An Analysis of the Cloud Computing Platform

MASSACHUSjETTS INSTITUTE
by OF TECHNOLOGY

Ratnadeep Bhattacharjee JUN 0 32009

LIBRARIES
Submitted to the System Design and Management Program
in Partial Fulfillment of the Requirements for the Degree of
ARCHVES

Master Of Science In Engineering And Management

Massachusetts Institute Of Technology
January 2009.

C 2009 Ratnadeep Bhattacharjee. All rights reserved.

The author hereby grants to MIT permission to reproduce and to distribute publicly paper
and electronic copies of this thesis document in whole or in part in any medium not known
or hereafter created.

Thesis Advisor

Prof. Michael Cusumano

SMR Distinguished Professor of Management & Engineering Systems
Head, Technological Innovation & Entrepreneurship Group
MIT Sloan School of Management

Signature of Author
Ratnadeep Bhattacharjee
Systems Design and Management Program, January 2009

Certified By
Michael Cusumano
Thesis Supervisor
aITSloa 1pf l a ent

Accepted By
_- Patrick Hale
Director
Systems Design and Management Program
Abstract
A slew of articles have been written about the fact that computing will eventually go in
the direction of electricity. Just as most software users these days also own the hardware
that runs the software, electricity users in the days of yore used to generate their own
power. However, over time with standardization in voltage and frequency of generated
power and better distribution mechanisms the generation of electricity was consolidated
amongst fewer utility providers. The same is being forecast for computing infrastructure.
Its is being touted that more and more users will rent computing infrastructure from a
utility or "cloud" provider instead of maintaining their own hardware. This phenomenon
or technology is being referred to Cloud Computing or Utility Computing.

Cloud computing has been in existence in some form or the other since the beginning of
computing. However, the advent of vastly improved software, hardware and
communication technologies has given special meaning to the term cloud computing and
opened up a world of possibilities. It is possible today to start an ecommerce or related
company without investing in datacenters. This has turned out to be very beneficial to
startups and smaller companies that want to test the efficacy of their idea before making
any investment in expensive hardware. Corporations like Amazon, SalesForce.com,
Google, IBM, Sun Microsystems, and many more are offering or planning to offer these
infrastructure services in one form or another. An ecosystem has already been created
and going by the investment and enthusiasm in this space the ecosystem is bound to
grow.

This thesis tries to define and explain the fundamentals of cloud computing. It looks at
the technical aspects of this industry and the kind of applications where cloud can be
used. It also looks at the economic value created by the platform, the network
externalities, its effect on traditional software companies and their reaction to this
technology. The thesis also tries to apply the principle of multi-homing, coring and
tipping to the cloud-computing platform and explain the results. The hurdles for both
users and providers of this service are also examined in this thesis.
Acknowledgements
I would like to thank Professor Michael Cusumano for his advice and guidance
throughout the process of writing this thesis. It was his course, "The Business of
Software", that got me very interested in the notion of platforms and software as a
service. I would like to express my gratitude to each of the following individuals for
agreeing to talk to me about their views and experiences on Cloud Computing: Lew
Tucker, Peter Coffee, Doug Hauger, Pete Koomen, Bert Armijo. I really learned a lot
from those interviews. Thanks to Andrew Kovacs and Sanjay Parthasarathy for arranging
some of the interviews.

I would like to thank Pat Hale, the director of the Systems Design and Management
program at MIT for supporting me while I charted a path tailored to my needs and
interests. I am really grateful for what I have learned from my professors and cohorts at
SDM. The last two years have been an eye opener for me.

I would like to take this opportunity to thank my manager, Mary Lautman, who gave me
the time and freedom to go and pursue my dreams. Without her support this would not
have been possible. I would also like to thank my colleagues at Sun Microsystems who
were very supportive of me during the entire course of the program.

Last but not the least I would like to thank my family, particularly my wife, Jayati and
our daughter Ipsita for bearing with me as I spent nights and weekends doing homework,
writing project reports, attending talks. I would also like to thank my parents for
providing me with the right foundation and instilling in me the right set of values that has
helped me all along.
Table Of Contents

ABSTRACT 2

CHAPTER 1: INTRODUCTION 5
Evolution 5

CHAPTER 2: CLOUD COMPUTING - THE TECHNOLOGY 12

The Push Towards Cloud Computing 12

Defining Cloud Computing 15

Infrastructure as a Service (IaaS) 15
Platform as a Service. 16
Software as a Service (SaaS) 17
Common Features 19

CHAPTER 3: THE VALUE OF CLOUD COMPUTING 23

Defining The Business of Cloud Computing 23

Value Creation In Cloud Computing 25

Analysis of Cost Differentiation 31

Software Development Lifecycle Revisited 32

The Stumbling Blocks 33

CHAPTER 4: CLOUD COMPUTING BUSINESS MODELS 39

Business Models 39

Platform Leadership in Cloud Computing 49

CHAPTER 5: THE ROAD AHEAD 62

The Hurdles 62

The Future of Cloud Computing 66

APPENDIX A: INTERVIEWS 70

APPENDIX B: SOURCES OF INFORMATION 89

BIBLIOGRAPHY 90
Chapter 1: Introduction
A major shift in the area of information technology is beginning to happen. It is being
driven by a few key factors:

a) Better software technology

b) Improved broadband access
c) Desire to be energy efficient
d) Desire to dwell on key competencies and outsource the rest to a specialist.

There is a trend towards computing moving out of infrastructure housed within the
company to a "cloud" that is hosted by a provider of raw computing infrastructure and in
many cases core services around this infrastructure. One might say this has been
happening for a long time. However, the nature of the recent change and the usage
patterns are very different. So much so that it has a profound effect on how companies do
business and make their infrastructure purchasing decisions. This has the impact on
hardware companies that have so far thrived on selling their wares to data centers of all
sizes. This also has an impact on small and medium software vendors and on startup
companies who would otherwise have had to spend a fair amount of their fixed costs on
infrastructure. This thesis examines all of these issues in details.

Before delving further into the investigation of the current technologies and their impact I
would like to talk a bit about the history of infrastructure as a service.

Evolution

Cloud computing in some form or another can be traced back to the early sixties. Time
sharing and multitasking - operating system features of all modern day computers can be
seen as the early enablers of utility or cloud computing. Without these features no form of
utility computing would have been possible.

During the early stages of the computer revolution IBM was considered the leader in this
field. The IBM mainframe was the first real and commercially viable system where the
users desktop machine did not perform much processing besides "dumb" terminal
functions. The mainframe in essence was the cloud and in general was hosted in a data
center located elsewhere from where the client was. Multiple users could use the
mainframe's resources at the same time. The mainframe could process multiple tasks at
the same time too. IBM is still a big player in the utility computing market today. There
were other corporations and technologies that could also be considered the early
innovators in this field. Some of them have changed their service offerings drastically to
survive while others have perished.

The seventies - Tymeshare Inc:

Tymeshare was an early mainframe and services company that rented out mainframes
space and compute power via telephone lines. The company provided high level
computational services to companies like Bank Americard, the IRS, etc. It was located in
Cupertino, CA. The company does not exist today. This could have been due to the
domination of IBM and DEC in the mainframe and mid range categories. There was no
need for the kind of service that Tymeshare provided.

Hosting service providers:

The eighties and nineties saw the birth or numerous companies that hosted large main
frames and mini-computer clusters. The hardware was rented out to corporations for their
data center processing at fixed monthly or yearly rates. The corporations were connected
to these data centers via dedicated ISDN lines. Along with computational power storage
services were also provided. Customers' data was backed up in tapes and restored on
demand. Usually demand for restoration of data had to be made via phone call to a
service desk. Email support was added later on. This allowed corporate data centers to be
hosted outside the premises and corporations did not need to manage the infrastructure
themselves. However, they had to pay a negotiated price for the infrastructure as well as
the service. The cost of service depended on the Service Level Agreement (SLA)
between the corporation and the provider.

Software development within the corporation was usually done on PCs in an environment
that simulated the actual datacenter. Once development and local testing was done the
software was "uploaded" to the real datacenter machines. A round of testing was done
with real data and the then the systems we made "live". The software systems were
unavailable during this process - at times the process could last days, depending on the
complexity of the operation.

Application Service Providers:

The eighties and nineties also saw the entrance of application service providers or ASPs.
As computation became more advanced and a lot more business applications were being
implemented using IT outsourcing of these applications also picked up. [1] suggests that
application service provision belongs to the third wave of outsourcing behind human
resources and hardware .

Application service providers (ASPs) at that time were defined as third-party service
firms that deploy, manage and remotely host software applications through centrally
located services in a rental or lease agreement. Initially ASPs provided compelling value
proposition to SMEs and start-up companies. It allowed small businesses to gain access
to technical expertise and "best-of-breed" applications at a lower cost of ownership.
Among other advantages for the customers offered by ASPs were scalability of
applications over time; access to better IT expertise and state-of-the-art technologies;
rapid implementation time; reduced downtime; and free upgrades. The pricing-model of
ASPs provided a predictable "cash-flow" because pricing was typically based on per user
per month. It also provided a scalable solution in a market place where rapid changes
were occurring in terms of technology as well as within business. The convergence of
software and IT infrastructure towards an open network centric environment enabled the
ASP concept to emerge. ASPs managed and delivered application capabilities to multiple
entities from data centers across a wide area network (WAN). Pricing, customer service,
SLAs and RAS (Reliability, Availability, Serviceability) were the parameters on which
ASPs were measured.

By the year 2000 they had become the next greatest thing since applications themselves.
ASP proponents were able to ride the wave of inflated expectations created by the
dot.com mania. Even the big telecommunications firms such as AT&T and Sprint -
jumped on the ASP bandwagon. Large IT service providers such as IBM Global Services
and EDS were pushing to both host and manage applications. Some software vendors
were getting more aggressive, too. Oracle, for example, expanded deployment options for
its E-Business Suite Online, a hosting service for ERP, CRM, and E-business
applications.

All of this failed to convince too many users. It turned out that ASPs were relatively
expensive to set up and maintain. A lot of employees were required to support the
customers using these applications as well as for maintaining these applications. The ASP
vendors hung their hat on a weak value proposition, namely the marginal cost savings
available from hosting someone else's applications. Users did the math and were not
impressed with what they saw. Given the marginal economics, the questionable business
models of many of the ASPs, and the consequences of failure, the lack of user enthusiasm
is understandable. Today the term ASP has assumed negative connotations and is on the
way to becoming a historical footnote. Most providers have exited the space. However,
some ASPs are hanging on, whether from hope or simply contractual obligations.

A majority of the large ASP vendors have moved their customer support off shore to cut
down service related expenses. Prime among them are IBM, EDS, and other software
consulting firms. This has enabled them to stay competitive. They have also instituted
SLAs with their customers and instituted sufficient risk management and mitigation
strategies.

Web Hosting:

The advent of the web saw the growth of numerous e-commerce and other web
applications. In a lot of cases these web applications are hosted on infrastructure that is
owned by some other entity. The application or content is hosted by one entity (the
provider) and owned exclusively by one other (the user organization). The resources are
off-premises, relative to the user organization, and are dedicated to it. Access to the
resources is available via the Web (that is, by HTTP). Some Web hosting uses fully fixed
resources allocated exclusively to a paying user organization (as in classic hosting). In
other cases, Web hosting is optimized, and some underlying resources may be shared
with some degree of elasticity. Thus, some web hosting extends into the cloud by virtue
of being elastic, although the resources in use remain dedicated to one user organization.
There are numerous small and large companies that still offer this hosting service. The
service essentially consists of a machine with a choice of wither Windows or Linux
operating system and an environment that supports either an application server where the
application can be deployed or engines that can handle scripting languages such as Perl,
Python, PhP, Ruby, etc. They also offer domain name registration, data backup and
similar value added services. The users are charged a fixed rate per month based on the
size of the hardware and type of software their web application needs.

Native Web Applications:

Gartner defines these as web-based applications in which the two parties are the provider
(for example, Google Search or Orbitz.com) and the individual. The resources (that is,
the applications and content) are off-premises, relative to the individual. The resources
are dedicated to only one user organization (the provider) and are on-premises relative to
it. In some notable cases (such as Google, Yahoo and Amazon), the provider offers
multiple applications and uses elastic, massively scalable methods of dynamically
allocating shared resources to its many applications. Thus, these are native cloud
applications. In other cases, (such as most e-commerce sites), the resources are fixed and
have a limited scale capacity - these are not part of the cloud, although they are
certainly Web offerings.

The first web based services started appearing shortly after the internet revolution in
1995-96. Email was one of the first real software available as a service. The Hotmail
web-based email service was founded by Jack Smith and Sabeer Bhatia and launched in
1996. Hotmail was one of the first free web-mail services. Hotmail was acquired in 1997
by Microsoft for an estimated $400M, who re-branded it as "MSN Hotmail". The current
version, "Live Hotmail" was officially announced on November 1, 2005 as an update to
Microsoft's existing MSN Hotmail service. Over time a host of other services became
available over the web. These included email, calendar, chat, talk, map services from
Yahoo, Google and other players.

A very important feature of these services is that apart from displaying content on
browsers these services also expose themselves using well defined Application
Programming Interfaces (APIs). This allows others to incorporate these services in their
web applications. This has created a slew of new ecommerce business categories.
Consider the market for aggregate services like vertical job search engines such as
indeed.com and simplyhired.com. These web applications scour other job related web
applications for relevant jobs and publish all the results on their web site. Other sites like
kayak.com aggregate data from other travel related sites like orbitz.com, travelocity.com,
priceline.com, etc. Zillow.com is a real estate web application that embeds maps showing
location of properties - this is done using Google Maps.

The revenue model for these web applications or web services tend to be quite simple.
Most of these services are available to retail users for free. For businesses that want to
use the APIs the web application provider generally charges a fee that varies from one
party to another.
These web applications are the precursor to Utility and/or Cloud computing which we
define and talk about in details in the next chapter. Chapter 3 talks about the economic
value of cloud computing, and the changes in software development methodologies
brought about by the cloud. Chapter 4 looks at the issue from a platform, and economic
point of view. It evaluates the current providers on various parameters to determine their
efficacy. In Chapter 5 we talk about the road ahead for this platform.
Chapter 2: Cloud Computing - The Technology

In this chapter we look at Cloud Computing from a technology point of view. The
industry has defined the technology in various ways. Industry analysts and bloggers tend
to differentiate between the terms cloud and utility computing. The definition' of utility
computing has focused on a business model that provides the computing services. Utility
computing service is one in which customers receive computing resources from a service
provider (hardware and/or software) and "pay by the drink", much as we normally do for
utility services such as electricity. Amazon Web Services is often used as one example of
such a service. Industry experts define cloud computing to be a broader concept than
utility computing and refer to it as the underlying architecture in which the services are
designed.

Sun's CTO of cloud computing, Lew Tucker, sees cloud computing as delivery of
applications across the network. He cites the example of document management system
like Google Docs that delivers an application over the network. Cloud computing is an
"all encompassing" term that includes any service that is delivered over the network, says
Tucker.

We have decided to go with a subset of these definitions to help explaining the

technology and how it is used. In the next chapter we define Cloud Computing from a
business perspective. Before introducing the definitions we talk about the technologies
that have been instrumental in enabling Cloud Computing as a viable business model.

The Push Towards Cloud Computing

Based on an article in gigaom.com written by Geva Perry, chief marketing officer at

GigaSpace Technologies. http://gigaom.com/2008/02/28/how-cloud-utility-computing-
are-different/
Improvements in technology, particularly virtualization, provisioning, and the use of
Application Programming Interfaces (API) have enabled the business of both cloud and
utility computing. We explain these technological terms and then define the term cloud
computing.

Virtualization allows the representation of computing horsepower as a logical entity. This

virtual computer could be a single physical machine, a series of physical machines
connected by a network, or part of a physical machine that has enough horsepower to be
shared by multiple applications in need of computing resources. Virtualization has
existed in one form or another since the days of the IBM mainframe. More recently
companies like VMWare created the new generation virtualization software. Others like
Xen provide a platform that companies like Sun, Citrix, Oracle have built their
virtualization software. The latest virtualization offering from these companies allows
different operating systems to be hosted on the same physical hardware. This has really
been a boon for cloud providers. Using virtualization they can meet the needs of a diverse
set of Operating System requirements with the similar hardware in their datacenter. This
helps reduce their cost of management of these datacenters. For these virtual machines to
talk to one another or to a network outside companies like VMWare have written
software that represent virtual routers. According to a recent New York Times
technology article [10] Cisco has even come up with a virtual switch that it's selling in
tandem with VMWare.

Provisioning is a general term used to describe the way machines are made usable by
installing an Operating System and other application software on top of it. In the area of
provisioning two important things happened over the last few decades that have
profoundly affected the way software was being deployed in the hardware: server
provisioning and resource provisioning. Server provisioning involves selecting a server
from a pool of available servers; loading the appropriate software (operating system,
device drivers, middleware, and applications); appropriately customizing and configuring
the system. This customization involves creating or changing a boot image for this server
and changing other parameters such as IP address, IP Gateway, etc.
Resource provisioning involves associating the provisioned server with an associated
network and storage resources. All of this can now be managed from a remote console.
There are many software products available to automate the provisioning of servers, from
vendors such as BladeLogic, IBM, Sun Microsystems, VMWare, HP among others.

The third important development that has helped the cause of cloud computing is the use
of Application Programming Interface (API). According to Lew Tucker, the CTO of
Sun's Cloud Computing division, this is an underrated aspect of cloud services. Internet
services are accessible to other services and applications because the former expose their
features as software APIs. What this means is that not only can we use a browser to
interact with a service provided over the internet, we can directly connect to the service
using these APIs. Technologies like SOAP, XML, WSDL have become standard formats
for data and logic interchange between disparate systems. HTTP protocol is usually used
as the transport mechanism. These have been adopted to build internet based services.
The term RESTful 2 API is used a lot these days. This refers to a concept called
REpresentational STate transfer which postulates that data within an internet service be
represented as resources each with a unique URI. Using the HTTP protocol these
resources can be created, read, updated, and deleted thereby manipulating resources on
the web.

According to Peter Coffee, Director Strategy, at Salesforce.com, many of the reasons for
moving to the cloud are not technical at all. The complexity of administering computing
is rising. This is partly because governance expectations have risen dramatically,
primarily due to regulations such as SOX and HIPPA. He also notes that the processors
of today they are not the fastest in single threaded chips. These are multi threaded at the
core. This means that tasks don't improve linearly simply by running on a machine
having these new multi core chips. shared data centers can continue to exploit the

2 REST refers to REpresentational State Transfer. The aim is to express every service
delivered over the web as a set of resources using universal resource identifiers (URI).
The client can perform create/read/update/delete operations on these resources during the
process of running the application.
improvements in chip technology that small organizations cannot hope to. The CAPEX
per dollar is tilting towards shared data centers.

Defining Cloud Computing

We have decided to define the Cloud Computing on the basis of how the technology is
delivered and used. We see three broad definitions:

Infrastructure as a Service (laaS)

This is generally defined as the delivery of computing infrastructure as a service. The

services include raw computing power, databases, networks, load balancers, etc. This
obviates the need for a business to have its own datacenter. Rather than purchasing
servers, software, data center space or network equipment, business can access these
resources from a cloud service provider. The service is typically billed on pay as you go
model and amount of resources consumed (and therefore the cost) will typically reflect
the level of activity. A typical bill plan may be 30 cents per GB of storage used per
month or 40 cents per CPU hour of compute resource used.

The infrastructure services are made available to clients via software APIs. Typically
RESTful and SOAP APIs are commonly used to access these infrastructures. IaaS is
heavily dependent on platform virtualization technologies for running client specific
virtual machines. One user may want to run an instance of the Windows Operating
System (OS) while another user may want to run an instance of the Solaris OS.

The implementation of IaaS may also include computer network including firewalls, load
balancers that one would need in a typical datacenter to provide security, high
performance and reliability to an application. Some cloud providers like Amazon Web
Services 3 have created a large community of users who have created language bindings
for the RESTful APIs that the web services expose. Other companies like RightScale 4

3 Amazon Web Services - http://aws.amazon.com

4 http://www.rightscale.com
provide a nifty browser based user interface that allows application developers to
assemble the infrastructure required to run an application by dragging and dropping
widgets representing CPUs, databases, storage devices, load balancers, firewalls, etc.

The various providers of this service differentiate their services in the number of OS
platforms they support, the software that comes with the OS, the pricing model and the
Service Level Agreements they have.

Platform as a Service.

Platform as a Service (PaaS) is a model of computing that makes all of the facilities
required to support the complete life cycle of building, testing, and delivering
applications and services over the Internet from a single cloud-with no software
downloads or installation for developers, IT managers or end-users. According to
Wikipedia this is also known as cloudware. PaaS is the next stage of what is already
available as mashups. Users can create applications by adding features from Google
Maps, Google Calendar, and other web services and make these features as part of their
applications. The next stage is to automate the process and that's what we see in the PaaS
model. In Platform as a Service the application developer does not have to write code per
se as the application can be developed by logic and visual customizations that the
platform provides.

PaaS offerings include workflow facilities for application design, application

development, testing, deployment and hosting as well as application services such as
team collaboration, web service integration and marshalling, database integration,
security, scalability, storage, persistence, state management, application versioning,
application instrumentation and developer community facilitation. These services are
provisioned as an integrated solution over the web. Salesforce.com's "Force.com" and
"VisulaForce" platforms are the poster child of the PaaS model. This model seems like
the web equivalent of Visual Basic development platform where developers could
visually assemble an application and add bits of code to customize it to their needs. In
this model the customization of the application is part of the development process.
According to Dave Mitchell, the CEO and founder of Bungee Labs 5, a PaaS Startup out
of Utah, the PaaS model will stop the current practice of developing "here" and
deploying "there". In the traditional model of software development applications are
written in one environment, tested in another, and redeployed to yet another for
production. In addition to the costs of building, configuring and maintaining these
separate environments, applications almost always need to be maintained, which incurs
even more costs along the way. In the conventional model, these costs and attendant risks
fall on the application owner, and are considered part of the cost of deploying a web-scale
application. In a PaaS environment, the entire software lifecycle is supported on the same
computing environment, dramatically reducing costs of development and maintenance,
time-to-market and project risk. PaaS should let developers spend their time creating
great software, rather than building environments and wrestling with configurations just
to make their applications run-let alone testing, tuning and debugging them.

Another characteristic of PaaS is the fact that it provides integration with other Web
Services some of which may not be part of its offerings. For instance an application
assembled using SalesForce.com's platform allows developers to connect to other native
and non-native services residing elsewhere in the cloud.

Given that the entire development environment is available on a cloud that is accessible
via the internet collaboration becomes easy. Needless to say the system is access
controlled so that a developer can collaborate only with those she chooses to.

Software as a Service (SaaS)

This is defined as the delivery of an application across the internet. The application is
hosted as a web application or service and is made available to users either through a
browser interface or via well known web service interfaces. Some of the services offered

5 http://www.bungeelabs.com
are free. Prime examples of such services are those provided by Google, Yahoo and other
providers. Most client server based applications can be made available using this model.
The browser acts as the client. Other services or non browser applications can also act as
clients. A service can be hosted on any datacenter as long as it is exposed to the internet.
Companies like Google, Microsoft, Yahoo provide common services like mail, calendar,
mapping, etc. from their datacenters. There are various other specialized services such as
CRM applications that are also offered as a service by companies like SalesForce.com.
The former services are usually free unless the user is expecting one or more value added
features. The latter type of services is usually offered at a monthly or annual subscription.
The SaaS model eliminates the need to install and run the application on the customer's
own computer, it also alleviates the customer's burden of software maintenance, and
support. On the flip side customers relinquish control over software versions or changing
requirements. One major problem with upgrading to newer versions of traditional
software is the problem of backward compatibility with the existing data. In the SaaS
model the data is stored along with the service in the provider's repository. It is the
provider's job to ensure that the new version of the service being rolled out works with
the existing data. SaaS reduces the upfront expense of software purchases through
monthly on-demand pricing. If software is sold as a service it cannot be pirated. Hence,
from the software vendor's standpoint, SaaS has the attraction of providing stronger
protection of its intellectual property. The SaaS software vendor may host the application
on its own web server. This job may be handled by an infrastructure service provider!

Application areas such as Customer relationship management (CRM), video conferencing

(Webex, Citrix), human resources, IT service management, accounting, IT security, web
analytics, web content management, e-mail, calendar are some of the initial markets
showing SaaS success. A new set of SaaS providers are coming up with newer
applications like Billing as a Service, and Monitoring as a Service. The distinction
between SaaS and earlier applications delivered over the Internet is that the latter were
developed specifically to leverage web technologies such as the browser and provide an
alternate user interface to the user. The browser solved the problem of thick clients
allowing applications to write a single front end that worked on all browsers. The user did
not need to install any client code. With SaaS the applications are now services that can
be accessed from a browser as well as via APIs from other services. According to Pete
Koomen, a product manager at Google, advances in SaaS were possible because of the
inroads made by rich web applications brought about by advances in web2.0
technologies. This increased the use of web-based applications like mail, calendar, maps
etc. that eventually led to these applications being exposed as services.

Common Features

All three of the cloud services defined above have some common characteristics:

Delivery Mechanism: A key feature of cloud based services is that there is no need to
install or update the application or software that is being delivered. The service is
essentially available "on demand". Application delivery is closer to a one-to-many model
than to a one-to-one model, including architecture, pricing, partnering, and management
characteristics. There is essentially one instance of the application running in the cloud.

Management: The activities of the cloud are managed from one or more central
locations rather than at each customer's site. This also means that the features of the
application are updated from a centralized location without the user having to do
anything at all. This obviates the need for downloadable patches and upgrades.

Measurable: The usage of cloud services must be measurable. Otherwise, they cannot be
billed. The way they are measured is by building telemetry as part of the service offering.
At any point in time the resource utilization for a given user can be measured. Also the
granularity at which resource utilization is measured is very important. This allows the
provider to change bill plans from time to time without changing the underlying metering
software. For instance, the service provider may choose to bill compute services at the
rate of 30 cents per CPU hour for a few months. Later it may want to change the bill plan
to 40 cents per hour during the day (8:00 am to 5: 00 pm, say) and at 20 cents per CPU
hour for the rest of the time. In order to charge the users in this way the service should be
measurable at least on an hourly basis.

Self-healing: To the user of cloud services there should be no disruption of service as far
as possible. In case of failure, there will be a hot backup instance of the application ready
to take over without disruption (known as failover). It also means that when a policy is in
place that says everything should always have a backup, when such a failure occurs and
the backup becomes the primary, the system launches a new backup, maintaining my
reliability policies.

Multitenancy: One key feature of all forms of Cloud Computing is the fact that different
clients are accessing the same service or using the same infrastructure without the
applications knowing about each other. This is known as multi tenancy. Multitenancy is
contrasted with a multi-instance architecture where separate software instances (or
hardware systems) are set up for different client organizations. With a multi-tenant
architecture, a software application is designed to virtually partition its data and
configuration so that each client organization works with a customized virtual application
instance. The concept is similar to having more than one tenant living in a house.

When it comes to "Infrastructure as a Service" multitenancy is the direct result of

virtualization. With both SaaS and PaaS the design of the data models and the
applications are specifically built with a 'multi-tenant' backend, thus enabling multiple
customers or users to access a shared data model. This further differentiates SaaS from
client/server or 'ASP' (Application Service Provider) solutions in that the former
providers are leveraging enormous economies of scale in the deployment, management,
support through multitenancy.

