Data Warehousing & Data Mining - TU Braunschweig · Data Warehousing & Data Mining Wolf-Tilo Balke Silviu Homoceanu ... • Assisted by artificial intelligence algorithms and pattern

5/29/2009

1

Data Warehousing & Data MiningWolf-Tilo BalkeSilviu HomoceanuInstitut für InformationssystemeTechnische Universität Braunschweighttp://www.ifis.cs.tu-bs.de

8. Building the DW

8.1 The DW Project

8.2 Data Extract/Transform/Load (ETL)

8.3 Metadata

DW & DM – Wolf-Tilo Balke – Institut für Informationssysteme – TU Braunschweig 2

8. Building the DW

• Building a DW, is a complex IT project

– A middle size DW-project contains 500-1000 activities

• DW-Project organization

– Project roles and corresponding tasks, e.g.:


8.1 The DW Project

Roles Tasks

DW-PM Project Management

DW-Architect Methods, Concepts, Modeling

DW-Miner Concepts, Analyze(non-standard)

Domain Expert Domain Knowledge

DW-System Developer System- and metadata-management,

ETL

DW User Analyze (standard)

• DW-Project usual tasks

– Communication, as process of information exchange between team members

– Conflict management

• The magical triangle, compromise between time, costs and quality

– Quality assurance

• Performance, reliability, scalability, robustness, etc.

– Documentation


8.1 The DW Project

• Software choice– Database system for the DW

• Usually the choice is to use the same technology provider as for the operational data

• MDB vs. RDB

– ETL tools• Differentiate by the data cleansing needs

– Analysis tools• Varying from data mining to OLAP products, with a focus on

reporting functionality

– Repository• Not very oft used

• Helpful for metadata management


8.1 The DW Project

• Hardware choice

– Data storage

• RAID systems, SAN’s, NAS’s

– Processing

• Multi-CPU systems, SMP, Clusters

– Failure tolerance

• Data replication, mirroring RAID, backup strategies

– Other factors

• Data access times, transfer rates, memory bandwidth, network throughput and latency


8.1 The DW Project

5/29/2009

2

• Project Timeline, depends on the development methodology, but usually– Phase I – Proof of Concept

• Establish Technical Infrastructure• Prototype Data Extraction/Transformation/Load • Prototype Analysis & Reporting

– Phase II – Controlled Release• Iterative Process of Building Subject Oriented Data Marts

– Phase III – General Availability• On going operations, support and training, maintenance and

growth

• The most important part of the DW building project is defining the ETL process


8.1 The DW Project

• What is ETL?

– Short for extract, transform, and load

– Three database functions that are combined into one tool to pull data out of productive databases and place it into the DW

• Migrate data from one database to another, to form data marts and data warehouses

• Convert databases from one format or type to another


8.2 ETL

• When should we ETL?

– Periodically (e.g., every night, every week) or after significant events

– Refresh policy set by administrator based on user needs and traffic

– Possibly different policies for different sources

– Rarely, on every update (real-time DW)

• Not warranted unless warehouse data require current data (up to the minute stock quotes)


8.2 ETL

• ETL is used to integrate heterogeneous systems

– With different DBMS, operating system, hardware, communication protocols

• ETL challenges

– Getting the data from the source to target as fast as possible

– Allow recovery from failure without restarting the whole process

• This leads to balance between writing data to staging tables or keeping it in memory


8.2 ETL

• Staging area, basic rules

– Data in the staging area is owned bythe ETL team

• Users are not allowed in the stagingarea at any time

– Reports cannot access data from the staging area

– Only ETL processes can write to and read from the staging area


8.2 ETL

• ETL input/output example


8.2 ETL

5/29/2009

3

• Staging area structures for holding data

– Flat files

– XML data sets

– Relational tables


8.2 Staging area data structures

• Flat files– ETL tools based on scripts, such as

Perl, VBScript or JavaScript

– Advantages• No overhead of maintaining metadata as DBMS does

• Sorting, merging, deleting, replacing and other data-migration functions are much faster outside the DBMS

– Disadvantages• No concept of updating

• Queries and random access lookups are not well supported by the operating system

• Flat files can not be indexed for fast lookups



• When should flat files be used?

– Staging source data for safekeeping and recovery

• Best approach to restart a failed process is by having data dumped in a flat file

– Sorting data

• Sorting data in a file system may be more efficient as performing it in a DBMS with Order By clause

• Sorting is important: a huge portion of the ETL processing cycles goes to sorting



• When should flat files be used?

