Documente Academic
Documente Profesional
Documente Cultură
VOL 7 NO 7
2009
SETLabs Briefings
Advisory Board
Gaurav Rastogi
Associate Vice President,
Head - Learning Services
Kochikar V P PhD
Associate Vice President,
Education & Research Unit
Raj Joshi
Managing Director,
Infosys Consulting Inc.
Ranganath M
Vice President &
Chief Risk Officer
Subu Goparaju
Vice President & Head,
Software Engineering &
Technology Labs
SETLabs Briefings
VOL 7 NO 7
2009
Perspective: Revenue and Customer Growth for ISVs using Cloud Offerings 61
By Ajit Mhaiskar and Bhavin Raichura
Virtualization is already in vogue and ISVs have to adapt themselves to suit the changing
model to be able to attract and retain customers. The authors bring up some pertinent
issues related to ISVs’ revenue growth and the evolving cloud space.
Index 93
“Cloud is designed to be available ever ywhere, all the
time. By using redundancy and geo-replication, cloud
is so designed that services be available even during
hardware failures including full data center failures.”
Amitabh Srivastava
Senior Vice President
Microsoft Technologies
Raghavan Subramanian
AVP & Head – Cloud Computing CoE
Infosys Technologies Limited
SETLabs Briefings
VOL 7 NO 7
2009
3
The horizontals describe the actors in CLOUD COMPUTING: THE RISE TO
each of these verticals. PROMINENCE
Cloud computing offers a new, better and
■ Enablers: OEMs, independent software economical way of delivering services and
vendors, etc. all the stakeholders will have to embrace the
dramatic changes to exploit opportunities to
■ Delivery Agents: Value added avoid becoming irrelevant.
resellers, communication, media and Following are some of the key trends/
entertainment service providers, last changes that we expect to see:
mile access suppliers, etc.
Movement across Domains and Competencies:
■ Consumer: Enterprise and retail users. Technology providers are realising that in order
to leverage economies of scale, it is essential
The complex relationship can be best to have competencies across hardware and
represented if we consider that the consumer software verticals. Nokia for example, operating
utilizes services and delivery agents act in the handset and telecom infrastructure
as intermediaries adding value to the raw space, has boldly ventured into the social
capabilities offered by the enablers [Fig. 1]. networking space with Ovi. The acquisition
It is important to realize that the roles for of US internetworking (an application service
the actors are not limited and the entities may provider offering managed hosted applications
span across domains and roles. With renewed like PeopleSoft and SAP) in 2006 has enabled
interest in decoupling technology from services AT&T to offer enterprise-class cloud services
and the falling price of bandwidth, the cloud labelled Synaptic Hosting.
model of operating will drive visible growth Acquisitions, although the preferred
and collaboration horizontally, vertically as route, need not be the only route, as with all
well as across quadrants. disruptive and emerging technologies, cloud
computing will drive the creation of alliances
spanning hardware, software and services, for
instance, HP - Intel - Yahoo [3]. NetSuite, a
leading vendor of on-demand enterprise services
has announced its partnership with BT to deliver
io n
rat
Su
Utility
Cost
Delivery
an
Agents
rt
ion
4
from InformationWeek says “It’s very hard increasingly rich as they become more and
to add a significant new capability to existing more service focused [11]. For the mass, cloud
products [7].” As we move up the triangle in computing is all about ubiquitous access to
Figure 1, hardware and software enablers and content. Thus, telecom and media companies are
delivery agents will collaborate and co-innovate reinventing themselves to become on-demand
to differentiate services. solution providers, aiming to provide complete
John Foley describes start-ups as having experience, as opposed to just providing services
the innate capacity to drive innovation and fill in silos. Citrix president Mark Templeton said,
niches, while pushing down costs and driving “Optimization of the user experience will happen
up performance [8]. There will be a delicate in the data centre, at the edge of the network
shift in the balance of power from traditional and in internet cloud, allowing IT to deliver any
enablers to start-ups that deliver ideas, the application to any user with the best performance,
likes of 3Tera, Appirio, Coghead and Kaavo. security and cost savings possible [12].” This is
This is reflected by the fact that venture capital evident from the increased investments in server
interest in the ‘cloud’ is high. While VC funding and storage consolidation. Cisco’s ‘medianet’
has dried up in many areas in this recession, suite, for one, has been built around advanced
some 25 startups in the cloud space garnered collaboration and entertainment, targeting both
more than $150 million in VC funding in the the business and home user.
past year [9].
Security and Legal Implications: It is
New Sales and Pricing Models: Delivery important to realise that even if data and
of cloud computing requires optimised applications are stored and accessed remotely,
infrastructure management costs and increased the responsibility of the security and integrity
operational efficiencies. This will have of data lies with the individual. Authentication
considerable impact on the way the enablers and authorization on the cloud, entangled with
realize revenues. The key trend to arise will be similar requirements offline will drive the need
the acceptance of the subscription model (opex for interoperable (across services and devices)
model) resulting in customers increasingly identity management. Single sign-on will be
transforming ‘my problem’ into ‘your problem.’ an area of investment. Ventures in this field
With the subscription model, the enablers include Microsoft Active Directory (within the
will require an upfront investment. However, enterprise) and OpenID. Also, issues around
the breakeven will be hazy depending on the privacy and the way individuals exercise
volume of subscribers. The traditional delivery control over personal data stored remotely will
mechanisms (brick and mortar, media like need to be addressed.
CDs and DVDs) are on the decline and not Hardware and software services available
transforming will mean sudden death, as is through the cloud may span geographies and
evident from Blockbuster’s partnership with cloud providers may soon subcontract their
TiVo in an attempt to ward off Netflix [10]. services. All the three actors will have to
understand the implications of having sensitive
Data/Content Driven Innovation: The data on the cloud and regulatory compliances
application delivery platforms will become viz., SOX, HIPPA, etc., governing the same.
5
ROLE ENTITIES: CLOUD TRENDS on the PDA while on the move or on high-
The following sections look at the consequences definition TVs at home. This mandates higher
of the cloud for each of the roles defined in the investment in product development but does
previous section in greater detail. not necessarily allow a longer concept-to-
market cycle.
The Enablers To support the increased demand and
Enablers provide resources that drive and adoption of cloud computing, the enablers
support the creation of solutions in terms of are aligning their resources to provide
both hardware and software that the consumer multi-tenanted architectures, virtualization
utilizes. Following are the buzz words in the technologies along with support to highly
enabler’s arena: scalable and elastic services. Virtualization
technologies span platforms, resources and
Consolidation and Integration: With the markets applications and the likes of VMware’s Mobile
changing rapidly, it is imperative for players virtualization platform are steps in that direction.
to find new opportunities. Some of the recent In fact enterprises are already reaping benefits
acquisitions highlight the clear horizontal of this. Westar Aerospace & Defence Group has
expansion across hardware and software been successful in slashing their data centre size
towards services. For instance, with its purchase by 50% and power and cooling costs by 30% with
of Sun, Oracle has become a true cloud player a server virtualization solution from Dell [14].
with services now ranging from operating
systems, programming/development platforms, Environmental Sustainability and Data Centres:
ERP, CRM and other support utilities, giving Environmental awareness will further drive
Oracle an edge over its competitors and allowing enterprises towards cloud computing as it
it to offer the entire gamut of computing services allows considerable reduction in energy costs.
required by any enterprise. Gartner estimates that over the next five years,
Examples of integration within the most enterprise data centres will spend as
domain include Adobe acquiring Virtual much on energy (power and cooling) as they
Ubiquity - developer of online word processor; do on hardware infrastructure [15]. To quote
Google acquiring FeedBurner - leader in RSS VMware, “Gartner estimates that 1.2 million
services; and AT&T acquiring Ingenio - live workloads run in VMware virtual machines,
search and commerce application provider, to which represents an aggregate power savings
name a few [13]. of about 8.5 billion kWh—more electricity than
is consumed annually in all of New England for
Ubiquity and Virtualization: The fact that the heating, ventilation and cooling [16].” Cloud
consumer would demand seamless access to enabling technologies like virtualization and
content, impacts both the enablers as well as server consolidation can help enterprises reduce
the delivery agents (providers in the software energy costs by as much as 80%.
vertical, a little more than anyone else). The Data centre consolidation will be driven
challenge being, developing applications that by cost, space and energy savings. HP, for one,
are ‘portable’ and offering seamless content is replacing 85 data centres with just six located
delivery – whether on the office laptop or in America. According to IDC, America alone
6
has more than 7000 data centres and predicts important for the delivery agents to weigh
that the number of servers will grow to 15.8 pros and cons before investing in the platforms.
million by 2010. In driving the cloud data In the retail space Microsoft and Google can
centres, Linux complemented by open source emerge as dominant players due to the inertia
solutions will be at the forefront. IDC expects keeping consumers tied to its suite of products.
Linux spending to boom by 21% in 2009 [17]. Supporting them will be hardware players
Cloud computing is also driving the (a near monopoly of Intel) and virtualization
usage of netbooks or laptops that are enhanced providers like Citrix and VMware. The situation
for mobility, compromised on computing is complicated in the enterprise space, driven
capacity with a reduced storage capacity. by leaders like Amazon, Oracle, IBM and
Therefore, there will be an increased demand Google. Cross platform compatibility and ease
for transfer processing and storage in data of migration demanded by the consumer will
centers. IDC reported that netbooks accounted require the delivery agents to understand long
for 30% of all laptop sales in Europe during the term strategies.
fourth quarter of 2008, with 3.6 million netbooks
sold [18]. Death of the System Integrators: System
integrators, as we know them today, will have
Marginalization of Fringe Players: Desktop to take a second look at their model of operation.
based utilities and tools like MS Office and With the rising popularity of subscription
Norton antivirus will see a reduction in their based applications like Siebel On-Demand and
installed user base and will ultimately be SalesForce.com, the demand for customised on-
marginalized, as the same services will be premise will decrease, taking away with it the
available online. The traditional fringe players biggest market of the SIs. In the long term, IT
will have to re-invent themselves to align services providers will have to increase efforts
with the new modes of delivery, warranted to provide end-to-end management of the IT
by the cloud. Adobe is already providing an estate (or whatever little would be left of it)
online version of its graphics editing program or work along with the product companies to
called Photoshop. Appistry is one of the offer technical support to their customers. Once
more innovative companies and has recently cloud computing technology reaches the critical
launched the CloudIQ platform, offering mass, there will be an increased demand from
enterprises the capability to port nearly any enterprises to migrate data, applications and
enterprise application to the cloud [19]. content to the cloud. In the short term, service
providers need to ready their arsenal to deliver
The Delivery Agents consulting services across technology and
Delivery agents are value added resellers of the human resource domain.
capabilities offered by the enablers. Following
are the key changes that we foresee in this Last Mile Connectivity: When push comes to
domain: shove, availability will triumph over utility.
Internet service providers (ISPs) and last
Collaboration, Partner Driven Work Environments: mile access supplier will have to ramp up
Industry alliances are being forged and it is their offerings rapidly to meet the increasing
7
requirements of the bandwidth hungry content delivery through the cloud – a clear shift
content and applications, with fibre being the in perspectives [20].
predominant technology for last mile access.
The Consumers
New Pricing and Delivery Models: Sales channels Consumers are the demand side of the cloud
will also have to evolve to provide ubiquitous equation and following are the trends for them:
delivery models and the revenues are going to
be long-tailed as the sales model will shift to a Convergence, On-Demand: The retail customer
subscription based service, which will imply will now, more than ever, come to expect on-
that customer retention and loyalty becomes demand everything - be it multimedia content,
all the more important. So all players will have applications, gaming or storage. AMD’s new
to reinvent, be it the telecom operators who campaign ‘The Future is Fusion’ is again
are shifting focus to value added services or reflective of the changing times. For the retail
the internet media houses that have to come user, it is all about bringing together convergent
up with variants of their web pages that can multimedia solutions on any screen supported
be accessed from mobile devices offering a with advanced graphics capabilities; for the
consistent user experience, along with richer enterprise user it is delivering enhanced server
interactive applications to keep the customers and powerful virtualization capabilities [21].
hooked on.
Collaboration and Social Networking: Cloud
Piracy: With the onset of the cloud, the users based platforms like Facebook and Twitter
will no longer be required to download or will become destinations for collaboration,
install applications in the traditional sense. e-commerce and marketing. Enterprises are
In the online world, controlled access implies already planning to listen to the voice of the
that piracy will become increasingly difficult, if customer using such tools.
not impossible. Case in point being the online Collaboration and virtual workspace
documentation services offered by Zoho, since solutions will see increased investments. A
there is no application that has to be installed key player in this space is WebEx, acquired
at the users’ end, there is no chance of having by Cisco in 2007 for $3.2 billion – again an
a pirated version of the application. example of a hardware player moving to the
Likewise with online gaming, the software cloud domain. Another promising
problem of pirated copies of the games being technology is IBM’s Bluehouse, based on Lotus
spread around, resulting in millions of dollars Notes. This enables employees among business
worth of revenue loss can be curbed. OnLive partners or within the same organization to
is one of the pioneers in this field and has share documents and contacts, collaborate on
signed contracts with major video game content joint project activities, host online meetings and
providers like Warner Brothers, Electronic build social-networking communities.
Arts and Epic Games. What is interesting is
that Nvidia, a provider of high end graphics Back to Core Competencies: The cloud enables
processors and cards, primarily in the desktop businesses to focus on their core competency
segment, has welcomed the initiative of game and cloudsource the IT estate enabling the
8
consumers to transfer risk. ‘My problem’ now REFERENCES
becomes A look at an IDC study makes it clear 1. Frank Gens, What User Want from IT:
that businesses want the cloud because of the Speed, Relevance, Information and
cost benefit [22]. Innovation, IDC exchange, March 2008.
Available at http://blogs.idc.com/
Decentralization of Management: The traditional ie/?p=141
view of management and governance of IT 2. Gartner Highlights 27 Technologies
resources through standards and frameworks in the 2008 Hype Cycle for Emerging
like ITIL, Sarbanes Oxley, HIPPA, etc., will Technologies, Gartner, 2009
change. As much as the technological impacts, 3. h t t p : / / w w w . h p . c o m / h p i n f o /
the challenges for enterprises will also be to newsroom/press/2008/080729xa.html
manage employee expectations working in a 4. http://www.netsuite.com/portal/
decentralised and distributed manner. Many press/releases/nlpr04-22-08b.shtml
legacy IT system integrations will break and 5. http://www.dmtf.org/about/cloud-
enterprises need to clearly understand and incubator
estimate the risks of losing visibility and control 6. h t t p : / / b l o g s . z d n e t . c o m /
over critical data. Hinchcliffe/?p=303
7. http://www.informationweek.com/
CONCLUSION news/software/hosted/showArticle.
Cloud computing promises different things to jhtml?articleID=210602537
different players in the IT ecosystem. It offers 8. http://www.informationweek.com/
a radical way of collaborating, delivering news/software/hosted/showArticle.
applications and content. More importantly it is jhtml?articleID=210602537
here to stay. So it is easy to see why the enablers 9. http://www.ciozone.com/index.php/
are paving the way for massive adoption of Cloud-Computing/Unraveling-the-
the cloud and why are the delivery agents Cloud-Ecosystem.html
leveraging their positions to catch the cloud 10. http://www.informationweek.com/
demand. As for the enterprise, it reduces the cloud-computing/article/showArticle.
TCO of the IT infrastructure while increasing jhtml?articleID=216300432
agility. 11. h t t p : / / c o m m u n i t y . z d n e t . c o . u k /
It is important to realise that the complete blog/0,1000000567,10008269o-
shift to the cloud is not imminent, but enterprises 2000458459b,00.htm
will be better off with a long term vision for 12. h t t p : / / c o m m u n i t y . z d n e t . c o . u k /
technology, people, information, legality and blog/0,1000000567,10008269o-
security to leverage capabilities offered by 2000458459b,00.htm
cloud computing. The delivery agents, more 13. h t t p : / / s t a r t u p . p a r t n e r u p .
than any other players, need to reassess their com/2008/01/02/2007-acquisitions-
role in enabling and delivering cloud computing web-internet-technology/
to consumer for lack of innovation and not 14. http://whitepapers.techrepublic.com.
keeping pace with the growth will result in com/abstract.aspx?docid=360865
marginalization. 15. Rakesh Kumar, Eight Critical Forces
9
Shape Enterprise Data Center Strategies. 19. h t t p : / / g i g a o m . c o m / 2 0 0 9 / 0 3 / 0 9 /
Available on www.gartner.com appistry-opens-the-cloud-to-almost-
16. http://www.vmware.com/solutions/ all-apps/
consolidation/green/ 20. http://www.edge-online.com/news/
17. h t t p : / / w w w . e c o n o m i s t . c o m / nvidia-onlive-a-net-positive-us
b usiness/displaySt ory .cfm?st ory_ 21. http://news.softpedia.com/news/
id=11413148 AMD-Launches-039-The-Future-is-
18. N e t b o o k T r e n d s a n d S o l i d - S t a t e Fusion-039-Campaign-93711.shtml.
Technology Forecast, Consumer 22. Frank Gens, IT Cloud Services User
Behavior Report. Available at https:// Survey, pt.3: What Users Want from
mr.pricegrabber.com/Netbook_ Cloud Services Providers, October
Trends_and_SolidState_Technology_ 2008. Available at http://blogs.idc.
January_2009_CBR.pdf com/ie/?p=213
10
SETLabs Briefings
VOL 7 NO 7
2009
11
Business Business Business
Provision Solution 1 Solution 2 Solution n
Business Resolve Required
Requirements Plan Infrastructure
Service
Service Procurement
Request Service Management Cloud
Service Infrastructure
Portal Provisioning
Assign Workflow
Consumer
Requested
Service
Platform Templates
Cloud Management
Repository
and Billing
12
infrastructure that can be requested on control of the IT organizations makes
demand. it dangerous for some mission critical
applications.
■ Pay-as-Use: Organizations can leverage
the pay-as-use model of cloud computing ■ Vendor Lock-in: Cloud computing
to ensure optimum utilization of services offered by different vendors
available resources. are not governed by any standards as
of today. Depending on the vendor, the
PUBLIC CLOUDS AND CHALLENGES applications have to undergo changes to
Public clouds like Amazon AWS, Microsoft adapt to the service.
Azure, Google AppEngine offer infrastructure
and platforms as services over the internet. In ■ Leveraging Existing Investment: Most
public clouds, resources and costs are shared large organizations that have already
by users who use them over the internet on pay invested in their own data centers would
per use model. see a need to leverage those investments
This model appeals especially to startups as an important criterion in adopting
and small organizations that have not invested cloud computing.
in hardware resources and are looking for ways
to avoid the large capex involved in procuring ■ Corporate Governance and Auditing:
infrastructure upfront. Even though there are Performing governance and auditing
several benefits like cost savings, faster time to activities with the corporate data
market, etc., from this model, there are a few abstracted in the public cloud poses
challenges listed below that are preventing wide challenges, that are yet to be addressed.
scale adoption of public clouds.
■ Maturity of the Solutions: Some of
■ Security: The biggest roadblock is the the PaaS offering like AppEngine offer
potential security issues due to multi- limited capabilities like only a subset of
tenant nature of public clouds. There JDO API.
are security and privacy concerns with
sharing same physical hardware with ENTERPRISE PRIVATE CLOUDS
unknown parties that need to addressed. In order to overcome these challenges,
organizations are looking at enterprise
■ Reliability and Performance: private cloud offerings. Enterprise private
Performance and availability of the cloud solutions help organizations leverage
applications are important criteria the existing IT environment and create a
defining the success of an enterprise’s cloud computing platform in the private
business. However, the fact that internal network. This model overcomes
organizations lose control over IT several challenges faced in public cloud
environment and important success adoption. Enterprise private clouds are
metrics like performance and reliability, seen as a natural progression of initiatives
and are dependent on factors outside the like virtualization already taken up by
13
several organizations. Enterprise private Automation
cloud solutions add capabilities like self- The private cloud solution should have certain
service, automation and charge back over the traits -
virtualized infrastructure.
Figure 3 provides the recommended ■ A provisioning engine that automates
logical architecture for an enterprise private the provisioning of the infrastructure
cloud. ■ Workflow driven with built-in approval
mechanisms enabling governance
Self Service ■ Enable user management and integration
The private cloud solution should have a with enterprise authentication and
self service portal that enables users request authorization mechanisms
infrastructure and platforms as a service. ■ Enable enforcing enterprise policies on
It should contain a service catalog that lists resource allocation through a rules engine
the categories and the services available, the ■ Enable capturing the common
associated SLAs and costs. deployment patterns using templates.
The service portal should enable
reserving as well as requesting the services on Self-service and automation helps reduce
demand. the time-to-market so that users can request
Virtualization Layer
NAS SAN
Physical Infrastructure
14
for infrastructure as a service and can get it holds true. Not just the infrastructure, even
provisioned on demand. the internal clouds are to be built and managed
by the IT team. Moreover, as the underlying
Management and Monitoring infrastructure is limited, it is likely to be less
The private cloud solution should also have scalable as compared to the immensely robust
an integrated monitoring and management and scalable infrastructure of cloud providers.
platform that should have the following The model also does not benefit from the lower
components — upfront capital costs and less management
overheads that are otherwise possible with
Monitoring and Management: Track various public clouds.
metrics at the software and infrastructure level So, the recommended approach is to
adopt a hybrid one where both public and
Metering & Chargeback: Track the usage of the private clouds are used for different categories of
various services and allow to charge back applications. With this approach, organizations
mechanisms to be plugged in can reap the benefits of both public and private
cloud models. This approach allows enterprises
SLA Management: Enable, define and monitor to adopt the public clouds partially, deploying
SLAs for the services only those services that are suitable for public
clouds. The private cloud helps apply the cloud
Patch Management: Enable patches to be rolled computing model internally as well. Thus the
out to the various software components used hybrid approach brings together the best in both
worlds of public and private clouds.
Reports: Generate reports on usage, SLA As technology matures, there will be
adherence, etc. better options for creating such an enterprise
cloud. There are already solutions available
Incident Management: Generate alerts when there that provide abstractions over infrastructure
are issues and provide ticketing mechanism to available internally through virtualization
track and resolve incidents. software like vmware ESXi, Xen, HyperV
and public clouds like AWS. Also there are
Virtualization VPN solutions available that can help create a
The private cloud solution should have secure network spanning infrastructure across
virtualization layer that virtualize the key enterprise data centers and public clouds.
infrastructure components including compute, There are still challenges to be addressed like
storage and network. latency, automated routing and load balancing,
end-to-end SLA management, etc., before such
ENTERPRISE CLOUD: HYBRID APPROACH solutions become enterprise ready.
Private clouds help overcome some of the
challenges associated with public clouds but TYPICAL USE CASES FOR PRIVATE
they are not as cost effective as public clouds CLOUDS
since the traditional model of owning, i.e., Scenarios where there are only intermittent
buying and managing the infrastructure, still usages of infrastructure are ideal for cloud
15
computing. Also, scenarios that involve operating system but also the software stack,
sensitive data and processes, or mission thus enabling creation of virtual appliances that
critical applications are better suited for can be provisioned on-demand.
enterprise private clouds. Some of the typical
use cases where enterprise private clouds can Cloud Burst: With a computing stack that
be leveraged include - provides abstraction over the underlying cloud
infrastructure and enables applications and data
Development and Test Platforms as Services: There to reside together on both private and public
are studies that indicate that around 30% of clouds, when there is sudden spike in usage
the infrastructure at large enterprise is used and the in-house private cloud environment
for development and testing. These resources is not able to support the requests, additional
are not always utilized as development and infrastructure can be provisioned from a public
testing are activities that happen occasionally. cloud without affecting the service quality.
These resources can be provisioned through an
enterprise private cloud so that the resources High Performance Grid: Enterprise private clouds
can be shared and utilized better and also the can also be used to create grid environments so
time to provision can be reduced. that the infrastructure that would otherwise
have got dedicated only for specialized grid
Public Cloud Emulation Environments: Private applications can be utilized better.
clouds can be used to emulate a public cloud
environment and can be used as a development TYPICAL USE CASES FOR PUBLIC CLOUDS
and test platform while developing the Some of the initial services that can be moved
applications to be deployed on the particular into public cloud are those that are not business
public cloud. The design, architecture and or mission critical or do not deal with the
the actual code can be validated using the sensitive data. Some of the typical use cases
private cloud environment. Further, the where public clouds can be leveraged include.
same environment can also be used to test
the developed applications for functionality BPOs: Business productivity online (BPO) suite
as well as validations before it moves to the applications are one of the first applications of
production at a public cloud. An example of public clouds in enterprises. There are several
this is usage of the open source Eucalyptus vendor offerings like exchange online and Google
framework to emulate some of the Amazon Apps that offer messaging and collaboration
AWS functionality. It can be used to create software as services on subscription model
the development and test environment that that can reduce the overheads associated with
emulates AWS EC2 and S3 environments. maintaining such application on premise.
Virtual Appliances: Private clouds can be used Data Backup and Archival: Cloud storage
to create virtual appliances that leverage is cheaper and offer storage on demand. So,
commodity hardware to create specialized enterprises are looking at public cloud storage
devices like load-balancers, storage devices, etc. solutions for their data back up and archival
A machine image is created with not only the needs.
16
Cloud Application Layer
CRM SaaS Financial Service Health Care Community
Application Application Service Application Portal Cloud Tools
Admin Tools
Cloud Platform Layer
Cloud Application Frameworks
Social Service Development
Multi-tenant Batch Frame Analytics
Commerce Composition Tools
Web Framework Work Framework
Framework Framework
17
applications developed using the cloud Startups You Should Know, Information
platform services. Week, September 2008. Available at
http://www.informationweek.com/
CONCLUSION news/software/hosted/showArticle.
In the current economic climate where the jhtml?articleID=210602537
expectations of efficiencies and cost savings 3. Private Cloud Computing for Enterprises:
are growing from IT organizations, enterprise Meet the Demands of High Utilization
private clouds provide a good opportunity to and Rapid Change. Available at http://
get started with cloud computing and reap www.cisco.com/en/US/solutions/
the associated benefits of agility, cost savings collateral/ns340/ns517/ns224/ns836/
and on-demand services while meeting the ns976/white_paper_c11-543729.html
stringent enterprise security, performance and 4. Daniel Nurmi, Rich Wolski, Chris
reliability requirements. Grzegorczyk Graziano Obertelli, Sunil
Soman, Lamia Youseff and Dmitrii
REFERENCES Zagorodnov, The Eucalyptus Open-
1. James Staten, Deliver Cloud Benefits source Cloud-computing System,
Inside Your Walls, Forrester, April University of California, Santa
2009. Available at http://www. Barbara. Available at http://open.
forrester.com/Research/Document/ eucalyptus.com/documents/nurmi_
Excerpt/0,7211,54035,00.html et_al-eucalyptus_open_source_cloud_
2. J o h n F o l e y , 2 0 C l o u d C o m p u t i n g computing_system-cca_2008.pdf.
18
SETLabs Briefings
VOL 7 NO 7
2009
19
Stora behind an open and standardized cloud
n APIs Security ge ng
utatio Brokeri
Com
p
Billing
ePayment
interface. Thus a key driver of the unified
Transactions
Provisioning
Management Lifecycle
Management
cloud interface (UCI) is to create an API
of VMs Regulatory Compliance
Replication
Deployment on
about other APIs.
Network unicati
Platform
QoS Comm
Logging ice Service Registries /
ic Serv
Monitoring Dynam reements
Level A
g Modeling
Ontologies ■ It is a singular abstraction/programmatic
ccess
Accounting Job Scheduling Migration of Vms
n t
Data A point of contact that encompasses the
e Dis
anagem tributed Programming
Load M Cloud Services (SOA) Query Processin
g entire infrastructure stack as well as
Policies Notification
Workflow mic Sc
aling Pr Business
ocess Modelin emerging cloud centric technologies
Autono g IaaS
20
layer that is agnostic to any cloud API, platform all allocated resources and running VMs. A
or infrastructure. The architecture comprises component on the left side of Figure 2 is used
of layers and components with a use case to depict this. [6].
described at the UCI project requirement page
[5]. The architecture abstracts the usage of any Approach 2: Enterprise Cloud Orchestration
cloud API and unifies them in one layer. This is Platform /Orchestration layer
done with the help of semantic web and OWL According to IDC, in virtually every industry,
which has a pool of resources semantically thousands of companies are trying to simplify
understood and described. This enables the the speed and adoption of their products and
user to use these resources irrespective of services by transforming them into cloud
whether these resources are being allocated services. We see that the race to the cloud is
from provider Amazon EC2 or Enomaly accelerating [7]. The scenario that is unfolding
platform, etc. Having a unified interface with is that there will not be just one cloud but
common definitions of these resources helps numerous types -- private clouds and public
to do operations like allocation, de-allocation, ones. These will further get divided into general-
provisioning of virtual machines or managing purpose and specialized ones. Similar to the
them through the UCI layer using the agent way that internet is a network of networks,
component. Assuming that the interface to InterCloud means a federation of all kinds
UCI is provided to the user via a web browser of clouds. All these clouds will be full of
or UCI cloud client, the UCI should provide applications and services. It will not be possible
a kind of a dashboard that shows the state of to use these without some type of orchestration.
External Cloud
XML RDF
Query Engine VMS Provision Layer
21
The initiatives of some of the early adopters Eli Lilly and the various cloud services
towards Cloud Orchestration are discussed it subscribes to. This layer should be
below. provided by another vendor and not
Eli Lilly itself and should comprise
■ Vendors like Cordys advocate the of various algorithms that determine
need for a layer in the cloud that the best cloud service for a particular
provides assembly and orchestration job based on factors like highest
for enterprises, which helps to deliver performance, lowest cost or other
useful business advantages [9, 10]. requirement. This approach will help
Cordys delivers an enterprise cloud Eli Lilly and other users to write to a
orchestration platform that helps single API rather than many and help to
enterprises to quickly adopt new ways optimize service usage. Eli Lilly also sees
of running their business and reaching the potential of using cloud computing
their customers. for external collaboration. It is already
doing some of this, but foresees that
■ Rightscale is another vendor that going forward, the cloud will become
provides an orchestration layer/ a point of integration between Eli Lilly
cloud management platform. A single and outside researchers. They have work
management platform is provided to going on at present that starts to fit into
conveniently manage multiple clouds this collaborative scheme. This gives an
that facilitates businesses to migrate example of how standardization needs
deployments [11]. It helps businesses to are driven both by vendors as well as
manage and scale cloud deployments as end users.
well as facilitate application migration
and management across multiple clouds. ■ CSC has recently announced cloud
Similarly organizations like Suntec are orchestration services for cloud services
looking at building an orchestration integration. This provides clients with
layer for billing infrastructure. features like service level management,
data transparency, remote monitoring,
■ Eli Lilly, a pharmaceuticals company auditing and reporting [13]. These
uses Amazon web services and other services also provide automated
cloud services to provide high- arrangement, management, federation,
performance computing to hundreds coordination, security and operation
of its scientists based on need. In future, of public, private and hybrid cloud
it foresees the possibility of using cloud computing environments, supporting
services from many different vendors industry-specific compliance, etc.
and wants to avoid a scenario where
Eli Lilly has to configure and manage Figure 3 illustrates how a client can
each of those separately [12]. Eli Lilly consume the services offered by more than
describes the need for an intermediate one cloud service provider (CSP) via an
orchestration layer that is in-between orchestration layer.
22
■ Note that the client uses only one single
Cloud 1 Cloud 2 Cloud 3 API offered by the orchestration layer
5. Execute 7. Execute
Service 3 3. Register and thus is insulated from the different
Service 1
Service 3
6. Execute 2. Register APIs offered by different CSPs.
Service 2 Service 2
23
service, the platform support required by to use brokers/adapters for interoperability.
the service could also be a limiting factor. New users however will be able to natively
use the standard API. There will also be
■ Others: Apart from the above, vendors developing orchestration layers to
there could be other challenges like build business processes/workflows using the
security, regulatory compliance, data cloud services provided by different vendors.
transparency, etc. With some of the major vendors like Microsoft
and Amazon rejecting the CCIF agenda and
IMPORTANT INTEROPERABILITY pursuing their own interoperability agenda,
FACTORS this makes standardization and consensus
This section discusses the emerging scenario more difficult and could lead to multiple
and other important interoperability factors standards. This could lead to a scenario in the
from different viewpoints. long run where multiple standards co-exist
We see that there are multiple initiatives and customers using brokers/adapters for
by stakeholders from industry, academia and interoperability for using services from multiple
users. This does help the problem or parts of the cloud service providers.
problem being addressed by multiple standard It is also important to look at standards
bodies/forums/consortiums in parallel and also required from the perspective of different
provide diverse view points. But it is important industry verticals. For example, HIPAA
for the standard bodies, vendors and users to sit compliance could be important for healthcare
together, discuss and arrive at a consensus on services, SOX compliance could be important
the standards and APIs in different areas and for financial services, etc. This requires active
share information. This is all the more essential participation from different vendors and users
due to the duplication and overlaps among the from these verticals in standard bodies. It will
various groups involved. The flip side of the also be good if different vertical specific groups
story is that this could lead to the possibility of are setup in order to focus and discuss the
several standards emerging and possible lack vertical specific requirements and come up with
of consensus. It is important for the standard standards that are vertical specific.
bodies/forums/consortiums to have balanced Another challenge is that since there are
representation of interests in order to avoid bias many models of cloud computing (SaaS, PaaS,
towards certain stakeholders’ agenda. IaaS), standards are required for particular
Though initiatives like OGF’s OCCI models and not just one set. There is a need
are trying to come up with standards in a to prioritize and concentrate on core set of
quick timeframe, it takes time for standards standards to start with and then expand to
to mature and for reference implementations other areas. It is important to note that over
to become available. Till then the users will specification inhibits innovation. Patents and
use APIs/platforms from cloud computing intellectual property could be a hurdle for
vendors, whichever they feel is most suitable standardization process. Unlike Sun’s open
for their requirements. When standards cloud platform APIs, it will be interesting to
emerge and these vendors want to use the see if other vendors give their cloud APIs and
services of other vendors, then they will need protocols to the community.
24
When applications are migrated from the way towards realizing the true potential/
one cloud to another, apart from functionality, benefits of cloud computing.
it is also important to ensure that non-functional
requirements (NFRs) are satisfied as well in REFERENCES
the new migrated environment. This requires 1. Cloud API Propagation and the Race to
standards for defining and exchanging meta Zero (Cloud Interoperability), January
information regarding the application between 2009. Available at http://www.
the cloud service providers to check for elasticvapor.com/2009/01/cloud-API-
compliance of NFRs before actual migration of propagation-and-race-to-zero.html
the application via VM migration. The scenario 2. The Cloud Computing Interoperability
could be complex considering the fact that there Forum. Available at http://www.
could be several NFRs pertaining to security, cloudforum.org/
availability, reliability, performance, scalability, 3. Unified Cloud Interface Project (UCI).
etc., that requires compliance. Available at http://groups.google.com/
group/unifiedcloud?hl=en
CONCLUSION 4. UCI Architecture. Available at http://
Interoperability and standardization have code.google.com/p/unifiedcloud/
huge impact on the cloud adoption and usage wiki/UCI_Architecture
and thus the industry is witnessing high 5. U n i f i e d C l o u d I n t e r f a c e ( U C I )
amount of energy and thrust towards these Requirements. Available at http://code.
from different stakeholders viz., users, vendors google.com/p/unifiedcloud/wiki/
and standard bodies. Standardization will UCI_Requirements
increase and accelerate the adoption of cloud 6. CCIF’s Unified Cloud Interface Project.
computing as users will have a wider range Available at http://code.google.com/p/
of choices in cloud without vendor lock-in, unifiedcloud/
portability and ability to use the cloud services 7. F o r e c a s t f o r C o m p u t i n g : C l o u d y ,
provided by multiple vendors. This will also iStockAnalyst, December 2008.
include the ability to use an organization’s Available at http://www.istockanalyst.
own existing data center resources seamlessly. com/article/viewiStockNews/
Standardization further promises to help articleid/2904589#
towards complexly developed business 8. Unified Cloud. Available at http://code.
applications on the cloud to be interoperable google.com/p/unifiedcloud/wiki/
and ensure data and application integration UCI_Architecture
across clouds. It also provides business 9. C O R D Y S , T h e I n t e l l i g e n t C l o u d
opportunities to users to choose and use Platform. Available at http://partners.
services provided by many different cloud cordys.com/cordysportalpartners_com/
vendors based on various criteria. On the other cloud_solutions.php
hand it helps vendors to provide additional 10. C O R D Y S - E n t e r p r i s e C l o u d
higher level services like orchestration, apart Orchestration. Available at http://
from normal cloud services that are needed www.cordys.com/cordyscms_com/
by the users. Standardization will thus pave enterprise_cloud_orchestration.php
25
11. Cloudonomics: Article RightScale Adds whats_next_in_t.html;jsessionid=35MV
Amazon EC2 Europe to List of Supported LYFIRJL2GQSNDLRSKHSCJUNN2JV
Clouds. Available at http://apache.sys- N?catid=cloud-computing
con.com/node/841086 13. CSC Announces New Family Of Cloud
12. Eli Lilly On What’s Next in Cloud Services. Available at http://www.csc.
Computing. Available at http:// com/banking/press_releases/27609-
www.informationweek.com/cloud- csc_announces_new_family_of_cloud_
computing/blog/archives/2009/01/ services.
26
SETLabs Briefings
VOL 7 NO 7
2009
27
Though virtualization techniques provide as premium, gold and silver. This classification
security and isolation guarantees, virtualization is based on the amount of business generated
overheads and interference effects adversely from the respective customers. This often means
affect the QoS parameters such as response that high net worth customers are classified
time and throughput agreed upon in SLAs of into premium category. This necessitates
applications co-hosted on the same physical box that the service providers guarantee higher
[3, 4]. However, not much research has been quality of service to the customers belonging
done to identify and understand the impact of to premium segment. Additionally, the service
the virtualization overheads and interference providers must have an understanding of the
effects on these QoS parameters. resource consumption pattern of different
To benefit from cloud computing, types of requests generated from such premium
enterprises are also migrating their applications customers. For example, browsing interactions
from existing dedicated on-premise hosts to may not be as resource demanding as the
private/public cloud computing platforms. payment interaction. These factors further
This migration activity is known as on- increase the complexity in fulfilling the SLAs.
boarding. Currently, this activity is a very To make the above on-boarding activity
specialized process executed by the SMEs. This more effective and efficient, it is important
specialized process helps in identifying the to design algorithms that can translate the
system requirements of an application, based application’s QoS and SLA requirements to
on workload experienced by the application system level specifications. Further, there is
and the client’s QoS. The understanding of the a need to investigate new set of mathematical
system requirements helps to frame appropriate models that can accurately predict response
policies specific to the application and enter times and throughputs even when they are
into service level agreements with clients. This co-located with other applications on the same
in turn helps the service provider to manage physical box. These models should also address
the entire utility data-center autonomically the virtualization overheads and consider the
(i.e., autonomic data-centers) without manual interference effects. We make an attempt to
intervention. In this process, there is no define a framework to address these challenges.
comprehensive understanding of the system
requirements of the application without PERFORMANCE ENGINEERING MODELS
precisely understanding how assured QoS of As shown in Figure 1, the existing performance
one application is affected by the co-location engineering models are categorized into four
of another application on the same host. classes. These are:
Service providers not only face the risk of over-
provisioning during low demands but they Single Host Operating System Models: These
also run the risk of under-provisioning during models deal with issues related to allocation
peak loads. Also, if the interference effects are of computing resources to multiple competing
overlooked, they face the risk of overpromising applications executing on the same server.
on the QoS promised in the SLA. Typically, the CPU (if the server has only one
Further, the service providers often CPU) is apportioned amongst these applications
classify their customers into different classes such on a time sharing basis.
28
Performance Engineering
Models (QoS and SLA
Optimization)
Single Server Performance Models: These ■ the database tier for handling database
models deal with QoS/SLA issues related access requests involving lookup for
to servicing multiple client requests for a non-cached data.
web-application, for instance, an e-commerce
application hosted on a single server. In general, Performance Models for Non-virtualized
an overwhelming majority of models deal with Shared Hosting Environments: These models
issues related to web-servers. deal with resource allocation and QoS/SLA
issues for scenarios where multiple applications
Performance Models for Clusters and Farms: run on single host that is a non-virtualized
These models deal with resource allocation system.
and QoS/SLA issues for scenarios where a The above performance models attempt
particular tier of an application is replicated to address questions related to capacity
across multiple physical machines. For instance, planning and load balancing. The models help
the architecture of an e-commerce application in understanding the trade-offs of different
typically consists of three tiers: architectural choices and aid in identifying
potential bottlenecks that may degrade system
■ the front-end tier for handling static performance. These models also provide
web requests composed of simple HTTP performance estimates by predicting key
(HTTPS) requests; performance metrics such as response time and
throughput. However, the models assume that
■ the application tier for handling sufficient amount of computational resources,
complex dynamic requests involving as needed to service requests, are available
execution of java servlets, scripts and at all times. These premises do not hold
classes; and true when an application is hosted on cloud
29
platform. The very premise of a cloud platform involves building component profiles at
is to make capacity available to applications different workloads and for different user and
on demand. The performance may degrade request category. This requires subjecting the
in times when sufficient computing resources application to synthetic workloads for different
are not made available to an application categories. The component profiles are then
whenever the workload on the application suitably adjusted to reflect the overheads
increases. The increase/decrease in computing of the virtualization technologies being
resource allocations to an application should used in the cloud platforms. The resources
be proportional to the increase/decrease in allocated to different components are varied
workload experienced by the application. This and detailed performance characteristics for
in essence, requires an intricate understanding each component are collected. The profiling
of the computational resource requirements of technique is repeated for each category.
the different components and of the various Statistical techniques are then used to derive
tiers of a typical three-tier application at analytical relationship between performance
different workloads. It is interesting to note at metrics of a component as a function of
this point that the workload and the resource resource allocations (CPU, memory, Network
requirements are not just functions of the I/O, etc.). The statistical equations are suitably
number of requests but also of the nature/ modified to account for the virtualization
type of requests. It is therefore pertinent to overheads depending on the type of technology
additionally gain a fine grained understanding used and a random variable denoting the
of the resource consumption patterns of interference effect.
different types and classes of requests. SLA Once the relationship between the
aware on-boarding of applications should resource requirements and the workload
take into account the above mentioned factors. is established, it is important to predict the
Automating the SLA aware on-boarding of response time and throughput of an application
applications is a two step process that involves: in the presence of other applications on the
same host. Consider an application A that
1. Translation of high level service level is co-located with other applications B and
objectives into system level thresholds
called SLA decomposition [5, 6]
2. P r e d i c t i o n o f r e s p o n s e t i m e a n d Type of
Virtualization
throughput at different workload mixes, Technology
Used Virtualization
accounting the virtualization overheads Overhead
Effects
and interference effects. CPU/
Memory/
Computational Network
Requirements I/O at
The modeling of the SLA decomposition Workload
Analysis Workload
Requests
requires capturing the relationship between the Categorization
(Request/
high level performance goals mentioned in the Service)
30
3. Using the SLA decomposition techniques
Virtualization
Overhead to identify the resource requirements
Effects
of different components at different
Requests
Workload
Performance
Predicted
Response
workloads for each request category.
Categorization
Engineering Time and
Request
Models Throughput
Service
4. Establishing an analytical relationship
Interference between the resource requirements of
Effects
the component and the workload.
Figure 3: Performance Prediction in the Presence of
Interference Effects and Virtualization Overheads 5. Dividing the time horizon into multiple
Source: Infosys Research
epochs. During each epoch, predict the
workload and the resource requirements
of the application in the next epoch. Repeat
C on the same physical host. The resource the steps from 1 to 5 or all applications that
requirements and the response time of requests are co-located with this application.
and throughput of application A will be
impacted by the resource consumption pattern 6. Predicting the response time and
of applications B and C co-located with it. The throughput of an application when it
performance engineering model should be is co-located with other applications on
tweaked to be able to capture this interaction. the same box. This in essence helps to
Overview of the performance prediction in the account for the interference effects.
presence of virtualization overheads and the
interference effects is shown in Figure 3.
The proposed approach to SLA aware
on-boarding of application onto cloud platforms Requests
31
7. Using the results of the performance elimination of the manual work further reduces
testing in step 5 to revise the resource the cost of operation for the service providers.
requirements of the application in step 3.
REFERENCES
The overall interaction between the SLA 1. Gartner Data Center Summit 2009, 5 – 6
decomposition technique and the performance October 2009, Royal Lancaster Hotel,
models for identifying and quantifying the London, UK, europe.gartner.com/
interference effect is shown in Figure 4. The datacenter
approach presented helps in accounting for the 2. S Balasamo et al., Model-based Performance
interference effects while deciding the resource Prediction in Software Development: A
requirements of the applications. Survey, IEEE, Transactions on Software
Engineering, 2004
CONCLUSION 3. P Barham et al., Xen and the Art of
SLA aware on-boarding of application is very Virtualization, ACM SIGOPS Operation
critical for the successful adoption of cloud Systems Review, 2003
platforms. The need for new performance 4. Y Koh et al., An Analysis of Performance
modeling techniques in this context has been Interference Effects in Virtual
explained in detail. A broad approach based Environments, IEEE International
on component profiling has been proposed Symposium on Performance Analysis,
to address the challenges associated with 2007
satisfactory performance of application on 5. Y Chen et al., SLA Decomposition
cloud platforms. The proposed approach can Translating Service Level Objectives
significantly improve the understanding of to System Level Thresholds, 4th
the application characteristics once deployed International Conference on Automatic
on cloud platforms. Additionally, it helps the Computing, 2007
service providers to provide more aggressive 6. G Jungy et al., Generating Adaptation
and practical deadlines for migrating the Policies for Multi-tier Applications in
applications from the enterprise owned data Consolidated Server Environments, 5th
centers to managed service provider’s (MSP) International Conference on Autonomic
data centers. The shorter schedules and Computing, 2008.
32
SETLabs Briefings
VOL 7 NO 7
2009
33
Sometimes sensor data might not be of Weather Monitoring and Forecasting System
interest or sufficient to the consumers. The Weather monitoring and forecasting system
event of interests can be more important than typically includes the following steps –
raw sensor data. An event can be a simple or a
composite event. Events such as temperature 1. Data collection
> 50 or humidity < 80 come under simple 2. Data assimilation
events. Events like fire or explosion detection 3. Numerical weather prediction
which is a combination of two or more simple 4. Forecast presentation [5].
events come under composite event. These
events are detected by considering readings Typically each weather station is
from multiple sensors. equipped with sensors to sense the following
To summarize, integrating WSNs with parameters — wind speed/direction, relative
cloud makes it easy to share and analyze real humidity, temperature (air, water and soil),
time sensor data on-the-fly. It also gives an barometric pressure, precipitation, soil
added advantage of providing sensor data moisture, ambient light (visibility), sky cover
or sensor event as a service over the internet. and solar radiation.
The terms Sensing as a Service (SaaS) and The data collected from these sensors is
Sensor Event as a Service (SEaaS) are coined huge in size and is difficult to maintain using
to describe the process of making the sensor the traditional database approaches. After
data and event of interests available to collecting the data, assimilation process is done.
the consumers respectively over the cloud The complicated equations that govern how
infrastructure. the state of the atmosphere changes (weather
We propose, a content-based publish/ forecast) with time require supercomputers to
subscribe platform to utilize the ever expanding solve them.
sensor data for various next generation
community-centric sensing applications. Intelligent Transport Monitoring System
This platform masks and virtualizes different Traffic congestion has been increasing as a
WSNs and allows seamless integration of result of increased automobiles, urbanization,
WSNs with the conventional cloud. This will population growth and density. Congestion
shift the paradigm from the conventional reduces efficiency of transport infrastructure,
sensor networks model to SEaaS sensor and increases travel time, air pollution and fuel
networks model. In this architecture - sensor, consumption. Intelligent transport monitoring
people and software are treated as individual system provides basic management systems
objects that can be used to build community- like navigation systems, traffic signal control
centric sensing applications where people systems, automatic number plate recognition
can share and analyze real time sensor data and complex management systems like
on-the-fly. surveillance systems, systems that integrate
data from other sources such as parking lot,
APPLICATION SCENARIOS weather, etc. [6].
We consider WSNs deployed for two different Different sensors involved in this
applications. system are — inductive loops, Magneto meters,
34
CCTV, GPS, etc. These sensors are mounted on
Client 1 Client 2 Client m
vehicles, roads and buildings. Data available
...
from sensors is acquired and transmitted
for central fusion and processing. Predictive
techniques can be developed in order to allow
advanced modeling and comparison with WSN – Cloud
Computing
historical baseline data. This data can be used Platform
in a wide variety of applications. Some of
the applications are — vehicle classification,
parking guidance and information system, WSN 1 WSN 2 ... WSN N
collision avoidance systems, electronic toll
gates and automatic road enforcement.
Figure 1: System Model
In the above scenarios, both the Source: Infosys Research
applications require storage of data and
huge computational cycles. They also require
analysis and mining of data to generate events. Cloud provides on-demand
Access to this data is limited in both the cases. computational and storage resources to the
Integrating these WSN applications with the consumers. It provides access to these resources
cloud computing infrastructure will ease the through internet and comes in handy when
management of storage and computational there is a sudden requirement of resources or
resources required. It will also provide an situations where it is not easy to assess the need
added advantage of providing access to the in advance.
application data over the internet through web.
A WSN CLOUD COMPUTING PLATFORM
SYSTEM MODEL We propose content-based publish/subscribe
The system model depicted in Figure 1 consists platform, where the publishers are different
of WSNs deployed for different applications, WSNs deployed across geographical locations
cloud infrastructure and the consumers. and subscribers are those who consume the
Consumers are those who seek services from information published. Publish/subscribe
the system. WSN consists of physical wireless model gives an added advantage of publishers
sensor nodes to sense different modalities. being loosely-coupled with subscribers and
Each sensor node is programmed with the is scalable. The proposed platform consists
required application. Apart from the application of WSN virtualization manager (WSNVM),
program, sensor node also consists of operating computation and storage manager (CSM),
system components and network management subscription registry manager (SRM), service
components. On each sensor node, application provider (SP), metering and accounting
program senses the modalities and sends back to manager (MAM) and SaaS/SEaaS application
gateway (in the cloud) directly or in multi-hop interfaces.
through other nodes. Routing protocol plays a WSNVM masks the lower level details of
vital role in managing the network topology and each WSN cloud in terms of different platforms,
to accommodate the network dynamics. sensors being used, data being generated, etc.
35
It also provides a unified view of different
SaaS / SEaaS Application Interface
WSNs. CSM provides required computational
Subscription Registry Manager
cycles internally to process the data emanated
Computation Metering and
from the sensors. It also maintains the historical and Storage Service Accounting
Manager Provider Manager
sensor data of different WSNs. SRM manages
the users’ subscriptions and credentials. SP Command Interpreter Data Processor
36
system for various sensor data required. For Metering and Accounting Manager
each application, registry component stores user This module operates on a base assumption that
subscriptions, sensor data and sensor event types all the services of the WSN cloud are contracted
the application is interested in. Each application to the consumer via SLA mentioned above.
is associated with a unique application ID along Consumer uses signed web service requests to
with the service level agreement (SLA). SLA access the data.
provides basis for metering and accounting of Figure 3 depicts the UML sequence
services to be used, by covering all the attributes diagram that describes the role of MAM module
of the service customs. This agreement provides in the WSN cloud platform.
details concerning:
■ Request from the consumer to consumer
■ The type of contract — limited time, long web service
term, unlimited time, ad hoc, etc. ■ Subscription registry manager checks the
■ The time model to be used — everyday, credentials of the service request using
monday to friday, etc. the supplied signature and gives a fault
■ The amount model that defines limits to response (2.1) in case of unauthorized
the amounts of service to be provided request
■ Security — signatures and certificates for ■ Service request is sent to the MAM
encryption and authentication module, requesting it to start counting
■ Start dates and expiration dates of the (3.1) the web service access
contract. ■ The requested service is executed on the
WSN cloud
Service Provider ■ Service execution is completed
The service provider module is divided into ■ Request message is sent to the MAM
two sub components — analyzer component module, requesting it to stop counting
and disseminator component. the web service access
5
Disseminator Component: It receives the data or 6
event of interest from the analyzer component
and delivers the data through SaaS/SEaaS
Figure 3: Sequence Diagram
interface to the subscribed applications.
Source: Infosys Research
37
■ Message sent to the consumer indicating relay the data to the gateway to which they
that the service is completed, returns are connected. Once the data is available to the
the result. WSN cloud platform, it takes care of the rest,
right from processing to dissemination of the
SaaS/SEaaS Application Interface data (or event).
The interfaces built with Web 2.0 technologies Once this system is in place, the consumers
gives access to the WSN cloud platform web might be interested in the following services —
services. Consumers can consume the services
through web services that are often referred to ■ Temperature of particular location/city
as internet application programming interface periodically (e.g., one hour or one day)
(IAPI). This allows the users to access the ■ Weather forecast of particular location/
remotely hosted services over network, such city periodically (e.g., one hour or one
as internet. Consumers can build their custom day)
applications by weaving the required services ■ Notify me when the rainfall in a particular
from the WSN cloud platform. location is above some threshold (e.g.,
The services are delivered to the >2cm)
consumers in the following ways. They are: ■ Notify me if some vehicle jumps over
traffic signal
Continuous: As and when the requested data ■ Notify me if there is any fire event in
is available, it is sent to the consumers. The the forest
best example for this is fleet tracking with GPS ■ Notify me when particular bus reaches
sensor system. The vehicle position information particular bus stop.
is sent to the consumers continuously.
PLUGGING WSN INTO LEGACY CLOUD
Periodic: The data is delivered to consumers at COMPUTING PLATFORMS
regular periodic intervals. A good example is The proposed WSN cloud computing platform
to send across the temperature in the city at is a software platform that can be used on any
regular intervals of time to news agencies. of the legacy cloud computing infrastructure.
Two cases in this scenario are depicted
Event-based: The data is delivered when in Figure 4. In case 1, WSN can be integrated
some event of interest occurs. This is often with the legacy cloud infrastructure and the
information deduced from the raw data such proposed software platform co-exists with the
as detecting fire from temperature, humidity cloud management software platform such as
and light in the forest. load balancing algorithm, metering algorithms,
etc. In the second case, proposed software
Query-based: Consumers can query for a specific platform co-exists with any other applications
data from the WSN cloud platform. running on the cloud infrastructure as well as
with the cloud management software.
The two application scenarios described earlier Since most of the existing cloud
in the paper are evaluated with the proposed computing platforms (hardware and software)
WSN cloud platform. Deployed WSNs will in the market provide web services to access
38
Journal, January 2008. Available at
WSN Cloud
Software http://cloudcomputing.sys-con.com/
Platform
node/612375
Cloud
WSN 2. L M Vaquero, L Rodero-Merino, J
Case 1 Caceres and M Lindner, A Break
in the Clouds: Towards a Cloud
Definition, ACM SIGCOMM Computer
Cloud
WSN Communication Review, Vol 39 No
1, 2009. Available at http://delivery.