Service Level Agreement (SLA)

As cloud services are being used more and more users are asking for some sort of
accountability from the service providers. The service providers are responding by
introducing a certain minimum level of service. These guarantees are enshrined in the
Service Level Agreements (SLAs) that the providers agree to when the user signs up for
the service. The system is dynamically managed by service-level agreements that define
policies such as how quickly responses to requests need to be delivered. If the system is
experiencing peaks in load, it will create additional instances of the application on more
servers in order to comply with the committed service levels - even at the expense of a
low-priority application. Sometime the users are made to pay extra for a certain level of
service over and above the "free" SLA. If the cloud provider violates the SLA the user is
credited either financially or by offering free service for a certain period of time.
Google has an SLA for mail, calendar, docs clubbed under the Google Apps SLA 6 . The
agreement starts with the following paragraph:
"Google Apps SLA. During the Term of the applicable Google Apps Agreement, the Google
Apps Covered Services web interface will be operational and available to Customer at least
99.9% of the time in any calendar month (the "Google Apps SLA"). If Google does not meet the
Google Apps SLA, and if Customer meets its obligations under this Google Apps SLA, Customer
will be eligible to receive the Service Credits described below. This Google Apps SLA states
Customer's sole and exclusive remedy for any failure by Google to provide the Service."

Amazon has a service commitment for its Elastic Computing Services (EC2) 7 which
states the following among other things:
"AWS will use commercially reasonable efforts to make Amazon EC2 available with an Annual
Uptime Percentage (defined below) of at least 99.95% during the Service Year. In the event
Amazon EC2 does not meet the Annual Uptime Percentage commitment, you will be eligible to
receive a Service Credit as described below."

Most cloud providers are also making the availability of their systems available via APIs.
This is being used to report the status of their cloud in other websites like
cloudstatus.com:

6 Thefull SLA can be found at: http://www.google.com/apps/intl/en/terms/sla.html

7 Amazon EC2 SLA can be found at: http://aws.amazon.com/ec2-sla/
 0 !r eitk

Outage Dashboard
Amazon Service Summary
This dashboard displays the last week of health status for selected remote computing serv
Elastic Compute Cloud (EC2)
services with recent outages, a health bar is shown. Given no recent outages in a providel
Simple Storage Service (S3) are shown. Click a Service in the left panel for detailed service health status, metrics, and
Simple Queue Services(SQS) Amazon Web Services
Simple D8 (SDOB)
Simple Storage Service
Flexible Payment Service (FPS)
Googla App Engine Summy
Engine
Datastore
Simple Queue Service
memcache
URLFetch

t Service alerts on Twitter

wHyperic is hiring
Google App Engine
Health
Sign-up for CloudStatus Updates.

Figurel: Website showing the outage dashboard of Amazon and Google App Engine.
(Source: http://www.cloudstatus.com)
Chapter 3: The Value of Cloud Computing
Cloud computing looms largeron corporatehorizon
Sunday December 21, 3:17pm ET
By Michael Liedtke, AP Technology Writer
Cloud computing: Pie-in-the-sky concept or next big breakthrough on tech
horizon?

SAN FRANCISCO (AP) -- Todd Pierce recently put his job on the line.

To meet the computing needs of 16,300 employees and contractorsat

Genentech Inc., Piercetook a chance and decided not to rely entirely on
business softwarefrom Microsoft, IBM or anotherlong-established
supplierthat would have let Genentech own the technology. Instead,
Piercedecided to rent these indispensableproductsfrom Google Inc.

The above clipping is from a recent article published by Associated Press8 . The case of
Todd Pierce and Genentech illustrates one of the ways in which businesses are adopting
cloud computing.

In this chapter we define cloud computing as a business and look at the value of cloud
computing. We look at the values created by each of the three forms of cloud computing
defined in the previous chapter - SaaS, PaaS, and IaaS. We also look the issues that could
offset these values.

Defining The Business of Cloud Computing

Cloud computing is a business model where IT services, both infrastructure and software,
are provided by external service providers, called cloud providers. These services are
either free or users pay for each service based on actual usage. This allows users to
concentrate on the actual task they need to accomplish leaving the all other dependent
tasks, such as installation, upgrades, maintenance of software, on the cloud provider. This
allows users to spend less on non-core expenditures. The service providers tend to have

8 http://biz.yahoo.com/ap/081221/tec_cloud_computing.html?.v=2
many low paying customers as opposed to a few high paying ones. Owing to a large
number of customers the service providers are able to amortize their cost over a large set
of users bringing down the cost of serving each customer.

Cloud computing has been in the limelight in 2008. Companies like Amazon, Google,
and Salesforce.com have led the way in terms of mindshare and revenue. However,
revenue from cloud based services for Google and Amazon still do not come close to
their other sources of revenue. Amazon Web Services provide infrastructure and core
services like content management that commercial websites need. Their income from
these services as reflected by the "other income" line on their income statement was 22
Million Dollars during the quarter ending September 2008 as compared to loss of 3
Million for the same period last year9.

Google doesn't disclose the results of its business applications division, but it's relatively
small, too. According to Associate Press the Mountain View-based company's non-
advertising operations generated combined revenue of just $540 million during the past
four quarters, while Google's advertising sales totaled $20 billion. After studying cloud
computing trends, Sanford Bernstein analyst Jeffrey Lindsay predicted Google's cloud
based services (Google Apps) will rake in revenue of about $1.5 billion by 2012, a small
share next to the estimated $18 billion for Microsoft's desktop office software.

Microsoft recently announced Azure, their vision of cloud computing. This does not seem
to be a pure cloud offering as that would preclude usage of their existing revenue
generating server and office software business. Azure allows users to run software either
locally on their machine or remotely in a cloud offering from Microsoft.

There are numerous smaller players offering a variety of services using different business
models - these will be discussed in the next chapter.

9 Source: Amazon's financial statements publicly available on their website.

Value Creation In Cloud Computing

In order to analyze the business value of cloud computing we ask the following question:
Why should a business move its computational needs to the cloud? The paragraphs below
explain a set of possible reasons to move to the cloud.

Better Utilization of Resources:

As mentioned in the previous chapter Virtualization is a key technology that fueled the
move towards cloud computing. Non-virtualized enterprise datacenters are not 100%
utilized. According to Gartner in the case of non-virtualized datacenter the utilization rate
can be as low as 20% to 30% at times. This is because an application running on a given
hardware configuration may not utilize all of the resources of the hardware. Also, the
application may not need to be up and running all the time. Different applications may
need different hardware because of the way the datacenter is organized. A cloud provider
of infrastructure services would typically run a virtualized datacenter where there is better
utilization of hardware. Fewer or no physical servers in the data center leads to reduced
power and cooling costs, not to mention savings on server hardware and related
maintenance over time. Application, backup, and disaster recovery testing is now
completed in a fraction of the time it takes with purely physical servers. According to Ian
Pratt l o, the chief architect at Xen, a leading provider of open source virtualization
software, one of the features of the technology is the ability to transfer the location of
virtual machines between different physical machines. The hosting companies love this
because it means they can support more users from the same hardware by dynamically
moving virtual machines around to balance the load. If there are a few users making
heavy use of their virtual machines on one server, the hosting company could arrange to
move them to different physical servers and then rebalance that over time. Rather than
selling their server capacity to (say) 100 users, this extra "statistical multiplexing game"

10 Gartner Research - publication # G00159774: Gartner Interviews Ian Pratt,

Virtualization Visionary.
would mean they could sell it to 200 users. These ideas have been brought to the cloud
arena leading to much better data center utilization in cloud computing business.

Economies of Scale and Economies of Scope:

According to Peter Coffee, director of strategy at Salesforce.com, the number of

regulations that businesses need to comply with has increased over the years. As more
and more services are being offered using software the types of compliance at the
enterprise level is increasing as well. There is SOX and HIPPA, for instance. Besides,
users' expectations from software have also increased a lot over the years. Users now
expect their software to be more secure, responsive, and data oriented. Users of a mail
service for instance would want the service to take care of controlling spams as well as
preventing Trojan Horses from infecting their PCs. Users of CRM software from
Salesforce.com would not like their data to be visible to a competitor who is using the
same service from Salesforce.com. Enterprises that offer their services over the web - a
search service, for instance, would like to prevent denial of service attacks that could
bring their business to a grinding halt. All of this requires extra investment that the
business would want to do without. A public cloud provider who serves more than one
business has the necessary scale to employ the best of breed solutions and amortize the
cost across a large set of users and/or businesses thereby giving their customers more
"bang for the buck". Small businesses may simply not be able to afford a security
specialist or extra hardware that could take care of securing their infrastructure.

Peter Coffee also mentions that cloud providers would get better chip utilizations than
traditional data center customers. This is because the new breed of chips are essentially
getting better performance from their predecessors by virtue of chip level multithreading.
Applications performance does not scale linearly as these applications may not have been
written in a way to take advantage of the new chipset's architecture. However, cloud
providers who expose their infrastructure via an API can implement this API in a way
that makes better utilization of the chipset's architecture.
New Datacenter Requirements:

For companies to derive more value out of their IT spending the enterprise IT
professionals are now expected to provide a richer set of services which require
knowledge of not just software and hardware but other subjects like Statistics,
governance, etc. Datacenter employees need to engage business units in more of their
data-related tasks, to provide strong policy leadership in information security and
business process governance, and to ramp up their initiatives in data analytics and search.
These are very specialized tasks that require specialized skills that companies may not be
able to invest in. Cloud providers, on the other hand, would have the capability and the
know how to provide these complex services. Take for example, the analytics 11 service
being offered by Google. More and more businesses are using this service to get rich
insights into their website traffic. This is not easy to build in house - the business would
have to invest in new employees with the right skill set. They would also have to delay
their time to market as this software would take some amount of time to be implemented.
All of this can be avoided by subscribing to the service offered by Google or some other
provider.

Lowering the entry barrier for new businesses:

New businesses that use software would typically have to set aside money for purchase of
hardware, software packages that would be installed in the hardware, and human
resources that can install and maintain that software. This is a sunk cost that companies
can do without, especially since they do not know whether their business will succeed or
not. If the business fails the money spent on this infrastructure would be wasted. If the
business takes off really well the company may not have the infrastructure to support the
all their customers. This may cause dissatisfaction amongst their customers causing the
company to lose business. They may decide to spend more money to augment their
infrastructure only to realize that the demand was very short lived. The crux of the

1 http://www.google.com/analytics/
problem is that it is hard to forecast how a new business would shape up and if
companies have an opportunity to use and pay for computing infrastructure based on the
needs of their business it would work out to be more economical for them. This reduces
the money needed to start a new business thereby reducing the barrier of entry for many
businesses.

Easier to install and test new software.

In the traditional software development model the enterprise needs to set aside hardware
and software resources for the purposes of testing. These resources are a sunk cost to the
organization in that they cannot derive much value from these once the testing is over and
the next round of testing is yet to begin. In a pay as you go model enterprises can use as
much infrastructure resources for testing purposes as is needed. However, immediately
after the testing is over they can let go of these resources. No additional costs are incurred
and there is no idle hardware/software cost involved.

Software upgrades are also easy. The enterprise can provision a new set of servers, load
the latest software and "turn it on" once the testing phase is over. While turning on the
next set of servers they can turn off and de-provision the existing set of servers. This is
much less cumbersome than upgrading a piece of software while it is running.

Dynamic and Granular Sourcing of Infrastructure:

According to Thomas Bittman, a datacenter analyst with the Gartner Group, enterprises
can source their infrastructure needs differently now. He predicts that traditional
outsourcing and hosting is not dead, but it is dying. Economies of scale are replacing
skills, and speed is replacing static contracts. Once enterprises decide to use infrastructure
as a service they would base their choice of vendor based on various factors - price,
reliability, availability of service, support etc. Even within these constraints it is possible
for enterprises to go with one provider for a certain set of infrastructures services and use
another provider for a different set of needs. For instance, an enterprise could choose to
use storage services from Amazon, computing services from Google and use
Salesforce.com as their CRM software provider. All of this is possible because the
services are accessed via standard APIs. Although cloud providers have different APIs
fir accessing their services there are cloud based middleware companies that are building
a uniform API layer fronting these various cloud providers. These APIs allow datacenter
applications to stay the same while switching between cloud providers. While it would be
hard for an enterprise to move from Salesforce.com to some other CRM service provider
they could easily move to a different provider for storage and compute services. IT
managers can make sourcing decisions a project at a time, and dynamically based on
changing workloads, priorities and costs. This is a huge benefit that enterprises did not
have before.

Reduced Risk from Project Failure:

Software projects often face the risk of delay or non-completion. Enterprise managers
may realize that the project they started a few quarters back is not relevant in a new
business environment. It may also be that the implementation did not go as planned and
the enterprise is forced to reevaluate its choice of software, hardware, human resources
assigned to this project. Risk of IT project failure is a huge concern for many business
buyers that make them not want to buy any software, SaaS or on-premise. SaaS has an
advantage in that if a project fails six months into a program, the SaaS buyer has invested
very little and can easily switch providers, whereas the on-premise buyer has already
spent millions of dollars for licenses and infrastructure and consultants.

Process Improvements:

One of the key characteristics of cloud services, particularly SaaS and PaaS, is that it
reduces the amount of time users spend on doing non core activities that traditional
software users have to contend with - installation, configuration management, upgrades
and other change management. This allows enterprises to revamp their processes to be
much leaner as compared to enterprises that use traditional on-premise software. This is a
very profound change based on my experience developing traditional software for the last
15 years. One cannot underestimate the relief of not having to continuously go through
testing, debugging, reinstallation cycles. That does not mean there is no learning curve
for SaaS based applications but once that knowledge has been acquired things would be
much more smooth sailing. Enterprises can have a leaner workforce, as some of the work
mentioned above is not there anymore. The extra workforce could be retrained to perform
some other task more useful the core business of the company.

