Analysis Services en SQL Server 2008

Analysis Services in SQL Server® 2008
Eduardo Castro SQL MVP,MCDBA , MCSE, MCAD, MCSD
ecastro@mswindowscr.org
Comunidad Windows Costa Rica

Data Warehouse System Components
Data Warehouse
User
Data Access
Data
Sources
Staging
Area
Data Marts
Data Input
Data Access

Understanding Data Warehouse Design
The Star Schema
Fact Table Components
Dimension Table Characteristics
The Snowflake Schema

The Star Schema
Employee_Dim
EmployeeKey
EmployeeID
...
Product_Dim
Time_Dim
ProductKey
TimeKey
ProductID
...
TheDate
...
Shipper_Dim
Customer_Dim
ShipperKey
CustomerKey
ShipperID
...
CustomerID
...
Dimension Table
Fact Table
Sales_Fact
TimeKey
EmployeeKey
ProductKey
CustomerKey
ShipperKey
Sales Amount
Unit Sales ...

Fact Table Components
Measures
time_dim
134 1/1/2000
DimensionTables
sales_fact Table
customer_dim
Foreign Keys
201 ALFI Alfreds
customer_key
product_key
time_key
quantity_sales
amount_sales
product_dim
201
25
134
400
10,789
25 123 Chai
The grain of the sales_fact table is defined by the lowest level of detail stored in each dimension

Dimension Table Characteristics
Describes Business Entities
Contains Attributes That Provide Context to Numeric Data
Presents Data Organized into Hierarchies

The Snowflake Schema
Defines Hierarchies by Using Multiple Dimension Tables
Is More Normalized than a Single Table Dimension
Is Supported within Analysis Services

Defining OLAP Solutions
OLAP Databases
Common OLAP Applications
Relational Data Marts and OLAP Cubes
OLAP in SQL Server 2005

OLAP Databases
Optimized Schema for Fast User Queries
Robust Calculation Engine for Numeric Analysis
Conceptual, Intuitive Data Model
Multidimensional View of Data
Drill down and drill up
Pivot views of data

OLAP in SQL Server
Microsoft Is One of Several OLAP Vendors
Analysis Services Is Bundled with Microsoft SQL Server 2005 and SQL Server 2008
Analysis Services Include
OLAP engine
Data Mining technology

Dimension Fundamentals

Cube Measures
Are the Numeric Values of Principle Interest
Correspond to Fact Table Facts
Intersect All Dimensions at All Levels
Are Aggregated at All Levels of Detail

Relational Data Sources
Star and Snowflake Schemas
Are required to build a cube with Analysis Services
Fact Table
Contains measures
Contains keys that join to dimension tables
Dimension Tables
Must exist in same database as fact table
Contain primary keys that identify each member

Applying OLAP Cubes
Defining a Cube
Querying a Cube
Defining a Cube Slice
Working with Dimensions and Hierarchies
Visualizing Cube Dimensions
Connecting to an OLAP Cube

Defining a Cube
Atlanta
Chicago
Market Dimension
Denver
Grapes
Cherries
Detroit
Melons
Apples
Products Dimension
Q4
Q1
Q2
Q3
Time Dimension

Fact Sales
Atlanta
Chicago
MarketsDimension
Denver
Grapes
Cherries
Dallas
Melons
ProductsDimension
Apples
Q4
Q1
Q2
Q3
TimeDimension
Querying a Cube

SQL Server 2008 Analysis Services Enhancements
Multidimensional Analysis with SQL Server Analysis Services
Data Mining with SQL Server Analysis Services

Multidimensional Analysis with SQL Server Analysis Services
Cube Wizard
Dimension Wizard
The Attribute Relationship Designer
The Aggregation Designer
AMO Warnings

Cube Wizard
More efficient interface
Create a cube based on a single de-normalized table
Create a cube based on a data source that has only linked dimensions