WSN Cloud
Software Other acm.org/10.1145/1500000/1496100/
Platform Software
Platform
p50-vaquero.pdf?key1=1496100&k
Case 2 ey2=5282660521&coll=GUIDE&dl=
Figure 4: WSN and Cloud Platform GUIDE&CFID=47965963&CFTOK
Source: Infosys Research EN=10653164
3. Weiss, Computing in the Clouds.
netWorker, Vol 11 No 4, 2007
4. F Akyildiz, W Su, Y Sankarasubramaniam
data and computing infrastructure, WSN and E Cayirci, Wireless Sensor Networks:
virtualization manager uses them to store and A Survey. Computer Networks, Vol 38
retrieve the data from the cloud. Other services No 4, 2002
like registry, metering, service provider of the 5. Weather Forecasting, Wikipedia
proposed platform together will run as an 6. Intelligent Transportation System,
application instance over the existing cloud Wikipedia
computing platform. 7. G Xu, J Ma and T Huang, A XML-
based Composite Event Approach. In
CONCLUSION Proceedings of the First international
Cloud computing has been used as an extension Workshop on interoperability of
of parallel processing. Coordinating various Heterogeneous information Systems,
computing resources to achieve bigger task Bremen, Germany, November 04 - 04,
is the key of cloud computing. In wireless 2005
sensor network computing facility is available 8. Harvard Sensor Network Testbed,
with each sensor node. Using the processed MoteLab. Available at http://motelab.
data from this intelligent sensor and using eecs.harvard.edu/
computing facility of the cloud will add another 9. www.citysense.net
value to this domain. We believe it will shift 10. M M Hassan, B Song and Eui-Nam
the operational paradigm of the collaborative Huh, A Framework of Sensor-
business process. Cloud Integration Opportunities
and Challenges. In ICUIMC ’09:
REFERENCES Proceedings of the 3rd International
1. J Geelan, Twenty-one Experts Define Conference on Ubiquitous Information
Cloud Computing, Cloud Computing Management and Communication,
39
New York, USA, January 2009, ACM. hassan.pdf?key1=1516350&key2=23126
Available at http://delivery.acm. 60521&coll=GUIDE&dl=GUIDE&CFID
org/10.1145/1520000/1516350/p618- =49153557&CFTOKEN=72600633.
40
SETLabs Briefings
VOL 7 NO 7
2009
Jitendra: Cloud computing is all new and operating system based platform
organizations do not see a value yet in terms of approach provides two benefits:
leveraging the platform. Do you see a complete
change in perspective and attitude from the ο First is cost. The OS efficiently owns
standpoint of an IT organization? How do and manages all the computing
businesses look at the concept of considering resources and also automates all
two worlds now? How do you visualize the management functions. This helps
shift in perception, as one has to look at data us drive the costs in the data center
in datacenters as well as data in cloud? down, both capex and opex.
Amitabh: Cloud holds a lot of promise for
the computing industry but the word ‘cloud’ ο Second is agility. Cloud is a complex
is not well defined or even agreed upon. environment with tens of thousands
Everyone has their own definition of the of computers operating in data centers
word ‘cloud’ and sometimes hype takes over across the globe. The OS masks the
reality. So let me start with our definition complexities by providing a rich set
of cloud and then describe our approach to of abstractions that developers can
provide the best value to our customers. We use to write their cloud applications.
define cloud as a massive geo-distributed This allows developers to focus only
computer consisting of commodity machines, on their business logic and quickly
load balancers and switches that are spread take their application to market.
across the globe. Cloud computing presents
this massive geo-distributed computer as a ■ Cloud is an extension of the on-premises
utility service. IT. Cloud and IT are not an either-or
So, our approach to cloud computing has option. Unlike some who believe that
the following key facets: everything will move to the cloud, we
■ An operating system (OS), Windows believe customers should have the choice
Azure, manages this massive geo- to decide what runs in their IT and what
distributed cloud computer. Our runs on the cloud. Many customers will
41
continue to rely on their on-premises experience. This will enable developers
IT for some class of applications. For to innovate and bring new generation of
example, some data has to be kept on- applications quickly to market.
premises due to issues like compliance,
security and privacy. Applications Jitendra: I do agree that the newer applications
that require special hardware or have exploiting convergence, bringing different
special connectivity and bandwidth channels and serviced by a common platform
requirements for performance reasons is definitely one workload that can exploit
will continue to be on-premises. At the cloud well. But would you tell us, what are
same time there are many workloads the existing workloads in an enterprise setup
that will benefit from cloud. So, rather that can be moved to cloud and be leveraged
than forcing customers to pick cloud for cost optimization? Also, what are the
or IT, our approach is to make cloud additional business advantages of moving to
a seamless extension of IT and let the the cloud?
customers decide what to run where. Amitabh: Let us look at the characteristics of
We will make it simpler by providing cloud.
value services that allow applications to One of the key features is elasticity.
communicate securely between the cloud Hardware for applications are generally set up
and on-premise IT and make it easier for to handle peak load. For example, a service may
services to federate ID. require 1000 machines to handle peak load but
on an average it only needs 30 machines. So in
■ Developer’s existing skills transfer to a traditional environment we have to provision
cloud. On Windows Azure, we use the for 1000 machines. Such applications that are
same Windows programming model, elastic in nature are suitable for the cloud
so the APIs are still Win32 and the where one can easily add capacity on demand,
same development tools still work on only pay for what is used. So, on the cloud you
cloud. Windows Azure supports all will only provision the 30 machines and then
languages, and by providing command- provision more machines as the load increases.
line interfaces and REST protocols it can In addition, the machines can be returned when
interface with all tools and interoperate peak load subsides.
with other platforms. Of course, there Another aspect is the globally distributed
are certain aspects one needs to learn facet of business, where cloud is designed to be
about cloud, but majority of skills simply geographically distributed across continents.
transfer. Cloud provides a convenient way of migrating
data seamlessly across geographically
Easy development of new generation distributed centers. Cloud benefits applications
of applications that will span across three that are global in nature.
screens (PC, phone and TV) supported by IT Enormity of scale is one of the key
and cloud. With the same Windows platform characteristics of cloud. Cloud is designed
on the three screens, IT and cloud we are by using commodity machines in a highly
striving to provide a uniform and integrated distributed environment. If there is any
42
application that requires massive scale, cloud Jitendra: Amitabh, as you said you are
is designed to handle it. building a platform that is horizontal. What in
Availability is another important feature your terms defines building vertical solutions
of cloud. Cloud is designed to be available to our customers?
everywhere, all the time. By using redundancy Amitabh: Windows Azure is a general platform
and geo-replication, cloud is so designed that that is designed to enable easy development of a
services be available even during hardware wide range of applications. Our partners, ISVs,
failures including full data center failures. Our system integrated, etc., will build the various
platform goes further to make services available solutions. Partners, ISVs, etc., with domain
even during updates OS and the application knowledge in specific areas will build the
itself. verticals on our platform. We will help lower
Many of this ultimately translates into their costs and help and provide them with a
savings in cost. For this, it is important to rich platform that lets get to the market quickly.
measure the total cost of ownership. This should
include not only the hardware costs but also Jitendra: Thanks Amitabh for your time.
management and operations cost. It has been truly wonderful talking to you
and knowing your thoughts about how our
Jitendra: What will be the guidance to the customers can benefit from the Azure platform.
customers who are looking at moving to the
cloud? What do they need to do to be prepared About the Interviewer
to move to cloud? Jitendra Pal Thethi is a Principal Architect with
Amitabh: Take a thoughtful approach. The Infosys and anchors presales activities for Infosys
first is to not panic and just rush into the solutions and IP built on disruptive technologies
cloud. Using my favorite American football in the areas of Cloud Computing, Collaboration,
analogy, I’d say that cloud is only 5 minutes Data Virtualization, Call Center Optimization and
into the first quarter. You should first try Mobility. Jitendra has more than 14 years of experience
the cloud. It is important to understand the in IT Industry as a Solution Architect and Technology
different features the cloud offers, see how you consultant.
will integrate it into your environment. Then
review the architecture of your application to About the Interviewee
see if your application is taking full advantage Amitabh holds 14 patents and has published a
of the cloud. There is temptation to quickly variety of papers. His paper on ATOM with Alan
take the application ‘as is’ to the cloud. It Eustace in PLDI 1994 received the Most Influential
is like ‘outsourcing you hardware’ but you PLDI Paper Award in June 2005. He is the author of
will not enjoy the full benefits of the cloud. OM, ATOM and SCOOPS software systems, which
Cloud provides many benefits that will lead have resulted in products for Digital Equipment and
to very substantial cost saving and give you Texas Instruments on the Alpha and PC platforms.
agility in your application development, and He led the design and development of Vulcan, a
these gains will easily make up for any initial second-generation binary transformation system, at
investment you make in taking a thoughtful Microsoft. Vulcan is the foundation of a wide variety
approach. of tools developed at PPRC.
43
Amitabh earned a Bachelor’s degree in Electrical received the 2003-2004Distinguished Alumnus Award
Engineering from the Indian Institute of Technology, from the Indian Institute of Technology, Kanpur, and
Kanpur, India and a Master’s degree in Computer was selected as the 2004 Outstanding Engineering
Science from Pennsylvania State University. He Alumnus at Pennsylvania State University.
44
SETLabs Briefings
VOL 7 NO 7
2009
45
dynamically reassigned; and services started or
Provisioning
decommissioned or re-started. So, as traditional
Password Proliferation
IDM, merely managing users and services is Maintenance of On-demand
Task User Ids
not sufficient. When a deployment or service
or machine is decommissioned, the IDM has Policies Entitlements
to be informed so that future access to it is
revoked. IDM should ideally store its details Provisioning/
Life Cycle
Management
De-provisioning Deactivation
till it becomes active. Meanwhile access to its
relevant stored data has to be monitored and
granted by the defined access level for that Customer
Delegation Service Centers/
mode as mentioned in SLA. Traditional IDM is Self Help Link
not directly amenable for cloud computing due Administrative
to these peculiarities of cloud.
Figure 1: The Identity Life cycle Management
Today’s cloud requires dynamic Source: Infosys Research
g o v e r n a n c e o f t y p i c a l I D M i s s u e s l i k e,
provisioning/de-provisioning, synchronization,
entitlement, lifecycle management, etc. stands for real time de-provisioning. Just-in-
time provisioning indicates the federation of
IDENTITY LIFECYCLE MANAGNEMENT user accounts without sharing prior data, based
Lifecycle management incorporates an on some trust model. Service Provisioning
integrated and comprehensive solution Markup Language (SPML) provides XML based
for managing the entire lifecycle of user structures for representing provisioning or
identities and their associated credentials de-provisioning requests intended for identity
and entitlements. Functionally, it is divided lifecycle management [2]. SPML can make use
into two components — the provisioning of Service Administered Markup Language
component and the administrative component. (SAML) assertions and facilitate a complete
Administrative component defines delegations trust model between senders and receivers.
rules, providing self-service components to SAML defines an XML based framework for
change personal details or make requests to exchanging security information for enabling
the users. Delegation of administrative rights SSO or identity federation regardless of the
to local group or process-in-charge is crucial for underlying architecture. OASIS Security
a volatile and dynamic cloud based scenarios. Services is currently working on developing
Decentralizing the tasks will reduce the load a SAML 2.0 profile for SPML. SAML can help
on the authenticator component and also save SPML to establish trust and quantity, a subject
time in making access control decisions. Figure against which the SPML provisioning request is
1 illustrates the various components of lifecycle targeted. This makes just-in-time provisioning
management. and real time de-provisioning possible.
Real time de-provisioning of a user
Provision and De-provisioning account has to synchronize instantaneously
In cloud, provisioning means just-in-time or with all participating service providers. Any
on-demand provisioning and de-provisioning delay in de-provisioning could lead to security
46
vulnerability. Some of the issues like — ways in CLOUD ARCHITECTURE
which de-provisioning of one user affects the Cloud architecture plays an important role
other federated identities in cloud are matters of in choosing your IDM, SaaS or the all-in-
judgment on the functionality of the application one Platform-as-a-Service (PaaS) [6]. SaaS
deployed on the cloud. requires only application access, whereas PaaS
will require system access (for accessing the
Entitlement underlying platform) as well as application
Entitlement refers to the set of attributes that access (for accessing the hosted application
specify the access rights and privileges of on the underlying platform). Both require a
an authenticated security principal. Lack of common IDM that can integrate well into the
interoperable representation of this information existing authentication mechanism. The third
poses a challenge as the information needs to be type of cloud architecture is Infrastructure-
exchanged among different cloud based service as-a-Service (IaaS), which is not mentioned
providers. In the absence of interoperable explicitly, since the IDM requirement of PaaS
format, expensive and customized syntactic and IaaS are comparable. Consider one of
translation components are needed. The the most common SaaS IDM implementation
semantic aspect still remains to be tackled. using ping identity. Ping identity works by
While some applications like SalesForce deploying the technology behind the firewall
have built-in control for entitlement and and making the identities exportable [7].
authorization control for multiple attributes, This IDM mechanism allows integration of a
others require the help of OAuth or similar such number of authentication mechanisms such
technologies [3]. as Microsoft Windows based authentication,
LDAP authentication, CA site minder, etc. It is
Proliferation of On-demand User ID deployed on top of the existing authentication
Proliferation of on-demand user ID is a big infrastructure and the deployment is quite
concern in cloud computing IDM as the efficient and fast. It uses SAML to transfer
occurrence of multiple identities for the same credentials. It can be perceived as a layer
user in multiple service providers’ security of abstraction over the traditional IDM that
repositories cannot be ruled out. A simple way fights the challenges of IDM. This aspect of it
to overcome this problem is by the adoption makes this IDM architecture easy to deploy
of OpenID mechanism [4]. OpenID works and dynamic.
by making one primary user id as the key to PaaS is commonly defined as the delivery
authenticate a single end user with multiple of a computing platform and solution stack as
service providers. However, the difficulty in a service. It includes workflow capabilities for
this approach lies in the trust propagation and application design, application development,
development of trusted relationships [5]. as well as application services such as team
Synchronization services help expedite collaboration, web service integration, etc. PaaS
the roll-out and expansion of federated identity IDM automatically scales up to include all these
management capabilities by enabling services features. This is illustrated in Figure 2 overleaf.
in cloud to federate accounts and other data PaaS IDM has to address various
necessary to build up trust relations. functional modules like source control, test
47
premise segments. In addition to all these, IDM
Federated Sample handles the SaaS based challenges of federated
Space of End Users
user space.
Due to vender lock-ins, the primary
PaaS Cloud
limitation with PaaS happens to be a fact that
APP1
APP2
APP3
Developer Collaboration,
Communities, Bug Trackers.
Synchronization of USER CENTRIC ACCESS CONTROL
Maintenance Activities The traditional model of application-centric
Tester Communities Testbeds, access control, where each application keeps
Shared Testing Infrastructure
track of its collection of users and manages
Source Code Control
Versioning Change Tracker
them, is not feasible in cloud based architectures.
Synchronization to Source This is more so, because the user space maybe
Code Repository
shared across applications that can lead to data
Figure 2: PaaS IDM replication, making mapping of users and their
Source: Infosys Research privileges a herculean task. Also, it requires the
user to remember multiple accounts/passwords
and maintain them. Cloud requires a user centric
modules, development communities, etc. For access control where every user request to any
the sake of simplicity, the PaaS IDM could service provider is bundled with the user identity
adopt a Role-Based Access Control (RBAC) and entitlement information [8]. User identity
system to handle each of this and its user space. will have identifiers or attributes that identity
An RBAC system for source control will allot and define the user. The identity is tied to a
minimum set of privileges to the developer domain, but is portable. User centric approach
accounts and essential services, depending leaves the user with the ultimate control of their
on the interdependency of the applications digital identities. User centric approach also
hosted on the platform. For test communities, implies that the system maintains a context of
IDM manages tester accounts, privileges, auto- information for every user, in order to find how
run test suites and knowledge collaboration best to react to in a given situation to a given
portals of the tester communities required user request. It should support pseudonyms and
for hosting a test bed. In case of development multiple and discrete identities to protect user
communities, IDM manages the collaboration privacy. This can be achieved easily by using
of developer communities, access and privilege one of the open standards like OpenID or SAML.
of each group of developer, the bug tracker
system, etc. The cloud could also expect IDM to FEDERATION OF IDENTITIES
handle the database challenges, by controlling On the internet, it is likely that each user ends
the access and synchronization with the in- up with multiple credentials and multiple
48
access permissions across different applications federated world. Currently it is based on
provided by different service providers. These policy files framed by the local authority,
fragmented logins present a challenge to depending on various factors like the domain
the users and service providers, in forms of trust information automatically fed in by
synchronization of shared identities, security, the trust authorities. This is not a scalable or
etc. There is a strong need for an intrinsic flexible model that can meet cloud computing
identity system that is trusted across the web demands. Cloud scenarios require dynamic
and within enterprises and unambiguously trust propagation and dynamic authorization.
identifying users.
Federation of identities maintained VOLATILITY OF CLOUD RELATIONS
by the multiple service providers on the In a traditional model, the IDM is based on the
cloud is very critical to cloud based service long-term relation of a user to an organization
composition and application integration. An or trust domain. In cloud, which represents the
expected issue in this regard is the naming current e-commerce world, the relationships
heterogeneity. Different SPs use different change dynamically and quickly, and the IDM
factors for authentication like account number, has to incorporate all that. Any retrieval or cache
email ID, PayPal ID, etc. Also, when transactions of the volatile data has to be done cautiously.
traverse multiple tiers of service hosted in The possible damage of using old data should
clouds, the semantics of the context of identity be studied. Like, if the user has changed his
information has to be properly maintained, password login with old password, it should
constrained and relaxed as per specific needs. be restricted and locked in all the applications
Consider a complete transaction cycle for an that are participating in the identity federation.
e-bay purchase, based on PayPal account. Live data fetching, domain name resolution,
It traverses from e-bay to supplier, through canonicalization of the data like URL, account
various tiers in supplier’s domain to get IDs, etc., are the challenges.
approvals, release and shipping. Then it goes
through PayPal to approve, validate, release the SCALABILITY
pay, bill the amount to the customer, etc. For Cloud requires the ability to scale to hundreds
each step, the federation authority decides the of millions of transactions for millions of
essential attribute of the customer to be shared identities and thousands of connections – with
with each department. short/rapid deployment cycles. Performance
The user identity mapping in the has to be N+1 scalable across the globe and
previous environments have been one-to-one, deployments agile and quick (weeks not
or in other words, user ID to single user profile. quarters/years). With the software today it
In cloud architectures the mapping challenge is takes ~6 months to make a single SAML/
many-to-one, one-to-many and pseudonyms. SSO connection and it doesn’t address the
Pseudonyms are for privacy protection details, access control and compliance issues. Open
when a user does not want his identity to be Cloud Manifesto states that clouds have
tracked as he crusades various domains. to dynamically scale up and down, so that
Another issue is the trust relation nobody needs to hoard resources to handle
setup between the service providers of the peak hours [9].