– Filtering

• Using grep-like functionality

– Replacing text strings

• Sequential file processing is muchfaster at the system-level than itis with a database



• XML Data sets

– Used as common format for bothinput and output from the ETLsystem

– Generally, not used for persistentstaging

– Useful mechanisms

• XML schema (successor of DTD)

• XQuery, XPath

• XSLT



• Relational tables– Using tables is most appropriate especially when there

are no dedicated ETL tools

– Advantages• Apparent metadata: column names data types and lengths,

cardinality, etc.

• Relational abilities: data integrity as well as normalized staging

• Open repository/SQL interface: easy to access by any SQL compliant tool

– Disadvantages• Sometimes slower than the operating file system



5/29/2009

4

• How is the staging area designed?

– Staging database, file system, and directory structures are set up by the DB and OS administrators based on ETL architect estimations e.g., tables volumetric worksheet


8.2 Staging area storage

Table

Name

Update

strategy

Load

frequency

ETL Job Initial

row

count

Avg

row

length

Grows

with

Expected

rows/mo

Expected

bytes/mo

Initial table

size

Table

Size 6

mo. (MB)

S_ACC Truncate/

Reload

Daily SAcc 39,933 27 New

account

9,983 269,548 1,078,191 2.57

S_ASSETS Insert/

Delete

Daily SAssets 771,500 75 New

assets

192,875 15,044,250 60,177,000 143.47

S_DEFECT Truncate/

Reload

On

demand

SDefect 84 27 New

defect

21 567 2,268 0.01

• ETL

– Data extraction

– Data transformation

– Data loading


8.2 ETL

• Data Extraction– Data needs to be taken from a data

source so that it can be put into theDW

• Internal scripts/tools at the data source,which export the data to be used

• External programs, which extract the data from the source

– If the data is exported, it is typically exported into a text file that can then be brought into an intermediary database

– If the data is extracted from the source, it is typically transferred directly into an intermediary database


8.2 Data Extraction

• Steps in data extraction

– Initial extraction

• Preparing the logical map

• First time data extraction

– Ongoing extraction

• Just new data

• Changed data

• Or even deleted data


8.2 Data Extraction

• Logical map connects the original source data to the final data

– Most important part is the description of the transformation rules


8.2 Data Extraction

Target Source Transformation

Table Column Data Type Table Type DB

Name

Table Column Data Type

EMPL_

DIM

E_KEY NUMBER DIMENSION NUMBER Surrogate key

EMPL_

DIM

COUNTRY VARCHAR(

75)

DIMENSION HR_SYS COUNTRIES NAME VARCHAR(75) Select c.name from

employees e, states s,

countries c where e.state_id =

s.state_id and s.country_id =

c.country

…

• Building the logical map: first identify the data sources– Data discovery phase

• Collecting and documenting source systems: databases, tables, relations, cardinality, keys, data types, etc.

– Anomaly detection phase• NULL values can destroy any ETL process, e.g., if a foreign

key is NULL, joining tables on a NULL column results in data loss, because in RDB NULL ≠ NULL

– If NULL on foreign key then use outer joins

– If NULL on other columns then create a business rule to replace NULLs while loading data in DW


8.2 Data Extraction

5/29/2009

5

• Data needs to be maintained in the DW also after the initial load

– Extraction is performed on a regular basis

– Only changes are extracted after the first time


8.2 Ongoing Extraction

• Detecting changes (new/changed data)

– Using audit columns

– Database log scraping or sniffing

– Process of elimination



• Detecting changes (new/changed data)

– Using audit columns

• Store date and time a record has been added or modified

• Detect changes based on date stamps higher than the last extraction date



– Log scraping

• Takes a snapshot of the database redo log at a certain time (e.g., midnight) and finds the transactions affecting the tables ETL is interested

• It can be problematic when the redo loggets full and is emptied by the DBA

– Log Sniffing

• Pooling the redo log capturing thetransactions on the fly

• The better choice: suitable also for real-time ETL


8.2 Detecting changes

– Process of Elimination

• Preserves exactly one copy of each previous extraction

• During next run, it compares the entire source tables against the extraction copy

• Only differences are sent to the DW

• Advantages

– Because the process makes row by row comparisons, it is impossible to miss data

– It can also detect deleted rows



• Detecting deleted or overwritten fact records

– If records or incorrect values are inserted by mistake, records in ODS get deleted or overwritten

– If the mistakes have already been loaded in the DW, corrections have to be made

– In such cases the solution is not to modify or delete data in the DW, but inserting an additional record which corrects or even cancels the mistake by negating it



5/29/2009

6

• Data transformation

– Uses rules or lookup tables, or creating combinations with other data, to convert source data to the desired state

• 2 major steps

– Data Cleaning

• Mostly involves manual work

• Assisted by artificial intelligencealgorithms and pattern recognition

– Data Integration


8.2 Data Transformation

• Extracted data can be dirty. How does clean data look like?

