A Gentle Introduction to Microsoft SSAS

Shows you how to build an OLAP cube in Microsoft Analysis Services

2. Analysis Services Overview
•Included with the SQL Server License
•Special version of Visual Studio
•Microsoft’s application for creating multidimensional OLAP databases
which are queried with the MDX language
•Microsoft’s powerful data mining platform. Includes sophisticated
algorithms that can operate on relational or OLAP data
•This presentation demonstrates the creation of an OLAP database, also
called a cube
2
Part of the Business Intelligence Development Studio

3. Why Store Data in a Cube?
3
Analysis services, like the SQL Server relational database, is a platform for storing
data. But unlike the relational database, it does not store data in tables–it is
stored in other types of structures comprising a cube.
Why store data in cubes instead of tables? There are a number of reasons:
• better query performance
• fast, optimized aggregations calculations
• more efficient storage of data through its simplified read-oriented design
• richer calculation possibilities, supporting stored measures,
calculated measures, and key performance indicators (KPI)
• an easier to understand data model for the end user.
OLAP data is engineered to provide dimensional data views, hierarchical
browsing, attribute-based breakouts and filtering. And OLAP data does not
require or even use table joins!

4. Creating an Analysis Services (OLAP)
Database
•Dimensions and their associated elements
–Hierarchies
–Attributes
•Measures and their associated elements
–Stored Measures
–Calculated Measures
–Measure Groups
–Key Performance Measures (KPI)
–Measure Profiles
4
Creating an analysis services database consists of creating the database
structure, then populating those structures with data from external data
sources. The structures comprising an OLAP cube include:

5. Development Steps
1.Create Analysis Services Project
2.Create Data Source
3.Create Data Source View
4.Create cube object definitions
5.Deploy definitions to OLAP server and load
in data
6.Specify partitions and aggregations
5

6. Open BIDS and create a new Analysis
Services Project…
6

7. Create a data source and a data
source view
7
In the data source you identify the relational data
warehouse (DW) that hosts the source data and
provide the connection information.
In the data source view, you specify which the tables in
the DW you want to use to supply data to the cube.
If the DW has been designed using the classic star
schema or snowflake approaches, the SSAS cube
wizard can examine the DW’s structure and derive a
cube definition from it, making your job easier.
When the initial design from the wizard is complete
you can go back and make modifications, for example,
you can add or modify hierarchies and attribute
relationships.
Once you have the cube design you want, you load the
data in.

8. In this illustration, all tables from the star schema DW are used, only
the sysdiagrams table (which holds the database diagram data) remains
unselected.
8

9. 9
Once you have defined the data source view, you can inspect the
schema. By right clicking on a table, you can browse the data and
define derived columns. But at this point, we still have no OLAP
database. The creation of the OLAP database begins with creating a
cube.

10. Creating the Cube Structure
10
1. With the data source view in place, you are now ready to
create the cube. Right click on the cubes folder and select
“New Cube.” This launches the cube wizard.
2. Make sure the wizard has
correctly identified which are
dimension and which are fact tables
and tell it which table contains the
data for the time dimension.

11. Setting the Time Dimension
Parameters
11
Time is a unique dimension with
inherent assumptions about how it
should work.
You identify the time dimension as
such so that the MDX functions
(such as PrevMember and
ParallelPeriod) specific to it will
work.
You also tell which of the source
data columns map to well known
time concepts such as years,
quarters, and months.

12. 12
After defining the time
properties, the wizard displays
the measures that it defined.
Here you can select which ones
you want to keep. I am keeping
all of them in this illustration.
The “Fact Units Count”
measure counts the number of
records in the source table. This
information is used in
optimizing the aggregation
process.
In the Review New Dimensions
panel the dimensions that were
created along with their
hierarchies and attributes are
displayed.

13. A cube has now been defined. This panel lets you review the data model
(UDM) that has been created.
13

14. Completing the wizard, you give the cube a name.
14

15. The solution browser on the right side of the screen now shows the
cube and dimensions that were defined. These objects can be modified
by clicking on them.
15

16. Click on the Product dimension to observe that no hierarchy was
defined by the wizard. I will create a hierarchy for it.
16

17. A new hierarchy is created by dragging the an aggregate attribute into the center panel. I
use Category Code and Dim Product to form the hierarchy; Category and Item provide
the labels the user will see (they are mapped to the name Property).
Next I define attributes between these hierarchy levels. This is so the aggregation process
adds the data from the level directly beneath it instead of always going to the leaf level,
which would be a less efficient process entailing significantly more calculations.
17

18. 18
In the case of the time dimension,
there is an additional step. We
want the months to display in
chronological order, not
alphabetical order, so we assign a
value to the dimension’s
OrderByAttribute property. The
data source has a column called
CHRON_ORDER that contains
this ordering information.