49
INTEROPERABILITY infrastructure is secured with respect to some
The mass expects the cloud to provide a IDM requirements and the customers are looking
solution that can interoperate with all existing for a different set of security. The important
IT systems and existing solutions as such or with aspect is to see that the cloud provider meets
minimum changes. Seamless interoperation with the security requirements of the application
different kinds of authentication mechanism and this can be achieved only through 100%
such as the Microsoft Windows authentication, transparency. Open Cloud Manifesto exerts
SSO, LDAP, SAML, OPENID and OAUTH, stress on transparency in clouds, due the
OpenSocial, FaceBookConnect, etc., is what is consumer’s apprehensions to host their
expected of cloud. The syntactical barriers have applications on a shared infrastructure,
to be bridged. It requires an authentication on which they do not have any control [9].
layer of abstraction to which any model of Transparency can be achieved by complete
authentication can be plugged in and off audit logging and control.
dynamically.
PATTERNS IN CLOUD IDM
TRANSPARENCY Based on the insights gained so far three
Security measures assumed in the cloud must patterns in cloud IDM can be concluded.
be made available to the customers to gain their The ideal scenarios for each pattern are also
trust. There is always a possibility that the cloud mentioned.
Authenticated User
Enterprise on Security Domain A
APP 2 Security
APP 1
Domain B
Authenticator
IDP
Decryption
Firewall
Cloud
APP 3 Security
Domain C
LDAP
Firewall
Domain Name
Resolver
User trying to
Encryption of
authenticate by
Credentials
submitting credentials
to system
IDM Management
User Browser
50
Trusted IDM Pattern user can be shared using some mechanism like
This pattern is intended for a smaller or even for SAML. Authorization can be effectively handled
a private cloud that requires security. Scalability by XACML. A basic model of this pattern is
is definitely not a feature of this cloud. But illustrated in Figure 3 on page 50.
Google App Engine (appengine.google.com)
that follows this pattern assures that the External IDM
scalability is not a major concern at the moment This pattern is very similar to the initial
as the number of requests that could be tunneled pattern but for the fact that the credentials
through simultaneously is quite large. The main are submitted directly to the authenticator
feature of the pattern is that the authentication [Fig. 4]. The credentials can be collected by a
is always performed within the firewall. The different browser window, channeled by SSL.
credentials are submitted to the IDM component The pattern is intended for a public cloud. The
and it takes care of encrypting and tunneling IDM concentrates only on domain resolution
the credentials through a secure channel to and triggering of the authenticator to resolve
the authenticator. IDM is independent of the the authentication. This is the architectural
authentication mechanism. Hence deployment pattern adopted by ping identity. In ping
and integration is fast and efficient. Once the identity, domain resolution is done by referring
user is authenticated in by any authentication to a spreadsheet of valid users that is always
mechanism, then rest of the participating kept updated. It can also be achieved through
servers trust the user. The attributes of the other mechanisms like standard domains name
Domain B
Authenticator
IDP
Firewall
Cloud
APP 3 Security
LDAP Domain C
Firewall
User Browser
51
Open ID Request to Authenticate
Request to Access Service B
User
Open ID O Auth
Provider IDP
Lookup Provider Details
REST IDP
SOAP
SOAP
IDM SPML/
Component
Administration SAML
Translator
APP 2 Security Provisioning/
Domain C de-provisioning
delegation
synchronization, etc.
resolution, discovery or YADIS protocol, or web world where there are multiple service
XRDS query, etc., depending on the underlying providers based on a common user space.
technology used. The same drawback of The central identity system, understands
pattern 1 exists in pattern 2 also. Scalability is all technologies used for authentication like
an issue. Symplified (www.symplified.com) is SAML, OpenID, OAuth, etc. Let us assume that
vendor on cloud IDM, whose solution has close the central identity system to be collection of
resemblance to this pattern. modules, each handling a technology, taking to
a common user space and a policy database. The
Interoperable IDM Pattern information is converted to different formats,
This pattern illustrates a cloud to cloud depending on the technology used like OpenID,
scenario, using OpenID and OAuth. The or SAML, or WS-Security and conveyed to the
identity mechanism used, will understand and participating service providers [Fig. 5].
interoperate multiple identity schemes. OpenID A brief comparison of the three patterns
is an open and decentralized standard for user is shown in Table 1.
authentication and access control, by allowing
users to logon to multiple services with the CONCLUSION
same digital ID. Any service provider can Of the emerging technologies cloud computing
authenticate the user in to the system. OAuth has a lot of substance. The huge set of challenges
is again an open protocol that enables a user to it has brought with it has to be captured and
grant permission to a consumer site to access a tamed to produce more benefits. Choice of IDM
provider site without any sharing of credentials design for any cloud should be tailored to suit
[10]. SPML is used for XML based IDM LC. the definition of that particular cloud and open
This is extremely useful for an e-commerce to any kind of enhancements the cloud is bound
52
Features Trusted IDM Pattern External IDM Interoperable IDM
Examples of this Google App Engine's SDC Ping Identity Proposed Design
Pattern
53
8. C h r i s t i a n E m i g , F r a n k B r a n d t , Architecture, Lecture Notes in
Sebastian Kreuzer and Sebastian Abeck, Computer Science, 2007. Available
Identity as a Service – Towards a on http://www.springerlink.com/
Service-Oriented Identity Management content/5865u474424qw751/.
54
SETLabs Briefings
VOL 7 NO 7
2009
55
Developer Cloud Service Enterprises System
ISVs (Publishers)
Community Providers (Subscribers) Integrators
Create excitement Get competitive Low investment Faster time-to- New revenue
and enthusiasm in edge by service realization – market leveraging channel for service
developer accelerating switch from existing services development and
community adoption of cloud company owned maintenance
Flexible pricing
infrastructure to
Facilitate a channel Demonstrate models resulting in New solution
cloud services
for individuals to industry leadership lower TCO. No offering in terms of
earn online money through innovation Pay-As-You-Grow development, test or payment billing and
by adopting cloud flexibility in pricing hosting investment subscription
Create non-linear
technologies models guaranteed management
revenue channel No software,
QoS
Penetrate developer leveraging existing hardware and New solution
community investments Lower TCO non- vendor lock-in offerings in terms of
linear revenue service and project
Aligned with future
channel for service management
IT transformation
offerings
trends
Benefits
56
Register Publish Search Subscribe Pay
57
present a history view of the transactions Also, the business offering needs to
related to the subscriptions. consider various customer segments like
– individuals, ISVs and enterprises - and
Pay demonstrate huge flexibility in terms of the
The pay-per-use case addresses the monetary pricing and service models.
aspects of the service realization. It will be The problem for the key decision makers
consumed by all — publishers, subscribers to realize the business of service exchange
and the service exchange host. It will present is to balance the investment with potential
the information and alerts related to payments, growth and also having support for flexible
consolidated and comprehensive reports for pricing models - cloud computing simplifies
financial transactions. It will also have interface this problem.
with external systems for payment realization. These value propositions from cloud
computing facilitates and makes decision
CLOUD COMPUTING VALUE makers comfortable with the initial investment
PROPOSITION required to start an innovating offering and
Service exchange @ cloud has a great potential scale-up the infrastructure on-demand as the
to become another success story similar to business grows using pay-as-you-grow pricing
App Store, eBay or YouTube. From technology models.
perspective, cloud computing technology Also, for large enterprises, consider the
brings the following value: above benefits to existing Ariba deployments
to understand how it adds value by bringing
■ Dynamically scalable infrastructure Ariba as SaaS on Cloud platform.
(on-demand) Although, the benefits sound interesting
■ Guaranteed quality-of-service in terms of and promising, there are multiple challenges in
performance, scalability and availability realizing it, viz.,
of hosted services.
■ Lack of standardization across large
From business perspective, cloud players
computing brings attractive pricing models ■ Lack of maturity of existing solution and
for individuals, start-ups or enterprises: service offerings
■ Lack of appropriate business case and
■ Lower initial investment in terms of success stories to convince C-level
capital expenditure (capex) executives and
■ Flexible pricing and IT service models ■ Lack of clarity on security, data and IP
(opex). ownership in cloud based deployment
scenarios.
Service exchange is comparatively an
innovative business idea and there will be KEY STAKEHOLDER BENEFITS
constraints on the budget to experiment. At Service exchange @ cloud has something
the same time, huge infrastructure support is for everyone in the value-chain. Figure 3
required to manage scale and quality-of-service. articulates the value proposition of the cloud
58
Publisher
Service Commerce
Platform The publishers get a low investment platform
with high quality of service (QoS) services that
Service Service
Publisher Subscriber can be consumed by enterprises in production
scenarios. It creates a non-linear revenue
Revenue Model
channel for small and medium ISVs to sell their
Service publisher Service subscribes and
develops and publishes consumes the web services to a large service exchange marketplace.
the web services: services: Service exchange also provides flexible
Currency Converter Internet Facing Web
Service Site /Portal pricing models to attract more business and
Unit Conversion Custom Mobile
Service, etc. Application, etc.
offer competitive pricing. It also offers flexible
investment models to facilitate pay-more-as-
Figure 3: Key Stackholder Benefits you-grow and start with low capex.
Source: Infosys Research
Subscriber
The subscribers get ready-to-use services
computing technology and benefits to key from service exchange that can significantly
stakeholders. Service exchange realization influence the time-to-market new services from
can happen in multiple deployment scenarios subscriber’s perspective. It helps promoting
— over internet, over private clouds or over the enterprise reuse in private cloud scenario
extranet (partner network). We will articulate that helps reducing the total cost of operation
benefits to each stakeholder in different (TCO). The subscribers (enterprises, corporate,
business scenario. individuals) have multiple options of service
providers, the payment and pricing models
Developer Community and service models to choose from and select
Service exchange @ cloud over the internet the best-aligned for reuse. Also, all this comes
scenario provides opportunity to individual without any software, hardware, vendor or
developers to develop and deploy services investment lock-in that gives tremendous
to earn online money. It gives a great business agility for the decision makers.
opportunity to talented freelancers to earn
money online. System Integrator
Service exchange opens up new traditional
Cloud Service Provider application development and maintenance
The cloud service providers such as Microsoft, (ADM) opportunities around service
Amazon and Google can achieve competitive development, deployment, maintenance,
edge by promoting cloud adoption by driving management, monitoring and configuration.
developer community and enterprise to the The innovative solution and service
proprietary service exchange. offering around billing, payment and
The service exchange product offering subscription management can create non-
suitable to enterprises or government for linear revenue channel for system integrators
private cloud offering can open a new revenue for enterprise, government and other private
channel for non-liner growth. cloud or enterprise SOA scenarios.
59
Large Enterprises and the government can conceptualize service
Large enterprises having a vision to implement exchange in the private cloud scenario to
enterprise SOA can benefit from enterprise wide implement enterprise SOA while consolidating
reuse of the services through service exchange. IT infrastructure to reduce TCO.
It presents significant cost saving opportunities
for capital expenditure as well as operational REFERENCES
expenditure. It will act as a key enabler for 1. http://seekda.com/
enterprise SOA implementation. 2. webservicesx.NET. Available at http://
www.webservicex.net/WCF/default.
Government aspx
For the government IT landscape, service 3. Zuora: Z-Commerce Platform
exchange can act as a catalyst for lean IT 4. http://www.zuora.com/products/
transformation and IT consolidation for zcommerce/
significant cost savings and reducing TCO 5. Bhavin Raichura and Rajat Kumar,
through private cloud realization. Semantic SOA – IT Catalyst for
Business Transformation, AMCIS 2007
CONCLUSION Proceedings, AIS Electronic Library,
Service exchange @ cloud is a highly scalable Colorado - USA, 2007
monetizing platform. Cloud service providers 6. Bhavin Raichura and Shaurabh Bharti,
can and should promote the adoption of cloud Achieve Dynamic Integration & Runtime
offerings. The ISVs can offer various software Process Optimization using Semantic
features as services. Just as enterprises can SOA, ICSD 2007, Document Research
accelerate time-to-market new services, system and Training Centre (DRTC), Bangalore,
integrators can create new business and revenue 2007-02
channels and individuals can make money 7. Bhavin Raichura and Vijay Rao, Lean
online. The success of such a business model is IT Transformation, ebizQ.net, 2009-03.
also well tested and proven as Apple App Store Available at http://www.ebizq.net/
is to promote iPhone. Also, large enterprises topics/saas/features/11121.html.
60
SETLabs Briefings
VOL 7 NO 7
2009
There are around 75,000 independent cloud computing and the commoditization
software vendors (ISVs) worldwide that drive of business intelligence provide unique
approximately $250 billion of the software opportunities to ISVs to do more with less. Table
industry revenue. These ISVs produce, package, 1 overleaf shows the various opportunities
sell, deliver and update software. The market available to ISVs, enabled by these recent
share in the ISV industry is highly skewed, technology trends.
wherein the top 2% ISVs garner about 80% of Virtualization is one of the top trends
the industry revenue. This top 2% (about 1,700 in the industry today and provides important
ISVs) includes all ISVs with over $10 million benefits to ISVs.
in software revenue [1]. The remaining 98% of
the ISVs have very limited resources in terms BENEFITS OF VIRTUALIZATION TO ISVS
of ability to spend on software development, Most of the ISVs today offer solutions to their
marketing, sales, software distribution and customers in an on-premise model or in a hosted
deployment. In this paper, we focus on how model. Virtualization has already become a
cloud computing offers the large number of major trend in the IT industry, resulting in
small ISVs unique opportunities for revenue ISVs and large enterprises reaping substantial
and customer growth with significantly lower benefits from adoption of virtualization
capital and operating investments. We also technologies in their infrastructure. ISVs
discuss the new service offerings that small as that have not adopted virtualization yet can
well as large ISVs can bring to the market by certainly consider adopting it for the significant
leveraging cloud computing. benefits it can provide.
A good example is of ICICI bank, the
ISVS AND CURRENT INDUSTRY TRENDS largest private bank in India. The bank used
The existing trends in the industry like Web virtualization to consolidate 230 physical
2.0, social commerce, SOA, SaaS, virtualization, servers to just 5, running a little under 650
61
Technology ISV Opportunities
Trend
Reduce Grow Improve Improve Competitor Transform to Virtualized
TCO Business Customer Agility Differentia Servers
Satisfaction tion
(in data center)
Web 2.0 Low Medium High Low High
Social Low High Medium Low High Transform to
Commerce
SOA High Medium Medium High High Virtualized
Physical
SaaS High High Medium High Medium Servers
Servers
(on the cloud)
Virtualization High Low Medium High Medium
Cloud High High Medium Medium Medium
Computing
Business Medium Medium Medium Medium High Figure 1: Virtualization and Cloud Computing
Intelligence
Source: Infosys Research
Table 1: Various Opportunities Available to ISVs
Source: Infosys Research
62
or fully dedicated (private cloud) to a client continuity, cloud security, cloud
and managed by the cloud service provider. applications and storage.
The cloud service provider and the client can
negotiate the terms for pricing, QoS, SLA and ■ Salesforce.com provides customer
operations level agreement (OLA). Billing is relationship management (CRM)
done based on usage (computing based billing solution to businesses over the internet
- $/CPU/hr or storage based billing - $/GB using the SaaS model and was one of
or data transfer based billing - $/Mbps or $/ the pioneers in offering SaaS solutions.
Gbps).
Companies like Amazon, SalesForce.com ■ Facebook offers its infrastructure to
and Google are the pioneers in offering cloud developers to leverage social services.
based services. Amazon has the first mover
advantage in the cloud computing area and ■ IBM’s Blue Cloud and Microsoft Azure
has generated an estimated 500 million dollars are the new offerings on the block. IBM
from cloud offerings alone [5]. The following is recently announced LotusLive Engage,
a partial list of cloud offerings from different an integrated social networking and
vendors – collaboration cloud service designed for
businesses of all sizes.
■ Amazon’s elastic compute cloud (EC2)
is a web service that provides resizable ■ There are also many VC-funded startups
compute capacity in the cloud that is in the area of cloud computing (Coghead,
designed to make web-scale computing Bungee, LongJump, EngineYard,
easier for developers. RightScale, etc).
■ Amazon’s simple storage service (S3) ■ Virtualization solution leaders like Citrix
is an online storage web service that and VMware have also presented visions
provides unlimited storage through a of cloud infrastructures.
simple web services interface and has
been one of the pioneers in the area of With big players like Microsoft, IBM
offering highly-scalable cloud based and Google now entering the cloud computing
storage for a price. and storage provider market by making big
investments, the cloud services provider space
■ Google’s AppEngine offers users the is maturing fast and getting commoditized. It
ability to build and host web applications will be prudent for most ISVs, to desist from
on Google’s infrastructure. entering the cloud services provider market
and instead focus on building new solutions
■ Akamai is extending its content around offerings from big players like Amazon,
delivery network (CDN) to offer Microsoft, Google and IBM.
cloud based services. Akamai Table 2 overleaf shows a high-level
offers optimization services for comparison of various cloud service providers
cloud acceleration, cloud business in the context of the ISV market.
63
Considerations
Cloud
Service
Provider Offering Market ISV Focus Platform Competitor
Faster
Lower Costs
Maturity Adoption Capabilities Differentiation Time-to-market
64
what they currently support and scale quickly software solutions comprising primarily
depending on the success of the solutions that of Microsoft Office and Windows.
they provide.
■ Offer on-demand versions of existing
Faster Time-to-Market web-based solutions.
With the help of cloud service offerings,
ISV developers have to worry less about Cloud computing will enable ISVs to
scalability and focus on aspects like solution take risks with significantly lower investments
functionality and performance of key use cases. in capital and operating expenditure, but still
By leveraging the ready-made services and being able to scale up quickly to meet peak
plumbing provided by cloud service vendors, processing demand without over investing.
ISVs can bring their solutions to market much Callidus, a leader in the sales performance
faster with significantly reduced investments. management (SPM) software market, spent
ISVs should start looking for ways to about three years building an on-demand
quickly build, deploy and take advantage of the version of its existing products. More than one-
flexibility that cloud computing environments third of the customers today use the on-demand
can bring. Vendors like IBM, Microsoft, Google model and this number is likely to grow to more
and Amazon are making it easier for software than half the customers using it in a few years.
developers to build solutions based on open With the addition of on-demand offerings,
standards that are well supported by a vast Callidus has been able to successfully open up
array of technical resources. the market and add several new customers by
offering lower prices [6].
Offer New Solutions
Two interesting ways in which ISVs can Offering New Web-based Solutions Leveraging Cloud
leverage cloud services to offer new innovative Services: In general, SaaS brings business value
solutions to their customers are by: in terms of a flexible and economical business
model rather than a real technology value.
■ Extending existing solutions Cloud computing technology complements
■ Offering new web-based solutions SaaS by helping to realize these flexible business
leveraging cloud services. models by offering utilization-based pricing for
computing and storage resources.
Extending Existing Solutions: ISVs can extend ISVs can build new web-based solutions,
existing solutions in various ways – take them at the global level fast and scale them
very quickly to meet global demands using
■ Offer online services for existing cloud services. ISVs can also improve customer
software solutions that are currently satisfaction by leveraging Web 2.0 and social
deployed to desktops. Microsoft is commerce concepts in an innovative manner to
promoting a similar strategy called offer new solutions for horizontal and vertical
Software + Services to offer online markets.
service extensions to its vast array ISVs can offer new solutions in different
of highly successful desktop-based verticals like manufacturing, healthcare,
65
financial services, retail, energy management, tools that can help manage existing
etc., by leveraging cloud services datacenter deployments as well as cloud
ISVs can also offer horizontal solutions deployments in an integrated manner
in the areas of business intelligence and will greatly help.
analytics, compliance, managed services, etc.
About an year and a half back, Siemens ■ New security, compliance and
started looking at next-generation data centers management solutions can be built to
and examining where unified communications extend existing cloud service offerings
(UC) fit into the picture. Gradually, the
company developed a strategy to port its ■ ISVs can offer lift-and-shift services or
existing unified communications software to solutions to customers who are interested
Amazon’s Elastic Compute Cloud (EC2). The in virtualizing their existing applications
ultimate goal for Siemens is to give partners and putting them on the cloud.
and customers a front-end portal that allows
them to pick and choose the UC services needed ■ ISVs can also offer new kinds of managed
in a flexible manner [7]. services built around cloud service
In the context of web-based ISV solutions, offerings from other vendors.
the challenge is to balance IT investment for a
global expansion of the solution with actual CHALLENGES AND RISKS ASSOCIATED
growth through sales. Even for large ISVs, it is WITH CLOUD SERVICES
almost impossible to proactively plan scalability For ISVs, cloud services are not without risks.
to enable global operations. It is also impractical Some of the associated risks are -
to block large investments in terms of hardware,
software and people while the operation size is ■ Most of the cloud service providers
relatively small. What is needed is a dynamic today offer no guarantee of data and
and on-demand scalability of IT assets and can also suffer occasional outages which
related services as the solution adoption grows could impact business. Few vendors
globally. Cloud computing offers this dynamic allow security or process compliance
and scalable infrastructure to facilitate quick audits of their cloud infrastructure.
growth in an economical manner.
■ Most vendors today have implementations
Extend Cloud Service Offerings of Other which will result in a significant vendor
Vendors lock-in, even though they talk about
ISVs have a good opportunity to build new standards compliance.
solutions extending existing cloud service
offerings from vendors like Amazon, Google, ■ ISVs have traditionally built hosted
Microsoft, IBM, etc. Some key areas where the solutions or desktop based solutions.
existing vendor offerings can be extended are – Making the shift from the current
mindset of delivering desktop or web-
■ There is a need for better tools to manage based software to delivering services
cloud deployments. Management using utility computing will be very hard
66
work and will require skills that most CONCLUSION
of the small ISVs do not have currently. In the tight economy prevalent today, companies
While some ISVs will be able to take are spending much less on IT and ISVs will have
advantage of cloud services, the vast to take growth wherever they can find it. Cloud
majority of ISVs will have a very difficult computing is a double edged sword which
time making this switch. presents a significant challenge as well as an
important opportunity for ISVs. ISVs offering
Daryl Plummer from Gartner says that pure-play hosting services will really struggle in
ISVs are not positioned well to become the fending off the big cloud vendors. ISVs offering
next generation of Cloud Service Providers on-premise software will be forced to innovate
(CSP) or even SaaS providers. He says, some and build extensions to their software which
ISVs will either change their business entirely, uses cloud-based services. The cost of deploying
or go out of business if cloud computing software in the cloud will keep reducing at a
becomes the mainstream norm for delivery brisk pace, potentially leading to innovative ISV
of systems [8]. offerings built around the cloud infrastructure
Table 3 shows some of the key challenges resulting in increased competition that is very
for ISVs associated with cloud services. fast-moving. Cloud computing innovations
Challenge Details
Potential Competition Most ISVs will have to partner with cloud service providers like Amazon, Microsoft, Google, IBM,
from Cloud Service etc., to deliver their solutions. If these ISV services are offered as part of a bigger service
Providers offering, the ISVs will be opening doors to potential competition. Pricing will also become key as
profits will have to be shared with the cloud services provider. Building a high level of trust and
credibility with the cloud services provider will be critical. ISVs will also need to bring in significant
differentiation in their solution offering.
Introduce New Pricing ISVs will have to change their pricing model to include software, computing, storage and service
Models price. This new model will be challenging and could impact profit margins.
Maturity of Offerings The cloud computing technology is still maturing and many of the cloud computing offerings are
not yet production ready. There are also open issues around data security, compliance, data
ownership and standardization which need to be addressed.
Higher Adoption Risk Unless the open issues around data security, compliance, data ownership and standardization
are addressed, adoption of cloud services could be low. This low adoption could increase the
implementation risk for ISVs. ISVs will need to display a significant amount of courage, passion
and leadership to make their cloud-based offerings successful.
Extending Existing Extending existing ISV solution offerings to the cloud will be challenging as it will involve
Solutions Using Cloud significant enhancements and risks. Costs involved could also be significant if the existing
Services solution is a pure desktop-based solution.
Handling Cloud Service Most cloud vendors today don't provide availability assurances and SLAs are mostly non-
Outage existent. Cloud vendors also don't allow embedding of security and management agents or
monitors. Occasional outage of services from providers like Amazon, Google, IBM, Microsoft,
etc., is a distinct possibility and recent outages have only provided fodder to this thought. ISVs
will have to devise a plan to keep customers informed about such outages and assuage them if
such outages occur.
Lack Of Geographic With the exception of Akamai and Layered Technologies, no cloud vendor allows the placing of
Coverage an application in a specific geography on the cloud. Most cloud service providers today don't
have geographic coverage. Lack of geographic coverage could lead to significant performance
challenges.
67
happening in the industry are certainly a major Cloud, April 2009. Available at http://
point of inflection for the ISV market. ISVs that www.thevarguy.com/2009/04/02/
are able to innovate and navigate through these siemens-channel-partners-testing-
shifts will stay on to fight another day, while unified-communications-in-amazons-
those who fail to innovate will perish. cloud/
8. Daryl Plummer, Delivering Cloud
REFERENCES Services: ISVs - Change or Die or both!
1. Dan Lohmeyer, How does Microsoft Gartner, November 2008. Available
work with ISVs? Available at at http://blogs.gartner.com/daryl_
http://blogs.msdn.com/msftisvs/ plummer/2008/11/06/delivering-
archive/2007/08/22/how-does- cloud-services-isvs-change-or-die-or-
microsoft-work-with-isvs.aspx both/
2. Real CIO World, December 15, 2008 9. h t t p : / / w w w . f o r r e s t e r .
3. Bridget Botelho, VMware Entering com/imagesV2/uplmisc/
Final Phase of Virtualization Evolution: CloudComputingWebinarSlideDeck.pdf
C l o u d C o m p u t i n g , I T K n o w l e d ge 10. James Staten, Cloud Computing for the
Exchange, May 2008. Available at Enterprise, Forrester Research, February
http://itknowledgeexchange.techtarget. 2009. Available at http://www.
com/server-virtualization/vmware- forrester.com/imagesV2/uplmisc/
entering-final-phase-of-virtualization- CloudComputingWebinarSlideDeck.pdf
evolution-cloud-computing/ 11. R W a n g , I B M S t o r m s I n t o C l o u d
4. http://www.gartner.com/it/page. Computing With an ISV Friendly Pricing
jsp?id=868812&tab=agenda Model, Forrester Blog, October 2008.
5. h t t p : / / w e b 2 i n n o v a t i o n s . c o m / Available at http://blogs.forrester.
money/2008/04/22/amazon-web- com/appdev/2008/10/ibm-storms-
services-on-its-way-to-surpass-500m-in- into.html
sales-this-year/ 12. IBM to Deliver Software via Cloud
6. T h e C l o u d S h i f t : D o n ’ t f o r g e t Computing With Amazon Web Services.
ISVs. Available at http://www. Available at http://www-03.ibm.com/
o n - d e m a n d e n t e r p r i s e . c o m / b l o g s/ press/us/en/pressrelease/26673.wss
The_Cloud_Shift_Dont_Forget_ 13. Azure for ISVs. Available at http://
ISVs_31079814.html www.microsoft.com/azure/isv.mspx
7. Siemens Channel Partners: Testing 14. Windows World, Demystifying Cloud
Unified Communications In Amazon’s Computing, May, 2008.
68
SETLabs Briefings
VOL 7 NO 7
2009
69
This leads us to the questions such as — (32, 64 bit), memory capacities and hard disk
what are the types of parallel problems? What storage. Moreover, the variety is available in
are the line-of business (LOB) applications that as many numbers as required. This makes it
are best suited for clouds? What is the suitability suitable for a wider range of existing on-premise
of the existing software and hardware HPC HPC applications.
techniques on the cloud? And most importantly,
how suitable cloud computing is for HPC? Versatile Support for Parallel Computing Styles:
The availability of uni-core instances, multi-
WHY IS CLOUD GOOD FOR HPC? core instances and cluster of instances from
Let us identify the aspects that make cloud the public cloud vendors makes it amenable
computing an attractive proposition for HPC for different architectures – shared (as in SMPs,
workloads. CMPs) and distributed memory (clusters/grids)
and programming models – shared (OpenMP,
Versatile Support for Elastic Parallel pThreads, etc.) and distributed address space
Computing Execution Environment (MPI, PVM, etc.). The existing commercial HPC
Cloud is attractive for HPC primarily because applications and libraries written using these
a well conceptualized cloud platform (public architectures and programming models can be
or private) provides a wide range of parallel ported on to cloud infrastructures for deriving
computing options on it. As is well known, the additional benefits such as ROI and scaling.
high performance in HPC is achieved through
parallel computing techniques [6]. Inherent Versatile Support for Workload Variety
parallelism (at bit, instruction, task levels) in Compute intensive tasks that exhibit significant
target application exhibiting various degrees data parallelism such as Monte Carlo simulation
of parallelism (fine, coarse and embarrassingly over large data sets for generating risk analysis
parallel) is exploited by computation of the reports in finance, BLAST searches in Life
parallel portions on multiple processors (CPUs, Sciences, N-body simulation, etc., can be
multi-cores or nodes on a cluster), custom executed on a cluster of high power CPU
hardware platforms and accelerators (GPUs, instances provisioned from the cloud. Monte
FPGAs [1]) using different architectures and Carlo simulations also require significant
programming models (shared and distributed caching requirements that can be provided by
memory). distributed caching across multiple compute
At a basic level, suitability of a cloud instances. Memory intensive tasks limited by
platform for an HPC application would memory latency such as dynamic programming,
be determined by the support provided sparse matrix computations, etc., in many
by the platform for the execution platform financial applications are suitable for running
requirements — CPU (speed and numbers), in instances that have higher RAM capacities
latency and bandwidth of memory hierarchy and multi-core instances sharing the same
(cache, RAM, disk) and the network. Typically, physical node. For compute, memory and
cloud platforms provide a reasonably wide communication intensive tasks such as, dense
variety of compute units in terms of CPU speed, linear algebra (DLA) computations as in oil and
number of cores, frequency, architecture type gas exploration and simulation applications
70
that require small size message exchange can examples in this area. Cloud computing enables
perhaps be executed on a fewer multi-core flexibility not at the cost of optimal resource
instances provisioned from the cloud rather allocation, but in consonance with it.
than more number of low end uni-processor
instances. The shared memory model would Freedom from Performance Clippers and
obviate memory latency and bandwidth issues. Achievement of Better Architectural Match
For example, for tasks exhibiting coarse The trade-off in cloud computing is between
grained or are embarrassingly parallel, such as cost and performance, unlike the on-premise
web page search, indexing, machine learning, case where there are hard limits to the
etc., low speed network interconnects are not available horsepower and hence performance
a problem and distributed memory model gains. Often parallel computing application
is appropriate for scaling. Some of these architectures encounter bottlenecks in specific
applications can experience performance gains portions resulting in sub-optimal provisioning
by exploiting data parallelism on a larger and performance. For example, the master in
number of low strength processors having a a master-slave configuration often becomes
larger amount of distributed memory (total a bottleneck due to its centrality in the
RAM across the cluster). Frameworks such as architecture, similarly certain nodes in an
MapReduce are popular on cluster of nodes HPC cluster responsible for reading/writing to
provisioned from the cloud. However, it is data sources/sinks become bottlenecks due to
important to note that MapReduce is just one I/O latency and bandwidth limitations. Cloud
of the many different categories of parallel makes possible better matching of architectural
computing models [7]. Offline batch workloads requirements.
where the batch data can be uploaded on the
cloud storage space are very suitable for clouds. Availability of Feature rich HPC Frameworks
The advent of the Elastic MapReduce framework
Flexible yet Optimal Provisioning by Amazon has heralded the beginning of the
Cloud computing infrastructure platforms availability of HPC frameworks tailored to
coupled with dynamic provisioning features cloud computing infrastructures. Job scheduling
enable flexible ramp up and down of resources and resource provisioning are closely tied to the
based on SLA requirements. Policy aware topology of the cloud infrastructure and can be
provisioning enables specification of thresholds optimized by the cloud provider. For example,
and scenarios for resource ramp up and down provisioning the MapReduce cluster from
to handle unexpected workload fluctuations. the same subnet or physically proximal set of
This enables low variation in performance hardware can result in obvious performance
and scalability in true sense. For constant gains which only the cloud provider can make
workloads such as drug discovery and protein possible. Moreover, the cloud HPC user does
synthesis this does not matter, however for not have to handle the onerous tasks of setting
variable workloads faced by financial analytic up clusters, provisioning adequate capacity
applications this would be of great help. Amazon nodes (for example, high end compute node,
cloud‘s auto scaling feature and the provisioning I/O capacity node for masters in a master-slave
and management solutions from RightScale are configuration).
71
Clouds for Real time Workloads CHALLENGES FOR HPC ON CLOUD
HPC workloads such as extreme transaction Cloud computing based HPC is at a nascent
processing, distributed query processing, stage and holds great promise as indicated in
complex event processing, streaming data the earlier section. However, there are quite
applications, real time analytics applications a few challenges that need to be overcome
are more suitable for private clouds (in the henceforth.
current state of maturity of public clouds).
These applications are characterized by the Virtualization Related: Some of these arise due
need for online or real time responses from high to the basic issues pertaining to virtualization
performance computation on large on-premise and its effect on the absolute performance
data, often generated in real time. that can be expected as well as the variability
and instability in performance. There could
HPC Data Grid be unexpected performance variations when
Cloud infrastructure is appropriate for storing scaling to larger number of instances and
huge data sets for HPC computations, such as cores. Another issue is the possibility of
databases in BLAST searches in life sciences loss of performance due to the time taken in
applications, financial market data from third bringing up new instances as well as ramp up
party providers (such as data from Reuters in virtualized infrastructure.
Market Data System and the Reuters Tick
Capture Engine, etc.) for options pricing Cloud Management Services: Dynamic and
applications, etc. Amazon’s offer to host public policy based provisioning features to ensure
data sets on AWS is an initiative in this direction auto-scaling and load balancing are important
that makes things simpler, faster and cost to ensure reliability and expected throughput
effective for service users. Performance of cloud of HPC workloads. The solutions in this space
can be improved in the presence of data grid are still in nascent stage with many open issues.
middleware enabling sharing of data among the Solutions from Amazon for its own cloud and
participants in the cloud. Data grids reduce the independent solutions from vendors such as
I/O – blocking calls that an application might RightScale [8], 3Tera would go a long way in
incur when writing to files. making cloud platform (public and private)
In memory data grid (IMDG), distributed effective for HPC.
file systems (DFS) and distributed caching
strategies are the options in this area. The Public Cloud Related: The second category of
availability of cloud databases such as Amazon’s problems arises in the case of public clouds.
SimpleDB, Google’s BigTable, Microsoft’s SQL There are studies that indicate inordinately
Server Data Services, etc., that store data as key high latency of large size data uploads, storage
value pairs are worth exploring as the data tier costs associated with storing large amounts of
of the HPC application. This enables availability basic and derived data in the cloud. Most of
of durable and pervasive data handling the current public cloud infrastructures run
mechanisms across multiple compute nodes using high latency network and low bandwidth
and the ability to move workloads effectively interconnects. HPC clusters usually require
across machines. extreme low latency and high bandwidth
72
interconnects (such as Myrinet, Infiniband) for HPC application for cloud and could force
parallel tasks that are inter-task communication the architects to think of innovative options.
intensive and I/O intensive. Another aspect is Yahoo!’s Pig, IBM’s JAQL, and Facebook’s
the upload of large data sets to the cloud on Hive, MapReduce implementations such as
internet. Uploading a terabyte of data over a from Greenplum and Aster Data are efforts in
1.5Mbps T1 broadband line takes more than 80 this direction.
days. Hence, offline data transfer on physical
disks by courier service is to be considered. HPC ARCHITECTURES FOR THE CLOUD
This has implications in terms of security and Analysis of various cloud providers and
related issues. other participants in the cloud ecosystem, led
to a representation of the cloud based HPC
Benchmarking: As mentioned in an earlier application as given in Figure 1 overleaf. The
section, suitability of a cloud infrastructure for architecture provides many features that are
a specific HPC workload will be determined common in HPC systems and are described in
by the workload characteristics and its match the following sections.
with the declared and observed performance Cloud architecture can be used for HPC
of the compute infrastructure. Published workloads like scientific computations and in
performance data and benchmarking results for most cases can perform at the same level of
both uni-processor performance such as HPC efficiency as that provided by a dedicated grid.
Challenge and parallel computing performance It has to be noted that the network interconnect
benchmarks such as the NAS PB for the cloud between the machines may not be as fast as
computing infrastructure will be necessary in a dedicated grid and can cause performance
matching process. Without such benchmarking degradation when the nodes share a large
related inputs the cost-benefit analysis would amount of data.
be a faulty one.