Speed and Agility of Development:

As mentioned in the previous chapter one of the key characteristics of Cloud Computing
in general and SaaS in particular is that the software is upgraded much more frequently
than its traditional counterpart. Oracle, for instance may release newer versions of its
CRM software once a year. Salesforce.com, on the other hand deliver updates every
quarter. According to John Martin 2, CTO and SVP Strategy for IQNavigator, a SaaS
provider for on demand services procurement, the business buyer is faced with a choice
of the following:
a) can their implementation team logs in today to a hosted test environment to start the
implementation process, or
b) would they rather wait for the IT department to finish making an environment ready
with multiple servers, the Oracle database, an application server, a half-dozen ancillary
setups, and Internet browser access for their business partners.
The wait for option 2 is usually 3-6 months (assuming the business buyer can prioritize
the project high enough for IT's involvement and mid-six-figure investment), so
inevitably the business buyer would choose the hosted approach because it delivers value
much faster than on-premise.
He also states that Cloud Services have several advantages in evolving with the
customer's process requirements. SaaS is designed to be inherently configurable, with
switches to alter functionality for a single customer in real-time. Because SaaS operates

12 As stated in his blog:

http://buildingsaas.typepad.com/blog/2006/08/what_businesses.html
on a single code base and releases new capabilities every few months, new software
features and innovations are quickly available to every customer without the multi-year
product cycles or upgrade delays of on-premise software. In many cases, customizations
can be added within a product cycle or two for customers that require functionality
specific to them. In my personal experience of working with a Billing as a Service
provider we saw the company incorporate new features in their product within a span of
just 2 months. This helps users of the SaaS based software to change their applications
sooner.

Analysis of Cost Differentiation

Providers of cloud computing services and systems integrators believe that the following
cost reductions can be achieved by moving enterprise software to the cloud 13:

* Costs savings for servers, storage, networking, software licenses, power &
cooling, real estate.
* System administration labor - with the enterprise software running on an external
cloud.
* Costs savings due to decreased provisioning time.
* Leaner software development workforce due to decrease in amount of software to
be written.
* Shortening of application lifecycle and decreased time-to-market of new products
and services.
* Increased reliability at lower costs.
* Future costs that one encounters during maintenance, downtime, upgrades or
hardware, infrastructure applications, etc.
These can be quantified by the industry or vertical function in question.

1: Based on ongoing discussions on the "cloud computing" Google group.

http://groups.google.ca/group/cloud-computing
Software Development Lifecycle Revisited

The difference between cloud computing and on-premise software development is bound
to have some impact on the traditional software development methodologies.

For one, an enterprise's traditional software development lifecycle, agile or waterfall, can
become shorter once cloud computing is adopted. This is because some of the
finctionalities that were performed in house would now be provided by the cloud vendor.
This includes installation, patching, upgrades, etc. Most of the issues surrounding
performance and scalability will rest with the cloud service provider too, especially if the
software was initially being developed in house.

The quality assurance process also becomes shorter in some cases. Particularly if an
enterprise which initially had the software written in house decides to use a cloud service
provider for the same. The cloud provider usually supplies a sandbox to perform system
testing as a result of which no time is wasted in creating and maintaining test beds.
Enterprises who are moving their application to a cloud can create an on-demand test
environment on the cloud very easily without having to buy any hardware. For example,
if the application were to be eventually deployed on the EC2 platform, one can easily
deploy the application on as many EC2 images as needed to perform the testing and then
let go of that environment once the testing is over. This saves a lot of time.

According to Peter Coffee, Director Platform Strategy, at SalesForce.com, his integration

partners now have a 90 day cycle for deploying CRM applications based on
SalesForce.com. This is divided into three 30-day phases. In the first phase the
application is designed and implemented. In the next 30 days the integrator sits with the
enterprise and addresses visual and business issues as they show up - there is no
rebuilding etc. to be done. Over the next 30 days his team adds additional features that
are "nice to have" from the enterprises perspective. This was unthinkable with the
traditional software development model. Platform as a Service, like force.com, by their
very nature allow applications to be developed without too many lines of code bring
written. An enterprise simply has to customize the platform to its needs. This obviates the
need for some of the major steps in traditional software development.

Between the waterfall and agile models of software development the latter tends to lend
itself more to the new technology especially with SaaS. Agile chooses to do things in
small increments with minimal planning, rather than long-term planning. Each such
iteration involves a full software development cycle, including planning, requirements
analysis, design, coding, unit testing, and acceptance testing when a working product is
demonstrated to stakeholders. This is well suited to a cloud environment as the
infrastructure to test, deploy or validate is already taken care of by the provider. Cloud
services, at least on paper, are always available and this makes the agile goal of always
having software that's working a distinct possibility. That does not mean the waterfall
development model is going away - I am involved in a project where we have chosen to
build an application using a third party software as a service and have used the waterfall
methodology of requirement analysis, design and development. Needless to say, the
process of integration simply goes away with SaaS as any code that is written is
integrated with the SaaS application right from the beginning.

For software to be deployed on the cloud an enterprise has to think about the deployment
strategies early in its development life cycle - at the design stage. This is because the
cloud provider would have exposed its environment via a fixed set of APIs and any
software deployed in the cloud has to conform to those APIs. The software should have
algorithms in place that adds new virtual machine instances or new storage volumes as
more data is generated by the application or more requests are being served by the
application. This is in contrast to the traditional software development model where the
situation is much more adhoc.

The Stumbling Blocks

According to the Associated Press Cloud Computing has already swelled into an
estimated $36 billion market this year, representing roughly 13 percent of global software
sales. The big question now is whether it has reached its moment of reckoning, will it
sweep the likes of Microsoft, Oracle, and other software industry staples into
obsolescence? Not yet, in our opinion. There are several real and perceived problems that
need to be overcome.

Performance Bottlenecks:

Performance of an application residing in the cloud is still a question that is being

evaluated. According to Forrester Research 40 percent of companies consider SaaS
application performance a key concern. One thing is sure, there is some amount of
latency in the transit of data from the cloud to the client accessing the data which does
not exist for applications running in in-house datacenters. At issue is the protocol for data
interchange. Most cloud delivered applications rely on the traditional get/post mechanism
of the HTTP protocol. While great at document delivery, that architecture was never
really designed for dynamic user interfaces. We have added various layers on top of the
basic delivery mechanism to make things easier for but the underlying delivery
mechanism is still the same. Thus, we're really using well-designed, well-delivered Web
sites when using SaaS applications, not true, dynamic native interfaces.
A few things can be done to improve the situation - make sure the pipe between the user
and the server is as wide and short as possible. This can be achieved if the cloud provider
has multiple data centers and clients are served from the data center closest to them. In
some cases a geographically disbursed company may end up using several data centers or
the services of content delivery networks like Akamai to provide the best performance.

Switching Costs:

The traditional software model has its own set of problems. However, it would be safe to
say that enterprises have a set of applications in "steady state" that are running on
traditional software packages from companies like Oracle, SAP, Microsoft etc.
Enterprises would have to incur a huge cost in moving these applications to the cloud.
They would have to undergo organizational changes in order to move to a cloud based
computational model - fewer support personal would be required, the application
development lifecycle would change. Most organizations may not be ready for such
changes.

Also, what happens to all of the existing data center hardware if applications are moved
to the cloud or enterprises start using software services such as CRM from
Salesforce.com? These would have to be disposed off at a discount, something companies
may not be willing to do.

Business-software powerhouses Microsoft, IBM Corp., Oracle Corp. and SAP are all
maneuvering to protect their existing, lucrative software franchises while also setting up
their own online services to compete with the industry upstarts. Even Genentech, the
biggest U.S. company to buy Google's applications package so far, isn't ready to abandon
Microsoft entirely. It's still licensing Microsoft programs like Word for writing
documents and Excel for creating spreadsheets.

Single Point of Failure:

Corporate data centers, websites, and ISPs are prone to security threats on a regular basis.
These happen in various ways - denial of service attacks, virus infection, machine break-
ins etc. If a single corporate data center is attacked the effect is felt only by that
organization and the few that do business with it. If a large website like Microsoft.com is
attacked the effect is felt by a larger community - all users who want to download
patches or updates to existing Microsoft software, for instance. Cloud providers are not
immune to such attacks. However, they are playing host to a much larger set of customers
and possibly some enterprises whose business systems have completely moved to the
cloud. A failure of any sort will have a much larger impact. Cloud providers try to
minimize this by having multiple datacenters replicating user data and applications.
Hence, total shutdown may not be imminent in the case of failure in one of the data
centers. However, application performance will definitely be affected.

Privacy and Security:

The privacy and security of applications is a major issue. Many enterprises would hesitate
to move their data outside the firewall due to the nature of the data or the nature of their
business. Various countries or groups of countries have their own regulations around
where enterprises can store their data. Canada for instance, bars businesses from storing
data in the US due to the latter's draconian Patriot Act. EU has norms which mandate that
certain businesses store data within EU countries. Cloud providers are getting around
these by establishing data centers in more parts of the world. However, there isn't enough
distribution of the datacenters to overcome these regulatory constraints. One would
imagine that it will be hard to do business in China at all for a cloud provider that does
not have a local data center in China.

Lack of proper Service Level Agreements:

Currently cloud providers use uptime as the only measure in their service level
agreements (SLA). There are various other factors that cold come into play here. For
instance, customers who are willing to pay a premium price may want to get a lower
response times for their requests. According to InformationWeek 4 , Microsoft, through its
Azure cloud platform, is trying to create additional SLAs. These are still in their infancy.
The other problem with SLAs is that it is difficult to express and implement these SLAs
at a technical level. For instance, if a cloud provider promises a maximum length of time
for a certain transaction, the provider cannot guarantee it simply because part of the
transaction involves data flowing through the internet over which they really don't have
much control.

Lack of variety of services:

One other problem that cloud consumers face is the lack of all types of services. A case in
point is that of Northrop Grummen, the defense contractor. They wanted to use

14 http://www.informationweek.com/cloud-
computing/blog/archives/2008/1 0/willmicrosoft_2.html
SalesForce.com's CRM application. However, the tool they used for project estimation
and management, SEER developed by Galorath, was not available on the cloud. Being a
very large customer they were able to convince Galorath to make SEER available as a
service. Galorath ported SEER to the SelseForce.com platform thereby making it
available to Northrop as a service 15. The fact is that not all applications that enterprises
use on a daily basis are available on the cloud. Until that happens it would be difficult for
many enterprises to move to the cloud without considerable disruption to their business.

Vendor Lock-in:

Although the move towards cloud computing has accelerated there are still a handful of
large cloud providers - Amazon, Google, Saleforce.com. There are a whole bunch of
smaller players like RightScale, Hyperic, Mosso, Elastra - these startups are playing a
vital role in furthering the cloud ecosystem. However, this being the inception phase of
the technology some of these companies may not survive because of poor execution, or a
flawed business model. Others may be acquired by larger companies. What happens to
the user's applications and data in such cases? In the case of traditional software
companies one always has a tangible piece of software in hand whether the company
selling the software perishes or not.

High Initial Investment:

Most of the cloud providers are not able to reap the kind of profits that traditional
software companies make. Mark Benioffs Salesforce.com is now the largest cloud
computing service for businesses, with a market value of $4 billion, about 52,000
customers and revenue totaling $1 billion in its past four fiscal quarters. Yet according to
Associated Press, Salesforce's income of $37 million during that time translates into a
measly $3.70 profit on every $100 in sales. That looks anemic alongside Oracle's net
margin of about $24.80 for every $100 in sales in the comparable period. San Mateo-

1:5 http://www.galorath.com/wp/how-galorath-quantified-the-salesforcecom-platform-with-seer-
for-software-seer-sem.php
based NetSuite still hasn't eked out its first quarterly profit after a decade in business,
despite steady growth that boosted its revenue during the past four quarters to $143
million. The slim profit margins reflect the expenses cloud computing providers must
absorb to build big data centers and hire the engineers to run their software applications,
while they charge relatively modest fees to use their service. The other problem is that
they don't require their customers to pay additional money for product updates and
maintenance -- a gold mine for traditional software makers.

In the next chapter we look at how cloud providers can capture value, the various
business models that are in play or might come into being in future. We also look at how
network externalities may be a factor in this business.
Chapter 4: Cloud Computing Business Models

In this chapter we look at existing cloud computing business models. We also look at
cloud computing as a platform and try to apply the levers of platform leadership to a few
well-known companies in this space.

Business Models

According to Doug Hauger, General Manager of Marketing and Business Strategy at

Microsoft, we are in the very early stages of business models using this technology. One
major factor favoring the cloud platform is that end users seem to be fine with the idea of
keeping their personal data in datacenters managed by others. Most, if not all, of us have
a Yahoo, MSN, GMail, or some other public email account. A lot of us use the internet
for banking, bill payment, and shopping and readily provide our credit card number,
social security number, PINs, to these web sites in return for convenience. The
convenience comes in the form of availability and ubiquity of services, ease of use, etc.
On the other hand enterprise users are forced to either be within their firewall or connect
from outside via VPN (Virtual Private Network) to access their office email, calendar etc.
Individuals are beginning to ask for the same level of ubiquity for their enterprise data
and applications as they have for their personal accounts. This may push more enterprises
for cloud based solutions. Companies like Google are actively courting enterprises by
explaining the aforementioned virtues of the cloud.

Cloud computing can be looked as a stack of services as shown in the figure below -
each layer of the stack offers services that are built on layers below it. There are buyers
for these services at different levels of the stack depending on the needs of the business,
the size, and competency of those running the business. Each layer in the stack has the
potential to offer a different business model. There are some core services that span
across the various stacks. For instance, companies that supply metering and billing
services are needed for all types of cloud services. So are management services.

Fig: Stack of cloud Services.

Infrastructure as a Service (IaaS) providers such as Amazon, Google (with their App
Engine), Sun Microsystems, EMC, offer their service in ways that differ in terms of what
the user gets and how she is able to access the services. Amazon for instance makes a
machine instance available to developers. They have to choose the operating system
(Linux, Solaris, or Windows). Each such machine is referred to as Amazon Machine
Instance (AMI). Once the user gets an AMI she is free to install her own applications.
Amazon also makes storage, database, content management and other core services
required for someone to build a commercial web application for conducting business
online.

