Exadata Business
Case Analysis
Ram S. Mohan
Procter & Gamble
AGENDA
Company Background
Exadata Proof-of-Concept (POC)
POC Results - Reason To Believe
Business Requirements
Exadata Approach
Monthly Dashboard
System Landscape
Financial Model
Financial Due Diligence
Risk Mitigation
Opportunities
• Founded in 1837
• HQ in Cincinnati
• $76.7 Billion annual sales
• 135,000 employees
• More than 300 brands sold in
180 countries
• Operations in more than
80 countries
• Three Global Business Units - Beauty and Grooming, Household Care and Health and Well-Being
• Market Development Organization, Global Business Services and Corporate Function.
• 22 Billion Dollar Brands
P&G at a G
lance
William Procter
James Gamble
Exadata Proof of Concept (POC) Overview
• POC was conducted on Exadata V1• Testing Approach was designed over 2 months• >1 billion row DB was used• Designed queries based on business scenarios• Intentionally included queries that we CAN’T reliably run today
(e.g. > 8 hrs to complete)• All queries were run against base detail, no summarization layer
was involved.• No DB design tuning was performed.• Indices were NOT used to get the results.
POC Results – Reason To Believe
• Load Data - approximately 2 hours for 1.25 billion rows(!!!!)
• Run Baseline Queries – results (drum roll):
22X faster was the SLOWEST improvement
• >144X faster (conservatively!) seen for greatest improvement
• True ad hoc queries, that typically take 30 minutes in our environment, returned in 5 seconds.
• One of our pricing reports (complex and high data volume), which takes 6 hours today, ran in 5 minutes.
• We saw queries touching 1.25 billion rows, completing in under 3 minutes.
Business Requirements
• Deliver DW scale that matches the data scale that is over-running us.
• MUCH more head room for capacity challenged applications
• Produce business answers that we cannot today.
• Move from “custom created” Data Warehouse, to commodity delivered, “black box” appliances that are priced accordingly.
• Create an integrated data capability that gives us agility we sorely need.
• Dramatically expand what is an otherwise shrinking batch window.
• Create a Shopper-centric data eco-system (100TB-ish) that becomes the basis for our real-time business sensing.
Exadata Approach
Build, Operate and Govern a single Exadata Retail Trade Platform
• To perform data acquisition for all data coming from the Trade – POS, Market
Measurements, Distributor and others
Migrate applications from other Oracle platforms (RAC and Non-RAC)
Redesign back-end data loading and front-end reporting to take full advantage of infrastructure and capabilities
Drive Operational Excellence
Perform health checks once/quarter
System Management and Alerting – Oracle Enterprise Manager
8 Compute Nodes X86 64 Intel Cores
Storage Cells 14
Number of Disks 168 SAS
Disk Capacity (base 1024) 559GB
Total Storage 93.9 TB
Space for disk failure coverage 3.35 TB
Usable storage after ASM mirroring 43.59TB
System disk group (OCR, Voting disk, DBFS) 2.04 TB
Recovery Disk Group 40% 16.62 TB
Data Disk Group 60% 24.93 TB
Database Overhead (System, SysAux, Redo
log)
2 TB
UNDO Tablespaces 2 TB
TEMP Tablespaces 10% of DB 2.49TB
20% reserve in DGDATA 3.68TB
Space available for user data – DGDATA 80%
full (before compression)
14.71TB
Production Full Rack
Alpharetta, GA
4 Compute Nodes X86 32 Intel Cores
Storage Cells 7
Number of Disks 84 SATA
Disk Capacity (base 1024) 1.86TB
Total Storage 156.5 TB
Space for disk failure coverage 3.353 TB
Usable storage after ASM mirroring 67.1 TB
System disk group (OCR, Voting disk, DBFS) 1.02 TB
Recovery Disk Group 40% 26.4 TB
Data Disk Group 60% 39.6 TB
Database Overhead (System, SysAux, Redo
log)
1 TB
UNDO Tablespaces 1 TB
TEMP Tablespaces 10% of DB 3.96 TB
20% reserve in DGDATA 6.72 TB
Space available for user data – DGDATA 80%
full (before compression)
26.92 TB
Non-Production Half Rack
DEV & UAT Instances
Cincinnati, OH
Financial Model
Current
State
Key Deliverables
In-Scope &
Out-of-Scope
Future State Financial
Analysis
Stand alone
(Non-RAC)
Shared RAC Strategic Exadata
Servers Box CPU Utilization Business value Hosting
Storage Storage Architecture fit Hardware
Operations Support Operations Support Integration Software
Production Production Scalability Support
Non-Production Non-Production Lease vs. Buy Appl. Migration
Technology
Obsolescence
SQL* Server Storage Global vs.
Regional
Infrastructure
Integration
No head room SQL Cubes Co-existence Payout / ROI
Financial Due Diligence
Budget
� Be clear on savings – Hard savings, Cost Avoidance
� Manage Expectations – Best/Average/Worst case financial model – Use Average
Money in / Money out
� Minimize dual infrastructure – Decommission infrastructure ahead of schedule
� Streamline application migration – Fail Cheap!! --- Move, Learn & Reapply
� Store Less – Use compression. Implement ILM strategy
� Deliver on commitments – deliver early. Don’t start on the most complex system migration.
� Communicate value adds in business terms
Risks Mitigation
Issues
1. Minimize negative business impact from new technology deployment– HW, OS and DB
2. Avoid rework – Do it right the first time
3. Technology obsolescence
4. Decommission timing, if delayed, hurts financials
Action Steps
1. End-to-End SINGLE VENDOR support model. Oracle HW/SW integrated with Oracle Managed Services support
2. Leveraged Oracle Expert Services for Exadata implementation
3. Financial model built on 3-year straight line depreciation
4. Use expedited approach� Forklift application #1 (10gR2 schema)
� Migrate application #2 to this schema
Opportunities
• InfiniBand cabling for backup media management servers
• Support for Non-Exadata storage
• Consider Exadata Cloud – POC, New release/patch tests
• Exadata for SAP OLTP and Data Warehouse?
• Greater 100 TB database performance
• High Availability alternatives