Your SlideShare is downloading. ×
0
Data Warehousing (DAY 2) Siwawong W. Project Manager 2010.05.25
Agenda Data Warehouse: DW Structure/Modeling 10:00 – 10:30 09:00 – 09:15 Registration 09:15 – 09:30 Review 1 st  Day class...
1 st  Day Review
Dare Warehouse: Introduction Review <ul><li>Data vs Information </li></ul><ul><li>OLTP vs OLAP </li></ul><ul><li>Data Ware...
RDBMS & SQL: Review <ul><li>DBMS vs R-DBMS </li></ul><ul><li>Keys & Relationship </li></ul><ul><li>SQL Basic </li></ul><ul...
Data Warehouse:  Data Warehousing
Data Warehouse: Data Warehousing <ul><li>DW Architecture </li></ul><ul><li>Loading, refreshing </li></ul><ul><li>Structuri...
Data Warehouse:  Architecture
Data Warehouse Architecture Data Warehouse  Engine Optimized Loader Extraction Cleansing Analyze Query Metadata Repository...
Components of the Warehouse <ul><li>Data Extraction and Loading </li></ul><ul><ul><li>aka: ETL (Extract, Transform and Loa...
Loading the Warehouse Cleaning the data before it is loaded
Source Data  <ul><li>Typically host based, legacy applications </li></ul><ul><ul><li>Customized applications, COBOL, 3GL, ...
Data Quality – In Reality <ul><li>Tempting to think creating a data warehouse is simply extracting operational data and en...
Data Integration Across Sources Trust Credit card Savings Loans Same data  different name Different data  Same name Data f...
Data Transformation Example Data Warehouse encoding unit field appl A - balance appl B - bal appl C - currbal appl D - bal...
Data Integrity Problems <ul><li>Same person, different spellings </li></ul><ul><ul><li>Agarwal, Agrawal, Aggarwal etc... <...
Data Transformation Terms <ul><li>Extracting </li></ul><ul><li>Conditioning </li></ul><ul><li>Scrubbing (*) </li></ul><ul>...
Data Transformation Terms <ul><li>Extracting </li></ul><ul><ul><li>Capture of data from operational source in “as is” stat...
Data Transformation Terms <ul><li>Enrichment </li></ul><ul><ul><li>Bring data from external sources to augment/enrich oper...
Loads <ul><li>After extracting, scrubbing, cleaning, validating etc. need to load the data into the warehouse </li></ul><u...
Load Techniques <ul><li>Use SQL to append or insert new data </li></ul><ul><ul><li>record at a time interface </li></ul></...
Load Taxonomy <ul><li>Incremental  versus  Full  loads </li></ul><ul><li>Online  versus  Offline  loads </li></ul>
Refresh <ul><li>Propagate updates on source data to the warehouse </li></ul><ul><li>Issues: </li></ul><ul><ul><li>when to ...
When to Refresh? <ul><li>periodically (e.g., every night, every week) or after significant events </li></ul><ul><li>on eve...
Refresh Techniques <ul><li>Full Extract from base tables </li></ul><ul><ul><li>read entire source table: too expensive </l...
How To Detect Changes <ul><li>Create a snapshot log table to record ids of  updated rows  of source data and timestamp </l...
Data Extraction and Cleansing <ul><li>Extract data from existing operational and legacy data </li></ul><ul><li>Issues: </l...
Scrubbing Data <ul><li>Sophisticated transformation tools. </li></ul><ul><li>Used for cleaning the quality of data </li></...
Scrubbing Tools <ul><li>Apertus -- Enterprise/Integrator  </li></ul><ul><li>Vality -- IPE </li></ul><ul><li>Postal Soft </...
Structuring/Modeling Issues
Data: Heart of the Data Warehouse <ul><li>Heart of the data warehouse is the data itself! </li></ul><ul><li>Single version...
Data Warehouse Structure <ul><li>Subject Orientation </li></ul><ul><ul><li>customer, product, policy, account etc... A sub...
Data Warehouse Structure <ul><li>base customer (1985-87) </li></ul><ul><ul><li>custid, from date, to date, name, phone, do...
Data Granularity in Warehouse <ul><li>Summarized data stored </li></ul><ul><ul><li>reduce storage costs </li></ul></ul><ul...
Granularity in Warehouse <ul><li>Can not answer some questions with summarized data </li></ul><ul><ul><li>Did Anand call S...
Granularity in Warehouse <ul><li>Trade-off  is to have dual level of granularity </li></ul><ul><ul><li>Store summary data ...
Vertical Partitioning Frequently accessed Rarely accessed Smaller table and so less I/O Acct. No Name Balance Date Opened ...
Derived Data <ul><li>Introduction of derived (calculated data) may often help </li></ul><ul><li>Have seen this in the cont...
Schema Design <ul><li>Database organization </li></ul><ul><ul><li>must look like business </li></ul></ul><ul><ul><li>must ...
D/W Schema
Dimension Tables <ul><li>Dimension tables </li></ul><ul><ul><li>Define business in terms already familiar to users </li></...
Fact Table <ul><li>Central table </li></ul><ul><ul><li>mostly raw numeric items </li></ul></ul><ul><ul><li>narrow rows, a ...
Star Schema <ul><li>A single fact table and for each dimension one dimension table </li></ul><ul><li>Does not capture hier...
Snowflake schema <ul><li>Represent dimensional hierarchy directly by normalizing tables.  </li></ul><ul><li>Easy to mainta...
Fact Constellation <ul><li>Fact Constellation </li></ul><ul><ul><li>Multiple fact tables that share many dimension tables ...
De-normalization <ul><li>Normalization in a data warehouse may lead to lots of small tables </li></ul><ul><li>Can lead to ...
Creating Arrays <ul><li>Many times each occurrence of a sequence of data is in a different physical location </li></ul><ul...
Selective Redundancy <ul><li>Description of an item can be stored redundantly with order table  -- most often item descrip...
Partitioning <ul><li>Breaking data into several physical units that can be handled separately </li></ul><ul><li>Not a ques...
Why Partition? <ul><li>Flexibility in managing data </li></ul><ul><li>Smaller physical units allow </li></ul><ul><ul><li>e...
Criterion for Partitioning <ul><li>Typically partitioned by  </li></ul><ul><ul><li>date </li></ul></ul><ul><ul><li>line of...
Where to Partition? <ul><li>Application level or DBMS level </li></ul><ul><li>Makes sense to partition at application leve...
Data Warehouse vs. Data Marts What comes first?
What is Data Mart? <ul><li>Data Mart: a subset of a data warehouse that supports the requirements of particular department...
From the Data Warehouse to Data Marts Data Information Departmentally Structured Individually Structured Data Warehouse Or...
R easons for creating a data mart <ul><li>To give users access to the data they need to analyze most often </li></ul><ul><...
Data Warehouse and Data Marts OLAP Data Mart Lightly summarized Departmentally structured Organizationally structured Atom...
Characteristics of the Departmental Data Mart <ul><li>OLAP </li></ul><ul><li>Small </li></ul><ul><li>Flexible </li></ul><u...