Google on the other hand makes core functionalities of its systems available only via a
Python based software interface. Google does not provide access to an OS image or any
standard database. They provide APIs to access services like mail, a proprietary
datastore, images, users etc. Users are expected to write applications using these APIs.
That way Google is much more restricted than Amazon's offerings. SalesForce.com's
"Force.com" is similar to Google in that they allow the developer to build an application
using their workflow engine and visual authoring tools and leave the underlying
implementation of the application to the cloud.

Each of the two offerings above sustains a different group of users. Small and medium
businesses that do not want to spend money on data centers and yet want to control how
they configure their systems would go to cloud providers such as Amazon. There are
others who need to create a web site that needs services like mail, user management, a
data store and don't really care about the inner working of these services. For them the
Google App Engine would suffice.

According to Lydia Leong, an analyst with Gartner, virtualized hardware alone is not
enough for companies to realize the true value of the cloud. A managed, automated way
to architect, deploy, scale and maintain software which supports applications in
virtualized environments is needed, along with infrastructure software that enables
organizations to actually use the cloud to its fullest potential. The industry has created
numerous "secondary cloud" providers who rely on the services of infrastructure
providers like Amazon. They provide value by supplying management software that
makes it easy for applications to provision cloud infrastructure dynamically. Without
these services customers would have to write additional software to manage the
deployment of their applications to the cloud - software to add or remove compute
resources depending on the load on the application. Most of them provide an application
stack and/or a graphical user interfaces that developers can use to easily assemble a
virtual datacenter to run their application on a cloud. These secondary cloud service
companies tend to charge a premium price from their customers. For instance, if
Amazon's EC2 charges user 30 cents per "raw" CPU hour of usage, Elastra charges 50
cents for a CPU hour of a virtual machine that comes preinstalled with an application
server.

Quite a few of these secondary cloud companies are funded by companies that already
have large datacenters and are using the cloud to expand their market. One of them is
GoGrid, a provider of cloud servers supporting a variety of Linux and Windows
operating systems with preinstalled software, such as IIS, Apache, PHP, Microsoft SQL
Server, and MySQL. It is a subsidiary of ServePath a leading provider of server and
network hosting services. Similarly Mosso, another IaaS provider is backed by
Rackspace, an industry leader that currently manages over 30,000 servers for customers
all over the world.

Some of these secondary companies are also funded by companies like Amazon. Elastra,
for instance, is partially funded by Amazon. The company helps deploy and manage
applications on Amazon's EC2 and S3 platforms. RightScale is another company that
creates web solutions that run on Amazon's cloud. This has two benefits for Amazon - it
provides them with additional customers who come to Amazon by virtue of using
Elastra's management software, it also furthers the cause of Amazon's APIs if more
businesses like Elastra build their management software to work on the Amazon's cloud.
Next we look at how the Cloud Computing business model is different from the
traditional software model. We also look at how the various cloud providers differentiate
themselves in the marketplace.

High volume, low subscription fee:

One of the values of this technology discussed in the previous chapter was that, yes,
service providers can spread their costs over a large number of users. However, the
payments from each user are likely to be small, so this is a very different and potentially
more risky business model than the traditional product business. A common cliche used
by industry analysts is that there are likely to be a million customers paying $100 per
month for using cloud services like Amazon EC2 and S3 as opposed to a 100 million
Dollar customers. There is some truth in this cliche. This is one reason cloud providers
are trying to reach out to enterprise customers where they can sign contracts for a specific
period and be assured of revenue during that period. Enterprise users are also likely to
buy support services which is something that has high margins.

Revenue Source:

The revenue model for cloud providers is quite different from traditional software
providers in that the former do not charge an upfront cost for selling the software. There
are no additional license renewal fees either. While customers like Genentech are buying
enterprise licenses for cloud services like Google Apps, the bulk of the consumers of
these services are non-paying individuals, educational institutions, charitable
organizations. Companies like Google, Yahoo, Microsoft (with its Live Platform) makes
money by displaying relevant advertisement on the same browser window where the mail
or the calendar is being viewed. Google for instance, uses technology to parse the text of
the emails received and displays advertisements related to the contents of the email. In
fact, Google, which does not sell any traditional software, had its revenues from
advertisements cross 20 billion Dollars over the last 4 quarters (as reported by Associated
Press). In summary, advertisement is a major source of revenue for SaaS providers who
offer their services for free. In most cases these same services are also offered at a price
with no advertisement to small and medium businesses. The same holds true for
enterprises users who usually have to buy licenses for each user in the enterprise.

Other software services like maps, analytics, etc. are also available to enterprises for a
price. The price covers customer support and technical help. This is a small segment of
the revenue model. Companies like Kayak.com, Trulia.com, Tripadvisor.com all use
Google Map's enterprise addition for a fee 6.

Companies like Salesforce.com provide commercial applications like CRM as a service.

They are now the largest cloud computing service for businesses, with a market value of
$4 billion, about 52,000 customers and revenue totaling $1 billion in its past four fiscal
quarters. Other companies in this space are CogHead.com and Bungee Labs.

Pricing Models:

The way cloud services are priced differs from one provider to another. As mentioned
above a large chunk of the software delivered as a service, GMail, Google Calendar,
Search, Maps, Yahoo Mail, etc., are available to retail customers for free. There are
corporate customers who pay for the services. In fact Google has been actively trying to
push these services under a single bundle, Google Apps, among corporate customers for
$50 per license per year.

In the case of Platform as a Service (PaaS) Slaesforce.com, the pricing for unlimited
application usage of those applications is $50 per user per month. However, the company
has faced price pressures from smaller startups like Bungeelabs.com, coghead.com, etc.
and has been forced to come up with alternative revenue models. According Ariel
Kelman, Salesforce.com senior director of platform product marketing, as quoted on
ZDnet, there are a wide variety of apps on the Force.com platform, some that have high

16 http://www.google.com/enterprise/maps/success_stories.html
usage, such as recruiting, and others such as vacation request, employee surveys and
expense reports, are widely distributed but less frequently used. To address that segment
more equitably, salesforce.com is charging $5.00 per login, with a maximum of 5 logins
per user per month. They also set the metered pricing at a promotional price of 99 cents
per login throughout 2008.

The predominant pricing structure for Infrastructure as a Service (IaaS) seems to be a

"pay as you go" model. Users of the services only pay for the resources and services they
use. Amazon EC2 charges for resource usage as shown in the screenshots below. The
user pays for the machine on a per CPU-hour basis.

Pricing
Pay only for what you use. There is no minimum fee. Estimate your monthly bill using AWS Simple Monthly Calculator.

Instances

Standard Instances Unm/UNIX Windows

Small (Default) $0.10 per hour $0.125 per hour
Large $0.40 per hour $0.50 per hour
Extra Large $0.80 per hour $1.00 per hour

High CPU Instances inux/UNIx windows

Medium $0.20 per hour $0.30 per hour
Extra Large $0.80 per hour $1.20 per hour

Pricing is per instance-hour consumed for each instance type. Partial instance-hours consumed are billed as full hours.
Data Transfer
Internet Data Transfer
The pricing below is based on data transferred "in" and "out* of Amazon EC2.

Dat Transfer in
All Data Transfer $0.10 per GB

Data Tmsfer Out

First 10 TB per Month $0.17 per GB
Next 40 TB per Month $0.13 per GB
Next 100TS per Month $0.11 per GB
Over 150 TB per Month $0.10 per GB

Data transferred between two Amazon Web Services within the same region (i.e. between Amazon EC2 US and
another AWS service in the US, or between Amazon EC2 Europe and another AWS service in Europe) is free of charge
(i.e., $0.00 per GB). Data transferred between AWS services in different regions will be charged as Internet Data
Tansfer on both sides of the transfer.

Usage for other Amazon Web Services is billed separately from Amazon EC2.

Amazon Elastic Block Store

Amazon EBS Volumes

a $0.10 per GB-month of provisioned storage
n $0.10 per 1 million /110requests

Amazon EBS Snapshots to Amazon S3 (priced the same as Amazon S3)

" $0.15 per GB-month of data stored
- $0.01 per 1,000 PUT requests (when saving a snapshot)
* $0.01 per 10,000 GET requests (when loading a snapshot)

Elastic IP Addresses
No cost for Elastic IP addresses while In use
* $0.01 per non-attached Elastic IP address per complete hour

* $0.00 per Elastic IP address remap - first 100 remaps / month

* $0.10 per Elastic IP address remap - additional remap / month over 100

(Amazon EC2 i sold by Amazon Web Servkles LLC)

Figure: Amazon EC2 Pricing Plan. Price as on 01/10/2009.

(Source: http://aws.amazon.com)
As can be seen from the figures above the developer pays for CPU, data transfer, and IP
address mappings beyond a certain limit. There is a similar pricing model for storage
services from Amazon. Secondary providers of cloud services, companies like Elastra,
charge a slightly higher because of the additional value they provide. They charge higher
for the virtual computer instances that come with a predefined application stack such as
an application server.
The table below provides an example of how pricing for our solution works:

Oalbautara 1 -50 S.50

Appioeson Serwvr I .50 5.50 1 50

InstWcen

SO LardLa 5 .50 S.13 1 5,13

Bancew Inalance

AWS Sinail Server 5.10

LUnIUNIX
Instance"
$40
.SO
AW$ Lage Server
LnsmKVUNX
Ingance"

EkVC Biock Sime 100 5.10 $10.00 $.014

Dae Tranfer MC00 so so$

$44

Elast pvdes up to ITS of Data Tansfer per onth at NO CHARGE

Fig: Elastra's price list. (Source: www.elastra.com)

Leveraging existing business:

Cloud computing, being at the very early phase of a normal business life cycle, comes
with its own set of risks. The cost of setting up a data center is quite high. What happens
if you don't get enough customers after you have invested heavily in a data center? As a
result most companies who have entered this business either had an existing data center
business or had a cloud centric internal data center for their own business as in the case of
Amazon. The company had built a vast swath of infrastructure technologies to support
their booming internet retailing business. Initially this platform was made available to
other vendors who wanted to sell their wares from Amazon.com. With time, sections of
the platform were opened up as services like EC2, S3 etc. Salesforce.com seems to be the
first exception. Huge investments in data centers might be one reason why their profit
margin is so much lower compared to companies like Oracle and Microsoft. Salesforce's
income of $37 million during the year 2008 when it had total revenue of $1 billion
translates into a measly $3.70 profit on every $100 in sales. That looks anemic alongside
Oracle's net margin of about $24.80 for every $100 in sales in the comparable period.

Google already had a number of very large data centers to provide search results. These
were replicated for ads, mail, calendar services etc. With the App Engine, Google opened
up its datacenter to the outside world using a subset of the APIs that their internal
services used. Companies like Sun, HP, EMC, already run data centers for their
customers. Second tier cloud service providers are either backed by large server hosting
companies or are funded by mainstream cloud providers like Amazon as described earlier
in this chapter. As the technology gathers traction we see more of the same trend.
Companies like Verizon, AT&T who have a significant amount of data center assets are
getting into the cloud business. It is unlikely that someone not connected with
datacenters will enter the cloud business anytime soon.
Platform Leadership in Cloud Computing

In this section we look at Cloud Computing from a platform point of view and consider
the factors that could contribute to platform leadership in this technology. Platform
leadership involves driving industry wide innovation for an evolving system of separately
developed pieces of technology. We have seen numerous instances of platform leadership
in the recent past - Microsoft with its Windows Operating System, Google with its search
platform are some good examples. According to Professor Michael Cusumano and
Annabelle Gawer, there are four levers of platform leadership 17: scope, product
technology, relationship with complementors, and internal organization. We analyze a
few of the major players in this arena against these 4 levers.

Before that we need to understand if Cloud Computing has the potential to be a platform?
Most of the industry leaders we spoke with believe it is a platform or will be one much
like software operating systems (like Windows, for example.) Microsoft's Hauger
believes the technology is at the initial stages of becoming a platform.

We believe it can be - if we separate the services available on the cloud from the platform
on which these services reside. Products and services that bring together groups of users
in two-sided networks are platforms. They provide infrastructure and rules that facilitate
the two groups transactions and can take many forms. Cloud computing can be
envisioned as a two sided market that connects infrastructure providers to application
developers. So far though, many of these services are being made available by the
platform provider themselves. Amazon, for instance, is providing the compute, storage,
database services, content management services, etc. However, for some of these services
it has partnered with the vendors of the software. It has partnered with Sun to offer EC2
images of the Solaris Operating System. It also offers Windows images in partnership
with Microsoft. There are a slew of value added service providers who provide various
EC2 images with custom application stacks - these include MySQL database servers,

17 The Elements of Platform Leadership, Michael A Cusumano, Annabelle Gawer. Sloan

Management Review, Spring 2002.
application servers, content management systems, mail servers etc. These are turning out
to be quite popular among the developer community looking for specific application
stacks. We take these as the beginnings of a platform in this space.

Providers of platforms for two-sided networks are able to draw revenue from both sides.
In the case of cloud platform providers, that would be infrastructure providers like those
who add their own application stack to Amazon EC2 virtual machines and developers
who use these stacks. In this case it makes sense for Amazon to subsidize the supply side
of the platform as these providers of value added services bring more developers to
Amazon. This in turn gets more providers to deploy specialized stacks on Amazon EC2.

Since cloud computing is at its infancy and different players entered the market using
different technologies or domain expertise the current market has multiple platforms.
That is not to say we will not have platform providers who would seek market dominance
at some point. However, these days the IT community is much more aware of the tricks
of the trade in the business of software. Customers for instance understand the value of
interoperability, and the need for open APIs. Also, platform providers are trying to woo
customers who are already invested in other cloud based products. For instance,
businesses that use the CRM service from SalesForce.com are being wooed by Amazon
and Google. They are doing so by supporting Salesforce.com's APIs on their platform.
This shows that the various platform providers are sharing the platform, at least for now.