19. The information entered to define the hierarchies and attributes is stored in
XML files. The database doesn’t actually know what you have done yet. You
must “process” the dimension. This brings that information into the OLAP
database, which it then uses to create its internal structures.
19

20. Once a dimension has been processed, you can inspect it in the
browser pane to verify the hierarchy has the expected structure. Note
that the months show in chronological order.
20

21. After all the elements have been defined and the cube has
been completely processed, you can inspect the data in the
SSAS data browser.
21

22. Partitioning and Pre-Aggregating

23. 23
The cube has been defined and created. The leaf level data from the data warehouse
was loaded. We inspected the data in the multidimensional data browser.
We can tweak the physical design of the cube to improve scalability and query performance.
Three primary mechanisms for doing this are:
•Selecting ROLAP/HOLAP/MOLAP data storage options
•Partitioning
•Pre-aggregating
MOLAP is probably the most commonly used data storage option and is the default. It means
all the data will be stored in the multidimensional data cube. The illustration one the
succeeding slides will used MOLAP.
At the opposite end of the spectrum, is ROLAP where all data comes from relational data
tables. With ROLAP, the SSAS database is only providing metadata structures for presenting
information in the dimensional style. You can expect performance to be much slower. This
mode is used in situations where the source data is not static, changes frequently and you want
the reports to reflect those changes immediately.
HOLAP is a hybrid approach where all the aggregate data is in the SSAS database except the leaf
level data which resides in relational tables.
When the volume of data is very large, it can helpful to chop the data store into pieces, or
partitions. Storage mode is selectable per partition. This example uses a small amount of data
and only one partition will be employed.

24. 24
Aggregates are summary level data that are computed from the leaf level data that was
loaded from the source. Often the aggregates are totals and subtotals, but other summary
statistics such as averages or maximum values can also be used.
Pre-calculating and storing the aggregate values normally improves query performance (at the
cost of the storage space and time required to compute them.) The default is to do no pre-
aggregates. You can see this from the partitions panel shown below.
The data display shown earlier from the SSAS data browser included many aggregate data
values. Those aggregates were all calculated on the fly.
You can pre-calculate all aggregates or only some of them. If you are going to pre-calculate only
some, there are different strategies that can be employed to determined which are chosen for
calculation. You’ll see this ahead.

25. Let’s go through the aggregations process. Click the “Design Aggregations”
hyperlink to bring up the wizard. In the first panel of the wizard, push the count
button to compute the statistics that are used to drive the aggregation optimization
process.
25

26. 26
After a few seconds, the source
data has been analyzed. It
counts the number of records
per partition.
Once the statistics have been
computed, you can ask the system
to identify a set of aggregations to
perform. You can direct the
system to aggregate until A) a
certain amount of storage has been
used, B) a certain level of
performance gain has been
achieved, C) you click stop, or D)
don’t do anything
In this illustration, I am asking it
to aggregate until it reaches a
performance gain of 75%. The
system will run an optimization
algorithm to determine the best
ones to use.

27. 27
The system generates a chart telling
what percent (of the total possible
number of) aggregations it has
identified and what level of
performance gain would be
achieved by computing them.
At the completion of that step, the
wizard has identified which
aggregations to compute. You may
elect to have it compute them now
or you can defer the calculations till
later. (They could take a while.)

28. 28
Selecting “Deploy and Process
now” and pushing Finish, you
arrive at this screen.
Push the RUN button to
perform the calculations.
When it finishes, you get
a message heralding the
successful completion of
the deployment.
The information under
the Aggregations tab will
be updated.

29. Creating
Derived Measures and KPIs

30. Different Kinds of Reporting Data
•Calculated measures
–Percents
–Shares
–Differences
•Key Performance Indicators (KPIs)
30
Thus far, all the measures that have been constructed have been displays of
stored data or aggregates either stored or calculated on the fly. There are
other kinds information that can be made available to an end user.
Calculated measures are calculated on the fly using MDX expressions. KPI’s
are measures with associated goals and graphics. I will show an example of
both.

31. 31
In this example, I create a calculated measure that gives difference between
the data value at a given time and its value the previous time period. The
calculation is defined from the Calculations tab. It is given a name and an
MDX expression. In this example I make use of the PrevMember function.

32. 32
Displaying the Units measure and the Units Increase measure side-by-
side demonstrates that the calculated measure correctly computes the
difference between the current value and the one a month ago.
In the next series of slides I will use this calculated measure to construct
a KPI.

33. What is a Key Performance Indicator? (KPI)
33
Every KPI starts life off as a measure, presumably, a measure that is an indicator of company
performance. With each KPI, we assume that the company has established a target value –
goal – of what that indicator should be. For instance, sales revenue might be a performance
indicator. The goal might be to sell at least $100,000 in a given quarter.
The KPI will calculate the difference between the goal and the actual result. We assume the
company can assess those differences declaring them as either good, so-so, or bad. For
instance, the company may say, revenue > 100,000 is good, 90,000 to 100,000 is so-so, and
revenue less than 90,000 is bad.
This brings us to an essential distinguishing feature of the KPI: a graphical icon, known as an
indicator that is displayed to communicate the status of the KPI to the end user. That graphic
might be a happy face to show good, a neutral face to show so-so, and a frowning face to show
bad. Traffic lights with green, amber, and red are often used. The choice of graphics is up to
the client.
Setting up a KPI in Analysis Services entails computing a value of status. The difference
between the indicator and the goal is calculated, and the differences that are “good” are
mapped to the number 1, so-so to 0 and bad to -1. That number is the KPI’s status.
Optionally, you can define a trend for the KPI. The trend shows if, over time, the
performance measure has been moving upwards or downwards.

34. KPI Summary
•Begin with a measure indicating company performance
•Have goals associated with that performance measure
•Translate the difference between performance and goal into
its status with values of 1, 0, -1 (corresponding to good, so-so,
and bad)
•Display the status of the performance measure to the user as a
graphic
34

35. 35
You define KPIs from the KPI tab of the Cube browser. In this simple
illustration, our calculated measure, “Units difference” is the performance
indicator, and the goal is a constant value of 180. MDX expressions can be
used to provide more complex goal statements.

36. 36
Once you have defined the KPI, it may be inspected in the browser tab of the
KPI tab. Here you see the performance metric has a value of 179, just under
the target value. This is “so-so” and you see the neutral face showing.