CLOUD COMPUTING AND HPC
Security: Security of large data sets imported WORKLOADS
onto public clouds from an enterprise’s internal The versatility of the cloud computing platform
systems is definitely a concern from security enables its mapping with a variety of HPC
point of view. Most of the Byzantine fault workload patterns [Table 1 on page 75].
tolerance issues are handled by the cloud
infrastructure but the application architectures Applications that have Seasonal Workloads
also needs to account for such faults. Scalability at low cost is the unique selling
proposition for the cloud. Consumer facing
Transaction and RDBMS Related: Cloud applications such as retail systems that face
storage is mostly non-relational and most of the huge demand during festive seasons are
legacy enterprise HPC applications have data well suited for cloud architecture. Cloud
hosted in RDBMS. This disconnect could lead to infrastructure functions as a load balancer
obvious migration and porting issues of legacy at a high level and distributes the incoming
HPC applications for cloud infrastructures. request to one of the nodes in the cloud. We
This would determine suitability of a particular classify this workload as a single job getting
73
Applications Applications Applications Applications
Internet
Scaling and SLA
Cloud Architecture
Services (Auto
Management)
Value Added
ComputeServers Persistence
executed in a single machine. The workload can be run in one of the machine [Fig. 2 on
is executed by one single machine/node and page 76]. This effectively is data parallelism
hence when more jobs arrive, if more nodes in use to get the required throughput. The
are allocated, the application can easily scale performance of such workload is excellent
and cater to the increase in demand. Such when there is minimal data sharing between
workloads are highly suitable for cloud the participating nodes. Long running batch
model. applications working on large data volumes
are suitable for cloud environments.
High Throughput Workloads
Applications that run against a huge data High Performance Computing Workloads
volume and that has shorter time window for Under this category, we have grouped
execution can leverage cloud models. Such application workloads as — compute intensive
kind of workloads can be classified as single scientific calculations; embarrassingly
job multi machine workloads. The input can parallel logics like Monte Carlo; low latency
be cut in to smaller pieces and each piece requirements for systems like trading and
74
Application Application Worked Type Distribution of Example
Cloud Key Factor
Sterotypes Characteristics Suitability
Job Unit of Works
Adaptive Systems with Single Job 1 Job Very High Load Balancing Internet facing e-commerce retail
Systems Seasonal run in corresponds to (at Job Level) sites. tax processing system,
Demands Single 1 Unit of between the Regulatory systems that are required
Machine Work/Unit of Participants to operate in response to an event
Work run on a
Single Machine
High Long Running Single Job 1 Job = Many Very High Data Parallelism Purchase order systems, updation of
Through- Batch Systems run in Units of Work/ from the stock in a retail industry. Billing in
put Multiple A Unit of Work Application Side telecom. Back office risk analysis
Systems Machines run on a Single Data Sharing batches in financial firm
Machine. between
Results of Unit Machines
of Work Minimal. No
assemble later Task Parallelism
Search Engines Single Job A Unit of Work High Map Reduce Distributed information processing,
High run in is run in Many Algorithm Petabyte data processing — searching
Performa- Multiple Nodes in Two Implementation - for field to get a particular value
nce Data Machines Phases — Map might require a
Mining Reduction and Reduce High Global Parallel High Performance Data Analytics and
Algorithms (Data Phases File System mining in Telecom industry – real time
Dependencies in information for law enforcement??-
Set of Records) data load and retrieval on a columnar
database can improve the overall
throughput
Single Job 1 Unit of Work Medium Performance will Life science modeling for drug
Compute
High run in spreads across not be as good discovery and simulations
Intensive –
Performa- Scientific Multiple Multiple as a Dedicated
nce Machines Machine (with HPC Cluster
Computing Calculations Varying (Data + Task
Degrees of Parallelism)
Data Sharing)
Embarrassingly High Can Leverage Pricing application for a financial
Parallel Logics MPI and Open derivatives
— Monte Corlo MP Libraries
front office analytics. Cloud can be used for HPC Data Analytics Frameworks
these workloads however; performance would The advent of the Elastic MapReduce framework
depend on extent of data and sharing between by Amazon has heralded the beginning of the
the nodes. availability of HPC frameworks tailored to
For low latency applications, the current cloud computing infrastructures. Job scheduling
cloud architecture needs to evolve and will and resource provisioning are closely tied to the
have to support high speed networks, have topology of the cloud infrastructure and can be
connectivity to data providers and provide optimized by the cloud provider. For example,
infrastructure required to support complex provisioning the MapReduce cluster from
event processing capabilities. At this juncture, the same subnet or physically proximal set of
these applications are suited for in house hardware can result in obvious performance
deployment. gains which only the cloud provider can make
75
High Throughput Workloads — Cloud is highly suitable for divide and conquer approaches
A set of records
for batch processing
Node Node
Node Node
1 2 Clouds can be used as pay per use service for this
kind of workloads
Node Node
Node Node
Clouds with Grid Middleware are suitable for this kind of workload. Performance can be improved if records can be
routed to nodes where relevant data is already available. Master- Worker patterns can be used to achieve the same.
possible [Fig. 3]. Moreover, the cloud HPC for HPC jobs will emerge for workloads
user does not have to handle the onerous tasks such as extreme transaction processing,
of setting up clusters, provisioning adequate complex event processing, etc. Extensions
capacity nodes (for example, high end compute and customization of MapReduce frameworks
node, I/O capacity node for masters in a master- would emerge for different types of HPC
slave configuration). workloads and industry verticals. Parallel
computing libraries benchmarked on specific
CONCLUSION public cloud platforms would evolve. More
We foresee emergence of more mature cloud comprehensive performance benchmarks
provisioning and management solutions of popular public clouds would become
increasing the throughput of HPC jobs available. Customized HPC application stack
through enhanced resource allocation, images would become available for specific
scheduling and reliability. Public cloud public cloud platforms easing setup of HPC
with faster inter-node interconnects, that applications on the cloud. Industry vertical
are good for communication intensive specific SaaS HPC platforms would emerge
HPC jobs, will emerge over a span of time. and performance benchmarked with respect
Specialized on-premise private clouds to specific public cloud platforms.
76
High Performance Data Mining- Cloud with map reduce algorithm implementation can be leveraged
Intermediate data
arranged in key value
pair {k.v}- kept ready for
further reduction
Input data broken to
multiple pieces
Very large input data
K1
having some
dependencies Map logic K2 Map logic
K3
Output
Data
Map logic Results
Parallel
File
Systems
Ki
Map logic Map logic
Kj
Performance considerations
Network interconnect between the machines-nodes hosting the map logic intermediate data and reduce logic volume of
data transferred to intermediate storage. Performance can improve if map phase can implement local reduction
77
183.html known-applications-of-mapreduce/
7. Known applications of MapReduce, 8. RightScale. http://www.rightscale.
http://www.dbms2.com/2008/08/26/ com/products/.
78
SETLabs Briefings
VOL 7 NO 7
2009
79
Option Type Model CPU RAM Disks RHEL cost VMware Total Cost
Support (USD)
2 Limited Scalability for Servers and no scope for More Flexibility for Server Resource Management and Scope
Processing Power Sharing between Servers of Processing Power Sharing between Virtual Servers
3 Islands of Limited Configuration within Budget Effective Higher Configuration at the Same Price within
Budget
6 Green Initiatives are not served appropriately Organizations Green Initiative Served Better
80
monitoring tools and will operate a ticketing
Physical Server Virtual Server
Infrastructure Infrastructure system for resolving user support issues.
This simple example thus conveys how
Physical an IT organization can reduce infrastructure
RHEL Additional
Server 1
Spare
WIN
costs, optimize resources and achieve better
Virtual manageability to move away from standalone
RHEL
Physical
RHEL
Server
dedicated infrastructure.
Server 2
Virtual
Server
RHEL CREATING AN INTERNAL VIRTUALIZED
Physical
RHEL INFRASTRUCTURE
Server 3
Virtual
RHEL
A number of organizations have already
Server
made large investments in the dedicated
Physical
Server 4
RHEL VMWARE infrastructure for multiple environments
because were needed at that point in time.
Figure 1: Physical Servers and Virtual Server Now these environments with respectable
Source: Infosys Research numbers are either in excess or under utilized.
It makes business sense to utilize the same
infrastructure for new upcoming applications
using the excess/spare capacities. This calls
perspective, we need to monitor three server for some internal changes for unlocking
units (physical or virtual) in both the cases. these capacities by modifying the internal
However the flexibility to allocate memory IT infrastructure canvas to transform it
to one partition on the fly is not possible into an internal cloud, using virtualization
in the standalone case. After monitoring in technology.
virtualized environment, if we find that the It will help the organizations to save on
application needs more memory, we can new purchases and power, and thus reduce
allocate some more from the buffer capacity overall capex. Note that a virtual and dedicated
that is already available with us. We have server infrastructure may co-exist based on
the flexibility to accommodate some more business criticality and organization’s overall
environments in the same box in the future strategy of moving to a cloud.
that saves cost of purchasing additional There are a number of prominent factors
servers. that an organization should consider and
Day-to-day infrastructure management evaluate before delving into virtualization.
and support (backups, vendor co-ordination, Factors that need to be considered are —
OS upgrades and patching, application
upgrades) is carried out by internal IT team in ■ Business benefits in term of savings/
both the cases. value delivered
In case the infrastructure landscape ■ Capabilities of internal teams managing
is huge, the team that manages either the the existing infrastructure
standalone or virtualized environment will ■ Flexibility in managing the virtual
use tools like traditional scripts or third party environment
81
■ Groups involved in supporting the IT Other factors of IT management that were
landscape done for the standalone infrastructure would
■ Readiness to work in tandem be applicable here too. Rather than a localized
■ Roles and responsibilities backup, centralized backup may come into
■ Policies, procedures, OLAs and SLAs picture and will drive the organization’s
affected. backup strategy.
Monitoring of the server infrastructure
They should be properly evaluated, in case of traditional commands/scripts/
planned and executed and need strong backing third part tools would remain the same.
and support from the senior management. However, the organization needs to closely
Other technological factors that will come into monitor if there is any change in the licensing
picture are – policies specific to the monitoring agents
and licenses that are deployed for the virtual
■ Mode of storage and data access for the servers. This also holds true for all the
application (central/NFS/SAN) software licenses that are installed on the
■ Compatibility and interfacing of existing virtual servers.
server commodity hardware We note that the organization’s internal
■ Booting processes virtualized infrastructure (or should we call
■ Network access an internal private cloud?) has a boundary
■ Distribution and interfacing of and can be extended upto the limit the server
applications resources are available in the organization.
■ Virtualization/cloud readiness for the Beyond this, the IT management would face the
application issue of provisioning more resources or adding
■ Ones to move and ones not to move to more servers to the internal private cloud.
the virtualized infrastructure. Thus, we understand that internal virtualized
infrastructure has limitation of scaling but is
Due to the limited dynamic ability of initially suitable for the organization that is
the virtualized infrastructure to provision planning to transition in the future to private
resources, there will be a significant change or public clouds.
in the way we look at the configuration
management data base (CMDB) from the CONTINUING THE JOURNEY FURTHER —
perspective of application usage and updating THE PRIVATE CLOUD
information. Organizations should look at Crossing over the boundary of the organization
this process of dynamically updating CMDB. towards a private cloud (or call it external
The internal ticketing process would also private cloud for location namesake) indicates
continue to be the same with the exception of that we are moving towards something that is
a few variations that are a result of dynamic provisioned as a service (pay-per-use) to us like
provisioning. a commodity like electricity or water.
Management in the internal virtualized It further makes sense to understand
infrastructure would be simplified with various perspectives of vendors, researchers and
reference to the provisioning of resources. experts on terminologies like cloud computing,
82
vendors of the cloud, cloud computing who manages the private cloud. The end
technology and services. Organizations also user admin can on the fly create a server by
draw out key considerations before embarking providing the operating system, number of
on cloud services and prepare a scorecard CPUs, memory and disk space. There are
based on the key considerations that has various other granular parameters that differ
weight, raw score and a weighted score before from vendor to vendor. The responsibility of
deciding on a value for go, hold or no-go. The the homegrown application tuning still lies with
same can be considered for public cloud too but the IT organization whose end users work on
that requires greater emphasis on factors like the cloud infrastructure.
organizational strategy, application criticality, There are a few aspects like application
federal norms, security aspects and compliance transition to the private cloud and inter-
issues. dependency of the applications that should be
Private cloud can be called as a private thoroughly tested before transitioning. There
computing facility provisioned for any are vendors who have a process defined as to
organization. All the resources like server’s how data should be transitioned to the cloud,
infrastructure and cloud computing software is scheduling the timeframes for backing up the
dedicated to the organization. Unlike the internal data in the cloud, scheduling of adhoc backups
cloud, private cloud is extensible on demand. and restoration options.
It has the feature of dynamic provisioning of From infrastructure monitoring
the virtual resources. Different vendors deploy perspective, the standard tools that the
different mechanisms for quick provisioning organization uses like scripting can be used.
but capacity planning aspect of the underlying Vendors also provide interfaces to display the
systems is closely monitored. End user identity standard monitoring parameters like CPU, disk
management and resource provisioning are the space utilization and processes. Third party
key factors that have impact on the security vendors and cloud vendors also are in the fray
aspect [1]. to provide the cloud performance parameters.
One aspect to note here is that there The web service Amazon CloudWatch that
is a separation between the person who is tracks and monitors Amazons Elastic Compute
demanding resources and the person who is Cloud (EC2) service provides real time
provisioning them. From an organization’s monitoring by using web service APIs or
perspective, although it is flexible to manage command line tools for cloud resources with
such separation, there should be an approval demand pattern parameters like resource
process set in the standard processes for utilization, operational performance, and CPU
dynamic provisioning of the resources. It might utilization, disk reads and writes, and network
so happen that more resources are consumed utilization [2].
and performance tuning aspect of the entities Various aspects like vendor
takes a back seat. dependency, switching flexibility from one
From infrastructure management vendor to the other, business continuity
perspective, the responsibility of managing and disaster recovery capabilities of the
the servers and dynamic provisioning has vendor, application and software support,
now shifted to the third party private vendor stability, security aspects, compliance, SLA
83
provisioning should be given consideration. Fluctuating and unpredictable load patterns are
There should be a process defined for also factors that decide how we should scale
integration of the vendors ticketing system up and down in the elastic cloud. A hybrid
with the organizations internal one in a cloud infrastructure (mix of private and public
seamless way so that call handling takes place cloud and local virtualized infrastructure) can
smoothly and SLA violations get recorded and also exist.
reported accurately. The journey from a standalone
infrastructure to the public cloud is shown in
FINAL DESTINATION — THE PUBLIC Figure 2. Note the reduction in the periphery
CLOUD and the infrastructure landscape within an
There are four key steps to cloud adoption organization reduce as we move towards the
strategy - access, validate, prepare and execute public cloud. This surely presents a challenge
that help the CIOs integrate cloud computing to handle IT Service Management (ITSM)
with the IT strategy of a firm. Understanding the processes.
key set of activities pertaining to infrastructure, As per the standard ITSM support
timeframes and challenges associated to processes, the users raise a ticket in the
move further would determine how early and ticketing system as illustrated in Figure 3 [3].
successfully the organization becomes cloud The ticket appropriately gets processed at the
ready. client system and is raised as an issue based
Support for the organization on the severity level (L1, L2 or L3) in the cloud
application landscape and if they are cloud- vendors ticketing system. Note that the users
deployable or cloud ready in the public and the administrators access the resources
domain should be evaluated. Here, although in the cloud on the encrypted network as a
the technical infrastructure continues to part of security and compliance. Based on the
remain like the private cloud, it is not a levels, the ticket traverses through the three
dedicated one but a shared one wherein levels. User does not have a visibility beyond
the cloud space is shared by a number of this since underlying hardware and resources
organization’s and their end users. The provisioned are managed by the cloud vendor.
underlying infrastructure that was visible to Based on the dynamic provision requested,
the end user gets limited to a set of virtualized usage and the SLAs adhered as per agreed
sets that can be monitored and tasks can be terms between the cloud vendor and the client,
scheduled remotely. billing will get calculated. Resource usage
Data security measures become statistics are visible to the administrator of
prominent in the public cloud and necessitate the client.
data encryption and security measures to be From the system administrators
deployed for flow of traffic across the network. perspective, the CMDB that holds the server
All the features provided in the private cloud configurations may change frequently (scale up
like creating the server on the fly, dynamic or scale down) based on dynamic provisioning
provisioning of resources, scheduling of available. The process whether to involve
the backups, restoring a earlier backed up the change advisory board (CAB) for every
snapshot are available in the public cloud also. provision needs to be given a thought based on
84
Server resource at cloud vendor
Can be created on-the-fly
Shared server resources (system / database) for IT organization
Further reduction in costs Public Cloud
SLAs / security concerns due to sharing should be addressed
appropriately
Vendor provisioned / Third party monitoring tools
frequency and variation on higher side since it internet and installing them as a plug-in should
is directly linked to costs. not hamper the performance of the application.
Some of them are Hyperic (prominently visible),
CONTROL IN THE CLOUD USING Nagios, Zennos and others.
MONITORING TOOLS The monitoring product Hyperic HQ
There are internal monitoring mechanisms for equips IT organizations with the ability to
cloud provisioned by the cloud vendor by a third securely monitor internal IT infrastructure
party. However, there are a number of open as well the cloud services [4]. HQ’s ability to
source tools available which can be evaluated automatically discover, monitor and manage
for usage. There are various considerations like software services, regardless of type or location,
operating systems support, support on thin enables organizations to easily create a unified
clients (being lightweight) since monitoring view of the performance and health of their
has to be done across the network or across applications is a definite plus.
85
Change Advisory Board (CAB) Usage Billing
Monitoring/
Change Management Scheduling/ SLAs
Access (encrypted)
Problem Management
L3 Problem Management
ITSM
Incident Management Visibility Incident Management
L2
to Client
Service Tickets
Organization End Cloud Vendor End
Figure 3: Cloud Vendor and Organization ITSM processes Source: Infosys Research
An open source monitoring tool Nagios memory utilization and disk space utilization are
is a powerful comprehensive monitoring system available in most of the products. Cloud vendors
that enables organizations to identify and also do provide an administrator interface and
resolve IT infrastructure problems before they the usage/billing statistics. The administrator
strike and offer visibility through web interface, can dynamically provision/remove the dynamic
alerts reporting and multi-tenant (access based server resource created in the cloud. Clarity
view) capabilities [5]. should be obtained on business continuity and
Zennos supports full operational disaster recovery provisioning, country of data
awareness by monitoring the entire IT storage (few countries require critical data to be
infrastructure through agentless collection stored within country), provisioning clause for
and control. Key features are autodiscovery, switching to a different vendor.
IT configuration database, alerting, fault Once the final stage is reached, the
management, availability and performance organization is free to concentrate on its core
reporting and a host of other features [6]. business competencies and outsource a large
Organization can select appropriate part of its IT operations to the cloud vendor.
cloud vendor monitoring option or a third party The internal IT hardware/license management
tool or a combination of both that would serve overheads will reduce. Does this mean that the
best for comprehensive monitoring. local IT organization will cease to exist? Not
All the standard utilization and really. IT organization will have to prominently
monitoring parameters like CPU utilization, play a role in the following:
86
■ Transitioning of the existing application to have few cloud vendors like regular IT
to and from the cloud vendors that it has today. ITSM processes of
■ Ensuring data management and security an organization should be drafted considering
aspects the dynamic set of provisioning resources and
■ Planning and developing new application monitoring tools should be adopted as per
and monitoring existing applications application compatibility. Future may also
and virtual servers necessitate interfacing between two different
■ Handle the dynamic CMDB for the cloud cloud vendors rather than operating in silos
which now gets tied to the revenue and having vendor dependency. Most of
outflow the vendors have their own offerings today
■ Handling complex interfacing handling and standards for cloud computing are
in the cloud taking shape with features that are suitable
■ Monitoring cloud resources usage to different set of organizations. There are
statistics and optimizing resource usage initiatives by organizations and groups in
that is extensible on the fly this direction to have defined processes and
■ Responsibility of ensuring a balance inter-operatibility between the cloud vendors
between resource demands and spending and this will largely shape the adoption of the
■ Handling a hybrid infrastructure mix cloud in the years to come.
(public, private and internal virtual
infrastructure) REFERENCES
■ Understand portability of the data being 1. Bernard Golden, Defining Private
hosted and alternate plans on getting Clouds, Part One, CIO. Available
off the cloud if service provider shuts at http://www.cio.com.au/
business [7]. article/304190/defining_private_
clouds_part_one
CONCLUSION 2. A m a z o n C l o u d w a t c h . A v a i l a b l e
Moving to a cloud requires small incremental at http://aws.amazon.com/
steps, proper planning, willingness from cloudwatch/
teams for adoption and a very strong senior 3. IT Service Management. Available at
management support. From a standalone http://www.itsm.info/ITSM.htm
infrastructure an organization can transition 4. Gartner Names Hyperic “Cool Vendor”
to a hybrid one targeting to completely be in Latest Research Report. Available
on cloud in the future. Moving towards the at http://www.hyperic.com/news/
cloud reduces the day-to-day IT operational releases/hyperic-named-cool-vendor-
management issues since they shift towards the by-gartner.html
vendors end. Yet the complexities of demand- 5. h t t p : / / w w w . n a g i o s . o r g / a b o u t /
cost management, application interfacing, features/
security and process management and overall 6. http://www.zenoss.com/product/
billing rests with the internal IT organization. network-management
As the cloud computing adoption 7. Mario Moreira, Infrastructure - on
unfolds, an organization may also prefer Premises or in the Clouds? Agile Journal,
87
April 2009. Available at http://www. column-articles/1469-infrastructure-
agilejournal.com/articles/columns/ on-premises-or-in-the-clouds.
88
THE LAST WORD
Cloud Computing —
A Disruptive Technology
In a freewheeling discussion on multiple facets of cloud
as a technology, business trend and other related issues
Dr. Srinivas Padmanabhuni poses a set of questions to
Jitendra Pal Thethi, Principal, Microsoft Technology
Centre and Raghavan Subramanian, AVP and Head of
Cloud Computing CoE at Infosys.
Srinivas: Raghu, what are your thoughts on because public cloud opens up the possibility
cloud as a disruptive technology? How is it to a lot of great software ideas that remained as
related to recent trends of grid, virtualization mere ideas so far to be realized now as software
and SOA? and disrupt well-entrenched software. There
Raghavan: The technologies that make cloud will also be a lot of disruption on how enterprises
possible have been knocking at the doors of want to develop new applications and maintain
the research community and the common man and enhance their existing applications.
for quite some time now. Cloud is not new to Infrastructure engineers will advocate IaaS,
scientific research where there is a need for developers will push the evolution of PaaS,
massive computational and storage capabilities. while business-stakeholders might be keen
Cloud is not new to consumer software business on SaaS. Depending on an organization’s IT
either given the fact that Google, Amazon, culture, its IT portfolio could be a mix of these
Facebook and several others have been early different cloud delivery models. Disruption can
adopters of cloud. However what is new here also happen if in the enterprise world - a new
are the technologies enabling these cloud application like Facebook becomes the preferred
capabilities now available to enterprises small platform to write enterprise or industry specific
and large alike. Viewed this way cloud is both applications. The possibilities are limitless but
evolutionary and disruptive. Evolutionary the hype generated is working against it.
because organizations can try to aggregate
their own computational, storage and network Srinivas: Jitendra, how do you think cloud
needs and use the abstraction provided by can help enterprises in contrast to the popular
virtualization to reap cost, scalability, time-to- SMB segment?
market and several other benefits. Disruptive Jitendra: The value proposition of cloud applies
89
more to an enterprise segment than to an SMB must also help to identify the SaaS potential
in many different ways. Cloud certainly helps to of some of the custom applications within an
translate the investments from capital expense enterprise and help them in re-architecting
to operating expense which is becoming a a custom application into a multi-tenant,
good financial lever for enterprises. More customizable, pay-as-you-go metered SaaS
importantly, with a cloud based infrastructure application.
the speed at which a solution can be brought
to market is radically fast. With cloud, the Srinivas: Any thoughts on relative penetration
lead time to get the infrastructure in place of cloud in different verticals?
is translated from weeks to minutes that Jitendra: The penetration of the cloud in
further help organizations in getting their new different verticals is truly workloads driven.
solutions and ideas roll out quickly. This small Verticals like life sciences where there is a large
change is a huge catalyst to innovation in an data processing and computation involved are
enterprise. Organizations can quickly pilot a leveraging cloud for scaled-out architecture.
solution, measure its effectiveness and use it Verticals like banking and financial industries
create competitive advantage. are coming toward approaching cloud from
a cost structure and resource optimization
Srinivas: What are the business opportunities perspective. In the area of manufacturing and
from an IT services perspective? retail the emphasis is more towards leveraging
Raghavan: Enterprises always look up to the publically available infrastructure and
IT service providers for trusted opinions scenarios of partner integration.
on anything new. Enterprises would like
consultants to help them with their cloud Srinivas: What have been the typical business
strategy and implementation. In order to drivers considering the early adopters we have
do this IT service providers must be able to worked with?
understand an enterprise’s organizational Raghavan: Once again, the term cloud is broad
culture - the ecosystem in which it operates as it includes public, private, community and
and its IT portfolio before adopting cloud hybrid models of cloud delivery. From the early
strategy. IT service providers also need to have adopters of public cloud we see a few patterns
a firm grasp on the various cloud offerings, the emerge. Testing and other environments that are
potential and the limitations of the technology. not required through the year are being created
Enterprises need to understand the issues on a need basis on the public cloud. One-off
one can face and acquaint themselves with computational needs, like data crunching,
limitations and workarounds to overcome them. archiving, etc., are also happening on public
IT service providers have the role of tempering clouds. Some of the new applications with less
the marketing hype and provide a balanced and emphasis on security are getting developed in
unbiased view of the relevance of cloud and its a public cloud. On the private cloud front it
adoption for an enterprise. IT service providers is primarily to do with virtualization of data
have the role of helping enterprises migrate centers. So the applications are being moved
relevant legacy applications and develop new away from deployment models where they
applications using cloud. IT service providers used to run on dedicated hardware. This further
90
means that independent software vendors to cloud is not simple with the existence
(ISV) support for virtualization is becoming of multiple vendors that make the overall
an important factor for the future of ISVs. The migration cost and time high and therefore not
business drivers for the public cloud are time- much practicable.
to-market, pay-as-you use, less in-house staff,
opex instead of capex, standardization of IT Srinivas: What are your thoughts on cloud
and the ability for IT department to embrace an standardization and interoperability?
imminent future paradigm among various other Raghavan: Let us look at the standardization
things. For private or internal cloud the business needs at the IaaS layer, since PaaS and SaaS
drivers are very similar to virtualization i.e., bring a different set of problems to the picture.
abstraction of hardware resources, elastic IaaS needs a standard to ensure that an
scalability (limited), improved utilization of application and its dependencies can be
hardware, etc. extracted and abstracted from its current
running environment and deployed on to
Srinivas: What are the business and technology any target virtual machine, private or public.
inhibitors/showstoppers when it comes to This problem arises because cloud-players
cloud adoption? are providing point-solutions that address
Jitendra: Most of the business challenges this problem only in the narrow context of
are associated with the lack of trust on an their self-interest. Open virtualization format
externalized infrastructure. Data privacy, (OVF) is trying to address this problem through
security and compliance are pressures which standards, while companies like AppZero are
do not allow use of a cloud infrastructure addressing this through their technologies
crossing global boundaries. Ambiguity on (Virtual Application Appliance). The second
SLA definition and lack of end-to-end SLAs for problem comes from the proprietary storage
business services do not leave businesses to take formats used by cloud-players that would mean
a calculated risk to grab the cost advantages. either vendor lock-in or the need to write data
Lastly, the penalties agreed upon by the cloud extraction logic for vendor specific storage
providers for SLA breach is disproportionate to technologies. This problem has to be solved
business loss caused due to a possible outage. either through standards or by innovative
From a technology standpoint, the technologies.
heterogeneity of managing different cloud
vendors lead to integration challenges and Srinivas: What are the key takeways for
management overheads. Today, only a handful systems integrators?
of software licenses from vendors are available Raghavan: System integrators with a close
on a pay-by-use model and hence are not proximity to customers have to play a crucial
available as part of cloud provider. Third, role in taking the benefit of cloud computing
many cloud providers impose new models of to the enterprise customers. However their
development like use of column database rather approach to cloud computing will determine
than relational database that further creates a if they continue to retain their proximity to the
lot of ambiguity and design problems. Lastly, enterprises. Let me illustrate this point with one
the migration from an on-premises application example. Consider SLAs. System integrators
91
have to decide whether they will bundle the consultant. He can be reached at Jitendra_Thethi@
infra-offering and front the SLAs to provide a infosys.com.
single-point-of-contact for customers or whether
they partner with infra-players and carve out Raghavan Subramanian is an Associate Vice
different realms of SLAs thereby leading the end- President and the head of Infosys Cloud Computing
customer to deal with multiple players. There center-of-excellence.
are multiple ways in which system integrators Raghu was in the recent past with the Infosys
can retain and increase their proximity with SaaS unit where he was product manager for SaaS
their customers by making software (whether it social media platform and program manager for
is running on IaaS or PaaS infrastructure or if it Infynity, the Infosys Social Network.
were already consumed as SaaS) the focus of the He advises many clients in the areas of
interaction with the enterprise customers. There information security, Enterprise Application
are new models like process-as-a-service (PraaS) Integration and Object-oriented programming.
where process services are bundled along with He has co-authored two patent applications,
software to provide a higher level of abstraction has written several white-papers and articles in peer-
to customers. reviewed journals and magazines.
92
Index
93
Platform as a Service, also PaaS 13, 20, 24, 47-48, Service Level Agreement, also SLA 15, 27, 32,
89, 91-92 37, 46, 63, 67, 71, 74, 83
Quality of service, also QoS 20, 28-29, 56, 59, 63 Software as a Service, also SaaS 4, 17, 20, 24,
Sales Performance Management, also SPM 65 34-38, 45, 47, 48, 55, 58, 60-63, 65, 67, 76, 89-92
Sensor Event as a Service, also SEaaS 34-38 Total Cost of Ownership, also TCO 3, 9, 56,
Service Administered Markup Language, 59-60, 62
also SAML 46-52 Unified Cloud Interface, also UCI 20-21, 25
Service Provisioning Markup Language, Wireless Sensor Network,
also SPML 46, 52-53 also WSN 33-39
94
SETLabs Briefings
BUSINESS INNOVATION through TECHNOLOGY
Graphics & Web Editors SETLabs is an important part of Infosys’ commitment to leadership
Ankur Madnani in innovation using technology. SETLabs anticipates and assesses the
Srinivasan Gopalakrishnan
evolution of technology and its impact on businesses and enables Infosys
Program Manager to constantly synthesize what it learns and catalyze technology enabled
Abhoy K Jha
business transformation and thus assume leadership in providing best
IP Manager of breed solutions to clients across the globe. This is achieved through
K V R S Sarma
research supported by state-of-the-art labs and collaboration with industry
ITLS Manager leaders.
Ajay Kolhatkar PhD
Marketing Manager Infosys Technologies Ltd (NASDAQ: INFY) defines, designs and delivers
Pavithra Krishnamurthy
IT-enabled business solutions that help Global 2000 companies win in a
Production Manager flat world. These solutions focus on providing strategic differentiation
Sudarshan Kumar V S and operational superiority to clients. Infosys creates these solutions
Distribution Managers for its clients by leveraging its domain and business expertise along
Santhosh Shenoy with a complete range of services. With Infosys, clients are assured of a
Suresh Kumar V H
transparent business partner, world-class processes, speed of execution
and the power to stretch their IT budget by leveraging the Global Delivery
How to Reach Us:
Email: Model that Infosys pioneered. To find out how Infosys can help businesses
SETLabsBriefings@infosys.com achieve competitive advantage, visit www.infosys.com or send an email to
Phone:
+91-40-67048455
infosys@infosys.com
Post:
SETLabs Briefings,
B-19, Infosys Technologies Ltd.
© 2009, Infosys Technologies Limited
Electronics City, Hosur Road,
Infosys acknowledges the proprietary rights of the trademarks and product names of the other companies
Bangalore 560100, India
mentioned in this issue. The information provided in this document is intended for the sole use of the recipient
and for educational purposes only. Infosys makes no express or implied warranties relating to the information
Subscription: contained herein or to any derived results obtained by the recipient from the use of the information in this
setlabsbriefings@infosys.com document. Infosys further does not guarantee the sequence, timeliness, accuracy or completeness of the
information and will not be liable in any way to the recipient for any delays, inaccuracies, errors in, or omissions
Rights, Permission, Licensing of, any of the information or in the transmission thereof, or for any damages arising there from. Opinions and
and Reprints: forecasts constitute our judgment at the time of release and are subject to change without notice. This document
praveen_malla@infosys.com does not contain information provided to us in confidence by our clients.
NOTES
Authors featured in this issue
ADI MALLIKARJUNA REDDY V
Adi Mallikarjuna Reddy V is a Junior Research Associate in SETLabs, Infosys. He can be contacted at
Adi_Vanteddu@infosys.com.
AJIT MHAISKAR
Ajit Mhaiskar is a Principal Technology Architect with the Manufacturing business unit of Infosys. He can be
reached at Ajit_Mhaiskar@infosys.com.
AMIT WASUDEO GAWANDE
Amit Wasudeo Gawande is a Technology Lead at SETLabs, Infosys. He can be contacted at Amit_Gawande@
infosys.com.
ANJANEYULU PASALA
Anjaneyulu Pasala PhD is a Senior Research Scientist at SETLabs, Infosys. He can be reached at Anjaneyulu_
Pasala@infosys.com.
ANU GOPALAKRISHNAN
Anu Gopalakrishnan was a Product Technical Architect in SETLabs, Infosys.
ASHEESH CHADDHA
Asheesh Chaddha is a Project Manager with Performance Engineering and Enhancement practice, Infosys.
He can be reached at Asheesh_Chaddha@infosys.com.
ASHUTOSH AGARWAL
Ashutosh Agarwal is a Senior Project Manager and certified PMP with Infosys. He can be contacted at
ashutosha@infosys.com.
A V PARAMESWARAN
A V Parameswaran is a Senior Technology Architect with Infosys Performance Engineering and Enhancement
practice. He can be reached at Parameswaran_AV@infosys.com
BHAVIN JAYANTILAL RAICHURA
Bhavin Jayantilal Raichura is a Senior Technology Architect with the Manufacturing unit of Infosys. He can be
contacted at Bhavin_Raichura@infosys.com.
DEEPAK JOHN
Deepak John is a CRM Consultant with the Enterprise Solutions business unit at Infosys. He has a keen interest
in emerging technologies and can be reached at deepak_john@infosys.com.
GANESAN PANDURANGAN
Ganesan Pandurangan is a Senior Technology Architect working with System Integration unit of Infosys.
He can be contacted at Ganesan_Pandurangan@infosys.com.
KAUSTUBH VAIDYA
Kaustubh Vaidya is a Project Manager for shared database services with the Infrastructure Management
Services unit of Infosys. He can be reached at Kaustubh_Vaidya@infosys.com.
KUMAR PADMANABH
Kumar Padmanabh PhD is a Research Scientist and leads the wireless sensor networking (WSN) research group
at SETLabs, Infosys. He can be contacted at Kumar_Padmanabh@infosys.com
NIDHI TIWARI
Nidhi Tiwari is a Senior Technical Architect with SETLabs, Infosys. She can be reached at nidhi_tiwari@
infosys.com.
RAHUL BAKHSHI
Rahul Bakhshi is a Consultant with the Enterprise Solutions business unit of Infosys. He can be contacted at
rahulsantosh_b@infosys.com.
SHYAM KUMAR DODDAVULA
Shyam Kumar Doddavula is a Principal Technology Architect at SETLabs, Infosys. He can be reached at
shyamkumar_d@infosys.com.
SIVA PRASAD KATRU
Siva Prasad Katru is a Junior Research Associate in SETLabs. He can be contacted at SivaPrasad_Katru@
infosys.com.
SRINIVAS PADMANABHUNI
Srinivas Padmanabhuni PhD is a Principal Research Scientist in SETLabs, Infosys. He can be reached at
srinivas_p@infosys.com.
SUDEEP MALLICK
Sudeep Mallick PhD is a Principal Research Scientist with Infosys SETLabs. He can be contacted at
sudeepm@infosys.com.
SUMIT KUMAR BOSE
Sumit Kumar Bose PhD is a Senior Research Associate with Distributed Computing Lab of SETLabs. He can
be reached at Sumit_Bose@infosys.com.
Subu Goparaju “At SETLabs, we constantly look for opportunities to leverage
Vice President
technology while creating and implementing innovative business
and Head of SETLabs
solutions for our clients. As part of this quest, we develop engineering
VOL 7 NO 7 2009
methodologies that help Infosys implement these solutions right first
clouc computing
For information on obtaining additional copies, reprinting or translating articles, and all other correspondence,
please contact:
Telephone : +91-40-67048455
Email: SetlabsBriefings@infosys.com