According to Bert Armijo, VP of Marketing at 3tera.com, vendors often use APIs as a

lock-in mechanism, of-course, and the API wars for cloud are just getting started.
However, this strategy can backfire. First, customers see this coming. In cloud, for
instance, large users are already starting to push for standard APIs. Unfortunately
standards take time and agreement amongst the vendors. Second, introducing new APIs
implies that on-boarding software requires writing new code. He sites the example of
EC2 which, for the first two years of its availability, did not allow databases to run
reliably because its storage system was "unique."
He also believes that new APIs really aren't needed in all cases. Cloud systems that
support general purpose computing (as opposed to SaaS and PaaS) use virtualization to
allow running existing operating systems that already offer abstractions for memory,
storage and network. Although difficult, it is possible to leverage these abstractions.
Unfortunately, many cloud vendors chose to "augment" them with their own - hence the
APIs. 3tera's AppLogic has no API, so there's absolutely no code writing required to
make use of the system thereby reducing the barrier of entry.

Multi-homing costs can be quite high in the case of cloud platforms. For instance, just as
it would not make sense to write some of your applications using .Net, some in C++, and
some in Java, it would also be quite expensive to write, maintain, and deploy applications
on multiple cloud platforms. This is because the APIs are different from one provider to
another. A number of secondary cloud service providers are trying to provide solutions
that bring down multi-homing costs. It remains to be seen how the landscape changes.
The holy grail of Cloud Computing seems to be a desire among consumers to be able to
choose whichever cloud they want, whenever they want. Much like the electric grid
where your utility companies could choose to buy power from the cheapest source. Cloud
users would like to run their computation and store their data in the cheapest possible
location and possibly move them around as the price offer by service provides changes.

Even after a cloud provider does a good job of addressing pricing and winner-take-all
challenges he may face the danger of envelopment where one provider's services may be
"enveloped" by an adjacent provider that enters the market. Cloud platforms have
overlapping user bases. Salesforce.com's APIs are now available as part of Google's App
Engine API. Developers can deploy their applications on the App Engine and use
SalesForce.com's CRM features. There is nothing to prevent Google from making similar
features available to these customers in future. The same holds for the Azure platform
from Microsoft. Initially they may open up their platform to allow developers to use other
service providers but in course of time these could be closed down as Microsoft rolls out
its own services. Another envelopment threat is posed by second tier cloud providers -
the likes of Elastra who are providing an abstraction layer over Amazon's EC2 and
building their own internal cloud solutions. If these secondary platforms can gain enough
traction they have the capability to render the core IaaS providers as mere commodity
suppliers.

Levers of Platform Leadership:

As mentioned above the four levers of platform leadership are scope, product technology,
relationship with complementors, and internal organization. Let us briefly look at what
these mean in the context of cloud computing.

Scope:

Scope refers to the range of services and the level of services that a cloud provider might
offer. We defined cloud computing to include SaaS, PaaS, and IaaS. Companies could
operate in one, two or all three of these spaces. For instance, Google operates in PaaS by
virtue of its App Engine offering. It also has numerous SaaS applications: mail, calendar,
map, analytics, etc. Amazon on the other hand is perceived as operating only in the IaaS
space. Even within each type of service the scope can vary. For instance one provider
may provide a single service while another may provide a host of services.
SalesForce.com, for instance, specializes in CRM as a service and has a platform around
that. For everything else it depends on other cloud providers. The scope at which an
organization operates has a great impact on whether it has the potential to become a
platform leader.

Technology:

Cloud providers can establish platform leadership on the basis of their technology. One
aspect of this technology is the user interface. User interface to access the cloud can act
as a barrier to entry. Cloud services involve writing software slightly differently to take
advantage of the cloud. One has to make use of application programming interfaces
(APIs) to access the cloud. Usually SOAP or REST based APIs are supplied. Providing a
language binding (in PHP, Java, Ruby, Javascript, etc.) for the REST or SOAP based
APIs goes a long way in reducing the barrier to adoption of a cloud platform. In many
cases the providers themselves provide all the language bindings. In other cases, the
community creates language bindings and makes them available to other developers.
Google's App Engine seems to be a lone hold out in that it only allows usage of the
Python programming language. According to Microsoft's Doug Hauger, the Azure cloud
platform would allow developers to build their applications in most major languages
including PHP and Ruby. Coming from Microsoft this seems like a major change in
mindset. Yet another advantage can be having a Integrated Development Environment
(IDE) that automatically supports a particular cloud platform. Aptana, a software
company that offers an IDE for web2.0 development extended its features by allowing
users to store their source code in the cloud during development and also deploy their
completed application on their newly created cloud platform. Amazon, for instance, has
an EC2 plugin for Eclipse IDE.

Providing a set of services that application developers use to write traditional software is
another way of luring them to the cloud. For instance, most enterprise applications need
access to a database, identity services, messaging services, content management
especially if it's a web application. Providing these services gives a big boost to the
platform.

According to Lew Tucker, the value of a vendor's cloud depends both on applications
and platforms. He believes that platforms occur at multiple levels. One platform could
host another software platform. For example RightScale provides a management
framework that application developers can then use. Smugmug.com may use that
management framework for their application in the cloud. The old Independent Software
Vendors are slowly going to become "Independent Service Vendors" or "Internet Service
Vendors", says Tucker. One could have a "Ruby and Rails" service provider who
provides a platform for hosting Ruby applications on the cloud. This calls for a service
provider interface that other providers of services as well as platforms could leverage.
Technology could also be used to induce to a given cloud platform thereby increasing the
switching cost to other cloud vendors. One way of doing this is by storing data in a non
standard format. Amazon is already doing it by providing services like the simpleDB,
which stores data in a proprietary format. On the other hand if data is stored in the cloud
using MySQL or some other relational database it can easily be imported to another cloud
that also supports relational databases. The same holds true for the APIs. Each provider
has its own flavor of APIs to access its cloud. Hence, an application written to live in one
cloud cannot easily be ported to another cloud.

Relationshipwith complementors:

This is a vital part of establishing platform leadership. It would be foolish for cloud
providers to build all the services themselves. They should encourage the traditional ISVs
to enable their software on the cloud provider's platform. The good things about cloud
computing is that multiple implementations of the same service can exist on the cloud.
For instance, the same cloud can provide both MySQL and Postgres database services. In
fact, the more the merrier, as it gives application developers greater choice.

Internal organization:

The right internal structure for success in the cloud technology seems to be having
internal organizations that themselves believe in the cloud model. Specifically, if all
internal systems are deployed as cloud services it definitely helps the company develop
better understanding of the technology issues first hand. This also helps generate enough
knowledge base that can help gain developer mindshare outside the company.

In the next few paragraphs we analyze a few cloud providers against the four levers
described above.

Amazon Web Services.

Amazon is by far the most successful provider of Infrastructure as a Service in the world
today. The following types of businesses are leveraging Amazon's cloud:

Application Hosting
Backup and Storage
Content Delivery
E-Commerce
High Performance Computing
Media Hosting
On-Demand Workforce
Search Engines
Web Hosting

Clearly their customers cover a wide variety of businesses. However, their scope is
limited to providing infrastructure services only. They do not have application services
like the ones Google or Salesforce.com does. Amazon has a large number of developers
too. Currently there are over 29 billion objects stored in Amazon S3 vs. 22 billion at the
end of Q2 2008. That is sequential growth of 32 percent. On Oct. 1 2008, the service
peaked at over 70,000 requests per second to store, retrieve, or delete an object. Over
400,000 developers have registered to use Amazon Web Service18

From a technology point of view they have created a Firefox plugin, called ElasticFox,
allows developers to view details of their resource usage and other account information.
Their APIs have various language bindings supplied by members of a vibrant developer
community.

In order to get greater developer mindshare Amazon organizes various cloud camps
across the US and EU. They have also increased the scope of their offerings by making
EC2 and S3 services locally available to European customers.

18 Source: http://gigaom.com/2008/10/09/amazon-cuts-prices-on-s3/
In terms of service offerings, Amazon, being a web retailer, has made available some of
the key services that application developers need to build web services. These include
storage, compute service, messaging, content management, billing. They could add other
services like identity as part of the service.

Amazon has funded a few companies that build value added products leveraging their
cloud platform. They have also provided services like Amazon Devpay that allows
external vendors to provide value added services on EC2 and S3 without having to spend
any time managing metering and billing for these services. This has fueled the
availability of various application stacks on the EC2 platform.

Organizationally, the company has been using these services internally for its business
for a while. This infrastructure was also available to its partners. This gave them the
opportunity to hone their technology for prime time. There is no conflict of interest
within the various organizations as having a great cloud infrastructure helps their
ecommerce business as well.

Amazon has the potential to be a platform leader by being an early mover in this space
and doing the right things to establish platform leadership.

Google:

Google does not sell any of its software using the traditional model. It started off by
offering search as a service and over time built various services around it - primarily in
the area of search based advertising and analytics. Over time Google offered a few
pioneering SaaS applications: Google Maps which provided a very user-friendly web
interface to view maps. Then came Google Mail and its USP was the amount of free
storage allocated to each user - 1 GB! Google Docs was introduced thereafter as
collaborative way to create and share documents. To support all of its software as a
service offerings Google built massive datacenters with proprietary technology. These
datacenters helped them drive down the cost of computation andstorage. It also gave
them the opportunity to offer this same computational power to developers outside of
Google - using the Google App Engine.

As you can see Google started off with a very small set of services but over the last 8
years increased its scope to include both SaaS and PaaS. While Google doesn't disclose
the results of its business applications division, it's relatively small, too. The Mountain
View-based company's non-advertising operations generated combined revenue of just
$540 million during the past four quarters, while Google's advertising sales totaled $20
billion. While Google still makes most of its money from advertising Sanford Bernstein
analyst Jeffrey Lindsay believes Google's applications will rake in revenue of about $1.5
billion by 2012, a small share next to the estimated $18 billion for Microsoft's desktop
office software. In terms of scope Google has a large set of applications waiting to take
off and new ones are constantly being churned out in their labs.

From a technology point of view Google has been a pioneer in making the web richer and
more user friendly. From Google Maps, to Google Web Toolkit, to the scalability and
efficiency of their data centers they have lead the way in innovation. However, Google
offers a rather restrictive view of the cloud than, say, Amazon. This may not be suitable
for those enterprises looking for their choice of operating system and application stack.
This could come in the way of establishing them as a platform leader. Google has also
come out with a browser, Chrome. The browser could be used as a vehicle to showcase
their services by making these available as part of the browser's user experience.

Google maintains a very good relationship with the developer community by being very
open with its APIs right from the time it started. Its philosophy of "do no evil" has
endeared them to the community. The community has flourished and members have
contributed by providing various plugins for popular IDEs, tools that further reduces the
barrier to entry for developers. The architecture of its data centers is though a zealously
guarded secret. One other characteristic of Google is that the company builds its own
services. It does not have the notion of an SDK for service developers as such. Some
services were added by acquisitions of smaller companies. So, while on the demand side
of the platform they have a very large developer and user community on the supply side
its all Google. This may hamper them in the long run if ISVs start offering their services
in a competitor's cloud platform.

In terms of organization the company has been built from the ground up to think of
Software as a service. The company is relatively new and, hence, is not burdened by
traditional software business models.

Salesforce.com:

Salesforce.com was founded with a motto of "No Software" to reflect the alternate
delivery model of software as a service. The company started off as a provider of CRM
software as a service. In course of time it created a platform around this by establishing
AppExchange - a repository of applications that other companies build using
Salesforce.com's CRM APIs. Users are charged a flat fee for accessing these
applications. Recently it has also come up with it PaaS offering called Force.com that
allows developers to build and deploy their own application on Salesforce.com's cloud.
The scope of the platform has definitely increased over the last few years - from allowing
other vendors to create apps using the platform to allowing ordinary developers to create
and deploy applications.

In terms of technology Salesforce.com is fairly limited (as compared to Google or

Microsoft, say) due to their focus on CRM applications only. In order to encourage more
developers to adopt their platform, Salesforce.com have also announced a $1 million
dollar developer challenge as part of their "Tour De Force" visiting 20 cities as a global
developer road show. Also, Salesforce.com does not have any pre-existing technologies
that could have given them ready access to a set of customers. Other cloud vendors like
Google and Amazon Web Services are adding support for Salesforce.com's APIs from
their platforms. This may help the company be more ubiquitous.
Salesforce.com charges their customers on both sides of the platform. Just as developers
are expected to pay a fee, ISVs writing applications for the AppExchange platform are
also expected to pay a fee. This may lead to ISVs going to some other platform if those
platforms offer better financial incentives. Two of the major players in this business,
Oracle and SAP, have not yet started offering cloud offerings on a large scale. The
ERP/CRM SaaS marketplace may change considerable once that happens.

From an organization point of view they are similar to Google in that there is no
traditional software technology that the company has to unlearn in order to develop their
cloud based platform. This definitely helps.

Microsoft:

On the face of it Microsoft seems like the one company that would benefit the least from
pursuing a SaaS strategy. Bulk of their business is based on the traditional software
model and provides a large and steady stream of revenue. Their office suite generated 18
billion Dollars of their revenue last year. The only thing going against Microsoft is that
the growth in the cloud computing market is going to be huge and most of it will come at
the expense of traditional software market. According to a quote in sys-con.com, Merrill
Lynch analysts reckon that by 2011 the volume of cloud computing market opportunity
will amount to $160BN, including $95N in business and productivity apps (e-mail,
office, CRM, etc.) and $65BN in online advertising. All of this has prompted Microsoft
to recently launch their Azure platform. According to Doug Hauger of Microsoft, they
are not late in the game - it's just that their customers did not ask for a cloud based
solution all these years but are asking now.
 Fig: The Azure platform.
(Source: www.microsoft.com)

In terms of scope most technology analysts believe Azure lies somewhere in between the
offerings of Amazon and Google. While the platform does not provide direct access to a
virtual computer instance it does provide access to a host of infrastructure services from
Microsoft: Live services, SQL services, CRM services, .NET services and Sharepoint
Services. Azure hosted applications are expected to use these services. The platform also
comes with some "finished services" built on top of these services: Windows Live, Office
Live, etc. However, we must note that not all of these have been implemented yet.
Sharepoint and Microsoft Dynamics Services have not been implemented at all. The
other services are in various stages of preview. The platform, when completely
implemented, would be in a position to handle everything that one could do with the
traditional version of Microsoft's software. Hence, the scope is pretty broad.

From a technology point of view Microsoft has a great advantage in that being a platform
leader in the Operating Systems market it already has a large user base for these
technologies. Azure would simply make them available as a service. One of the problems
that Microsoft has always had is one of interoperability with other platforms. In fact this
was the basis of their dominance for a long time. In the cloud space they do not have a
first mover advantage - there are other entrenched providers in the market. The big
question is whether Azure based services can coexist with other non Microsoft cloud
services. For instance, can an application hosted on Azure use Microsoft Live Services
along with SalesForce.com's CRM application.

In terms of relationship with complementors, Microsoft has a large population of

developers and ISVs writing to Window's OS and services. Microsoft has managed to
have a great relationship with them in the past. It remains to be seen if they can leverage
that relationship to get more ISVs to build services on the platform. The Azure platform
has an SDK for ISVs - this is in contrast to Google's App Engine. According to Hauger,
providing the right level of expectation about service levels is also an important criteria
for a successful platform - the SLA does not have say 99.999% availability, however, it
must state the expected response time of the service accurately. At this point we are not
sure what those service levels will be.

One major problem that Microsoft is likely to face is how to organize itself to deliver on
the promise of the Azure platform without affecting their core businesses - specifically
Windows and Office. As long as the major chunk of their revenue is coming from
traditional software technologies those departments will have more clout. A strategy for
both these organizations to coexist with mutual respect has to be devised at the highest
levels of management. One other problem they might face is adjusting to the competitive
landscape in the cloud computing arena. Other players, especially Google and Amazon
have established themselves in the early movers.
Chapter 5: The Road Ahead

Cloud computing has made many strides in the last few years. As mentioned in the
previous chapters the business is slated to grow further in the coming years - in fact if
predictions are true cloud business will overtake the traditional software market.
However, there are still some impediments to its progress. We look at these issues, and
what companies can do solve these.

The Hurdles

Existing software and hardware infrastructure:

One of the greatest hurdles to increased adoption of cloud computing as a viable business
model is the capital that has already been spent on acquiring hardware and software
licenses for the enterprise. This coupled with the software that has been developed to run
on this infrastructure. One cannot change to cloud computing by flipping a switch. Most
of the software would have to be rewritten to use cloud APIs. Also, there may not be a
one-to-one replacement for all on-premise software running in an enterprise. In addition
the cost of hardware has decreased considerably over time - so much so that many of the
hardware suppliers are finding it hard to stay afloat.
Additionally, enterprises have generated volumes of data that are hosted on databases like
Oracle or in ERP systems from SAP, Oracle and other companies that do not have a
cloud product of their own nor do these companies provide a way to port this data to the
cloud. Companies would have to sell their existing infrastructure, rewrite a lot of their
code to be able to take advantage of the cloud. According to Microsoft's Hauger, any
new technology rarely eclipses the install base. This is especially true for cloud
computing in the absence of a clear migration path.

Change in software development methodologies:

In order to take advantage of the cloud platform enterprises have to re-evaluate their
software development methodologies. In the new scheme of things the choice of
programming language, architecture, and functionality, is restricted by the choice of the
cloud platform. Also, the build test deploy and maintain cycle is not the same as it used to
be. This calls for changes in program and project management that can take advantage of
the new paradigm. The software industry has to adopt a cloud based software
development model. It also calls on software engineers to change the way they look at
development. This change from the ground up cannot be done instantly. Some
organizations may not be able to make that transition at all leaving them to always
depend on traditional software development practices.

Lack of Frequently Used Tools:

Along with the core software that an enterprise has for running its software, there are
other enabling software that is vital for the smooth management of software development
and operations. These include among others, project management software, bug tracking
systems, customer quality measurement indicators. These have to reside in the cloud
along with enterprise applications. A case in point was that of Northrop Grumman's
desire to use the CRM solution from Salesforce.com. Initially they could not do so
because their project management software, SEER from Galorath, was not cloud enabled.
They worked with Galorath to make the software available on Salesforce.com's platform
only after which Northrop Grummen start using their CRM service.

Identity, Privacy and Security Concerns:

Lew Tucker, CTO of Sun's cloud services believes Identity Management is a still an
issue. Each cloud has its own identity solution. There is no single identity across multiple
clouds. While there have been attempts to solve this problem there is no clear solution
yet.
Increased demand for security and compliance is considered a good reason to move to the
cloud. According to Peter Coffee of Salesforce.com, only a cloud provider may be able to
spend that kind of money or amortize the cost of providing high security and auditability
across a large number of users. However, privacy is turning out to be a major concern
among enterprises. They are still leery of storing data outside their firewalls. What if
hackers get access to the data? There might be an SLA in place for handling these
scenarios but once the data has been compromised no SLA can undo that. Most
companies need to make their enterprise and data available for audit - this becomes an
issue if the data center is no longer in-house. Cloud providers are trying to get around that
problem by showing their datacenters to customers (after making them sign non-
disclosure agreements!). They are also getting their processes certified in order that their
customers can breathe easy.

Protocol to Handle Failure:

One of the issues Lew Tucker, CTO of Cloud services at Sun, brought up was around
rules of failure. What happens when an application is connected to 5 cloud services and
one of the services goes down? What is the protocol to handle failure? How should the
other services respond? Do we see the development of a new multi phase commit
protocol or can the two phase commit for transactions be applied to the cloud? These
issues have not been answered properly yet. While companies talk about high availability
they have not yet addressed the technical issues when the system is unavailable.

Bandwidth:

As more applications move to the cloud more bandwidth will be needed to transport data
between the cloud providers and the client accessing the application. This coupled with
the fact that media companies are also making their bandwidth hogging content available
on the net. Is there enough bandwidth to handle all the data movement? This may become
a serious concern going forward especially in areas of the world where the current
internet infrastructure is lacking. In addition network providers are challenging the Net
Neutrality act as they are not making as much money from the proliferation of broadband
usage as they had hoped. Any change in laws may alter the equation adversely affecting
cloud computing. Irrespective of that there needs to be more investment in building data
centers closer to the customer. There needs to be new networking technologies that can
optimize movement of data across networks and provide a stable response time to end-
users.

Cisco Systems recently introduced Nexus 7000, a new network switch for companies
burdened with rapidly growing Internet data transfers and the increased use of
applications that draw on remote data storage, known as cloud computing. The switch is
supposed to provide a sharp increase in traffic capacity over the company's current
products - to 15 trillion bits of data a second. To illustrate speed, the switch could
transfer all 90,000 Netflix movies in 38.4 seconds or send a two-megapixel digital image
to every human being on earth in 28 minutes. Better cloud services to serve the types of
files that are hosted by web 2.0 sites are also needed.

The HTTP Protocol as a Stumbling Block:

There are many tasks that cannot be accomplished with ease due to the limitations of the
HTTP protocol's request/response nature. In order to handle all types of messaging
across the cloud as is possible within an enterprise's data center we may need a better
protocol than HTTP. This is yet to happen.

Data Portability Issues:

Countries, and regions within countries, have their own rules around where data can
reside and how that data should be protected. Since 9/11 the US has introduced the
Patriot Act, which allows the US government access to private data. As a result
government's of other countries like Canada do not want data generated in their countries
to be stored in the US. The EU has similar rules which says data generated in the EU
should be stored within the region. This puts additional constraint on cloud computing.
Enterprises cannot move to the cloud if the cloud provider does not have a data center in
the region where they need to operate.

Lack of Channel Incentives:

Companies that sell traditional hardware and/or traditional software have various sales
channels to sell their products. Each channel has its own set of incentives. For instance, if
the company employs its own sales force to sell to an enterprise the sales personnel
generally get a share of the revenue. In cloud computing many customers currently buy
services from cloud providers by simply going to their website and supplying a credit
card number. There may not be a clear relationship yet between service metering and
billing and linking that to the sales person responsible for bringing in that business. This
has to be addressed wherever it hasn't been. Also cloud computing removes some of the
middlemen involved in the sales process. The hope is that as cloud service providers try
to acquire more enterprise customers they will involve enterprise sales force and possibly
other channel partners and provide the right incentives.

It was resellers that made Microsoft great and so, following in those well-trod footsteps,
Google has started recruiting resellers to push its Google Apps to businesses of all sizes
everywhere in the world, taking Microsoft on where it lives. According to sys-con.com,
so far Google has collected 50 "pilot partners." The authorized resellers are supposed to
be able to sell, customize and support Google Apps Premier Edition, creating new
revenue opportunities for themselves and easier access to Google's cloud services. And
they'd get a 20% discount off the $50-a-user-a-year price.

The Future of Cloud Computing

All the interviewers we spoke with were upbeat about the future of cloud computing.
Lew Tucker, CTO of cloud computing at Sun, is a believer in the "big switch" concept
described by Nicholas Carr. He thinks fewer and fewer enterprises will build traditional
datacenters of their own. There will be virtual datacenters. He also sees enterprises
getting a lot more interested in cloud computing and that too much earlier in the game
than one thought they would. This is because they private clouds are going to change how
data centers are run. These would be built much more around self-service. Developers in
various departments can do their own provisioning or internal requisition. The strategic
benefit for them is that through virtualization enterprises can dynamically shift their
resources to provide improved levels of service at different times in different geographic
locations. Hi sites a futuristic example of an investment bank that deals across continents
-- it can dynamically shift its resources to offer faster response times to APAC customers
for a part of the day, EMEA customers at some other time and US customers for the
remaining time, all with the same hardware resources. A cloud within an enterprise
would provide increased computational effectiveness at lower cost.

In the near term Lew sees a shift from web 2.0 to cloud. Startups would not buy any
hardware; they will deploy their application on a cloud. He also sees issues going forward
- we have not figured out quite a few things. As people try writing mashups that draw
upon multiple services new problems might show up. What happens when one of these
services are not available? We have to solve the issues surrounding dependency on large
disparate systems.

Peter Coffee of Salesforce.com, sees financial services, medical systems moving to this
model soon. Eventually R&D and government systems will also move to this platform.
They already have a customer in Dolby Labs who have built their entire bug tracking
system around Saleforce.com. Technicians running the sound systems in cinema halls can
go to the web and file a bug by simply logging on to the system and entering the product
ID of the system.

Doug Hager of Microsoft strongly believes that most companies will have moved their
business to a cloud of some form in 10 to 50 years! The install base will require an
evolution. In the shorter term we will have hybrid solutions, where on premise
applications written on Windows, Oracle, Sun platforms, for example, will come with a
level of abstraction into cloud based services. He sees this happening on 2 to 5 years. The
current economic crisis will accelerate the use of specialized services that are available
on the cloud.

We believe in the future of cloud computing as a viable technology and, hence, a viable
business too. One major shift in the near future will be towards private data centers. This
is because of the current industry landscape. Big players are getting into this space. Sun
recently announced the acquisition of Q-Layer, a Belgium-based infrastructure
management company that has technology to automate the deployment and management
of both public and private clouds 19 . Elastra is building a cloud solution that can be
deployed in a private data center. One recent revelation that may further help the
adoption of cloud computing in private datacenters is the story of Bechtel. Its CIO was
inducted into the CIO hall of fame last year for shifting its IT services towards a more
cloud centric model. He incorporated high-bandwidth networking practices from
companies such as YouTube, the standardized server approach of Google, extreme
virtualization techniques from Amazon, and the multitenant application support strategy
of Salesforce.com, among others. This helped provide a secure, ubiquitous, simplified
and rapidly deployable access to corporate and customer information for any Bechtel user
around the globe. Ramleth calls his approach the "consumerization of the computing
environment"-an internal cloud-computing infrastructure serving up in-house
applications on demand 20 . This shows the value enterprises can derive by moving to a
cloud based platform.

Among the provider of cloud computing services we see a trend towards providing one
company's cloud solution on another company's platform. Google's support for
SalesForce.com's platform via its Python APIs is a step in that direction. This is a trend
that will continue until companies begin to realize the risk of envelopment.

AMD recently announced that it was building a specialty supercomputer to deliver

gaming through a computing cloud. This news bolsters the cloud business model -

2 )http://gigaom.com/2009/01/08/for-sun-q-layer-is-a-smart-buy/
1"
http://www.cio.com/article/453214/Cloud Computing_to_the_MaxatBechtel
taking it beyond storage and run-of-the mill computing into the realm of specialty clouds.
Since gaming may involve mobile devices these cloud services may have to be optimized
to deliver services to the mobile platform too. This may be true for all clouds in future
with the convergence of PCs and mobile devices.

If the cloud computing industry takes the shape of electric utilities as many are
predicting, they may behave like the current utility providers. They may be reluctant to
change the architecture of their systems or incorporate greener and better technologies as
that would mean a complete overhaul of their data centers. Proper regulation must be put
in place so that data centers are forced to reduce per unit energy spend over time.

With the move towards cloud and the decrease in traditional datacenters, hardware
providers will have to find innovative ways to stay in business. They have to decide
whether they want to start their own clouds or continue to sell to other cloud providers
only. Their profits might decrease over time as we achieve a saturation level in data
center business. Needless to say this is will take some time to happen.

Over the last 10 years I have attended almost a dozen all-hands by the charismatic Scott
McNealy, former CEO of Sun Microsystems. He often talked about Sun being the
company that delivers "the big freaking web tone" (BFWT). I liked the vision but always
wondered why the world did not buy into that idea. With cloud computing the answer
seems quite obvious. He was ahead of his time. The cloud is a BFWT that comes with
applications that people need. It's the applications and the API to access this web tone
that adds value. The web tone delivers search results, images, blogs, calendars, emails,
customer relationship management software.

The promise of cloud computing is one that can revolutionize computing and change the
way we go about our business. Services would become more ubiquitous - available on
our mobile phones or from our computers. It is also a path towards better data center
utilization - this is essential as the need for computation increases in the face of growing
environmental challenges.
Appendix A: Interviews
Interview format:
Most of the interviews were conducted over the phone. We asked the following questions
to each of the interviewers:

* What are the technical improvements that have enabled cloud computing as a
viable business model?
* What in your view are the economic values of cloud computing?
* If cloud computing is a platform what are the platform levers? What strategies
could a platform provider adopt to dominate in this technology?
* What are the various business models that are cropping up as a result of this
technology?
* What are the impediments to the growth of Cloud Computing?

