Presentation is loading. Please wait.

Presentation is loading. Please wait.

SSAS 2012 Tabular Mode Best Practices

Similar presentations


Presentation on theme: "SSAS 2012 Tabular Mode Best Practices"— Presentation transcript:

1 SSAS 2012 Tabular Mode Best Practices
Itay Braun CTO & BI Architect, Twingo Manager of the Microsoft BI User Group in Israel

2 Target Understanding what is exactly SSAS 2012 Tabular Mode and when to use it Understanding the “small letters “ Exact Limitations

3 PowerView Great Visualization functionality Simple and intuitive BTW,
Part of SharePoint 2010 Enterprise Edition Can only query SSAS 2012 Tabular Model

4 PowerView Great Visualization functionality Simple and intuitive BTW,
Part SharePoint 2010 Enterprise Edition Can only query SSAS 2012 Tabular Model

5 Agenda General BI Solution Architecture What is SSAS 2012 Tabular Mode
BISM and other Main Features Direct Query DAX vs. MDX Processing Security Clients PoC – Lessons Learned

6 BI Solution Architecture

7 BI Solution Requirements
Business Requirements: single version of the truth unified data KPI and CPM Dashboards Reports… Technology is just an Enabler Main Technological Requirements: Easy to change and maintain Security Good performances Self Service BI Good visualization layer

8 Using a New Technology What does it do? When should I use it?
And when shouldn’t I use it? What’s it place in the technological road map? Two examples Similar Technologies Who are the biggest competitors? Which old technology replaced by the new one?

9 SSAS 2012 Tabular Mode Customers ask:
Does it replace good old MOLAP or is it just a better PowerPivot? From now on, should I develop in Tabular Mode only? We understand that in-memory column database is faster. How fast is Tabular compared to MOLAP? We understand that development at Tabular Mode is simpler. We are SSAS 2008 R2 experts and know nothing about DAX and Tabular. Would you still say Tabular is simpler option for us?

10 Is SSAS 2012 Tabular Mode Enterprise Ready?

11 What is SSAS 2012 Tabular Mode?

12 What is SSAS 2012 Tabular Mode
New DB Engine Tabular Mode Column Store In-Memory Compression (about 1:3; sometimes much better) All data copied (processed) to the Tabular model Direct SQL Queries translated to SQL, running directly against the relational SQL Server. Tabular Mode’s version of ROLAP

13

14 BISM – BI Semantic Model
BISM is a concept, not a technical term Users can easily query the intuitive data model No need to understand complex data source structure The BISM provides: Data Model Business Logic and Queries Data Access The implementation of the BISM are these three technologies: MOLAP SSAS Tabular Mode PowerPivot

15 Tabular vs. Multi Dimensional
Quicker development From detailed (granular, images) data to aggregate data Good performances Some features are missing Multi-Dimensional Rich Multi-dimensional functionality Known and well documented engine Suitable for very large cubes, beyond server’s memory size

16 Schema (SQL Server Data Tools)

17 Management Studio

18 Tabular Mode - Main Features
Understanding the main features of a technology is the key to understand when and how to use it Feature by Server Mode or Solution Type

19 Developing a Model Use the Visual Studio SQL Server Data Tools to build a BISM Open an existing PowerPivot Model using SSDT or Management Studio Deployment Immediate changes implementation at the SDDT Use small DB for development

20 Workspace Database Created during model authoring using SSDT.
Disappeared automatically when closing the project The workspace database resides in-memory. Workspace DB

21 Data Sources The model can use multiple data sources

22 Tabular Mode Schema vs. Data Source Views
Much simpler than a Data Source View Building the AdventureWorks DB using Tabular Mode is about x4 times faster.

23 Tabular Mode Schema Dimension = One Table
No Dimension Wizard to create Hierarchies

24 Tabular Mode Schema Measures and KPI are defined here. No cube Wizard.

25 Tabular Mode Schema Edit Table Properties

26 Tabular Mode Schema- Adding a column
Adding new calculated columns in the schema is possible. Either add calculated columns to the source DB or to the BISM

27 Adding a new columns Add new columns using DAX
At the OLAP Data Source View you write an expression in the source language, TSQL, PLSQL…

28 Schema - Limitations Self Joins are not supported
Two tables can have only one active relationships No role playing dimension Many to Many relationships allowed using DAX

29 Dimensions Active / non active relationships
Only ONE relationship can be active No Role Playing Dimensions The Date Dimension can be connected by either OrderDate or ShipDate or DueDate You’ll have to build many date dimensions

30 Main Features / Capabilities
Import Existing PowerPivot model Let users to quickly build model using Excel and upload it to the server in a click

31 Dimensions Dimensions Multiple Hierarchies

32 Category name from Dim Category with a measure works as expected
Snowflake Dimensions A dimension is based on one table / view only. A hierarchy can be based on one Table. Create a view on the data source which joins the tables. . Category name from Dim Category with a measure works as expected

33 Dimensions - Limitations
No member properties No Parent-Child No linked dimensions No need for Attribute Relationships The dimension is a table A dimension can be based on one and only one table Dimension names taken directly from the Large dimensions (million members ) performed better

34 Measures Sum, min, max, count, distinct count…or complex DAX Expressions A table can contain both attributes and measures.