Techniques for Creating Departmental Data Mart Sales Mktg. Finance <ul><li>OLAP </li></ul><ul><li>Subset </li></ul><ul><li...
Data Mart Centric Data Marts Data Sources Data Warehouse
Problems with Data Mart Centric Solution If you end up creating multiple warehouses, integrating them is a problem
True Warehouse Data Marts Data Sources Data Warehouse
D ata Marts Issues <ul><li>Data Mart Functionality </li></ul><ul><ul><li>the capabilities of data marts have increased wit...
D ata Marts Issues (Cont’) <ul><li>Data Mart Internet/Intranet access </li></ul><ul><ul><li>It’s products sit between a we...
Query Processing
Query Processing <ul><li>Indexing </li></ul><ul><li>Pre computed views/aggregates </li></ul><ul><li>SQL extensions </li></ul>
Indexing Techniques <ul><li>Exploiting indexes to reduce scanning of data is of crucial importance </li></ul><ul><li>Bitma...
Bitmap index <ul><li>A collection of bitmaps -- one for each distinct value of the column </li></ul><ul><li>Each bitmap ha...
Bitmap Index Customer Query : select * from customer where gender = ‘F’ and vote = ‘Y’ gender (f) vote (y) result vote gen...
Join Indexes <ul><li>Pre-computed joins </li></ul><ul><li>A join index between a fact table and a dimension table correlat...
Star Join Processing <ul><li>Use join indexes to join dimension and fact table </li></ul>Calls C+T  C+T+L C+T+L +P Time Lo...
Intelligent Scan <ul><li>Piggyback multiple scans of a relation (Redbrick) </li></ul><ul><ul><li>piggybacking also done if...
Parallel Query Processing <ul><li>Three forms of parallelism </li></ul><ul><ul><li>Independent </li></ul></ul><ul><ul><li>...
Pre-computed Aggregates <ul><li>Keep aggregated data for efficiency (pre-computed queries) </li></ul><ul><li>Questions </l...
Pre-computed Aggregates <ul><li>Aggregated  table can be maintained by the </li></ul><ul><ul><li>warehouse server </li></u...
SQL Extensions <ul><li>Extended family of aggregate functions </li></ul><ul><ul><li>rank (top 10 customers) </li></ul></ul...
IBM Red Brick Data Warehouse  <ul><li>Select month, dollars,  </li></ul><ul><li>  cume (dollars) as run_dollars,  </li></u...
On-Line Analytical Processing (OLAP)
Limitations of SQL <ul><li>“ A Freshman in Business </li></ul><ul><li>needs a Ph.D. in SQL” </li></ul><ul><li>-- Ralph Kim...
Typical OLAP Queries <ul><li>Write a multi-table join to compare sales for each product line YTD this year vs. last year. ...
What Is OLAP? <ul><li>Online Analytical Processing </li></ul><ul><ul><li>coined by EF Codd in 1994 paper contracted by  Ar...
The OLAP Market <ul><li>Rapid growth in the enterprise market </li></ul><ul><ul><li>1995:  $700 Million </li></ul></ul><ul...
Strengths of OLAP <ul><li>It is a powerful visualization paradigm </li></ul><ul><ul><li>e.g. Pivot </li></ul></ul><ul><li>...
OLAP Is FASMI <ul><li>Fast </li></ul><ul><li>Analysis </li></ul><ul><li>Shared </li></ul><ul><li>Multidimensional </li></u...
Multi-dimensional Data <ul><li>“ Hey…I sold $100M worth of goods” </li></ul>Dimensions:  Product, Region, Time Hierarchica...
Data Cube Lattice <ul><li>Cube lattice </li></ul><ul><ul><li>ABC   AB  AC  BC   A  B  C   none </li></ul></ul><ul><li>Can ...
A Visual Operation:  Pivot (Rotate) 10 47 30 12 Juice Cola Milk  Cream NY LA SF 3/1  3/2  3/3 3/4 Month Region Product
“ Slicing and Dicing” Product Sales Channel Regions Retail Direct Special Household Telecomm Video Audio India Far East Eu...
“ Slicing and Dicing” <ul><li>Slicing: Taking out the slice of a cube, given certain set of select dimension  (customer se...
Roll-up and Drill Down <ul><li>Sales Channel </li></ul><ul><li>Region </li></ul><ul><li>Country </li></ul><ul><li>State  <...
Nature of OLAP Analysis <ul><li>Aggregation -- (total sales, percent-to-total) </li></ul><ul><li>Comparison -- Budget vs. ...
Organizationally Structured Data <ul><li>Different Departments look at the same detailed data in different ways.  Without ...
Multidimensional Spreadsheets <ul><li>Analysts need spreadsheets that support </li></ul><ul><ul><li>pivot tables (cross-ta...
OLAP - Data Cube <ul><li>Idea: analysts need to group data in many different ways </li></ul><ul><ul><li>eg. Sales (region,...
Relational OLAP: ROLAP <ul><li>This methodology relies on manipulating the data stored in the relational database to give ...
Relational OLAP:  3 Tier DSS Store atomic data in industry standard RDBMS. Generate SQL execution plans in the ROLAP engin...
Advantage/Disadvantage: ROLAP <ul><li>Advantages   </li></ul><ul><ul><li>Can handle large amounts of data :  The data size...
Multidimensional OLAP: MOLAP <ul><li>This is the more traditional way of OLAP analysis .  In MOLAP, data is stored in a mu...
MD-OLAP: 2 Tier DSS MDDB Engine MDDB Engine Decision Support Client Database Layer Application Logic Layer Presentation La...
Advantage/Disadvantage: MOLAP <ul><li>Advantages </li></ul><ul><ul><li>Excellent performance :  MOLAP cubes are built for ...
Metadata Repository <ul><li>Administrative metadata </li></ul><ul><ul><li>source databases and their contents </li></ul></...
Metadata Repository   (Cont’) <ul><li>Business data </li></ul><ul><ul><li>business terms and definitions </li></ul></ul><u...
Recipe for a Successful Warehouse
For a Successful Warehouse <ul><li>From day one establish that warehousing is a joint user/builder project </li></ul><ul><...
For a Successful Warehouse <ul><li>Determine a plan to test the integrity of the data in the warehouse </li></ul><ul><li>F...
Data Warehouse Pitfalls <ul><li>You are going to spend much time extracting, cleaning, and loading data </li></ul><ul><li>...
Data Warehouse Pitfalls <ul><li>Your warehouse users will develop conflicting business rules </li></ul><ul><li>Large scale...
DW and OLAP Research Issues <ul><li>Data cleaning </li></ul><ul><ul><li>focus on data inconsistencies, not schema differen...
DW and OLAP Research Issues <ul><li>Warehouse Management </li></ul><ul><ul><li>detecting runaway queries </li></ul></ul><u...
References/External Links (1) Data Warehousing & Data Mining   S. Sudarshan Krithi Ramamritham IIT Bombay (2) Data Warehou...
