Handling Evolutions in Multidimensional Structures

1
Handling Evolutions in Multidimensional
Structures

• Presented by Jay Kothari

M. Body, M. Miquel, Y. Bédard, and A.
Tchounikine, Handling Evolutions in
Multidimensional Structures, ICDE, 2003.
2
Overview

• Multidimensional systems require gathering data
  from heterogeneous sources throughout time
• Data is integrated in Multidimensional structures
  organized around several dimensions.
• Structures are likely to vary over time

3
Problem for DW Designer

• Do you keep the trace of the evolutions?
• Do you map all data in a given version of the
  structure?

4
Tradeoffs for Problems

• Do you keep the trace of the evolutions?
• Limits the capability of comparison for analysts
• Do you map all data in a given version of the
  structure?
• Entails alteration and (potential) loss of data

5
Goal

• Tack History
• Compare data mapped into static structures

6
Introduction DW

• DW single site repository of information
  collected from multiple sources where the
  information is organized around major subjects
• DM Requirements
• Must be easy for the end-user to understand
• Must maximize efficiency of the queries

7
DW Models

• Multidimensional, or hypercube
• Represent measurable facts and the dimensions
  that characterize the facts
• Retail Example
• Facts price, amount of a purchase
• Dimensions product, location, time, customer
• Location City, State, Country
• Star Schema models data as a simple cube
• Hierarchical relationship is not explicit, but
  encapsulated
• Snowflake Schema normalizes dimension tables
• Represent hierarchies by identifying dimensions
  in various granularities
• Galaxy fact constellation, allows for the
  collection of stars when multiple fact tales are
  needed

8
Multi-tier OLAP Architecture

• Tier 1 Warehouse Server
• Tier 2 Data Mart
• Tier 3 OLAP Server
• Tier 4 OLAP Client

9
OLAP Architecture Warehouse

• Tier 1 Warehouse Server
• Tier 2 Data Mart
• Tier 3 OLAP Server
• Tier 4 OLAP Client

10
Multi-tier OLAP Architecture

• Tier 1 Warehouse Server
• Tier 2 Data Mart
• Tier 3 OLAP Server
• Tier 4 OLAP Client

11
Multi-tier OLAP Architecture

• Tier 1 Warehouse Server
• Tier 2 Data Mart
• Tier 3 OLAP Server
• Tier 4 OLAP Client

12
Multi-tier OLAP Architecture

• Tier 1 Warehouse Server
• Tier 2 Data Mart
• Tier 3 OLAP Server
• Tier 4 OLAP Client

13
Considering Time

• Multidimensional models consider
• Facts are dynamic
• Dimensions are static
• Time is recorded as a Time-dimension
• Often, changes occur on analysis structure

14
Considering Evolution Mapping

• OLAP systems map data into most recent analysis
  structure
• Allows for temporal comparison
• Loses data

15
3 Slowly Changing Dimensions

• Three possible ways of handling changes in a
  structure
• Update data structure
• Keep all versions of members
• All evolutions are kept inside members

16
Problems with SCDs

• Three possible ways of handling changes in a
  structure
• Update data structure
• Doesnt track history
• Keep all versions of members
• Comparisons across transitions cannot be made,
  since links are not kept, even if evolutions are
• All evolutions are kept inside members
• Overlapping between versions may occur and cannot
  be handled

17
Motivational Case Study

• Studying Institution Restructuring
• Fact table
• Amount Measure
• Hierarchical Time measure year
• Hierarchical Org. dimensiondiv.gtdept.

18
Dimensions and Data
19
Query 1 Total Amount by Year/Div.
20
Q2 Split Joness Dept
21
New Proposed Models

• Mendelson and Vaisman
• End-user needs
• temporal multidimensional model based on
  timestamps
• Define TOLAP that allows the end user to choose
  aggregation
• Allows end-user to choose b/w temporally
  consistent or last version
• Track links b/w transitions
• Only allows for last version reporting
• Eder and Koncilia
• Mapping functions to allow conversions b/w
  structure versions
• Store links across transitions (old and new
  member)
• Does not consider schema evolutions nor time
  consistent presentation

22
This approach..

• Handle Evolutions in multidimensional structures
• Solutions to exploit knowledge on evolutions in
  order to map data in a given representation

23
Changes occurring on a multidimensional schema
- Creation and deletion of a dimension. -
Creation and deletion of a hierarchy. - Creation
and deletion of a level. - Move of a level in
the hierarchical schema structure. - Creation
and deletion of a measure.
24
Evolutions on Instances of Dimensions 6 Operators

• Creation of a dimension member.
• Deletion of a dimension member.
• Transformation of a member (change of an
  attribute, its name or meaning...).
• Merging of n members into one member.
• Splitting of one member into n members.
• Reclassification of a member in the dimension
  structure.

25
Deducing Hierarchical Models

• Deduced from dimension instances
• No longer necessary to define schema structure
• Current MD models do not support complex
  hierarchical structures, and by not imposing a
  structure, the model gains genericity
• Consider non-onto, non-covering and multiple
  hierarchies
• Every evolution on schema has an impact on its
  instance
• Accounts both evolutions on schema and instances

26
Temporal Multidimensional Model

• Definition 1 Member Version
• Object or an abstract entity of special interest
• Definition 2 Temporal Relationship
• Establishes explicit hierarchy link between two
  member versions and represents a rill-up function
• Definition 3 Temporal Dimension
• Directed graph where nodes are member versions of
  a set and the arcs are the relationships of the
  Temporal Relationships
• Definition 4 Levels in a dimensions
• Set of member versions having the same level
  value if defined, or the depth in the DAG of a
  Temporal Dimension
• Shows how the temporal dimension evolved over
  time

27
Temporal Multidimensional Model 2

• Definition 5 Temporally Consistent Fact Table
• Keeps links across transitions given by the
  temporal dimensions, time dimension, and set of
  measures
• Definition 6 Confidence Factor
• Reliability of Data
• Used to distinguish between source and mapped
  data
• Definition 7 Mapping Relationship
• Allows for dealing with the problem of keeping
  links between transitions
• Example Gives the mapping from the previous
  department problem of percentages
• Definition 8 Temporal Multidimensional Schema
• Relates Temporal Dimension, Time Dimension,
  Mapping Relationships, and a Temporally
  consistent Fact Table
• Definition 9 Structure Version
• Defines valid time of a structure version and
  restricts it to its elements
• Valid and and unchanged structure over its given
  valid time
• Example What structures in the case study are
  valid during a given time frame (Paul and Bill)
  b/w 01/01 and 12/02

28
Last set of Definitions

• Definition 10 Temporal mode or Presentation
• Satisfies end-user needs of presentation of
  requests
• Definition 11 Multiversion Fact Table
• Calculated from temporal dimensions, Mapping
  Relations, and Temporally consistent fact table
• Definition 12 Data Aggregation
• Computes DA of a cube based on MultiVersion Fact
  Table and Temporal Relationships between Member
  Versions.

29
Evolution Operators

• Insert
• Inserts member version
• Exclude
• Excludes member versions from a dimension
• Associate
• Defines a new transition link b/w two versions of
  a member, and if consistent then it will be added
  to the set of Mapping Relationships
• Reclassify
• Changes the position of a member in the
  hierarchical structure by redefining its parents.

30
Using Evo Operations
31
Logical Temporal Model

• De Facto Multi Dim systems are only made up of
  dimensions and fact tables
• Temporal mode of representation
• Similar to a dimension and a confidence factor
  appearing as a measure

32
More on the Evo Ops

• Difference b/w member and relationships in a
  dimension
• First problem with current tools Hierarchical
  links are stored as keys in child attributes.
• Solution Interpreting change in hierarchical
  structure
• Problem Lots of changes O(bd)
• -gtso they create a new Reclassify operatorwhich
  inserts, excludes and associates.

33
Their Implementation (architecture)

• 3 Parts
• Temporal Datawarehouse
• Contains the Temporal Multidimensional Schema
• Mapping Relationships
• Multiversion Datawarehouse
• TMP can be preceeded and Multiversion fact table
  can be inferred from the Temporally Consistent
  fact table and Mapping Relations
• OLAP cube
• Built from the MVDW using aggregations and allows
  requests to integrate the temporal models of
  presentation

34
Adding MetaData

• Categories
• Related to versions of members
• Related to evolution of members
• Example of Type 2

35
Table of Mapping relationships

• Used to build the MultiVer Fact Table
• User has User can obtain transformations

36
Conclusion

• Focus on designers needs
• Handling the most evo ops
• Allowing to have nontrivial hierarchical
  structures in dimensions to match with real life
  cases
• Big contribution
• TMP
• Changes in analysis Framework
• Confidence facotrs
• Still flawed