In some cases we asked a few questions specific to the company they worked for.
Interview with Lew Tucker. Chief Technology Officer, Cloud Services,
Sun Microsystems.

Q: Where do you draw the line between what is cloud computing and what is not?
Answer: I see delivering of applications across the network as cloud computing. For
example a document management system like Google Docs delivers an application over
the network. Cloud computing is an "all encompassing" term that includes any service
that is delivered over the network.

Q: In your mind what are the technologicalbreakthroughs that have enabled cloud
computing?
Answer: Virtualization is the first thing that comes to mind. The idea that multiple
Operating Systems (OS) can be hosted on the same physical hardware has really helped
the cloud computing business. This increases server utilization. It also allows for the
growth of new kinds of appliances constituting an OS and a software stack. Before
virtualization one had to provision different machines, one for Linux, one for windows
server, one for an application server running on Ubuntu, for example. The other really
overlooked area has been the use of Application Programming Interfaces (APIs) on the
web. This has allowed web services to consume other services on the internet. Cloud
computing allows these APIs to be used by web applications. These are simple HTTP
based APIs. Saleforce.com was one of the players that ensured all their applications
always had APIs.

Q: What are the technical hurdles that still remain to be conquered?

Answer: I believe Identity Management is a still an issue. Each cloud has its own identity
solution. There is no single identity across multiple clouds. While there have been
attempts to solve this problem there is no clear solution yet.
The other issues are related to security. What is the right way to ensure security in the
cloud? Govt. regulation of where users data can reside and their desire to have control
over such data is another pain point.
Q: Do you see cloud computing as a platform?
Answer: The value of a vendor's cloud depends on applications and platforms. Note that
platforms occur at multiple levels. One platform could host another software platform.
For example RightScale provides a management framework that application developers
can then use. Smugmug may use that management framework for their application in the
cloud. The old Independent Software Vendors are slowly going to become "Independent
Service Vendors" or "Internet Service Vendors". You could have a "Ruby and Rails"
service provider who provides a platform for hosting Ruby applications on the cloud.

Q: What are your thoughts on private clouds?

Answer: Enterprises are a lot more interested and that too much earlier in the game than
we thought they would. This is because they private clouds are going to change how data
centers are run. These would be built much more around self service. Developers in
various departments can do their own provisioning or internal requisition. The strategic
benefit for them is that through virtualization enterprises can dynamically shift their
resources to provide improved levels of service at different times in different geographic
locations. Think of an investment bank that deals across continents - it can dynamically
shift its resources to offer faster response times to APAC customers for a part of the day,
EMEA customers at some other time and US customers for the remaining time, all with
the same hardware resources. A cloud within an enterprise would provide increased
computational effectiveness at lower cost.
There are barriers to the adoption of cloud technologies too. For one, this would require a
rewrite of existing applications to take the benefit of cloud APIs. It seems like existing
applications will only be able to extract the benefits of virtualization. However, any new
application can take advantage of the cloud.

Q: Do you see companies like Elastra,Hyperic, and others who arefronting cloud
services with their APIs reducing the clouds to commodities?
Answer: There will be big shifts in utility computing. However, as users start using a
cloud provider and store more and more of their data in that cloud it would be difficult
for them to move their data. It could be the case the case that data stays in the same cloud
but computation moves from one to the other depending on who is offering the cheapest
rates at a given point in time.

Q: If all cloudprovidersstartproviding the samefeatures how you customers

differentiate these providers?
Answer: If you look at the US telecom market there are only a handful of providers. Yest
cutomers find a way to choose between one or the others. I think there will be factors like
customer service, SLAs, connectivity to other clouds on which cloud providers will
compete. The container for sure would become a commodity.

Q: Where do you see Cloud Computinggoing in the near term? In the long term?
Answer: In the near term I see a shift from web 2.0 to cloud. Startups would not buy any
hardware, they will deploy their application on a cloud. In the long term I am a big
believer in the "big switch" concept as described by Nicholas Carr. Fewer and fewer
companies will build their own datacenters. There will be virtual datacenters.
I also see issues going forward - we have not figured out quite a few things. As people
try writing mashups that draw upon multiple services new problems might show up.
What happens when one of these services are not available? We have to solve the issues
surrounding dependency on large disparate systems.
Interview with Peter Coffee, Director of Platform Research at
Salesforce.com.

Q: What in your mind are the technical improvements over the last decade that has
enabled this new business?

First lets look at the negative aspects of today's data centers. Organizations are unhappy
with the state of things are actively looking for alternatives. The complexity of
administering computing is rising. This is partly because governance expectations have
risen dramatically, primarily due to regulations such as SOX and HIPPA. The industry
mindshare of data protection and governance has increased. The cost of implementing
and administering security has increased too. Would enterprises like to bear these costs
themselves or should it be amortized across thousands of users? Clearly it's the latter.
The industry is also revaluating where the tasks of administering security and data
protection can be better performed. If an organization has a massive data center that is
managing the data and computation resources of more than one enterprise they can spend
more money to hire the right talent and enforce the strictest policies. This is something
even large enterprises may not want to do.

The reason mentioned above is not technical. If you look at the processors of today they
are not the fastest in single threaded chips. These are multi threaded at the core. This
means that tasks don't improve linearly simply by running on a machine having these
new multi core chips. Shared data centers can continue to exploit the improvements in
chip technology that small organizations cannot hope to. The CAPEX per dollar is tilting
towards shared data centers.

Coming to the other improvements that have enabled cloud computing has been the
increase memory size, increased availability of bandwidth and the very high level of
interactivity among applications. Services on the other hand have become unproductive
and undifferentiated. The increased economy of processing and bandwidth is pushing
towards an increase of capacity.

Ever since September 2007, I have been asking enterprises to look at the rising cost of
capital. Does the cost of capital match the value derived from the capital. Enterprises
either over provision their data centers to meet anticipated demand or under provision
their datacenters and rush to meet increased demand later at the risk of losing customers.

Improved wireless connectivity has also helped. Users now expect to connect to their
applications not just from the desktop but also from virtually anywhere. That calls for an
infrastructure that can service up these applications with optimal performance from
anywhere.

Q: What according to you is the economic value of cloud computing?

Answer: According to the CIO of Bechtel who studied YouTube, Google and
Salesforce.com they found startling differences in the cost savings per unit of bandwidth
these companies get. The number of machines managed by a system administrator is far
higher in a company like Google as compared to Bechtel. According to the CIO any
enterprise that does not have a strategy to move to the cloud will end up on the losing
side. Clouds provide economic value by bringing down the cost of bandwidth to $20 per
mb/sec of b/w. The number of machines per systems administrators is very high in cloud
enables datacenters. Companies like Salesforce.com can push up to three updates to their
software each year and our customer's can accommodate that upgrade with ease as they
do not have to migrate their data or change their programs. There is a huge opportunity
cost. Cloud computing is a major tipping point.

Q: If cloud computing is a platform what are the platform levers?

Answer: For SalesForce.com the system had to scale to meet the computing needs of the
planet. To meet the needs of the business you had to have a customization story.
Salesforce.com has code that we run and metadata that customers have specific to their
application. It's the metadata that allows applications to customize. We asked the
following questions: suppose we were able to provide our customers with customization
of the logic - something that allows them to manipulate not only the descriptive stuff but
active parts of an application we could get users build the application on our platform
without having to leave the web. We also added customization of the visual element
using a technology we called Visualforce. This gives us more leverage with our
customers and acts as a platform lever. It allows users to develop their application much
more rapidly. The cost of project management reduced too.

Q: What are the various business models that are cropping up as a result of this
technology?

The main business model is software as a service. For us its CRM and software that
handles customer service and support. Collaboration services is also another area. We
also have applications that tread both Salesforce.com and other applications that run
outside of Salseforce.com.

We also allow businesses to go to market on their own by building their application or

infrastructure on SalesForce.com. We do the billing for them. Their customer does not
have anything to do with Salesforce.com. Its like these applications become an OEM on
our platform.

In Force.com our customers use their own URL and render their application completely
out of our own infrastructure.

Q: How can cloudproviders create value and how can they capture value?

Cloud computing creates value by allowing developers to innovate quickly. The cost of
adoption is low. There is scope for rapid improvement with cost reductions. I believe the
one can get superior Ease of Use in the cloud. One can get superior adaptability for a
global workforce. Multiple end user language cab be supported. The platform in general
provides more stability, security, governability and auditability. The platform is also very
predictable in my opinion. We just signed an enterprise license with Dell - they have
decided to build all their IT apps on Salesforce.com.

Q: I also want to get your thoughts on some specific questions around Platform as a
Service (PaaS).

a) What are the strategies one might adopt to beat companies like SAP and Oracle.

The major issue is industry's perception that customizations in a platform like

Salesforce.com is hard. We need to drive home the point that user experience delivery is
very quick.

b) Where do you see the PaaS model going from here?

I see the platform as a service being more widely adopted. System integrators can
implement projects faster and customize better to meet the requirements of their users in
a shorter period of time.

c) What in your opinion are the stumbling blocks to users moving to a PaaS model?

As I mentioned above the perception in the community is a major stumbling block. The
industry thinks that having the enterprise data center in their basement gives them more
control. This is not true. There are some latency issues that need to be resolved but this is
not a major issue. There is of course the issue of government regulations around data
protection and storage. We are working with various governmental authorities to simplify
these rules.

d) Besides CRM what other types of software do you see shifting a PaaS model?
I see financial services, medical systems moving to this model soon. Eventually R&D
and government systems will also move to this platform. We already have a customer in
Dolby Labs who have built their entire bug tracking system around Saleforce.com.
Technicians running the sound systems in cinema halls can go to the web and file a bug
by simply logging on to the system and entering the product ID of the system.
Interview with Bert Armijo, VP of sales, Marketing and Product
Management at 3Tera.

Q: What are the technical improvements over the last decade that has enabled this
new business.?

Answer: There are a few key developments that had to happen before cloud computing
could be realized - virtualization, plentiful bandwidth, powerful microprocessors,
distributed computing, and the web all play a role. The web provided a standard user
interface experience and broke the tight coupling between the user and the application.
We're now completely comfortable that applications run remotely. More powerful x86
processors and distributed computing ended our dependence on big-iron systems like
mainframes and SMP and their proprietary operating systems. Virtualization created a
separation between software and the hardware that runs it. And lastly, bandwidth is now
available to move large workloads around.

Q: What in your opinion are the economic values of cloud computing?

IT operations have gotten unbelievably complex over the past decade or so. One analyst
report published a couple years ago showed that while labor comprised 20% of IT
spending in the 90's it's now 80%. Just getting a new server in an enterprise data center
can take months and require numerous approvals. If you start a small web company
setting up servers, networking and storage in a traditional colocation facility will cost
more than your developers salaries. Yet, servers are all the same. Cloud computing cuts
through this morass, making resources available on-demand, over the internet.

This has several affects. For small companies, it puts world class data center resources
within their budgets for the first time. For enterprises it puts a price tag on server
operations; a benchmark that most found they simply couldn't come close to matching.
Plus, it puts resources mere minutes away, which makes many projects viable that
couldn't be implemented before because of the delay in deployment.

Q: If cloud computing is a platform what are the platform levers? What I meant to ask
was what would it take for cloud providers to retain their customers. Each cloud has its
own set ofAPIs to connecting to the cloud as well as proprietarydataformats like
Amazon's simpleDB. Seems to me one can move computation aroundfrom one cloud to
another.However, data seems to be hard to move around.How do you look at this aspect
of the cloud business?

Answer: Keep in mind that barriers work both ways ...

Vendors often use APIs as a lock-in mechanism, of-course, and the API wars for cloud
are just getting started. However, this strategy can backfire. First, customers see this
coming. In cloud, for instance, large users are already starting to push for standard APIs.
Unfortunately standards take time and agreement amongst the vendors. In the first two
gatherings to discuss standards Amazon, Google and Microsoft didn't bother to send a
participant. Second, introducing new APIs implies that on-boarding software requires
writing new code. For the first two years EC2 was available, it was nearly impossible to
run a database reliably because the storage system was "unique."

Moreover, new APIs really aren't needed. Cloud systems supporting general purpose
computing (as opposed to SaaS and PaaS) use virtualization to allow running existing
operating systems which of course already offer abstractions for memory, storage and
network. Although difficult, it is possible to leverage these abstractions. Unfortunately,
many cloud vendors chose to "augment" them with their own - hence the APIs.

3tera's AppLogic on the other hand has no API, so there's absolutely no code writing
required to make use of the system. We've made this choice specifically to allow as much
existing software to be used in the cloud as possible. This required a tremendous amount
of extra engineering, but as a result you can literally install software like Apache straight
from the rpm off their site. Even the Wordpress 5 minute install works unaltered.

So how do you keep your customers? Innovate. Produce a good product. Create a better
business model. Support your customers.

As revolutionary as AppLogic is, our approach to selling and our pricing model are as
much a strategic advantage as the product itself. But that's a subject for a different email.

Q: What are the various business models that are croppingup as a result of this
technology?

The one most people are familiar with are service providers who set up the data centers,
servers, etc and make the resources available to subscribers.

3tera is almost alone at the moment selling a platform for cloud computing but others will
likely follow. Our customers include service providers as well as enterprises.

There are a couple companies focusing on software packaging for cloud computing and
we're also seeing a few system integrators starting to offer cloud readiness services for
enterprises.

Q: How can cloud providers create value and how can the capture value?

Done right, cloud computing creates immense value for users.

Service providers create value by offering "resources on demand." Their value creation is
in efficient reliable operations. 3tera licenses our AppLogic system to service providers
or enterprises who want to create clouds. Our value to service providers is in enabling
them to enter the space and generate revenue from new customers. For enterprises our