• Data Quality characteristics:

– Correct: values and descriptions in data represent their associated objects truthfully

• E.g., if the city in which store 1 is located is Braunschweig, then the address should not report Paris.

–Unambiguous: the values and descriptions in data can be taken to have only one meaning


8.2 Data Cleaning

– Consistent: values and descriptions in data use one constant notational convention

• E.g., Braunschweig can be expressed as BS or Brunswick, by our employees in USA. Consistency means using just BS in all our data

– Complete

• Individual values and descriptors in data have a value (not null)

• Aggregate number of records is complete


8.2 Data Cleaning

• The data cleaning engine produces 3 main deliverables:

– Data-profiling results:

• Meta-data repository describing schema definitions, business objects, domains, data sources, table definitions, data rules, value rules, etc.

• Represents a quantitative assessment of original data sources


8.2 Data Cleaning

– Error event table

• Structured as a dimensional star schema

• Each data quality error identified by the cleaning subsystem is inserted as arow in the errorevent fact table


8.2 Cleaning Deliverables

Data Quality Screen:

- Status report on data quality

- Gateway which lets only clean data

go through

– Audit dimension

• Describes the data-quality context of a fact table record being loaded into the DW

• Attached to each fact record

• Aggregates the information from the error event table on a per record basis


8.2 Cleaning Deliverables

Audit key (PK)

Completeness category (text)

Completeness score (integer)

Number screens failed

Max severity score

Extract timestamp

Clean timestamp

…

5/29/2009

7

• Core of the data cleaning engine

– Break data into atomic units

• E.g., breaking the address into street, number, city, zip and country

– Standardizing

• E.g., encoding of the sex: 0/1, M/F, m/f, male/female

– Verification

• E.g., does zip code 38106 belong to Braunschweig?

– Matching


8.2 Cleaning Engine

• Types of enforcement

– Column property enforcement

• Ensures that incoming data contains expected values

• NULL values in required columns

• Numeric values outside the expected high/low ranges

• Columns whose lengths are unexpectedly short or long

• Columns that contain values outside of valid value sets

• Adherence to a required pattern

• Hits against a list of known wrong values (if the list of acceptable values is to long)

• Spell-checker rejects


8.2 Data quality checks

• Structure enforcement

– Focus on the relationship of columns to each other

– Proper primary and foreign keys

– Explicit and implicit hierarchies and relationships among group of fields e.g., valid postal address



• Data and Value rule enforcement

– E.g., a commercial customer cannot simultaneously be a limited and a corporation

– Value rules can also provide probabilistic warnings that the data might be incorrect

• E.g., boy named ‘Sue’ might be correct, but most probably it is a gender or name error and such a record should be flagged for inspection



• Overall Process Flow


8.2 Data cleaning

• Sometimes data isjust garbage

– We shouldn’t loadgarbage in the DW

• Cleaning data manuallytakes just…forever!!!


8.2 Data Quality

5/29/2009

8

• Use tools to clean data semi-automatic

– Open source tools

• E.g., Eobjects DataCleaner, Talend Open Profiler

– Non-open source

• Firstlogic both by Business Objects (now SAP)

• Vality both by Ascential (now IBM)

• Oracle Data Quality and Oracle Data Profiling


8.2 Data Quality

• Data cleaning process

– Use of regular expressions


8.2 Data Quality

• Regular expressions for date/time data


8.2 Data Quality

• Core of the data cleaning engine

– Anomaly detection phase:

• Data anomaly is a piece ofdata which doesn’t fit into thedomain of the rest of the data itis stored with

• “What is wrong with thispicture?”


8.2 Cleaning Engine

• Anomaly detection

– Count the rows in a table while grouping on the column in question e.g.,

• SELECT city, count(*) FROM order_detail GROUP BY city


8.2 Anomaly detection

City Count(*)

Bremen 2

Berlin 3

WOB 4,500

BS 12,000

HAN 46,000

…

• What if our table has 100 million rows with 250,000 distinct values?

