able to:

  • Create Columnstore indexes
  • Work with Columnstore Indexes

 

Module 5: Implementing an Azure SQL Data Warehouse

This module describes Azure SQL Data Warehouses and how to implement them.

 

Lessons

  • Advantages of Azure SQL Data Warehouse
  • Implementing an Azure SQL Data Warehouse
  • Developing an Azure SQL Data Warehouse
  • Migrating to an Azure SQ Data Warehouse
  • Copying data with the Azure data factory

 

Lab: Implementing an Azure SQL Data Warehouse

  • Create an Azure SQL data warehouse database
  • Migrate to an Azure SQL Data warehouse database
  • Copy data with the Azure data factory

 

After completing this module, you will be able to:

  • Describe the advantages of Azure SQL Data Warehouse
  • Implement an Azure SQL Data Warehouse
  • Describe the considerations for developing an Azure SQL Data Warehouse
  • Plan for migrating to Azure SQL Data Warehouse

 

Module 6: Creating an ETL Solution

At the end of this module you will be able to implement data flow in a SSIS package.

 

Lessons

  • Introduction to ETL with SSIS
  • Exploring Source Data
  • Implementing Data Flow

 

Lab: Implementing Data Flow in an SSIS Package

  • Exploring source data
  • Transferring data by using a data row task
  • Using transformation components in a data row

 

After completing this module, you will be able to:

  • Describe ETL with SSIS
  • Explore Source Data
  • Implement a Data Flow

 

Module 7: Implementing Control Flow in an SSIS Package

This module describes implementing control flow in an SSIS package.

 

Lessons

  • Introduction to Control Flow
  • Creating Dynamic Packages
  • Using Containers
  • Managing consistency.

 

Lab: Implementing Control Flow in an SSIS Package

  • Using tasks and precedence in a control flow
  • Using variables and parameters
  • Using containers

Lab: Using Transactions and Checkpoints

  • Using transactions
  • Using checkpoints

 

After completing this module, you will be able to:

  • Describe control flow
  • Create dynamic packages
  • Use containers

 

Module 8: Debugging and Troubleshooting SSIS Packages

This module describes how to debug and troubleshoot SSIS packages.

 

Lessons

  • Debugging an SSIS Package
  • Logging SSIS Package Events
  • Handling Errors in an SSIS Package

 

Lab: Debugging and Troubleshooting an SSIS Package

  • Debugging an SSIS package
  • Logging SSIS package execution
  • Implementing an event handler
  • Handling errors in data flow

 

After completing this module, you will be able to:

  • Debug an SSIS package
  • Log SSIS package events
  • Handle errors in an SSIS package

 

Module 9: Implementing a Data Extraction Solution

This module describes how to implement an SSIS solution that supports incremental DW loads and changing data.

 

Lessons

  • Introduction to Incremental ETL
  • Extracting Modified Data
  • Loading modified data
  • Temporal Tables

 

Lab: Extracting Modified Data

  • Using a datetime column to incrementally extract data
  • Using change data capture
  • Using the CDC control task
  • Using change tracking

 

Lab: Loading a data warehouse

  • Loading data from CDC output tables
  • Using a lookup transformation to insert or update dimension data
  • Implementing a slowly changing dimension
  • Using the merge statement

 

After completing this module, you will be able to:

  • Describe incremental ETL
  • Extract modified data
  • Load modified data.
  • Describe temporal tables

 

Module 10: Enforcing Data Quality

This module describes how to implement data cleansing by using Microsoft Data Quality services.

 

Lessons

  • Introduction to Data Quality
  • Using Data Quality Services to Cleanse Data
  • Using Data Quality Services to Match Data

 

Lab: Cleansing Data

  • Creating a DQS knowledge base
  • Using a DQS project to cleanse data
  • Using DQS in an SSIS package

Lab: De-duplicating Data

  • Creating a matching policy
  • Using a DS project to match data

 

After completing this module, you will be able to:

  • Describe data quality services
  • Cleanse data using data quality services
  • Match data using data quality services
  • De-duplicate data using data quality services

 

Module 11: Using Master Data Services

This module describes how to implement master data services to enforce data integrity at source.

 

Lessons

  • Introduction to Master Data Services
  • Implementing a Master Data Services Model
  • Hierarchies and collections
  • Creating a Master Data Hub

 

Lab: Implementing Master Data Services

  • Creating a master data services model
  • Using the master data services add-in for Excel
  • Enforcing business rules
  • Loading data into a model
  • Consuming master data services data

 

After completing this module, you will be able to:

  • Describe the key concepts of master data services
  • Implement a master data service model
  • Manage master data
  • Create a master data hub

 

Module 12: Extending SQL Server Integration Services (SSIS)

This module describes how to extend SSIS with custom scripts and components.

 

Lessons

  • Using scripting in SSIS
  • Using custom components in SSIS

 

Lab: Using scripts

  • Using a script task

 

After completing this module, you will be able to:

  • Use custom components in SSIS
  • Use scripting in SSIS

 

Module 13: Deploying and Configuring SSIS Packages

This module describes how to deploy and configure SSIS packages.

 

Lessons

  • Overview of SSIS Deployment
  • Deploying SSIS Projects
  • Planning SSIS Package Execution

 

Lab: Deploying and Configuring SSIS Packages

  • Creating an SSIS catalog
  • Deploying an SSIS project
  • Creating environments for an SSIS solution
  • Running an SSIS package in SQL server management studio
  • Scheduling SSIS packages with SQL server agent

 

After completing this module, you will be able to:

  • Describe an SSIS deployment
  • Deploy an SSIS package
  • Plan SSIS package execution

 

Module 14: Consuming Data in a Data Warehouse

This module describes how to debug and troubleshoot SSIS packages.

 

Lessons

  • Introduction to Business Intelligence
  • An Introduction to Data Analysis
  • Introduction to reporting
  • Analyzing Data with Azure SQL Data Warehouse

 

Lab: Using a data warehouse

  • Exploring a reporting services report
  • Exploring a PowerPivot workbook
  • Exploring a power view report

 

After completing this module, you will be able to:

  • Describe at a high level business intelligence
  • Show an understanding of reporting
  • Show an understanding of data analysis
  • Analyze data with Azure SQL data warehouse