Thank you for your attention! [email_address] www.blueballgroup.com
Upcoming SlideShare
Loading in...5
×

ITReady DW Day2

2,009

Published on

0 Comments
1 Like
Statistics
Notes
  • Be the first to comment

No Downloads
Views
Total Views
2,009
On Slideshare
0
From Embeds
0
Number of Embeds
2
Actions
Shares
0
Downloads
136
Comments
0
Likes
1
Embeds 0
No embeds

No notes for slide

Transcript of "ITReady DW Day2"

  1. 1. Data Warehousing (DAY 2) Siwawong W. Project Manager 2010.05.25
  2. 2. Agenda Data Warehouse: DW Structure/Modeling 10:00 – 10:30 09:00 – 09:15 Registration 09:15 – 09:30 Review 1 st Day class 09:30 – 10:00 Data Warehouse: Data Warehousing 10:30 – 10:45 Break & Morning Refreshment 10:45 – 12:00 DWs and Data Marts 12:00 – 13:00 Lunch Break 13:00 – 15:00 Query Processing 15:00 – 15:15 Break 15:15 – 16:00 OLAP & DSS
  3. 3. 1 st Day Review
  4. 4. Dare Warehouse: Introduction Review <ul><li>Data vs Information </li></ul><ul><li>OLTP vs OLAP </li></ul><ul><li>Data Warehousing & Data Mining </li></ul><ul><li>Application-Oriented vs Subject-Oriented </li></ul><ul><li>Advantage & Painful of Data Warehousing </li></ul>
  5. 5. RDBMS & SQL: Review <ul><li>DBMS vs R-DBMS </li></ul><ul><li>Keys & Relationship </li></ul><ul><li>SQL Basic </li></ul><ul><ul><li>DDL, DML & DCL </li></ul></ul><ul><li>Data Integrity </li></ul><ul><li>SELECT command and JOIN operators </li></ul>
  6. 6. Data Warehouse: Data Warehousing
  7. 7. Data Warehouse: Data Warehousing <ul><li>DW Architecture </li></ul><ul><li>Loading, refreshing </li></ul><ul><li>Structuring/Modeling </li></ul><ul><li>DWs and Data Marts </li></ul><ul><li>Query Processing </li></ul>
  8. 8. Data Warehouse: Architecture
  9. 9. Data Warehouse Architecture Data Warehouse Engine Optimized Loader Extraction Cleansing Analyze Query Metadata Repository Relational Databases Legacy Data Purchased Data ERP Systems
  10. 10. Components of the Warehouse <ul><li>Data Extraction and Loading </li></ul><ul><ul><li>aka: ETL (Extract, Transform and Load) </li></ul></ul><ul><ul><li>(via http ://en.wikipedia.org/wiki/Etl ) </li></ul></ul><ul><li>The Warehouse </li></ul><ul><li>Analyze and Query -- OLAP Tools </li></ul><ul><li>Metadata </li></ul><ul><li>Data Mining tools </li></ul>
  11. 11. Loading the Warehouse Cleaning the data before it is loaded
  12. 12. Source Data <ul><li>Typically host based, legacy applications </li></ul><ul><ul><li>Customized applications, COBOL, 3GL, 4GL </li></ul></ul><ul><li>Point of Contact Devices </li></ul><ul><ul><li>POS, ATM, Call switches </li></ul></ul><ul><li>External Sources </li></ul><ul><ul><li>Nielsen’s, Acxiom, CMIE, Vendors, Partners </li></ul></ul>Sequential Legacy Relational External Operational/ Source Data
  13. 13. Data Quality – In Reality <ul><li>Tempting to think creating a data warehouse is simply extracting operational data and entering into a data warehouse </li></ul><ul><li>Nothing could be farther from the truth </li></ul><ul><li>Warehouse data comes from disparate questionable sources </li></ul><ul><li>Legacy systems no longer documented </li></ul><ul><li>Outside sources with questionable quality procedures </li></ul><ul><li>Production systems with no built in integrity checks and no integration </li></ul><ul><ul><li>Operational systems are usually designed to solve a specific business problem and are rarely developed to a a corporate plan </li></ul></ul><ul><ul><ul><li>“ And get it done quickly, we do not have time to worry about corporate standards...” </li></ul></ul></ul>
  14. 14. Data Integration Across Sources Trust Credit card Savings Loans Same data different name Different data Same name Data found here nowhere else Different keys same data
  15. 15. Data Transformation Example Data Warehouse encoding unit field appl A - balance appl B - bal appl C - currbal appl D - balcurr appl A - pipeline - cm appl B - pipeline - in appl C - pipeline - feet appl D - pipeline - yds appl A - m,f appl B - 1,0 appl C - x,y appl D - male, female
  16. 16. Data Integrity Problems <ul><li>Same person, different spellings </li></ul><ul><ul><li>Agarwal, Agrawal, Aggarwal etc... </li></ul></ul><ul><li>Multiple ways to denote company name </li></ul><ul><ul><li>Persistent Systems, PSPL, Persistent Pvt. LTD. </li></ul></ul><ul><li>Use of different names </li></ul><ul><ul><li>mumbai, bombay </li></ul></ul><ul><li>Different account numbers generated by different applications for the same customer </li></ul><ul><li>Required fields left blank </li></ul><ul><li>Invalid product codes collected at point of sale </li></ul><ul><ul><li>manual entry leads to mistakes </li></ul></ul><ul><ul><li>“ in case of a problem use 9999999” </li></ul></ul>
  17. 17. Data Transformation Terms <ul><li>Extracting </li></ul><ul><li>Conditioning </li></ul><ul><li>Scrubbing (*) </li></ul><ul><li>Merging </li></ul><ul><li>Householding </li></ul><ul><li>Enrichment </li></ul><ul><li>Scoring </li></ul><ul><li>Loading </li></ul><ul><li>Validating </li></ul><ul><li>Delta Updating </li></ul>
  18. 18. Data Transformation Terms <ul><li>Extracting </li></ul><ul><ul><li>Capture of data from operational source in “as is” status </li></ul></ul><ul><ul><li>Sources for data generally in legacy mainframes in VSAM, IMS, IDMS, DB2; more data today in relational databases on Unix </li></ul></ul><ul><li>Conditioning </li></ul><ul><ul><li>The conversion of data types from the source to the target data store (warehouse) -- always a relational database </li></ul></ul><ul><li>Householding </li></ul><ul><ul><li>Identifying all members of a household (living at the same address) </li></ul></ul><ul><ul><li>Ensures only one mail is sent to a household </li></ul></ul><ul><ul><li>Can result in substantial savings: 1 lakh catalogues at Rs. 50 each costs Rs. 50 lakhs. A 2% savings would save Rs. 1 lakh. </li></ul></ul>
  19. 19. Data Transformation Terms <ul><li>Enrichment </li></ul><ul><ul><li>Bring data from external sources to augment/enrich operational data. Data sources include Dunn and Bradstreet, A. C. Nielsen, CMIE, IMRA etc... </li></ul></ul><ul><li>Scoring </li></ul><ul><ul><li>computation of a probability of an event. e.g..., chance that a customer will defect to AT&T from MCI, chance that a customer is likely to buy a new product </li></ul></ul>
  20. 20. Loads <ul><li>After extracting, scrubbing, cleaning, validating etc. need to load the data into the warehouse </li></ul><ul><li>Issues </li></ul><ul><ul><li>huge volumes of data to be loaded </li></ul></ul><ul><ul><li>small time window available when warehouse can be taken off line (usually nights) </li></ul></ul><ul><ul><li>when to build index and summary tables </li></ul></ul><ul><ul><li>allow system administrators to monitor, cancel, resume, change load rates </li></ul></ul><ul><ul><li>Recover gracefully -- restart after failure from where you were and without loss of data integrity </li></ul></ul>