– Use data sampling e.g.,

• Divide the whole data into 1000 pieces, and choose 1 record from each

• Add a random number column to the data, sort it an take the first 1000 records

• Etc.

– Common mistake is to select a range of dates

• Most anomalies happen temporarily


8.2 Anomaly detection

5/29/2009

9

• Data profiling

– E.g., observe name anomalies


8.2 Data Quality

• Data profiling

– Pay closer look to strange values

– Observe data distribution pattern

• Gaussian distribution


8.2 Data Quality

SELECT AVERAGE(sales_value) – 3 * STDDEV(sales_value),

AVERAGE(sales_value) + 3 * STDDEV(sales_value)

INTO Min_resonable, Max_resonable FROM …

• Data distribution

– Flat distribution

• Identifiers distributions (keys)

– Zipfian distribution

• Word ranking by frequency

• Web page hits follow this distribution

– Home page many hits

• Some values appear more often than others

– In sales, more cheap goods are sold thanexpensive ones


8.2 Data Quality

• Data distribution

– Pareto, Poisson, S distribution

• Distribution Discovery

– Statistical software: SPSS, StatSoft, R, etc.


8.2 Data Quality

• Data integration

– Several conceptual schemas need to be combined into a unified global schema

– All differences in perspective and terminology have to be resolved

– All redundancy has to beremoved


8.2 Data Integration

schema 2

schemaschema 4schema 3schema 3

schema 1schema 1

• There are four basic steps needed for conceptual schema integration1. Preintegration analysis

2. Comparison of schemas

3. Conformation of schemas

4. Merging and restructuring of schemas

• The integration process needs continual refinement and reevaluation


8.2 Schema integration

5/29/2009

10



schemas with

conflicts

modified

schemas

list of conflicts

integrated

schema

identify conflicts

resolve conflicts

integrate schemas

Preintegration analysis

Comparison of schemas

Conformation of schemas

Merging and restructuring

• Preintegration analysis needs a close look on the individual conceptual schemas to decide for an adequate integration strategy

– The larger the number of constructs, the more important is modularization

– Is it really sensible/possible to integrate all schemas?





Integration strategy

Binary approach n-ary approach

Sequential

integration

Balanced

integration

One-shot

integration

Iterative

integration

• Schema conflicts– Table/Table conflicts

• Table naming e.g., synonyms, homonyms

• Structure conflicts e.g., missing attributes

• Conflicting integrity conditions

– Attribute/Attribute conflicts• Naming e.g., synonyms, homonyms

• Default value conflicts

• Conflicting integrity conditions e.g., different data types or boundary limitations

– Table/Attribute conflicts



• The basic goal is to make schemas compatiblefor integration

• Conformation usually needs manual interaction

– Conflicts need to be resolved semantically

– Rename entities/attributes

– Convert differing types, e.g., convert an entity to an attribute or a relationship

– Align cardinalities/functionalities

– Align different datatypes



• Schema integration is a semantic process

– This usually means a lot of manual work

– Computers can support the process by matching some (parts of) schemas

• There have been some approaches towards (semi-)automatic matching of schemas

– Matching is a complex process and usually only focuses on simple constructs like ‘Are two entities semantically equivalent?’

– The result is still rather error-prone…



5/29/2009

11

• Schema Matching– Label-based matching

• For each label in one schema consider all labels of the other schema and every time gauge their semantic similarity

– Instance-based matching• Looking at the instances (of entities or relationships) one can e.g.,

find correlations between attributes like ‘Are there duplicate tuples?’ or ‘Are the data distributions in their respective domains similar?’

– Structure-based matching• Abstracting from the actual labels, only the structure of the

schema is evaluated, e.g., regarding element types, depths in hierarchies, number and type of relationships, etc.



• If integration is query-driven only Schema Mapping is needed

– Mapping from one or more source schemas to a target schema



Data

Source

schema S

Correspondence

Target

schema T

CorrespondenceMapping

compiler

Low-level mapping

High-level mapping

• Schema Mapping

– Abstracting from the actual labels, regarding element types, depths in hierarchies, number and type of relationships, etc.