35 Multi Grain Measure Groups
Support many “Measure Groups” with different dimensionality For ex. Internet Sales (product level), Sales Quota (Category Level)

36 KPI Key performance Indicators are the heart of any BI solution
Replaces existing Measures

37 KPI Easy to create using a wizard Value – existing measure
Target – measure or a number Status – using a GUI

38 Perspectives Same as in SSAS Multidimensional
Allows users to see a Database in a simpler way Can hide: Tables (dimensions) Columns (Attributes) Hierarchies Measures KPI

39 Partitions Same logic as in SSAS Multidimensional. Easy to create
Can be processed alone (like in Multidimensional)

40 What’s missing? Translations Sync Two Databases Actions
Custom Assemblies (SSAS Procedures) Self Join (Parent Child) Role Playing Dimensions

41 More Limitations Named Set Scopes Write Back

42 Direct Query

43 DirectQuery Mode DirectQuery mode uses data that is stored in a SQL Server database. Used for Real Time analysis No processing Changes to the underlying source data reflected immediately. no extra management overhead of having to maintain a separate copy of the data.

44 Direct Query Performances
No clear answer, depends on the source DB Still checking the efficiency of the generated SQL Queries. Security - Any security enforced by the back-end database is guaranteed to be enforced, using row-level security Unlike Multi Dimensional, you Can’t mix Direct Query and Tabular Partitions

45 DirectQuery Mode It is possible to have a model over data sets that are too large to fit in memory on the Analysis Services server.

46 Direct Query Limitations
Can only use one SQL Server DB as a source The entire model is either is either using DirectQuery or not. Limited use of DAX functions Client restrictions: Can only be queried by using DAX Excel can’t be used because it uses MDX

47 DAX vs. MDX

48 DAX vs. MDX MDX (Multi Dimensional Cube) vs. DAX
MDX (Tabular) vs. DAX (Tabular)

49 Querying a Tabular Model using MDX
Quick start – just use MDX to query the model You don’t have to learn DAX In general speaking, MDW performs well Fine tuning DAX queries

50 What is DAX Set of Excel-like formulas
Enables advanced data modeling inside the PowerPivot for Excel or Tabular Mode Two types of calculations calculated columns measures

51 Learn DAX QuickStart: Learn DAX Basics in 30 Minutes
Chris Webb’s blog. 6 lessons about DAX start here Converting MDX to DAX – First Steps

52 DAX vs. MDX DAX and MDX coexists peacefully
Tabular Mode can be queried using both DAX and MDX Use MDX for data analysis (group by, hierarchies) Use DAX for detailed reports MDX uses Crossjoin to join many columns. - Excel uses MDX, PowerView uses DAX, SSRS can use both

53 DAX vs. MDX More Limitations MDX can only query In-Memory Models
DAX can query both In-Memory and Direct Query Models. DAX cannot be used to query Multi Dimensional Cubes Microsoft are working on it

54 (Tabular + DAX) vs. (MOLAP + MDX)
The fact that VertiPaq is an in-memory database doesn't mean that it will perform much better than a multidimensional cube. Warm Cache MOLAP Cube query will probably perform better or just as good as DAX. Query results caching

55 (Tabular + DAX) vs. (MOLAP + MDX)
DAX on Tabular mode usually give you good performance without special tuning. DAX queries are never cached Second (warm) run is just as fast the first (cold) one. Distinct Count queries usually performed better using DAX + Tabular Overall good performance of DC even with MDX

56 When number of days increased – the runtime of MDX(CUBE) query is better then DAX (Tabular)

57 (and much faster from the warm cache) than in DAX.
Complex query that run on list of members (Products and Dates) and make aggregations – will return faster in MDX (Cube) (and much faster from the warm cache) than in DAX. MDX Query Same Query in DAX

58 DAX in XMLA Easy to manage parameters DAX Editor sample

59 Processing

60 Processing Highlights
Tables can be partitioned Process FULL Process ADD (incremental) One table (dimension) can be processed without processing the related measures

61 Processing

62 Processing Error It says “processed” but the error message and a closer look on the partitions shows the table has never been processed

63 Processing the dimensions but not the measure tables
Process the dimensions Process the measures Unlike MOLAP, It is possible to query tables (dimensions and measures ) while other tables haven’t been processed yet. Processing the dimensions but not the measure tables

64 Partitions

65 Partitions Every table can be partitioned.
Since Dimension = Table, a dimension can be partitioned too. Not sure this is a good idea.

