MDX A Practical View

Glenn Schwartzberg

These slides the opinions of the presenter and do not constitute official positions of

Oracle or any other organization. This material has not been peer reviewed

and is presented here in spite of the better judgment of the presenter

Disclaimer

Why this presentation?

This presentation will attempt to demystify MDX because

I sit through most presentations and at the end say Huh? (No offence Gary, most MDX presentations are over my head)

Because of that

I will attempt to show the basics with live practical examples.

The examples are simple enough that even I can understand them and use them.

Agenda

MDX Basics

What the heck are we talking about

Members

CurrentMember

Tuples

Sets

CrossJoins

Emptyset

(NonEmpty)

Properties

Agenda -

Continued

Examples

How to get Descendants

Solve Order

Rank

ISValid

(is it important?)

Time Balance

MTD/YTD functionality

Lag/lead beginning period

Rolling Average(single

and multiple time dims)

MDX?

A common Multidimensional Expression language that is common to OLAP products including:

Microsoft Analysis Services

Cognos

Oracle BI

Essbase

Its a query language -

Not a fully formatted

reporting language

Need to have some sort of front end program to format it.

Two uses with Essbase

Query data from a database

Block storage or Aggregate storage

Select {[Sales],[COGS]} on Axis(0){[Market].Levels(0).members} on axis(1)From ASOSamp.BasicWhere [Jan]

Create Member formulas

Aggregate storage only[Sales]/[Cogs]

A Few Basics

Members

Just like in Essbase a single item in a hierarchy

Represented as [MemberName] or

[Parent].[MemberName] or

[Dimension].[Parent].[MemberName] etc.

Note braces instead of double quotes

Last option necessary if you allow duplicate member names

Currentmember

Very powerful

One of the most important concepts in MDX

Only use it if necessary

[Jan] is more efficient than [Jan].Currentmember

Determines location for a starting point of a calculation Product.currentmember

Determines relative location for other functions Product.currentmember.parent

Can be used to loop through members, but for a calculation a single member is considered

Tuples

A combination of one or more dimensions to form a distinct intersection

(like Cross dim ->

in calc scripts)

Only one member from a dimension allowed

Every cell in the database is represented by a single tuple

with all of the dimensions represented

Enclosed in Parens

Examples: ([Sales],[New York],[100-10], [Actual],[Jan]) ([Profit],[Markets],Products])

Sets

Multiple members with the same dimensionality

Can include 0, 1, or multiple Tuples

Wrap in curly braces {} {[Period].levels(0).members} {([Scenario].[Actual],[Measures].[Sales]), ([Scenario].[Actual],[Measures].[Margin])}

Why 0 tuples?

Good for getting lists

Crossjoins

Combines multiple sets together to create a superset

of all the sets

Enclosed in Parens(CrossJoin({[Jan],[Feb]}, {[Y-2009]}))

= ([Jan], [Y-2009]) and ([Feb], [Y-2009])

You can nest cross joins by using commas and parens

Crossjoin(CrossJoin({[Jan],[Feb]}, {[Curr Year]}) ,{[Measures].[Units],[Measures].[Returns]})

Crossjoin

Example

SELECTCrossjoin(CrossJoin({[Jan],[Feb]},{[Curr Year]}) ,{[Measures].[Units], [Measures].[Returns]}) on Axis(0),

{[Payment Type].levels(0).members} ON axis(1)from ASOSamp.Sample

Properties

Used to get alternate member information

Member_Alias

(Default)

Member_Name

Gen_Number

Level_Number

Member_Unique_Name

IS_Expense

Comments (Hmmm -

cant do this with other tools)

Properties

Member names and Aliases

Return is pretty ugly

({[Geography].Levels(2).members})PROPERTIES [MEMBER_Name] ON axis(1)

Removing Empty Rows

We add the Non Empty directive to remove missing rowsNON EMPTY ({[Geography].Levels(2).members})PROPERTIES

[MEMBER_Name] ON axis(1)

Does not work well if you have multiple axis (I find you need to

cross join them instead)

Another way

Use a filter (which is probably better)

Like Where

syntax (just like SQL)

Different from MDX Where clause

Can be a member

Or a comparison

Can be crossjoins

Filter({[Geography].Levels(2).members},[Measures]. [Units] > missing)

More on NonEmpty

Version 11 offers new commands

NonEmptyMember

NonEmptyTuple

NonEmptySubset

Use for Formulas (same dimension(s) as in formula)

Can significantly speed up calculations especially complex calculations

Use NonEmptyTuple

if it takes more than one dimension to determine missing

Getting Descendants

In Calc formulas you could use @Descendants

There is not a direct equivalent in MDX

Instead we use Ancestors

Sounds confusing

How can we get descendants

looking at ancestors?

It is because

The order of the members in the statement makes a difference

IsAncestor([stores].currentmember,[Great Buys]) (If the current member is an ancestor of Great Buys)DOES NOT EQUALIsAncestor([Great Buys],[stores].currentmember) (If Great Buys is an Ancestor of the current member)

Lets take a look at a simple example:Case When IsAncestor([stores].currentmember,[Great Buys] ) Then 1 Else 2 EndCase When IsAncestor([Great Buys],[stores].currentmember ) Then

1 Else 2 End

Solve Order

Two Pass on steroids

Allows you to set the order of calculation

Best way to think of it is

this happens before that

Multiple types of Solve order

Dimension

Member

WARNING

-

The order of precedence is different

in different Essbase versions

In 7X-

the first equal wins in 9X and 11X the last

wins

Rank

No I dont mean the person sitting nearby that has not showered (or am I?)

Use a With clause to create the member set to compare

WITH MEMBER [Measures].[Sales_Rank] AS 'Rank(Product.CurrentMember, Product.levels(0).members)' SELECT {Sales, COGS, [Measures].[Sales_Rank]} ON COLUMNS, { Product.levels(0).members} ON ROWS FROM Sample.Basic;

ISValid

Checks for valid intersections

Using Prevmember, NextMember, Parent,Children

etc

Is this important?

Only if you want correct answers. Example:

Jan = 200, Feb = 300

Current month

prior month

Easy for Feb 300

200 = 100

What about Jan

Jan -

??? = 200??

Time.currentmember.firstchild

What if you are at the lowest level already?

ISValid

([Time].currentmember,[Units]) -

([time].currentmember.prevmember,[Units])

CASE WHEN (ISValid([Time].currentmember.prevmember)) Then([Time].currentmember,[Units]) -([time].currentmember.prevmember,[Units]) ELSE Missing END

Time Balance

In 11.X can do it natively, so why discuss it

Gary will tell you in his ASO presentation about not making accounts the accounts dim.

You will need another way to do it.

Tag Time balance members with UDAs

You can use similar logic for expense reporting or other special processing)

QTD/YTD functionality

Not built into ASO cubes

You could do this with alternate rollups

Downside. Problems if

You are using Time Balance (Periods needs to be stored)

You want to use certain functions like ranges

So how do you do it?

QTD/YTD functionality

Create a View or Analytic dimension

Must be a Dynamic dimension

Store the first member

Other members are calculated off of it

Does it matter if the members are tagged with + or ~

It depends. If dimension is label only then no, if not then yes

QTD/YTD Continued

QTD=SUM(PeriodsToDate([Periods].Generation s(2), [Periods].CurrentMember), [View].[Periodic])

YTD=SUM(PeriodsToDate([Periods].Generation s(1), [Periods].CurrentMember), [View].[Periodic])

Note

In Version 11, If you use timebalance use Aggregate instead of sum

Other COOL manipulations

Lead,Lag,Firstchild,Lastchild, FirstSibling, LastSibling, ParallelPeriod.ClosingPeriod, etc.