  21. 21. Load Techniques <ul><li>Use SQL to append or insert new data </li></ul><ul><ul><li>record at a time interface </li></ul></ul><ul><ul><li>will lead to random disk I/O’s </li></ul></ul><ul><li>Use batch load utility </li></ul>
  22. 22. Load Taxonomy <ul><li>Incremental versus Full loads </li></ul><ul><li>Online versus Offline loads </li></ul>
  23. 23. Refresh <ul><li>Propagate updates on source data to the warehouse </li></ul><ul><li>Issues: </li></ul><ul><ul><li>when to refresh </li></ul></ul><ul><ul><li>how to refresh -- refresh techniques </li></ul></ul>
  24. 24. When to Refresh? <ul><li>periodically (e.g., every night, every week) or after significant events </li></ul><ul><li>on every update: not warranted unless warehouse data require current data (up to the minute stock quotes) </li></ul><ul><li>refresh policy set by administrator based on user needs and traffic </li></ul><ul><li>possibly different policies for different sources </li></ul>
  25. 25. Refresh Techniques <ul><li>Full Extract from base tables </li></ul><ul><ul><li>read entire source table: too expensive </li></ul></ul><ul><ul><li>maybe the only choice for legacy systems </li></ul></ul>
  26. 26. How To Detect Changes <ul><li>Create a snapshot log table to record ids of updated rows of source data and timestamp </li></ul><ul><li>Detect changes by: </li></ul><ul><ul><li>Defining after row triggers to update snapshot log when source table changes </li></ul></ul><ul><ul><li>Using regular transaction log to detect changes to source data </li></ul></ul>
  27. 27. Data Extraction and Cleansing <ul><li>Extract data from existing operational and legacy data </li></ul><ul><li>Issues: </li></ul><ul><ul><li>Sources of data for the warehouse </li></ul></ul><ul><ul><li>Data quality at the sources </li></ul></ul><ul><ul><li>Merging different data sources </li></ul></ul><ul><ul><li>Data Transformation </li></ul></ul><ul><ul><li>How to propagate updates (on the sources) to the warehouse </li></ul></ul><ul><ul><li>Terabytes of data to be loaded </li></ul></ul>
  28. 28. Scrubbing Data <ul><li>Sophisticated transformation tools. </li></ul><ul><li>Used for cleaning the quality of data </li></ul><ul><li>Clean data is vital for the success of the warehouse </li></ul><ul><li>Example </li></ul><ul><ul><li>Seshadri, Sheshadri, Sesadri, Seshadri S., Srinivasan Seshadri, etc. are the same person </li></ul></ul>
  29. 29. Scrubbing Tools <ul><li>Apertus -- Enterprise/Integrator </li></ul><ul><li>Vality -- IPE </li></ul><ul><li>Postal Soft </li></ul>
  30. 30. Structuring/Modeling Issues
  31. 31. Data: Heart of the Data Warehouse <ul><li>Heart of the data warehouse is the data itself! </li></ul><ul><li>Single version of the truth </li></ul><ul><li>Corporate memory </li></ul><ul><li>Data is organized in a way that represents business -- subject orientation </li></ul>
  32. 32. Data Warehouse Structure <ul><li>Subject Orientation </li></ul><ul><ul><li>customer, product, policy, account etc... A subject may be implemented as a set of related tables. </li></ul></ul><ul><ul><ul><li>e.g., customer may be five tables </li></ul></ul></ul>
  33. 33. Data Warehouse Structure <ul><li>base customer (1985-87) </li></ul><ul><ul><li>custid, from date, to date, name, phone, dob </li></ul></ul><ul><li>base customer (1988-90) </li></ul><ul><ul><li>custid, from date, to date, name, credit rating, employer </li></ul></ul><ul><li>customer activity (1986-89) -- monthly summary </li></ul><ul><li>customer activity detail (1987-89) </li></ul><ul><ul><li>custid, activity date, amount, clerk id, order no </li></ul></ul><ul><li>customer activity detail (1990-91) </li></ul><ul><ul><li>custid, activity date, amount, line item no, order no </li></ul></ul>Time is part of key of each table
  34. 34. Data Granularity in Warehouse <ul><li>Summarized data stored </li></ul><ul><ul><li>reduce storage costs </li></ul></ul><ul><ul><li>reduce CPU usage </li></ul></ul><ul><ul><li>increases performance since smaller number of records to be processed </li></ul></ul><ul><ul><li>design around traditional high level reporting needs </li></ul></ul><ul><ul><li>Trade-off with volume of data to be stored and detailed usage of data </li></ul></ul>
  35. 35. Granularity in Warehouse <ul><li>Can not answer some questions with summarized data </li></ul><ul><ul><li>Did Anand call Seshadri last month? Not possible to answer if total duration of calls by Anand over a month is only maintained and individual call details are not. </li></ul></ul><ul><li>Detailed data too voluminous </li></ul>
  36. 36. Granularity in Warehouse <ul><li>Trade-off is to have dual level of granularity </li></ul><ul><ul><li>Store summary data on disks </li></ul></ul><ul><ul><ul><li>95% of DSS processing done against this data </li></ul></ul></ul><ul><ul><li>Store detail on tapes </li></ul></ul><ul><ul><ul><li>5% of DSS processing against this data </li></ul></ul></ul>
  37. 37. Vertical Partitioning Frequently accessed Rarely accessed Smaller table and so less I/O Acct. No Name Balance Date Opened Interest Rate Address Acct. No Balance Acct. No Name Date Opened Interest Rate Address
  38. 38. Derived Data <ul><li>Introduction of derived (calculated data) may often help </li></ul><ul><li>Have seen this in the context of dual levels of granularity </li></ul><ul><li>Can keep auxiliary views and indexes to speed up query processing </li></ul>
  39. 39. Schema Design <ul><li>Database organization </li></ul><ul><ul><li>must look like business </li></ul></ul><ul><ul><li>must be recognizable by business user </li></ul></ul><ul><ul><li>approachable by business user </li></ul></ul><ul><ul><li>Must be simple </li></ul></ul><ul><li>Schema Types </li></ul><ul><ul><li>Star Schema </li></ul></ul><ul><ul><li>Fact Constellation Schema </li></ul></ul><ul><ul><li>Snowflake schema </li></ul></ul>
  40. 40. D/W Schema
  41. 41. Dimension Tables <ul><li>Dimension tables </li></ul><ul><ul><li>Define business in terms already familiar to users </li></ul></ul><ul><ul><li>Wide rows with lots of descriptive text </li></ul></ul><ul><ul><li>Small tables (about a million rows) </li></ul></ul><ul><ul><li>Joined to fact table by a foreign key </li></ul></ul><ul><ul><li>heavily indexed </li></ul></ul><ul><ul><li>typical dimensions </li></ul></ul><ul><ul><ul><li>time periods, geographic region (markets, cities), products, customers, salesperson, etc. </li></ul></ul></ul>
  42. 42. Fact Table <ul><li>Central table </li></ul><ul><ul><li>mostly raw numeric items </li></ul></ul><ul><ul><li>narrow rows, a few columns at most </li></ul></ul><ul><ul><li>large number of rows (millions to a billion) </li></ul></ul><ul><ul><li>Access via dimensions </li></ul></ul>
  43. 43. Star Schema <ul><li>A single fact table and for each dimension one dimension table </li></ul><ul><li>Does not capture hierarchies directly </li></ul>T i m e p r o d c u s t f a c t date, custno, prodno, cityname, ... c i t y