66 Partitions It is possible to process only some of the partitions

67 Processing Options

68 Process Full Processes a partition or table and all the objects that it contains. Drops all data, and then processes the object. Any affected calculated columns, relationships, user hierarchies, or internal engine structures (except table dictionaries) are recalculated. This kind of processing is required when a structural change has been made to an object.

69 Processing Options Process Clear - Drops all the data in a database, table, or partition. Process Data - Loads data into a partition or table. Process Defrag - Optimizes the table dictionary (an internal engine structure) For ex. After deleting old partition, the dictionary still contains references to these deleted rows. The best practice is to run it frequently, based on your sliding window design

70 Processing Options Process Recalc - recalculates calculated columns, rebuilds relationships, rebuilds user hierarchies and rebuilds other internal engine structures. Must be issued after Process Clear or Process Data.

71 Processing Options Defrag Recalc Add Data Clear Full Default V DB
Table Partition

72 Handling Unknown Values
A dimension key in the Fact table might be missing in the dimension table. For ex. A sale of $ for a non-existing product It is not possible to configure the processing behaviour. All unknown rows merged into one “unknown’ member

73 Using Views in the Tabular Model
Use views to filter small amount of data at the development phase, later change the view definition to return all data. Using views let you do changes in the underlying DB without affecting the Tabular Model Schema.

74 ProcessAdd The simplest solution – add a new partition + Full processing It is possible to run process add to load new data. It is under the responsibility of the DBA to avoid duplication. You can change the query binding using the UI. Use views as much simpler query binding.

75 Parallel Processing It is possible to process many tables on parallel.
However, only one partition of a table at a time.

76 Security

77 Security Two main permissions: Create roles using SSDT.
Database Permissions Allowed Row Sets Create roles using SSDT. Use DAX to restrict access =Region[Country]=”USA” Dynamic security is also possible

78 Security Create a role

79 Add the role name to the connection string Always uses Visual Total
View filtered data

80 PoC – Lessons Learned

81 PoC – Lessons Learned Forget what you know about SSAS 2008, new design is needed No Migration Wizard Consider de-normalization. Server Configuration Server Properties (SSAS - Tabular) Memory paging allows models to be larger than the physical memory of the server

82 Project Properties - Query Mode
Defualt: In-memory Direct query Direct Query with In-Memory In-Memory with Direct Query Configuration

83 PoC – Lessons Learned Full process of a whole DB might consume too much memory. Consider running queries one by one.

84 Clients

85 PowerView A feature of SQL Server 2012 Reporting Services Add-in
Requires Microsoft SharePoint Server 2010 Enterprise Edition Interactive data exploration and visualization Intuitive ad-hoc reporting for business users A browser-based Silverlight application

86 Clients - PowerView For only SSAS 2012 Tabular Mode or PowerPivot.
SSAS 2012 Multi-dimensional as data source is planned for the near future. Watch demo here More info at MSDN

87

88 Clients – Pyramid Analytics
The Pyramid Analytics Suite is a highly scalable, enterprise data analytics application that brings sophistication with simplicity in a synergistic web-based interface for consuming Microsoft Analysis Services OLAP cubes. BioPoint Dashboards BioXL - A complete browser-based cube viewing solution

89

90 Clients – Excel 2010

91 Clients - Custom Build your own app using AMO

92 Monitoring Good monitoring is essential for Enterprise Ready solutions. Profiler Dynamic Management Views Performance Monitor No query log, though

93 Licensing Tabular Mode not in Standard Edition

94 Conclusion Multi Dimensional OLAP: Tabular Mode
Familiar, Huge Install base, wide knowledgebase, stable, large scale implementations, monitoring tools, many OLAP Viewers Tabular Mode New Technology Simple, easier development, sometimes faster Microsoft now focusing on this technology

95 Tabular vs. Multi Multi: Optimized disk system
Prallel processin gof part Building aggs Unary operators Block Computation Tuning IO Thread Pool – how many threads can query the partitions Multi User settings Coordinator query blanacing factor Coordinator query boost priority level

96 Tabular Memory, memory, memory Don’t let the system memory page out
DAX query performances, writing patterns No prallel processing of partitions Query performances vs processing Adjusting compression levles lets you favour processins time versus query time No NUMA issues

97 Links My Blog: http://blogs.microsoft.co.il/blogs/itaybraun
Cathy dumas’ blog: DAX Editor - Paul Te Braak’s blog AdventureWorks tutorial for tabular models DAX quick start guide

98 Thanks


Download ppt "SSAS 2012 Tabular Mode Best Practices"

Similar presentations


Ads by Google