Good for ANY dimension

Really good for time dimensions.

Lead and Lag are like @Next and @prior or @PrevSibling or @NextSibling

(New calcscript commands in 11X)

No easy equivalent for Firstchild and Lastchild in Calcscripts.

Lead and lag less efficient than FirstChild, ParallelPeriod, etc

Example

SELECT{Sales} on axis(0) ,{[Jan],[Jan].lead(1),closingperiod([year].levels(0))} on Axis(1)FROM [Sample.basic]

Rolling Averages (One time dim)

Easiest scenario

Just have to lead or lag a range(dont forget ISVALID)

with member [year].[rollavg] as 'AVG({[Jan]:[Jan].lead(11)}) 'SELECT{[sales]} on axis(0) ,{[year].[rollavg]} on Axis(1)FROM [Sample.basic]

Rolling Averages (two time dims)

Documentation says it skips missing values, but in this case did not

Use Avg and TupleRange to get members

Use Case (or IIF) to make sure you are at proper level (if required) Necessary in my case as I hard coded end points.

Hard coded end points is more efficient than using relative positioning (faster querying)

Case when (IsLeaf([time].currentmember) and not (IsEmpty([time].currentmember))) then

AVG ({TUPLERANGE(

([years].currentmember.lag(1), [Time].currentmember.NextMember), ([Years].currentmember.lag(1), [Dec])), TUPLERANGE(

([Years].currentmember, [Jan]), ([Years].currentmember, [Time].currentmember))}

, [Measures].[Units]) Else

MissingEND

Head or Tail

Used to get a subset of a range (another way to get range)

For example

I want the a 14 month rolling average with two time dimensions.

Needs to be dynamic based on the selected time period

Could span 3 years depending on what period I start with

Will not work well if you have alternate rollups in periods.

Head Example

SELECT {} ON AXIS(0),{HEAD(([FY2010], [Jan]) : ([Years].LastChild, ClosingPeriod([Periods].Levels(0))), 14)} ON AXIS(1)

FROM [KScope.Test]

To make dynamic change

[FY2010], [Jan] to

[Years].currentmember,

[periods].currentmember

Flow Members

New in 11X

A Flow tag can be assigned to Accounts dimension members bearing formulas

Assume you have Sales and Additions figures for all 12 months. You want to perform an aggregation to populate each months beginning inventory.

Sales Additions InventoryJan 5 1 50Feb 6 3 46Mar 4 2 43Apr 7 0 41

Could do Inventory = SUM(MemberRange(Jan:Time.CurrentMember), (Additions - Sales)) + Beg_Inventory

OR

(Addition Sales), and tag the member as Flow

Restrictions on Flow

Restrictions on Alternate Hierarchies

If alternate hierarchies are used in the aggregate storage time dimension, the following restrictions apply when using Flow and TB tags on the Accounts dimension:

The shared level among alternate time hierarchies must be level 0.

The order of members at shared level among alternate time hierarchies must be the same in all alternate hierarchies.

gschwartzberg@interRel.com

MDX A Practical ViewGlenn Schwartzberg These slides the opinions of the presenter and do not constitute official positions of Oracle or any other organization.This material has not been peer reviewed and is presented here in spite of the better judgment of the presenterWhy this presentation? AgendaAgenda - ContinuedMDX?Two uses with EssbaseA Few BasicsCurrentmemberTuplesSetsCrossjoinsCrossjoin ExamplePropertiesPropertiesRemoving Empty RowsAnother wayMore on NonEmptyGetting DescendantsSlide Number 20Solve OrderRankISValidISValidSlide Number 25Time BalanceQTD/YTD functionalityQTD/YTD functionality QTD/YTD ContinuedOther COOL manipulationsExample Rolling Averages (One time dim)Rolling Averages (two time dims)Slide Number 34Head or TailHead ExampleFlow MembersSlide Number 38Restrictions on FlowSlide Number 40