Business Models for Cloud-based High Perfomance Computing
Service Provision
Insights from the Swiss Higher Education Sector
Markus Eurich and Roman Boutellier
D-MTEC, ETH Zurich, Scheuchzerstrasse 7, 8092 Zurich, Switzerland
Keywords: Business Model, Cloud Computing, High Performance Computing, Higher Education, Public Sector.
Abstract: In 2013, the Swiss Federal Institute of Technology in Zurich (ETH Zurich) and the University of Zurich
jointly set up a Cloud stack in order to experiment with high performance computing (HPC) service
provision in higher education institutions and its corresponding business model alternatives. This project
demonstrated that Cloud-based service provisioning is possible for HPC and can be applied to big data
problems as well. On this basis and against the background of new public management reforms, this study
aims to foster the understanding of the business model aspects: value proposition and revenue mechanisms.
Therefore, 14 interviews were conducted on the potential use of Cloud HPC services and revenue
mechanisms. The results show that HPC service providers appreciate Cloud computing providing shorter
time to service and more customized services; and eventually becoming more transparent and efficient, i.e.
complying with new public management concepts. However, the service consumers do not see a real need
to consume Cloud-based services as there is hardly any "Cloud-only" application at the moment. Finally, the
three revenue mechanisms ‘pay per use’, subscription, and ‘pay for a share’ are discussed.
1 INTRODUCTION
The emergence of Cloud computing contributed a
great deal to the digitization of the public sector
(e.g., Chandrasekaran and Kapoor, 2011; Kundra,
2010; Lifka et al., 2013). In 2011, Norwich
University's College of Graduate and Continuing
Studies conducted a survey on Cloud computing
among government and higher education institution
professionals: the results show that almost half of
the respondents indicated that their organizations are
in the process of implementing Cloud computing
services (Norwich University, 2011). There are
already some successful cases of Cloud computing
in the public sector: Australia’s national science
agency virtualized its business applications so that
they can be managed and shared across all its
locations; the regional government of Castilla in
Spain is using Cloud-based services to accelerate the
rollout of e-government applications for taxes and
driving licenses; and the Chinese University of Hong
Kong centralized its data center and network
resources on a private Cloud platform (Macias and
Greg, 2011). However, the public sector is still
significantly lagging behind the private sector in
terms of Cloud deployments (Baldwin, 2012). In a
2011 study on the future of Cloud computing in the
public and private sectors, over 1,500 interviews
were conducted with professionals from
organizations in Europe, North America, and Asia.
The interviews showed that only 23% of public
sector organizations are using Cloud-based hosted
data or remotely hosted apps compared to 42% of
the organizations in the private sector. The study
indicates that European organizations are
particularly slow in adopting Cloud services and
appear to be behind Asian and US organizations
(Red Shift Research, 2011).
This study now aims to provide insights on, and
discusses some implications of, the use and
implementation of Cloud computing in the European
higher education sector as part of the broader public
sector. The goal is to foster the understanding of
business models for Cloud-based high performance
computing services in higher education.
Before the research questions are given, the
terms “Cloud computing”, “high performance
computing”, and “business model” are defined for
this paper.
101
Eurich M. and Boutellier R..
Business Models for Cloud-based High Perfomance Computing Service Provision - Insights from the Swiss Higher Education Sector.
DOI: 10.5220/0005050801010110
In Proceedings of the 11th International Conference on e-Business (ICE-B-2014), pages 101-110
ISBN: 978-989-758-043-7
Copyright
c
2014 SCITEPRESS (Science and Technology Publications, Lda.)
Cloud Computing refers to the use of computing
resources, which are available in a remote location
and accessible over a network. Therefore, the Cloud
is an operational model, a usage model, and a
business model. Cloud computing services are often
divided into three layers: software as a service
(SaaS), platform as a service (PaaS), and
infrastructure as a service (IaaS) (Eurich,
Giessmann, Mettler, and Stanoevska-Slabeva, 2011;
Vaquero, Rodero-Merino, Caceres, and Lindner,
2009; Weinhardt et al., 2009).
A Business Model describes an organization’s
value creation, proposition, and capture (McGrath,
2010; Osterwalder and Pigneur, 2010; Teece, 2010).
Value creation includes resources, activities, and
business partners. Value proposition refers to the
benefits that a customer can expect from the product
or service. Value capture is comprised of revenue
streams and pricing (Osterwalder and Pigneur, 2010;
Timmers, 1998). In this study, value capture focuses
on the revenue mechanism (cf. e.g., Hedman and
Kalling, 2003; Johnson, Christensen, and
Kagermann, 2008).
High Performance Computing (HPC) refers to
all computations that need high processing power or
memory capacity. HPC uses resources that are
optimized for a massive parallel workload
computing in the back-end. The HPC service
consumer typically interacts with the HPC resources
only via a front-end system (Calleja et al., 2010;
Reuther and Tichenor, 2006).
HPC assists researchers in solving complex
problems in a variety of different areas like weather
forecasts, earthquake simulations, biomedicine,
nanotechnology, materials science, environmental
modeling, and disaster simulation (Calleja et al.,
2010). This selection of HPC applications shows
that HPC can be important for the public sector.
However, deploying and maintaining HPC resources
is expensive and knowledge-intense. As most public
HPC services are consumed by members of higher
education institutions and as they also typically
possess the necessary knowledge and experience,
many higher education institutions run their own
HPC infrastructure. Today, HPC service
provisioning is almost exclusively organized at the
institutional level. Even though most higher
education institutions in Switzerland are directly, or
at least indirectly, controlled by the government and
paid for with tax money, they cannot provide each
other with HPC services. This can be a problem
when HPC infrastructures are specialized for
specific purposes: a researcher from institution A
cannot use the service from institution B although
institution B possesses the most adequate HPC
resources for A’s problem. The leading higher
education institutions’ resources are the most
exhaustive while some smaller universities cannot
afford any HPC resources at all. However, they
cannot just use or buy some of the resources that
belong to other higher education institutes. One
reason is that the institutional and funding structures
are very heterogeneous. There are two additional
major obstacles: first, the value proposition of
services provided via a Cloud solution compared to
the traditional way of service provisioning is
unclear. Second, there is no pricing mechanism to
charge other institutions. In order to discuss the first
obstacle, we describe the case of tests on a private
Cloud infrastructure that were conducted in the
course of the Swiss Academic Compute Cloud
project (Kunszt et al., 2013); to discuss the latter
obstacle, revenue mechanisms are discussed in this
study. The need for reasonable revenue mechanisms
will become increasingly important in the course of
new public management reforms. Higher education
institutions are typically public, non-profit
organizations. However, new public management
reforms require these institutions to adopt for-profit
management concepts in order to have higher
accountability, transparency, and efficiency (De
Boer, Enders, and Leisyte, 2007; Schubert, 2009). It
may be reasonable to assume that higher education
institutions could provide Cloud-based HPC services
to other public sector institutions in the long run. In
this way, the higher education institutions could
provide services to, e.g., biomedicine or disaster and
earthquake simulations, public organizations like
hospitals or regional governmental institutions.
We use two preconditions for the study: first,
Cloud-based service provisioning is possible for
HPC. For instance, SGI Cyclone is a supercomputer
on demand that provides elastic, scalable, and cost
transparent services and that gives a service
consumer immediate access to the resources and
computing capabilities (SGI, 2014). Some Clouds
support virtual machines that have several hundred
cores and a considerable amount of memory.
Second, the Cloud model can be applied to big data
problems as well. If a remote Cloud should be used,
the data transfer might be cumbersome or even
prohibitive, but a local Cloud can deal with large
data volumes or can even be explicitly designed in
order to manage big data problems through Hadoop
(Kunszt et al., 2014).
The business model part “value creation” is of
rather technical nature and is presented in detail in
Kunszt et al. (2013; 2014). Pilot tests were
ICE-B2014-InternationalConferenceone-Business
102
conducted in a private Cloud, which is a Cloud
infrastructure that is operated for a single
organization. In the course of the Swiss Academic
Compute Cloud project (Kunszt et al., 2013; 2014),
more sophisticated tests followed in a hybrid Cloud,
which is a composition of two or more (in our case
private) Clouds, which remain distinct entities but
are bound together (cf., Mell and Grance, 2011;
Sotomayor, Montero, Llorente, and Foster, 2009).
The Swiss Academic Compute Cloud project has
access to OpenStack Cloud installations at ETH
Zurich, the University of Zurich, SWITCH, and the
Zurich University of Applied Sciences (Kunszt et
al., 2013; 2014), which manage the Cloud
installations as self-run IT centers. As the Cloud
installations are operated by self-run IT centers, they
are capital intensive and a reasonable life-cycle
management must be applied to keep the IT
infrastructure up-to-date (Eurich, Tahar, and
Boutellier, 2011; Eurich, Calleja, and Boutellier,
2013). The current Clouds in the project are
relatively small: they range from 100 to 400 Central
Processing Unit (CPU) cores each, but are equipped
with quite decent memory and storage. The choice
of OpenStack as a reference Cloud software stack
has emerged from an evaluation done in the Swiss
Academic Compute Cloud project (Kunszt et al.,
2013).
The relation between technological innovations
and management decisions is complex, but both
aspects must be aligned and they can iteratively
influence each other (Hedman and Kalling, 2003):
the same technology commercialized in different
ways may result in different economic outcomes
(Chesbrough, 2010). Against the technological
background (Kunszt et al., 2014), this paper is
dedicated to the managerial aspects. The research
questions focus on two major business model parts:
value proposition and revenue mechanism.
Value proposition: What are the benefits of
scientific Cloud-based HPC services?
Revenue mechanism: How can Cloud-based
HPC services be priced?
These questions are discussed for both the service
consumer and the service provider.
To this end, this article is structured as follows.
The next section clarifies the research methodology.
The following two sections describe the results of
our study and are structured in accordance with the
research questions: value proposition (section 3) and
pricing mechanisms (section 4). After these
descriptions, the results are discussed in section 5.
The paper concludes with a summary and an outlook
on future research.
2 METHODOLOGY
The aim of this study is to analyze two major
business model components: value proposition and
revenue mechanism for Cloud computing services
that are meant to be jointly provided by and
accessible to several higher education institutions.
The study was conducted as part of the Swiss
Academic Compute Cloud project (Kunszt et al.,
2013).
For the analyses of value proposition and
revenue mechanism, the study was based on an
inductive qualitative research design (Bryman and
Bell, 2007; Creswell, 2013). Information was
gathered by means of semi-structured interviews.
Eight interviews were conducted with academic
service consumers, who are currently using HPC
service that are provided in a traditional manner.
They are the potential buyers of Cloud-based HPC
services and are in charge of making investments
and taking decisions. Six interviews were conducted
with service providers at ETH Zurich and at the
University of Zurich. Additional information was
used, which was collected from interviews with
representatives from the Swiss National
Supercomputing Centre, the Swiss Initiative in
Systems Biology (System X), and the Friedrich
Miescher Institute.
In a previous study (Eurich, Calleja, and
Boutellier, 2013), revenue streams of HPC services
were analyzed; and as a project-internal pre-
assessment step, three revenue mechanisms were
identified as being acceptable in terms of economic
sustainability and convenience: ‘pay per use’,
subscription, ‘pay for a share’.
The interviews with scientific service consumers
at ETH Zurich and at the University of Zurich were
all conducted in 2013; six were conducted face-to-
face and two via phone. Information was gathered
by interviewing research groups, which are currently
using central computing services. The interviewed
service consumers were asked what they would use
the computing capacity for; how they perceive the
pricing approaches from a service consumer
perspective ('pay per use', subscription, 'pay for a
share'); and what advantages and disadvantages they
expect from these approaches.
The interviews with the service providers at ETH
Zurich and at the University of Zurich were also
conducted in spring 2013: four were conducted face-
to-face and two by phone. These interviews also
followed an interview guide. The service providers
were asked in what way Cloud Computing resources
could improve service provisioning, how they
BusinessModelsforCloud-basedHighPerfomanceComputingServiceProvision-InsightsfromtheSwissHigher
EducationSector
103
perceive the pricing approaches from a service
provider perspective ('pay per use', subscription, 'pay
for a share'), and what are advantages and
disadvantages they expect from these approaches.
Obviously, some questions were the same for
both the service consumers and providers. This was
done on purpose in order to reveal a potential
difference in perception between consumers and
providers.
The data gathered from the interviews was
analyzed using open, axial, and selective coding
techniques (Urquhart, 2001). The extracted key
statements and assertions were then grouped along
the research questions and interviewee categories
(service consumers, service providers, HPC experts),
which resulted in a grid that allowed to identify
patterns (Campbell, 1966).
3 VALUE PROPOSITION
Several studies have analyzed the value propositions
of Cloud computing in the public sector (e.g., Forst
& Sullivan, 2011; Kundra, 2010; Macias and Greg,
2011). The identified benefits of Cloud computing
services for the public sector include amongst
others: simple scalability, labor optimization, capital
expenditure reduction, fast deployment, assured
service levels, access to up-to-date technology, and
reduced maintenance effort. However, it is
noticeable that the identified value proposition for
the public sector is more or less the same as for the
private sector (Baldwin, 2012).
With a particular focus on scientific Cloud
applications, a large-scale user survey revealed
several benefits of Cloud computing services like
computing elasticity, data elasticity, and rapid
prototyping (Lifka et al., 2013). Like the more
general studies on Cloud computing for the public
sector, this survey does not differentiate between
advantages for service consumers and for service
providers. A reason for this lack of discrimination
might be ascribed to the issue that public sector
institutions are typically only perceived as service
consumers of Cloud computing services.
In our case, however, the public sector
organizations are not only service consumers, but
also service providers. Therefore, we aimed to gain
insights into service consumers' (3.1) and service
providers' (3.2) perception of Cloud computing
benefits.
3.1 Service Consumer Perspective
Service providers need to understand the needs of,
and the number of potential service consumers. The
service consumers reported that they could mainly
use Cloud computing services for:
Testing and experimenting: So far, academic
service consumers see the major benefit of
Cloud service in conducting tests and
experiments on the Cloud infrastructure. In
this way, they would use Cloud services only
in a pre-phase of an actual research project.
With the tests, service consumers aim to
produce preliminary results that they can use
to write a fact-based research proposal in
order to get a grant to buy their own
infrastructure.
Training for students: In a scientific context,
senior staff is often not very pleased to see
juniors and students experimenting with their
high-end, sometimes fragile IT infrastructure.
Therefore, they would appreciate Cloud
computing services that are totally separated
from the operational computing resources.
Students could use this test environment in the
Cloud to gain some experience. For the same
reason, workshops and classes for students
could also be conducted on Cloud resources.
Cloud resources are particularly useful and
convenient when workshops take place
infrequently; in this case the teacher does not
need to spend much time for setting up a test
and demonstrating the IT environment.
Some special applications: Only a few
consumers see a need for Cloud computing
resources for particular applications. Cloud
applications mentioned in the interviews
include medical IT services, like on-the-fly
services during surgery or ultrasound image
recognition, or some sort of easy simulations.
Storage: Scientific service consumers are
particularly interested in Cloud-based storage
services, which would allow them to access
their data whenever and from wherever they
want. However, they would very much
appreciate a trustworthy and reliable European
storage service. Trustworthiness and reliability
are demanded because the users are worried
about their sensitive data. They fear a
potential data and knowledge leakage as well
as being spied on. Recent laws and
regulations, like the US Patriot Act, spurred
further unease and uncertainty among the
academic users.
ICE-B2014-InternationalConferenceone-Business
104
3.2 Service Provider Perspective
The results of the interviews show that the picture is
in fact different when you ask service providers
about the usefulness of Cloud services. Regardless
of the Cloud computing layer, they expect that
Cloud computing could deliver the following value
to the consumers:
Flexibility: By making use of a Cloud-based
service provision approach, the service
provider can decouple the service provisioning
and the actual hardware infrastructure that the
services depend on. The provider can develop
the underlying infrastructure with no or very
little impact on the services provided, which
results in a much higher service availability as
well.
Provisioning of additional services: Cloud
resources would give service providers the
opportunity to react faster to consumers’
demands and also to provide services that are
currently not in the provider's service
portfolio. With Cloud computing, service
providers could draw on ready-made solutions
from the Cloud, assembling their services
based on customer demand. For example,
applications that run only on certain operating
systems that are currently not supported can
be added with ease.
Time to service: Cloud resources have the
potential of improving the time to service for
the users. Users could get the service
immediately. Currently, the users have to wait
weeks or even months until the service is set
up, tested, and ready to be consumed. When
users require a specific service, it can take the
service provider a long time to provide it,
particularly when additional resources need to
be procured. Cloud computing could help to
bridge the time until the service is ready.
Self-serving aspect and increased automation:
Cloud computing solutions could be provided
directly to the users. At least some
experienced users could consume the services
from the Cloud, which could increase the level
of automation and reduce the time and money
for administrative overhead.
Elasticity: Service providers would be given
the chance to define the capacity of services
provided in accordance with the actual
demand in a short amount of time. Cloud
computing enables a dynamic scaling up and
down; when less capacity is needed or service
consumers do not require a particular service
anymore, the Cloud should provide the option
to give the capacity and resources back when
not needed.
Balance workload: Linked to the option to pay
for services only on demand, the workload
could be balanced better, i.e., Cloud resources
can be used for topping up capacity and for
boosting the capacity in times of peak demand
in the short run.
4 REVENUE MECHANISMS
In a previous study (Eurich et al., 2013), revenue
streams of Cloud computing services were analyzed;
in a project-internal pre-assessment three revenue
mechanisms have been identified as being
acceptable in terms of economic sustainability and
convenience:
'Pay per use': Service consumers are charged
a fee according to the time and volume of a
computing service that has been consumed.
Subscription: The service consumer pays a fee
on a regular basis for the usage of a service.
Subscriptions allow services to be sold in
bundles.
'Pay for a share': Service consumers buy a
share in order to get a corresponding amount
of service.
Both, service consumers (4.1) and service
providers (4.2) were asked for an assessment of the
three revenue mechanisms, 'pay per use',
subscription, and 'pay for a share'.
4.1 Perception of Revenue Mechanisms
by Service Consumers
4.1.1 Pay per Use
Pro
Service consumers appreciate the 'pay per use'
revenue mechanism as a fair approach. Costs are
transparent and you only pay for what you get.
Service consumers could imagine to 'pay per use' for
some kind of small jobs, like for testing or
experimenting with a service. Especially when a
research activity contains uncertainties, service
consumers do not want to spend their money
upfront. In this case, they neither want to buy a
package that lasts for a month (like in the
subscription model) nor do they want to spend
money upfront (like in the 'pay for a share' scheme).
The 'pay per use' scheme gives them the freedom to
initiate research activities and to immediately quit
BusinessModelsforCloud-basedHighPerfomanceComputingServiceProvision-InsightsfromtheSwissHigher
EducationSector
105
the ones they do not want to pursue any longer.
Service consumers perceive the possibility of
terminating the consumption and the payments of a
service advantageous to other schemes, e.g., the 'pay
for a share' scheme. They do not want to subsidize
other users due to being locked in a contract when
they do not need any further services. Finally, the
'pay per use' revenue mechanism sounds particularly
attractive to service consumers, who develop
applications themselves, i.e., who operate more on
the PaaS or IaaS layer. These researchers often come
from computer science or physics departments. The
'pay per use' model would give them the opportunity
to flexibly and quickly make use of just the services
they need.
Contra
On the other hand, service consumers do not want to
be bothered with billing and accounting. They do not
want to spend time on administrative tasks like
checking invoices and accounting activities. The
'pay per use' revenue mechanism is perceived as
expensive and not paying off for a longer time of
service consumption. Finally, academic service
consumers are concerned about acquiring money. In
the current academic budget allocation model it is
almost impossible to get money to be spent on a 'pay
per use' basis. Researchers cannot ask for money if
they do not know on what it will be spent. Currently,
they can only successfully receive grants when they
submit a proposal for funding in which they ask for
money that will be spent on a specific hardware or to
buy a share.
4.1.2 Subscription
Pro
Academic service consumers acknowledge
'subscription' to be fair. Like the 'pay per use'
scheme it is easy to understand and you pay for what
you get. Some interviewees intuitively liked it
because it is also the way they pay for their private
mobile or smart phone services and Internet
connections. They like the idea of having some
predefined packages from which they can choose
and would be fond of having the option to upgrade
to another package if necessary. Service consumers
consider this a flexible and easy (because there are
predefined packages) approach.
Contra
The same arguments as for the 'pay per use' scheme
hold also true for the subscription approach. There
would be fewer invoices and accounting activities
than in the 'pay per use' scheme, but it would still be
less convenient than the 'pay for a share' solution.
Problems with the current academic budget
allocation model would also arise when applying the
subscription revenue mechanism.
4.1.3 Pay for a Share
Pro
Among our three revenue mechanisms, the 'pay for a
share' is perceived as the most convenient by
academic service consumers. They only need to pay
once and then are all set for the next couple of years.
Researchers can focus on their actual research and
do not have to care about comparing prices, squaring
accounts, and other kinds of accounting tasks. An
interviewee stated, “maybe in the end it does not pay
off, but despite that you pay for convenience. You
do not need to pay every time you need a service. If
you have some capacity left you can do some extra
research”. Moreover, it is well aligned with current
academic budget allocation models and, therefore, it
might be relatively easy to get money for buying a
share for services in a Cloud stack.
Contra
Some of the arguments that favor the 'pay per use'
and the subscription approach can be interpreted as
disadvantages of the 'pay for a share' approach. For
example, a service consumer mentioned, “if you do
not make full use of your share, you subsidize others
and waste money for something that you do not
need”. In addition, there might be a loss in flexibility
to change the volume or the kind of service
consumption compared to the 'pay per use' and the
subscription approach. Finally, the renewal of a
share might be a problem because this means a
massive investment at one time.
4.2 Perception of Revenue Mechanisms
by Service Providers
4.2.1 Pay per Use
Pro
First, the providers of academic services
acknowledge that the ‘pay per use’ scheme could
provide an added value for their service users. They
find the ‘pay per use’ scheme a feasible approach
when users do not know their demand and when
service consumption is only required for a few
weeks. In the communication with the service
consumer, this approach is probably the easiest to
explain because everyone basically knows how it
works. Since it is very transparent what is provided
ICE-B2014-InternationalConferenceone-Business
106
and what is used, reporting and accounting is also
easily understandable.
Second, an advantage for the service providers
themselves could be a better handling of spare
capacity. This approach could be particularly
interesting in case of external organizations’
demand.
Contra
A disadvantage of the ‘pay per use’ scheme is that
service consumption is highly unpredictable. A
service provider mentioned that “if you want to give
your consumers the choice to consume a service
whenever they want, service delivery becomes
difficult to plan and schedule”. In the scientific
environment, the situation is probably even more
unpredictable than on the free market. Unlike the
free market, at a university with a self-run IT center
you may only have a limited number of users, which
may have a demand at the same time, e.g., right after
the examination period or at the beginning of a new
term. An interviewee described that “on the free
market, you may have a higher number of users
from different regions and with different needs, and
as a result demand might be better balanced. At the
university you need to invest a lot upfront into
hardware if you want to offer your consumers an
immediate response to their service requests”.
However, due to the high unpredictability, there is
no guarantee that the investment is amortized.
Like the service consumers, the providers also
assume an extra effort for billing and accounting.
Some algorithms need to be set up to monitor the
users’ consumption and to calculate the price of it.
However, once done successfully, billing and
monitoring could be automated and there is no extra
effort for this anymore.
4.2.2 Subscription
Pro
Compared to the ‘pay per use’ scheme, the
subscription model is characterized by a more stable
and predictable source of income. This helps service
providers in calculating and forecasting revenues
and expenses and provides them with a long-term
frame for planning, adjustments, and procurements.
Contra
The subscription scheme features some
unpredictability about users’ demand in terms of
quantity and type of services, which is a problem for
higher education institutions with capital intensive
self-run IT centers. Especially if users are given the
opportunity to flexibly up- or downgrade to another
package, some big upfront investments are necessary
to guarantee a high level of service availability and a
reduced service time. Another problem is that
service providers of higher education institutions
typically cannot move money from one year to the
next; this means that even if a massive increase in
service consumption is predicted, the service
provider can only invest in its IT infrastructure or
service portfolio once it has got the money.
4.2.3 Pay for a Share
Pro
The ‘pay for a share’ scheme is in line with the
current budget allocation model. In addition, service
providers appreciate that this model gives
shareholders a sense of ownership and community.
Compared to the ‘pay per use’ and the subscription
scheme, the ‘pay for a share’ approach guarantees a
certain degree of income that lasts for three to four
years, the typical life cycle of hardware.
Contra
On the downside, there is no transparent service fee.
It is not intuitively clear what and how much a
consumer gets. The service consumer may get
confused what the share is worth: it can be time on
the Cloud stack, performance, or another service
(like storage, consulting). The ‘pay for a share’
approach is subject to unpredictability. This
approach is particularly prone to drop outs. Each
shareholder contributes a considerable amount of
money at the beginning of a long-term shared
ownership. This initially paid sum is higher than a
monthly fee in the subscription model or a daily fee
in the ‘pay per use’ approach, because it is meant to
last for a much longer time; as a consequence, this
sum for the share can have quite some impact on the
service portfolio or on the Cloud infrastructure. One
problem is that a major chunk of revenues comes
from new professors, who may get a share as bonus
for joining the new university. However, it is quite
unpredictable whether they are able to pay for a
renewal once the lifecycle of the hardware is over.
5 SUMMARY AND DISCUSSION
Cloud computing has emerged as new way to
digitize the public sector. However, the public sector
lags considerably behind the private sector in terms
of Cloud deployments. To facilitate the Cloud-
enabled digitization of the public sector, research is
needed on both the service consumption and the
provisioning side.
The perceived advantages and disadvantages of
different models of Cloud-based HPC service
consumption/delivery are summarized in Table 1.
BusinessModelsforCloud-basedHighPerfomanceComputingServiceProvision-InsightsfromtheSwissHigher
EducationSector
107
Table 1: Perception of different models of Cloud-based
HPC service consumption/delivery.
pay per use subscription pay for a
share
Service
consumers
+ trans-
parent cost
– admini-
stration
+ fair and
well known
– budget
allocation
+ con-
venience
– subsidy
of other
users
Service
providers
+ easy to
sell
– not
predictable
+ stable
income
– big
upfront
investment
+ long-term
income
– renewal
of hardware
On the service consumption side, the interviews
show that scientific service consumers want to focus
on their research tasks. They expect an easy
approach and want the higher education institution
to clear obstacles from their paths. The service
consumers currently perceive Cloud services as a
playground for testing, experimenting, and training
students. Up until now, most scientific service
consumers could not see any useful Cloud
computing applications. Indeed, a really convincing
Cloud application that could make scientific
consumers move to the Cloud is missing. As there is
no "Cloud-only" application at the moment,
consumers do not see a real need to consume Cloud-
based services instead of traditional services.
Decisions are based on beliefs not on facts. “Cloud
computing doesn’t pay off for us,” reported an
interviewed service consumer, and several other
HPC service consumers join him in the preception
that Cloud computing services are too expensive for
the amount of CPU they need. Some service
consumers reported that they would use Cloud
services if they come at a reasonable price. In the
end, they care little about pricing strategies. The
three revenue mechanisms, ‘pay per use’,
subscription, and ‘pay for a share’, are perceived
differently among the interviewees and there is no
clearly preferred revenue mechanism. Many service
consumers do not know much about Cloud
computing and, in fact, do not even care much about
what computing resources are used and how they are
priced. An interviewed service consumer put it
straight: "I just need something powerful to run”
[my computations on].
On the service provisioning side the interviews
revealed that the service providers tend to perceive
Cloud-based services as an additional resource in the
short run, but not as a replacement of traditional
HPC service provision. A service provider assumes
that “Cloud provisioning is [currently] most useful
when a user needs 10,000 or more processors for
only a week to a maximum of three months”.
Service providers appreciate Cloud-based HPC
service provisioning for its flexibility, its elasticity,
its self-serving aspect accompanied with an
increased automation, the chance to provide
additional services, the potential to shorten the time
to service, and the opportunity to balance the
workload.
This study focuses on the private and hybrid
Cloud services provisioning of self-run IT centers of
higher education institutions. Recently, ETH Zurich
carefully opened up towards the public Cloud in
times of peak consumption even though there are
concerns about confidentiality and security.
Consuming additional computing power from the
public Cloud relieves the pressure from university IT
centers’ decision makers to purchase equipment and
manage the infrastructure. In case of high usage
unpredictability, decisions can be postponed. This
development promotes the pay-per-use pricing
scheme, which is particularly burdened with the
disadvantages of usage unpredictability in the case
of self-run private Cloud centers.
6 OUTLOOK AND CONCLUSION
Currently, Cloud-based HPC services are actually
not necessary from a service provisioning
perspective because powerful private infrastructures
already exist. Therefore, there is a lack of motivation
to establish and invest into Cloud-based HPC
services. It remains to be seen to what extent Cloud-
based HPC service provision can reap the benefits
that service providers expect. However, the Cloud is
real; it is here and it is growing. Higher education
instutions might be well advised to at least gather
some experiences with the Cloud because the IT
infrastructure has become an essential requirement
in attracting the best reseachers (Drucker, 2002). Not
uncommonly, applications only emerged years after
a new technology was introduced, e.g., computer
simulations (Drucker, 1999). We assume that
currently Cloud computing is only a means to
optimize service provisioning (cf., Pring, 2010)
while truely innovative applications on the basis of
the Cloud may only emerge later.
In order to reap optimization benefits,
government bodies must put some incentives in
place or enforce public institutions to move to the
Cloud by the means of new public management
reforms. However, the realization of cost and service
advantages of the Cloud requires a holistic approach.
Training has to be provided to both the service
consumers and the service providers. In addition,
government bodies need to support different pricing
ICE-B2014-InternationalConferenceone-Business
108
schemes. For example, at the moment it is almost
impossible to get a grant for a research project that is
based on a per-pay-use or subscription approach.
Finally, there needs to be some regulations, which
control the exchange of computing services for
money among public institutions. To conclude,
Cloud-based service provisioning is most
advantageous at an organizational level, but the
realization and acceptance depends on the
involvement and support from government bodies
and the service consumers.
This study sheds light on both the service
consumers’ and the service provider’s opinion on the
value of Cloud-based service provision. However,
the findings are limited to insights that contribute to
facilitate Cloud-based HPC service provisioning
among higher education institutions. ETH Zurich
has already received inquiries for the usage of HPC
services from public and private organizations.
Because of technical, legal, and regulatory issues,
none of these requests could have been granted.
Future research could focus on incentive schemes,
legal and regulatory aspects, and technological
requirements to enable service exchange among
organizations. In this context, some of the findings
of this study could be tested and transferred to other
public organizations.
We assume that the importance of services will
increase at ETH Zurich, especially in the area of
HPC (Eurich, Tahar, Boutellier, 2011). Therefore,
there might be an overemphasis on service
provisioning in the assessment of pricing schemes.
Future work could discuss the possible relation
between the specific use (e.g., experimentation,
storage) and the suitable pricing scheme.
Finally, it should be considered that HPC service
provisioning can no longer be subsidized the way it
used to be. The rapid increase in the demand of
computing resources has pushed higher education
institutions and also other public organizations to
their limits in terms of computing service
provisioning and its financing. Energy costs grow
exponentially: the current way of HPC service
provisioning must be rethought. Decision makers
need to reflect on the different types of consumers
and their ties to the national infrastructure.
REFERENCES
Baldwin, H. Public Sector Cloud Computing: The Good,
the Bad and the Ugly, Computerworld, May 9, 2012,
http://www.computerworld.com/s/article/9226932/Pub
lic_sector_cloud_computing_The_good_the_bad_and
_the_ugly, accessed June 2014.
Bryman, A. and Bell, E., 2007. Business Research
Methods, Oxford University Press. Oxford, UK, 2
nd
edition.
Calleja, P., Gardiner, C., Gryce, C., Guest, M., Lockley, J.,
Parchment, O., and Stewart, I. HPC-SIG Report 2010,
UK High Performance Computing Special Interest
Group, 2010, http://www.hpc-sig.org/Publications?
action=AttachFile&do=get&target=HPC-SIG_Report
2010.pdf, accessed June 2014.
Campbell, D. T., 1966. Pattern Matching as an Essential in
Distal Knowing. In Hammond, K. (ed.), The
Psychology of Egon Brunswik. Holt, Rinehart and
Winston. New York, pp. 81-106.
Chandrasekaran, A. and Kapoor, M. State of Cloud
Computing in the Public Sector – A Strategic analysis
of the business case and overview of initiatives across
Asia Pacific, Frost & Sullivan, May 11, 2011.
http://www.frost.com/prod/servlet/cio/232651119,
accessed June 2014.
Chesbrough, H., 2010. Business Model Innovation:
Opportunities and Barriers. Long Range Planning,
43(2-3), pp. 354-363.
Creswell, J., 2013. Qualitative Inquiry and Research
Design: Choosing Among Five Approaches. Sage
Publications. Thousand Oaks, CA, 3
rd
edition.
De Boer, H. F., