Case Studies in Cloud Computing

David Cearley and Gene Phifer

Cloud Adoption: Rapid Growth Predicted
Percentage of Users Relying on Cloud Model
30%
The timing of the adoption of hosting services
will be tied to company size
20%
(although exceptions will abound)
Large
10% Companies
MSE Companies
Small Companies
0%

2007 2008 2009 2010 2011 2012

• Why Now?
- Exploitation of cloud economics, especially in response
to global recession
- Acceptance of SaaS models
- Offerings and vendors at every turn
Key Issues
1. How have companies
Get Rid of
successfully used cloud system Applications?
infrastructure services, and
what lessons can be learned? Get Rid of
Middleware?
2. How have companies
Get Rid of
successfully used cloud Servers?
application infrastructure Business Services

Mgmt. and Security

services and what lessons can Information Services
be learned? Application Services
App. Infrastructure Services
3. How have companies
System Infrastructure Services Cloud
successfully used cloud Enablers
application services and what
lessons can be learned?
Cloud computing is a style of computing in which scalable and
elastic IT-related capabilities are provided "as a service" to
customers using Internet technologies.
Eli Lilly
Less-expensive, rapid, seamless access in/outside
• Drivers to cloud
- Traditional infrastructure deployment was inhibiting business
- Desire to move from fixed to variable cost model
• Solution
- Multiple cloud providers
- Google for consumerization
• Benefits
- Reduced provisioning cycle time
• New server: 7.5 weeks to 3 minutes
• New collaboration environment: 8 weeks to 5 minutes
• 64-node Linux cluster: 12 weeks to 5 minutes
• Key Lesson: Time to delivery is paramount
6
Large Energy Company — Experimenting
Wiki & Blog Site Product Evaluation
• Driver/Challenge • Driver/Challenge
- Team needed to provision a blog - Wanted to evaluate open-source
and wiki to support a business unit. content mgmt. system, but internal
- Project team was told to wait for 3 capacity was not available
months for server availability on • Process/Solution
internal systems
- Summer intern provisioned and
• Process/Solution installed Alfresco on Fedora in 1 day
- Provisioned a Fedora image on - Evaluation done, and system shut
Amazon in one afternoon. down after 2 months
- Two months later system went - AMI available for future use if needed
into production following a detailed
security review and penetration • Result/Benefit
testing - Agility, low cost, no capital outlay
• Result/Benefit
- Agility — respond rapidly to new
business needs

• Key Lesson: The cloud excels for temporary environments.

Large Energy Company — Ideal Workloads
Internet Facing Application Development and Testing
• Driver/Challenge • Driver/Challenge
- Analysis of experiments across - Numerous large SAP projects —
multiple labs Dev/Test constrained by internal
- Internal systems push data to this resources
system - Projects typically 6-18 months with
- Available to company and third parties need for additional short-lived
involved in research environments at short notice
• Process/Solution • Process/Solution
- Created parallel development and - Built Dev/Test environment on
production environments Amazon with large Amazon instances
- Started with a hardened Red Hat and permanent IP addresses
Linux build certified by security team. - All traffic is encrypted and local
- Common sign-on and backup to S3 authentication used
• Result/Benefit • Result/Benefit
- Agility, low cost - Agility, Cost, easier process

• Key Lesson: The cloud is an excellent environment for AD.