  44. 44. Snowflake schema <ul><li>Represent dimensional hierarchy directly by normalizing tables. </li></ul><ul><li>Easy to maintain and saves storage </li></ul>T i m e p r o d c u s t f a c t date, custno, prodno, cityname, ... c i t y r e g i o n
  45. 45. Fact Constellation <ul><li>Fact Constellation </li></ul><ul><ul><li>Multiple fact tables that share many dimension tables </li></ul></ul><ul><ul><li>Booking and Checkout may share many dimension tables in the hotel industry </li></ul></ul>Hotels Travel Agents Promotion Room Type Customer Booking Checkout
  46. 46. De-normalization <ul><li>Normalization in a data warehouse may lead to lots of small tables </li></ul><ul><li>Can lead to excessive I/O’s since many tables have to be accessed </li></ul><ul><li>De-normalization is the answer especially since updates are rare </li></ul>
  47. 47. Creating Arrays <ul><li>Many times each occurrence of a sequence of data is in a different physical location </li></ul><ul><li>Beneficial to collect all occurrences together and store as an array in a single row </li></ul><ul><li>Makes sense only if there are a stable number of occurrences which are accessed together </li></ul><ul><li>In a data warehouse, such situations arise naturally due to time based orientation </li></ul><ul><ul><li>can create an array by month </li></ul></ul>
  48. 48. Selective Redundancy <ul><li>Description of an item can be stored redundantly with order table -- most often item description is also accessed with order table </li></ul><ul><li>Updates have to be careful </li></ul>
  49. 49. Partitioning <ul><li>Breaking data into several physical units that can be handled separately </li></ul><ul><li>Not a question of whether to do it in data warehouses but how to do it </li></ul><ul><li>Granularity and partitioning are key to effective implementation of a warehouse </li></ul>
  50. 50. Why Partition? <ul><li>Flexibility in managing data </li></ul><ul><li>Smaller physical units allow </li></ul><ul><ul><li>easy restructuring </li></ul></ul><ul><ul><li>free indexing </li></ul></ul><ul><ul><li>sequential scans if needed </li></ul></ul><ul><ul><li>easy reorganization </li></ul></ul><ul><ul><li>easy recovery </li></ul></ul><ul><ul><li>easy monitoring </li></ul></ul>
  51. 51. Criterion for Partitioning <ul><li>Typically partitioned by </li></ul><ul><ul><li>date </li></ul></ul><ul><ul><li>line of business </li></ul></ul><ul><ul><li>geography </li></ul></ul><ul><ul><li>organizational unit </li></ul></ul><ul><ul><li>any combination of above </li></ul></ul>
  52. 52. Where to Partition? <ul><li>Application level or DBMS level </li></ul><ul><li>Makes sense to partition at application level </li></ul><ul><ul><li>Allows different definition for each year </li></ul></ul><ul><ul><ul><li>Important since warehouse spans many years and as business evolves definition changes </li></ul></ul></ul><ul><ul><li>Allows data to be moved between processing complexes easily </li></ul></ul>
  53. 53. Data Warehouse vs. Data Marts What comes first?
  54. 54. What is Data Mart? <ul><li>Data Mart: a subset of a data warehouse that supports the requirements of particular department or business function </li></ul><ul><li>The characteristics that differentiate data marts and data warehouses include: </li></ul><ul><ul><li>a data mart focuses on only the requirements of users associated with one department or business function </li></ul></ul><ul><ul><li>data marts do not normally contain detailed operational data, unlike data warehouses </li></ul></ul><ul><ul><li>as data marts contain less data compared with data warehouses, data marts are more easily understood and navigated </li></ul></ul>
  55. 55. From the Data Warehouse to Data Marts Data Information Departmentally Structured Individually Structured Data Warehouse Organizationally Structured Less More History Normalized Detailed
  56. 56. R easons for creating a data mart <ul><li>To give users access to the data they need to analyze most often </li></ul><ul><li>To provide data in a form that matches the collective view of the data by a group of users in a department or business function </li></ul><ul><li>To improve end-user response time due to the reduction in the volume of data to be accessed </li></ul><ul><li>To provide appropriately structured data as ditated by the requirements of end-user access tools </li></ul><ul><li>Normally use less data so tasks such as data cleansing, loading, transformation, and integration are far easier, and hence implementing and setting up a data mart is simpler than establishing a corporate data warehouse </li></ul><ul><li>The cost of implementing data marts is normally less than that required to establish a data warehouse </li></ul><ul><li>The potential users of a data mart are more clearly defined and can be more easily targeted to obtain support for a data mart project rather than a corporate data warehouse project </li></ul>
  57. 57. Data Warehouse and Data Marts OLAP Data Mart Lightly summarized Departmentally structured Organizationally structured Atomic Detailed Data Warehouse Data
  58. 58. Characteristics of the Departmental Data Mart <ul><li>OLAP </li></ul><ul><li>Small </li></ul><ul><li>Flexible </li></ul><ul><li>Customized by Department </li></ul><ul><li>Source is departmentally structured data warehouse </li></ul>
  59. 59. Techniques for Creating Departmental Data Mart Sales Mktg. Finance <ul><li>OLAP </li></ul><ul><li>Subset </li></ul><ul><li>Summarized </li></ul><ul><li>Superset </li></ul><ul><li>Indexed </li></ul><ul><li>Arrayed </li></ul>
  60. 60. Data Mart Centric Data Marts Data Sources Data Warehouse
  61. 61. Problems with Data Mart Centric Solution If you end up creating multiple warehouses, integrating them is a problem
  62. 62. True Warehouse Data Marts Data Sources Data Warehouse
  63. 63. D ata Marts Issues <ul><li>Data Mart Functionality </li></ul><ul><ul><li>the capabilities of data marts have increased with the growth in their popularity </li></ul></ul><ul><li>Data Mart Size </li></ul><ul><ul><li>the performance deteriorates as data marts grow in size, so need to reduce the size of data marts to gain improvements in performance </li></ul></ul><ul><li>Data Mart Load performance </li></ul><ul><ul><li>2 critical components: </li></ul></ul><ul><ul><ul><li>End-user response time and data loading performance: to increment DB updating so that only cells affected by the change are updated and not the entire MDDB structure </li></ul></ul></ul><ul><ul><ul><li>users’ access to data in multiple marts: one approach is to replicate data between different data marts or, alternatively, build virtual data mart  it is views of several physical data marts or the corporate data warehouse tailored to meet the requirements of specific groups of users </li></ul></ul></ul>
  64. 64. D ata Marts Issues (Cont’) <ul><li>Data Mart Internet/Intranet access </li></ul><ul><ul><li>It’s products sit between a web server and the data analysis product.Internet/intranet offers users low-cost access to data marts and the data WH using web browsers. </li></ul></ul><ul><li>Data Mart Administration </li></ul><ul><ul><li>Organization can not easily perform administration of multiple data marts, giving rise to issues such as data mart versioning, data and meta-data consistency and integrity, enterprise-wide security, and performance tuning. Data mart administrative tools are commerciallly available </li></ul></ul><ul><li>Data Mart Installation </li></ul><ul><ul><li>data marts are becoming increasingly complex to build. Vendors are offering products referred to as ”data mart in a box” that provide a low-cost source of data mart tools </li></ul></ul>
  65. 65. Query Processing
  66. 66. Query Processing <ul><li>Indexing </li></ul><ul><li>Pre computed views/aggregates </li></ul><ul><li>SQL extensions </li></ul>
  67. 67. Indexing Techniques <ul><li>Exploiting indexes to reduce scanning of data is of crucial importance </li></ul><ul><li>Bitmap Indexes </li></ul><ul><li>Join Indexes </li></ul><ul><li>Other Issues </li></ul><ul><ul><li>Text indexing </li></ul></ul><ul><ul><li>Parallelizing and sequencing of index builds and incremental updates </li></ul></ul>
  68. 68. Bitmap index <ul><li>A collection of bitmaps -- one for each distinct value of the column </li></ul><ul><li>Each bitmap has N bits where N is the number of rows in the table </li></ul><ul><li>A bit corresponding to a value v for a row r is set if and only if r has the value for the indexed attribute </li></ul><ul><li>Products that support bitmaps: Model 204, TargetIndex (Redbrick), IQ (Sybase), Oracle 7.3 </li></ul>
  69. 69. Bitmap Index Customer Query : select * from customer where gender = ‘F’ and vote = ‘Y’ gender (f) vote (y) result vote gender 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 M F F F F M Y Y Y N N N
  70. 70. Join Indexes <ul><li>Pre-computed joins </li></ul><ul><li>A join index between a fact table and a dimension table correlates a dimension tuple with the fact tuples that have the same value on the common dimensional attribute </li></ul><ul><ul><li>e.g., a join index on city dimension of calls fact table </li></ul></ul><ul><ul><li>correlates for each city the calls (in the calls table) from that city </li></ul></ul>
  71. 71. Star Join Processing <ul><li>Use join indexes to join dimension and fact table </li></ul>Calls C+T C+T+L C+T+L +P Time Loca- tion Plan
  72. 72. Intelligent Scan <ul><li>Piggyback multiple scans of a relation (Redbrick) </li></ul><ul><ul><li>piggybacking also done if second scan starts a little while after the first scan </li></ul></ul>
  73. 73. Parallel Query Processing <ul><li>Three forms of parallelism </li></ul><ul><ul><li>Independent </li></ul></ul><ul><ul><li>Pipelined </li></ul></ul><ul><ul><li>Partitioned and “partition and replicate” </li></ul></ul><ul><li>Deterrents to parallelism </li></ul><ul><ul><li>startup </li></ul></ul><ul><ul><li>communication </li></ul></ul><ul><li>Partitioned Data </li></ul><ul><ul><li>Parallel scans </li></ul></ul><ul><ul><li>Yields I/O parallelism </li></ul></ul><ul><li>Parallel algorithms for relational operators </li></ul><ul><ul><li>Joins, Aggregates, Sort </li></ul></ul><ul><li>Parallel Utilities </li></ul><ul><ul><li>Load, Archive, Update, Parse, Checkpoint, Recovery </li></ul></ul><ul><li>Parallel Query Optimization </li></ul>
  74. 74. Pre-computed Aggregates <ul><li>Keep aggregated data for efficiency (pre-computed queries) </li></ul><ul><li>Questions </li></ul><ul><ul><li>Which aggregates to compute? </li></ul></ul><ul><ul><li>How to update aggregates? </li></ul></ul><ul><ul><li>How to use pre-computed aggregates in queries? </li></ul></ul>
  75. 75. Pre-computed Aggregates <ul><li>Aggregated table can be maintained by the </li></ul><ul><ul><li>warehouse server </li></ul></ul><ul><ul><li>middle tier </li></ul></ul><ul><ul><li>client applications </li></ul></ul><ul><li>Pre-computed aggregates -- special case of materialized views -- same questions and issues remain </li></ul>
  76. 76. SQL Extensions <ul><li>Extended family of aggregate functions </li></ul><ul><ul><li>rank (top 10 customers) </li></ul></ul><ul><ul><li>percentile (top 30% of customers) </li></ul></ul><ul><ul><li>median, mode </li></ul></ul><ul><ul><li>Object Relational Systems allow addition of new aggregate functions </li></ul></ul><ul><li>Reporting features </li></ul><ul><ul><li>running total, cumulative totals </li></ul></ul><ul><li>Cube operator </li></ul><ul><ul><li>group by on all subsets of a set of attributes (month,city) </li></ul></ul><ul><ul><li>redundant scan and sorting of data can be avoided </li></ul></ul>
  77. 77. IBM Red Brick Data Warehouse <ul><li>Select month, dollars, </li></ul><ul><li> cume (dollars) as run_dollars, </li></ul><ul><li> weight, </li></ul><ul><li> cume (weight) as run_weights from sales, market, product, period t where year = 1993 and product like ‘Columbian%’ and city like ‘San Fr%’ order by t.perkey </li></ul>
  78. 78. On-Line Analytical Processing (OLAP)
  79. 79. Limitations of SQL <ul><li>“ A Freshman in Business </li></ul><ul><li>needs a Ph.D. in SQL” </li></ul><ul><li>-- Ralph Kimball </li></ul>
  80. 80. Typical OLAP Queries <ul><li>Write a multi-table join to compare sales for each product line YTD this year vs. last year. </li></ul><ul><li>Repeat the above process to find the top 5 product contributors to margin. </li></ul><ul><li>Repeat the above process to find the sales of a product line to new vs. existing customers. </li></ul><ul><li>Repeat the above process to find the customers that have had negative sales growth. </li></ul>
  81. 81. What Is OLAP? <ul><li>Online Analytical Processing </li></ul><ul><ul><li>coined by EF Codd in 1994 paper contracted by Arbor Software * </li></ul></ul><ul><li>Generally synonymous with earlier terms such as Decisions Support, Business Intelligence, Executive Information System </li></ul><ul><li>OLAP: Multidimensional Database </li></ul><ul><ul><li>MOLAP: Multidimensional OLAP (Arbor Essbase, Oracle Express) </li></ul></ul><ul><ul><li>ROLAP: Relational OLAP (Informix MetaCube, Microstrategy DSS Agent) </li></ul></ul>* Reference: http://www.arborsoft.com/essbase/wht_ppr/coddTOC.html
  82. 82. The OLAP Market <ul><li>Rapid growth in the enterprise market </li></ul><ul><ul><li>1995: $700 Million </li></ul></ul><ul><ul><li>1997: $2.1 Billion </li></ul></ul><ul><li>Significant consolidation activity among major DBMS vendors </li></ul><ul><ul><li>10/94: Sybase acquires ExpressWay </li></ul></ul><ul><ul><li>7/95: Oracle acquires Express </li></ul></ul><ul><ul><li>11/95: Informix acquires Metacube </li></ul></ul><ul><ul><li>1/97: Arbor partners up with IBM </li></ul></ul><ul><ul><li>10/96: Microsoft acquires Panorama </li></ul></ul><ul><li>Result: OLAP shifted from small vertical niche to mainstream DBMS category </li></ul>
  83. 83. Strengths of OLAP <ul><li>It is a powerful visualization paradigm </li></ul><ul><ul><li>e.g. Pivot </li></ul></ul><ul><li>It provides fast, interactive response times </li></ul><ul><li>It is good for analyzing time series </li></ul><ul><li>It can be useful to find some clusters and outliers </li></ul><ul><li>Many vendors offer OLAP tools </li></ul>
  84. 84. OLAP Is FASMI <ul><li>Fast </li></ul><ul><li>Analysis </li></ul><ul><li>Shared </li></ul><ul><li>Multidimensional </li></ul><ul><li>Information </li></ul>Nigel Pendse, Richard Creath - The OLAP Report
  85. 85. Multi-dimensional Data <ul><li>“ Hey…I sold $100M worth of goods” </li></ul>Dimensions: Product, Region, Time Hierarchical summarization paths Product Region Time Industry Country Year Category Region Quarter Product City Month Week Office Day Month 1 2 3 4 7 6 5 Product Toothpaste Juice Cola Milk Cream Soap Region W S N
  86. 86. Data Cube Lattice <ul><li>Cube lattice </li></ul><ul><ul><li>ABC AB AC BC A B C none </li></ul></ul><ul><li>Can materialize some Group-Bys, compute others on demand </li></ul><ul><li>Question: which Group-Bys to materialize? </li></ul><ul><li>Question: what indices to create? </li></ul><ul><li>Question: how to organize data (chunks, etc)? </li></ul>
  87. 87. A Visual Operation: Pivot (Rotate) 10 47 30 12 Juice Cola Milk Cream NY LA SF 3/1 3/2 3/3 3/4 Month Region Product
  88. 88. “ Slicing and Dicing” Product Sales Channel Regions Retail Direct Special Household Telecomm Video Audio India Far East Europe The Telecomm Slice
  89. 89. “ Slicing and Dicing” <ul><li>Slicing: Taking out the slice of a cube, given certain set of select dimension (customer segment) , and value ( home furnishings ..) and measures (sales revenue , sales units ..) or KPIs (Sales Productivity) </li></ul><ul><li>Dicing means viewing the slices from different angles. </li></ul><ul><ul><li>e.g.   Revenue for different products within a given state OR revenue for different states for a given product . </li></ul></ul>From: http :// www . executionmih . com / data - analysis / bi - slice - dice . php
  90. 90. Roll-up and Drill Down <ul><li>Sales Channel </li></ul><ul><li>Region </li></ul><ul><li>Country </li></ul><ul><li>State </li></ul><ul><li>Location Address </li></ul><ul><li>Sales Representative </li></ul>Higher Level of Aggregation Low-level Details Drill-Down Roll Up
  91. 91. Nature of OLAP Analysis <ul><li>Aggregation -- (total sales, percent-to-total) </li></ul><ul><li>Comparison -- Budget vs. Expenses </li></ul><ul><li>Ranking -- Top 10, quartile analysis </li></ul><ul><li>Access to detailed and aggregate data </li></ul><ul><li>Complex criteria specification </li></ul><ul><li>Visualization </li></ul>
  92. 92. Organizationally Structured Data <ul><li>Different Departments look at the same detailed data in different ways. Without the detailed, organizationally structured data as a foundation, there is no reconcilability of data </li></ul>marketing manufacturing sales finance
  93. 93. Multidimensional Spreadsheets <ul><li>Analysts need spreadsheets that support </li></ul><ul><ul><li>pivot tables (cross-tabs) </li></ul></ul><ul><ul><li>drill-down and roll-up </li></ul></ul><ul><ul><li>slice and dice </li></ul></ul><ul><ul><li>sort </li></ul></ul><ul><ul><li>selections </li></ul></ul><ul><ul><li>derived attributes </li></ul></ul><ul><li>Popular in retail domain </li></ul>
  94. 94. OLAP - Data Cube <ul><li>Idea: analysts need to group data in many different ways </li></ul><ul><ul><li>eg. Sales (region, product, prod.type, prod.style, date, sale.amount) </li></ul></ul><ul><ul><li>Sale.amount is a measure attribute, rest are dimension attributes </li></ul></ul><ul><ul><li>Group-By every subset of the other attributes </li></ul></ul><ul><ul><ul><li>materialize (pre-compute and store) Group-Bys to give online response </li></ul></ul></ul><ul><ul><li>Also: hierarchies on attributes: date -> weekday, date -> month -> quarter -> year </li></ul></ul>
  95. 95. Relational OLAP: ROLAP <ul><li>This methodology relies on manipulating the data stored in the relational database to give the appearance of traditional OLAP's slicing and dicing functionality . In essence, each action of slicing and dicing is equivalent to adding a &quot; WHERE &quot; clause in the SQL statement . </li></ul>
  96. 96. Relational OLAP: 3 Tier DSS Store atomic data in industry standard RDBMS. Generate SQL execution plans in the ROLAP engine to obtain OLAP functionality. Obtain multi-dimensional reports from the DSS Client. Data Warehouse ROLAP Engine Decision Support Client Database Layer Application Logic Layer Presentation Layer
  97. 97. Advantage/Disadvantage: ROLAP <ul><li>Advantages </li></ul><ul><ul><li>Can handle large amounts of data : The data size limitation of ROLAP technology is the limitation on data size of the underlying relational database . In other words, ROLAP itself places no limitation on data amount . </li></ul></ul><ul><ul><li>Can leverage functionalities inherent in the relational database: Often, relational database already comes with a host of functionalities. ROLAP technologies, since they sit on top of the relational database, can therefore leverage these functionalities. </li></ul></ul><ul><li>Disadvantages </li></ul><ul><ul><li>Performance can be slow : Because each ROLAP report is essentially a SQL query ( or multiple SQL queries ) in the relational database, the query time can be long if the underlying data size is large . </li></ul></ul><ul><ul><li>Limited by SQL functionalities: Because ROLAP technology mainly relies on generating SQL statements to query the relational database, and SQL statements do not fit all needs (for example, it is difficult to perform complex calculations using SQL), ROLAP technologies are therefore traditionally limited by what SQL can do. ROLAP vendors have mitigated this risk by building into the tool out-of-the-box complex functions as well as the ability to allow users to define their own function </li></ul></ul>
  98. 98. Multidimensional OLAP: MOLAP <ul><li>This is the more traditional way of OLAP analysis . In MOLAP, data is stored in a multidimensional cube . The storage is not in the relational database, but in proprietary formats . </li></ul>
  99. 99. MD-OLAP: 2 Tier DSS MDDB Engine MDDB Engine Decision Support Client Database Layer Application Logic Layer Presentation Layer Store atomic data in a proprietary data structure (MDDB), pre-calculate as many outcomes as possible, obtain OLAP functionality via proprietary algorithms running against this data. Obtain multi-dimensional reports from the DSS Client.