Dimension Wizard
Create dimensions more efficiently
Automatically detect parent-child hierarchies
Provide safer default error configuration
Set member properties while creating the dimension

The Attribute Relationship Designer
Flexible relationship
Rigid relationship
Manage Relationship through right-click options in the Attribute Relationship pane

The Aggregation Designer
New Aggregation Designer:
Aggregations designs are shown grouped by measure group
New view available for manual aggregation design
Improved Usage-Based Optimization Wizard:
Ability to append new aggregations to an existing aggregation design
Ability to modify storage settings for one or more partitions simultaneously
Improved Aggregation Design Wizard

AMO Warnings
Uses familiar symbols such as the wavy underline and a yellow triangle with an exclamation point
Non-intrusive warning messages appear when you pause the mouse over a warning
Available for logical errors in database design, when users depart from design best practices, and for non-optimal aggregation designs

Data Mining with SQL Server Analysis Services
Improved Data Mining Wizard
Separating Test and Training Data
Filtering Model Cases
Cross Validation of Mining Models
Data Mining Add-Ins for Microsoft Office Systems

Data Mining Enhancements
Improve both short-term and long-term predictions with the Microsoft Time Series algorithm, which has been enhanced to support both ARTXP and ARIMA algorithms
Enable drillthrough on a mining structure to allow queries about the cases used for both training and testing
Create column aliases to make it easier to understand column content. In the Data Mining Designer, the alias appears in parentheses next to the column usage label
Separate test and training data
Improve performance and analyze different scenarios by using Model Case Filtering
Create cross-validation reports
Utilize Data Mining Add-ins for Office

Separating Test and Training Data
Three ways to partition data into training and test sets:
The Data Mining Wizard
Modifying the properties of the mining structure:
If you did not create a test partition when you created the structure, you can modify the HouldoutMaxCases, HoldoutMaxPercent, and HoldoutSeed properties
DMX statement, AMO, or XML DDL

Filtering Model Cases

Limit the cases used in a model based on any attribute included in the model
Use to compare subsets of your data such as different regions

Cross-Validation of Mining Models
Additional method to test and view mining model accuracy
Data is partitioned into cross-sections which are used to train and test models against each of the other cross sections
One portion of the data is used to test, the remaining data is used to train the model
To define a cross-validation report, you must configure:
The Fold Count, which specifies the number of folds or partitions that the data is broken into for testing
The Max Cases, which specifies the maximum number of cases to use (a value of 0 specifies that all cases will be used)
The Target Attribute, which defines the column or attribute that you want to predict
The Target State, which defines a particular value that you want to analyze within the target attribute

Data Mining Add-Ins for Microsoft Office System
Data Mining Client for Excel – create, test, and manage data mining projects within Excel 2007
Table Analysis Tools for Excel – use powerful analysis tools such as analyzing key influencers, highlighting exceptions, and forecasting for data stored in spreadsheets
Data Mining Templates for Visio – render decision trees, regression trees, cluster diagrams, and dependency nets in diagrams created in Visio 2007
Important: Data Mining Add-ins for Microsoft Office System will be available for SQL Server 2008 when it releases to manufacturing.

Demonstration
Implementing Multidimensional Analysis

Optimizing Storage
What Are Sparse Columns?
How to Compress Data and Backups

What Are Sparse Columns?
Eliminate limitation of 1024 columns.
Efficient way to manage object models that frequently contain numerous NULL values.
CREATE TABLE products (product_numint, item_numint, price decimal(7,2), ..., color char(5) SPARSE, width float SPARSE...)

How to Compress Data and Backups
Data compression can be enabled on tables or views
Different compression types can be configured on a per-partition basis
The following data compression types can be defined:
Row compression
Page compression
Backup compression:
Normally decreases time required to perform a backup
Managed with Transact-SQL, Backup Task, Maintenance Plan Wizard, and Integration Services Backup Database task

Contact Information
Eduardo Castro
ecastro@mswindowscr.org
http://ecastrom.blogspot.com