Product

ProdID: Decimal

Product: VARCHAR(50)

Group: VARCHAR(50)

Categ: VARCHAR(50)

Product

ID: Decimal

Product: VARCHAR(50)

GroupID:Decimal

• Schema mapping automation– Complex problem, based on heuristics

– Idea:• Based on given schemas and a high level mapping between

them

• Generate a set of queries that transform and integrate data from the sources to conform to the target schema

– Problems• Generation of the correct query considering the schemas

and the mappings

• Guarantee that the transformed data correspond to the target schema



• Schema integration in praxis

– BEA AquaLogic Data Services

• Special Feature: easy-to-use modeling: “Mappings and transformations can be designed in an easy-to-use GUI tool using a library of over 200 functions. For complex mappings andtransformations, architects anddevelopers can bypass the GUItool and use an Xquery sourcecode editor to define or edit services. “



www.bea.com

• What tools are actually given to support integration?– Data Translation Tool

• Transforms binary data into XML

• Transforms XML to binary data

– Data Transformation Tool• Transforms an XML to another XML

– Base Idea• Transform data to application specific XML → Transform to

XML specific to other application / general schema →

Transform back to binary

• Note: the integration work still has to be done manually



5/29/2009

12

• “I can’t afford expensive BEA consultants and the AquaLogic Integration Suite, what now??”– Do it completely yourself

• Most used technologies can be found as Open Source projects (data mappers, XSL engines, XSL editors, etc)

– Do it yourself with specialized tools• Many companies and open source projects are specialized in

developing data integration and transformation tools– CloverETL

– Altova MapForce

– BusinessObjects Data Integrator

– etc…



• Altova MapForce– Same idea than BEA Integrator

• Also based on XSLT and a data description language

– Editors for binary/DB to XML mapping

– Editor for XSLtransformation

– Automatic generationof data sources, web-services, and transformation modules in Java, C#, C++



• The loading process can be broken down into 2 different types:

– Initial load

– Continuous load (loadingover time)


8.2 Loading

• Issues

– Huge volumes of data to be loaded

– Small time window available when warehouse can be taken off line (usually nights)

– When to build index and summary tables

– Allow system administrators to monitor, cancel, resume, change load rates

– Recover gracefully -- restart after failure from where you were and without loss of data integrity


8.2 Loading

• Initial Load– Deliver dimensions tables

• Create and assign surrogate keys, each time a new cleaned and conformed dimension record has to be loaded

• Write dimensions to disk as physical tables, in the proper dimensional format

– Deliver fact tables• Utilize bulk-load utilities

• Load in parallel

– Tools• DTS – Data Transformation Services

• bcp utility – batch copy

• SQL* Loader


8.2 Loading

• Continuous load (loading over time)

– Must be scheduled and processed in a specific order to maintain integrity, completeness, and a satisfactory level of trust

– Should be the most carefully planned step in data warehousing or can lead to:

• Error duplication

• Exaggeration of inconsistencies in data


8.2 Loading

5/29/2009

13

• Continuous load of facts

– Separate updates from inserts

– Drop any indexes not required to support updates

– Load updates

– Drop all remaining indexes

– Load inserts through bulk loaders

– Rebuild indexes


8.2 Loading

• Metadata - data about data

– In DW, metadata describe the contents of a data warehouse and how to use it

• What information exists in a data warehouse, what the information means, how it was derived, from what source systems it comes, when it was created, what pre-built reports and analyses exist for manipulating the information, etc.


8.3 Metadata

• Types of metadata in DW

– Source system metadata

– Data staging metadata

– DBMS metadata


8.3 Metadata

• Source system metadata

– Source specifications

• E.g., repositories, and source logical schemas

– Source descriptive information

• E.g., ownership descriptions, update frequencies and access methods

– Process information

• E.g., job schedules and extraction code


8.3 Metadata

• Data staging metadata

– Data acquisition information, such as data transmission scheduling and results, and file usage

– Dimension table management, such as definitions of dimensions, and surrogate key assignments

– Transformation and aggregation, such as data enhancement and mapping, DBMS load scripts, and aggregate definitions

– Audit, job logs and documentation, such as data lineage records, data transform logs


8.3 Metadata

– E.g., Cube description metadata


8.3 Metadata

5/29/2009

14

• Business Intelligence (BI)

– Principles of Data Mining

– Association Rule Mining


Next lecture

Data Warehousing & Data Mining - TU Braunschweig · Data Warehousing & Data Mining Wolf-Tilo Balke Silviu Homoceanu ... • Assisted by artificial intelligence algorithms and pattern

Documents