  100. 100. Advantage/Disadvantage: MOLAP <ul><li>Advantages </li></ul><ul><ul><li>Excellent performance : MOLAP cubes are built for fast data retrieval, and is optimal for slicing and dicing operations . </li></ul></ul><ul><ul><li>Can perform complex calculations: All calculations have been pre-generated when the cube is created. Hence, complex calculations are not only doable, but they return quickly. </li></ul></ul><ul><li>Disadvantages </li></ul><ul><ul><li>Limited in the amount of data it can handle : Because all calculations are performed when the cube is built, it is not possible to include a large amount of data in the cube itself . This is not to say that the data in the cube cannot be derived from a large amount of data . Indeed, this is possible . But in this case, only summary - level information will be included in the cube itself . </li></ul></ul><ul><ul><li>Requires additional investment: Cube technology are often proprietary and do not already exist in the organization. Therefore, to adopt MOLAP technology, chances are additional investments in human and capital resources are needed. </li></ul></ul>
  101. 101. Metadata Repository <ul><li>Administrative metadata </li></ul><ul><ul><li>source databases and their contents </li></ul></ul><ul><ul><li>gateway descriptions </li></ul></ul><ul><ul><li>warehouse schema, view & derived data definitions </li></ul></ul><ul><ul><li>dimensions, hierarchies </li></ul></ul><ul><ul><li>pre-defined queries and reports </li></ul></ul><ul><ul><li>data mart locations and contents </li></ul></ul><ul><ul><li>data partitions </li></ul></ul><ul><ul><li>data extraction, cleansing, transformation rules, defaults </li></ul></ul><ul><ul><li>data refresh and purging rules </li></ul></ul><ul><ul><li>user profiles, user groups </li></ul></ul><ul><ul><li>security: user authorization, access control </li></ul></ul>
  102. 102. Metadata Repository (Cont’) <ul><li>Business data </li></ul><ul><ul><li>business terms and definitions </li></ul></ul><ul><ul><li>ownership of data </li></ul></ul><ul><ul><li>charging policies </li></ul></ul><ul><li>operational metadata </li></ul><ul><ul><li>data lineage: history of migrated data and sequence of transformations applied </li></ul></ul><ul><ul><li>currency of data: active, archived, purged </li></ul></ul><ul><ul><li>monitoring information: warehouse usage statistics, error reports, audit trails. </li></ul></ul>
  103. 103. Recipe for a Successful Warehouse
  104. 104. For a Successful Warehouse <ul><li>From day one establish that warehousing is a joint user/builder project </li></ul><ul><li>Establish that maintaining data quality will be an ONGOING joint user/builder responsibility </li></ul><ul><li>Train the users one step at a time </li></ul><ul><li>Consider doing a high level corporate data model in no more than three weeks </li></ul><ul><li>Look closely at the data extracting, cleaning, and loading tools </li></ul><ul><li>Implement a user accessible automated directory to information stored in the warehouse </li></ul>From Larry Greenfield
  105. 105. For a Successful Warehouse <ul><li>Determine a plan to test the integrity of the data in the warehouse </li></ul><ul><li>From the start get warehouse users in the habit of 'testing' complex queries </li></ul><ul><li>Coordinate system roll-out with network administration personnel </li></ul><ul><li>When in a bind, ask others who have done the same thing for advice </li></ul><ul><li>Be on the lookout for small, but strategic, projects </li></ul><ul><li>Market and sell your data warehousing systems </li></ul>(Cont’)
  106. 106. Data Warehouse Pitfalls <ul><li>You are going to spend much time extracting, cleaning, and loading data </li></ul><ul><li>Despite best efforts at project management, data warehousing project scope will increase </li></ul><ul><li>You are going to find problems with systems feeding the data warehouse </li></ul><ul><li>You will find the need to store data not being captured by any existing system </li></ul><ul><li>You will need to validate data not being validated by transaction processing systems </li></ul><ul><li>Some transaction processing systems feeding the warehousing system will not contain detail </li></ul><ul><li>Many warehouse end users will be trained and never or seldom apply their training </li></ul><ul><li>After end users receive query and report tools, requests for IS written reports may increase </li></ul>
  107. 107. Data Warehouse Pitfalls <ul><li>Your warehouse users will develop conflicting business rules </li></ul><ul><li>Large scale data warehousing can become an exercise in data homogenizing </li></ul><ul><li>'Overhead' can eat up great amounts of disk space </li></ul><ul><li>The time it takes to load the warehouse will expand to the amount of the time in the available window... and then some </li></ul><ul><li>Assigning security cannot be done with a transaction processing system mindset </li></ul><ul><li>You are building a HIGH maintenance system </li></ul><ul><li>You will fail if you concentrate on resource optimization to the neglect of project, data, and customer management issues and an understanding of what adds value to the customer </li></ul>(Cont’)
  108. 108. DW and OLAP Research Issues <ul><li>Data cleaning </li></ul><ul><ul><li>focus on data inconsistencies, not schema differences </li></ul></ul><ul><ul><li>data mining techniques </li></ul></ul><ul><li>Physical Design </li></ul><ul><ul><li>design of summary tables, partitions, indexes </li></ul></ul><ul><ul><li>tradeoffs in use of different indexes </li></ul></ul><ul><li>Query processing </li></ul><ul><ul><li>selecting appropriate summary tables </li></ul></ul><ul><ul><li>dynamic optimization with feedback </li></ul></ul><ul><ul><li>acid test for query optimization: cost estimation, use of transformations, search strategies </li></ul></ul><ul><ul><li>partitioning query processing between OLAP server and backend server. </li></ul></ul>
  109. 109. DW and OLAP Research Issues <ul><li>Warehouse Management </li></ul><ul><ul><li>detecting runaway queries </li></ul></ul><ul><ul><li>resource management </li></ul></ul><ul><ul><li>incremental refresh techniques </li></ul></ul><ul><ul><li>computing summary tables during load </li></ul></ul><ul><ul><li>failure recovery during load and refresh </li></ul></ul><ul><ul><li>process management: scheduling queries, load and refresh </li></ul></ul><ul><ul><li>Query processing, caching </li></ul></ul><ul><ul><li>use of workflow technology for process management </li></ul></ul>(Cont’)
  110. 110. References/External Links (1) Data Warehousing & Data Mining S. Sudarshan Krithi Ramamritham IIT Bombay (2) Data Warehousing Hu Yan e-mail: [email_address] (3) Data Warehosing Concept: MOLAP, ROLAP and HOLAP http://www.1keydata.com/datawarehousing/molap-rolap.html
  111. 111. Thank you for your attention! [email_address] www.blueballgroup.com
  1. A particular slide catching your eye?

    Clipping is a handy way to collect important slides you want to go back to later.

×