William Henry

1
William Henry
Serendipity is finding something unexpected and
useful while searching for something else
entirely
Vietnam Veteran 1968 MVS systems programmer
AC Nielsen Oracle
Consulting Internet CRM company Oracle
Consulting Contractor

2
Cost Cutting and Intelligence Reporting Megatrends
One Vendor Solutions
KPIs
Reports
Were consolidating
BI Project cost models have changed Requirements
have not
3
Are Star Schema's Obsolete?
WHAT LEAD ME TO ASK THIS QUESTION? -- Latest
implementation utilized an all Oracle 10G
product stack. -- Initial requirements outlined
typical ODS and DW topology. -- 58 Data Sources
distributed throughout USA and Canada. -- Over
1/2 billion transactions (4 years historical)
. Tight budget and daily ODS update requirement
meant Techniques and products that were too
inefficient to the processing requirements
Fast, Accurate, and KISS were eliminated.
4
Softwar
LE Even if the software could be made to do 100
percent of the things customers say they need,
there is still a couple of serious problems that
remain. First, the customer may not fully
understand each and every one of its own needs,
and second, even if it does, those needs are
likely to change over time. A far better approach
is to buy a system that delivers 80 percent of
the benefits of a perfect system at 20 percent
of the cost. After 80 percent solution is
installed and delivering substantial benefits,
you then have the option of incrementally
enhancing your systems automation and
information capabilities a little at a time.
Going for a 100 percent solution requires perfect
understanding of your business before the
project starts and a lot of customization to
make the software fit. That is a very expensive
and risky approach you might end up with nothing.
5
Build it and they will come gets a bad rap see
Google and its cost efficient architecture
Blank Palette
Linux
Star Office MySQL Apache PHP
MySQL doesnt have the features required for
high volume BI implementations MS Office defacto
standard Apache/PHP Portal building too
complex No ad-hoc tools No Enterprise Reporting
tool
SATA Drives
And / Or
Windows
MS Office MySQL Apache PHP
Cygwin VIM
SATA Drives
6
Oracle BI
Build an ODS to eliminate OLTP joins Partition
by Time, Use Compress (sub-partitions dont
support compress) Build Summary tables from Fact
and Customer tables Use exchange partition
extensively (Otherwise Gather Table Stats
takes too much time) Build Materialized Views on
Summary Tables Hookup Portal, Discoverer,
Reports Single Source of Truth Fast and Accurate
7
You only run as good as your worst bottleneck
High volume data is totally insensitive to sales
pitches, programmer complex techniques, and
processing flexibility desires. Identify your
worst bottleneck no matter were it goes. Then
worry about political realities
8
Tough Bottleneck Politics
We already own it, you must use it SANs a
magic bullet? Or The lost art of I/O load
balancing. Infrastructure Architects and
Administrators BI Catch 22 Timely information
is important, change is not
9
Customers and Complexity's Pot of Gold
No Programming Required
Best of Breed
Distributed Reporting is still a big problem
Web Services
XML .NET Vendor Buzz Words Beans
10
Softwar
LE Building an integrated system out of several
different software products that were never
designed to work together is a very difficult
task. At Oracle, we used to sell and deliver
these best-of-breed systems,delivery was
hard. IBM has thousands of consultants eager to
help you make it work. The more complex the
integration project the more likely it will be
late and over budget It may fail completely. And
at the end its the customer who assumes the
financial risk for the success or failure of
these projects. The Best-of-Breed software
product assembly approach is absolutely unique
to the computer industry. If Detroit ran like
Silicon Valley, nobody would sell cars -- just
parts. Customers would have to figure out which
were the best parts. I know it sounds crazy ,
but thats how companies put together business
systems.
11
Star Schema Evolution
Move from cubes to RDBMSs (Open Systems) No
compression meant slim fact tables Sparse
calculation
Features that required coding Partitioning
(RPCs) Compression Materialized Views Portal
50,000 Products - What 208 Weeks
- When 1,000 CGP Retailers - Who 5,000
Locations - Where 10
12
Combinations
7
10 Million Row fact table 1x10
10
1 sparse 1x10 10,000,000,000 10 Billion
12
How many stars in the universe? There about 10
billion (10 ) stars in an average galaxy, and
there are about 10 billion galaxies that we can
observe in the universe, so the answer is 100
billion billion or 10
10
20
A CRM Implementation Customers
1,000,000 Products 10,000 Companies
1,000 Periods (daily) 1,480 Ads tracked
500,000 Households 900,000 Roughly 10
combinations
26
13
Todays Star Schema Best Practicies
The threat If not followed descent into chaos is
inevitable, it will be a project Death March.
Push/Pull/Dribble Job Schedulers Inferred
Domains Structure AnalysisDe-duplication
Conformed Dimensions Audit Dimension
Assembler Quality Screen Handler Error Event
Handler Database Independent Surrogate
Keys Slowly Changing Dimension Processor
Late Arriving Dimension Handler Fixed
Hierarchy Dimension Builder Variable
Hierarchy Dimension Builder Multivalued
Dimension Bridge Table Builder Junk Dimension
Builder Surrogate Key Pipeline Grain Fact
Table Loader Periodic Snapshot Grain Fact
Loader Accumulating Snapshot Grain Fact Loader
Late Arriving Fact Handler Aggregate
Builder Multidimensional Cube Builder
Real-time Partition Builder Dimension Manager
System Fact Table Provider System Job
Scheduler Workflow Monitor Recovery/Restart
System Parallelizing/Pipelinig System
Problem Escalation System Version Control
System Version Migration System Linage and
Dependency Analyzer Compliance
Reporter Security System Backup System
Metadata Repository Manager Project
Management ETL Tools and Best of Breed BI to
facilitate these topics.
14
Gap Analysis and ETL Techniques
OLTP technologies versus BI technologies One
Statement Reports versus ETL simplicity Use a BI
prototype to do Gap Analysis focus on texture
of data (business) not on elegant software
techniques Get to know the data better than
anyone else. There is more job security in
business knowledge than technology knowledge
15
Oracle 10G BI
LE If the Internet turns out not to be the future
of computing were toast.
But if it is, were golden
16
Portal, Business Intelligence, and Enterprise
Reporting
Portal and Web Discoverer are golden for Access
and Ad-Hoc Tools for Enterprise Reports are
still Best of Breed approach Portal
Reports Oracle Reports XML Publisher .NET and
Office ETC
Operational Portal EUL Portal KPI Portal
17
Softwar
LE Daily Business Intelligence cannot take the
place of good management judgement it simply
provides better data so managers can base their
decisions on hard facts rather than gut feelings.
Those hard facts are very difficult to come by
in most companies, because current computer
systems are a bunch of loosely connected process
automation systems that are terrible at providing
useful information about the business to
management. A single, unified process automation
database with integrated Business Intelligence
moves a company out of the dark and into the
information age. Just ETLing data to a ODS
doesnt turn data into Information See CRM
implementations without a good Customer
Master. DWs without ties to financials, KPIs,
or improved business processes.