Razorfish — An Easy Move
• Background/Driver
- Needed to improve their ability to respond quickly to customer
demands to support highly visible web campaigns
- Support high volume short run campaigns more cost effectively
• Process/Solution
- Using Rackspace as a cloud infrastructure platform.
- Build Blogs, Microsites, campaign-related pages for large companies,
such as Southwest Airlines, H&R Block
• Result/Benefit
- From 4-6 weeks and tens of thousands of dollars to set up to 24 to 48
hours and 3K-5K … on average 25% of cost
• Key Lesson: Look at your process models first. If you are moving
web-centric applications with solid security and management practices,
you can move them with little deviation to cloud infrastructure.
Wipro — Private First, Public Later?
Reduced provisioning time for projects and COEs
• Driver to cloud — Traditional provisioning for new
projects and COEs was excessive
• Solution — Private cloud
- Self-service portal for automation of server provisioning
• Benefits
- Internal server provisioning: 46 days to 35 minutes
- Utilization: <10% to 40%
- Reprovisioning of assets: nil to reduced capex 30%
- Average server cost from $2,000 to $800
- Improved license management, network bandwidth utilization and
energy costs
• Key Lesson: Time to delivery can be improved while
utilizing assets more effectively.
 Japan Post — Business-Critical Cloud
Background: Massive recently privatized
organization (20,000+ branches)
(Japan Post Service) (Japan Post Bank) (Japan Post Insurance)
Process/Solution
• Applications developed internally with
guidance from external consultancy
Contracted business
• Seven custom applications in production,
with five more under development
• One large project encompassed 57,000 users, (Japan Post Network)
with up to 25,000 concurrent requests. Sales agency services
• Optimizing internal development process to
match PaaS Source: Japan Post
SaaS APaaS
Results:
• Live nationwide since October 2007 Customer Customer
• Average development time of 3-4 months Information Complaint
(3 to 4 times faster than traditional Management Reporting
development) System System
• High user satisfaction (functional, performance)
• No performance or security issues so far

Key Lesson: Rapid deployment;. early stage prototypes generate user buy-in
Japan Ministry of Economy,
Trade and Industry — Consumer Site
• Driver/Challenge
- Need to build an application to support a new government
program targeting Japanese consumers in a short time.
- Must be available to public via the Web and support potentially large and highly
volatile transaction volumes
• Process/Solution
- Built a consumer exchange Web application on Force.com and salesforce.com sites
- Consumers can exchange old appliances for credits toward new appliances and
merchandise
- Simple interface for public use
• Result/Benefits
- Built in only 3 weeks
- 40 million consumers expected to access site at peak times
- Expected to support more than 20 million transactions … 510,000 transactions
first month
- Has helped boost sales of flat panel TVs and refrigerators
• Key Lesson: Cloud computing works well for high-scalability
requirements.
Presidio Health — Move it All to the Cloud
• Driver to cloud — Rapid business growth
restricted by infrastructure
• Solution
- Appistry for software; GoGrid for platform
- Homegrown apps. for physician performance management and
point-of-service collections
- No rearchitecting of on-premises apps. facilitated by front-ending apps.
with message broker
- Transient data in cloud; sensitive permanent data in traditional database
• Benefits
- PCI & HIPAA compliance
- No unscheduled downtime
- Flat costs for 50% more capacity
• Key Lesson: The cloud is ready for apps but perhaps not for
sensitive data.
 Packaged Shipping Company — Just the
Private Cloud, Please
• Driver/Challenge
- Support complex algorithms with lots of computing power to process large,
complex data streams from multiple sources
- Break large integrated batch processes into more discrete-linked processes
that can execute in parallel to improve response time
• Solution
- Implemented Appistry grid computing environment on internal systems to
create a private cloud grid infrastructure service initially for a single application
- Opportunistically migrate new or existing batch and transactional applications
• Benefits
- Able to develop new analytical application that was not economically feasible
using earlier infrastructure models
• Considering other areas where HPC and “scatter/gather” offers value
- 4 hour batch now runs in 20 minutes. Developing applications in 60% less time
• Lessons Learned
- Easy to break down existing batch jobs
- Need to change the mindset of developers & their approach to development
• From monolithic application all to use of core shared services in the cloud
• From batch/linear to parallel execution and scatter/gather
Author Solutions — Running
the Business in the Cloud
• Requirements/Challenge
- Automate self-publishing workflow for authors
and publishers
- Integrate disparate back- and front-office systems into a complete solution
• Process/Solution
- Created an end-to-end self publishing application using salesforce sites,
force.com and Amazon services
- Integrates with existing on-premises systems including crystal reports,
Microsoft Dynamics, Great Plains
• Result/Benefit
- Developed application in significantly less time and for lower cost than that
estimated for a traditional custom in-house application
- Lower ongoing operational costs
- 50%-75% reduction in time and cost to modify workflow and add products
• Key Lesson: Hybrid cloud/on-premises solutions are complex, but
they work.
David Allen Company — Business
Critical Cloud Services
• Background/Driver
- Needed to build a more unified CRM system to process
complex projects
- Replace numerous disparate legacy systems including a
notes-based
CRM element
- Needed to be able to adapt system quickly and easily
- Did not start with a desire to move external
• Process/Solution
- Provided operational specifications to a variety of APaaS vendors to develop proof of
concepts. Selected a Longjump based system
- Running the business on a cloud-based CRM application with customizations for all
groups that touch customers.
• Result/Benefit
- Significant development savings compared to traditional development.
- Lower operational cost compared to internal systems
- Developed and deployed system in a few weeks. Ongoing updates can be implemented
rapidly. More flexibility to match the system to changing needs
• Key Lesson: Its not always feasible to pick up existing
infrastructure and applications and simply drop them into the cloud.
SaaS Vendors Use the Cloud
• Driver/Challenge
- Business identified need to deliver a number of
HRIS applications in a SaaS model, but a
traditional approach would take 24 months to deliver
• Process/Solution
- Private implementation of Longjump APaaS
- Built cloud application services for HR to be delivered through their
partner channel.
• Result/Benefit
- Developed in 4½ months vs. 24 months
- Able to provide a highly customizable service to their customers
- Focus development on application design instead of infrastructure,
database and security model design
• Key Lesson: The cloud can be used to deliver
highly customizable services.
Fulcrum Saves Money Using Packaged
Integration for salesforce.com

Challenge Solution
• Fulcrum needed to synchronize data • Pervasive DataSynch for salesforce.com (a
between QuickBooks, saleforce.com multienterprise packaged integrating process)
• Did not want to do any custom • Leverages salesforce.com's APEX Web-
integration development work services-based APIs
• Was price-sensitive and preferred a • Includes prebuilt adapters, maps, and so on,
SaaS solution versus owning software and is paid for on a subscription basis
Results: + Two-way synchronization of business data (e.g., customer)
+ No data re-keying (and salesforce.com became master record)
+ No integration development work required
Key Lesson: PIPs can reduce deployment time and cost, and help
companies reduce custom integration efforts.
JohnsonDiversey —
Public + Private = JD Cloud
• Background/Driver
- Legacy on-premise systems were clumsy. Sharing
documentation and collaboration was painful.
- Storage limitations created inefficiencies
• Process/Solution
- Adopted Google apps – gmail replaced in-house e-mail and docs augments
Microsoft Office environment
- Using Google sites for internal for simple team and project collaboration
- Google app. engine used to build an internal talent review application
- Oracle CRM On Demand used for remote sales force.
• Result/Benefit
- Rapid rollout and adoption of applications across
• 3 ½ month for complete total project — Google docs rollout over a weekend.
- Bandwidth consumption for messaging and collaboration reduced by 20%
- Total investment pays back in 14 months. Reduced operating cost of
email/collaboration environment by 70%
- User satisfaction and use up more than 25%
• Key Lesson: At least two credible vendors must be available, must reduce
operating costs, must be extensible and there must be an exit strategy.
Avago Migrates to Google Apps.
Background Key Concerns
• 3,800 users • Who at Google would be able to view
• Bleeding-edge IT profile the mailboxes?
• Company facing 2007 migration • When a court issues a discovery
request, how would Google respond?
Results • How is non-Avago device access
• Migration took two weeks after six protected? What about orphan files?
months of planning/piloting
Lessons Learned
• Migration help from Appirio
• No synchronization with AD; GAPE • GAPE has proven to be a stable and
directory manually maintained functionally rich platform
• Since December 2008, Avago has • 1 person managing 4,100 mailboxes
called the Google help desk four times • E-mail costs were cut with a move to
• Users employing adjacent GAPE and storage woes alleviated
collaboration services differently • Network capacity had to be altered to
• Google meeting 99.9% uptime SLA support a cloud e-mail deployment
Lessons Learned and Best Practices
• Today's Cloud Computing Environment is Best For
- Applications that do not have much interaction with back-end systems
- Web servers
- Apps. where demand varies dramatically over a cycle (periodic peaks)
- Short-term use (AD, QA, campaigns)
- Rapid ramp up required (minutes/hours vs. days/weeks)
• Principles for early-adopter environments
- Use a risk-based approach for security
- Reuse existing processes/policies/learnings
- Automate where practical
- Consider differentiated support options
- Keep it simple for the user/customer
- Consider managing risk by with hybrid architectures that spread load across multiple
providers with a layer of abstraction.
• Issues Remain
- Security
- Data location, privacy, potential loss, portability abilities
- Management/governance
- Vendors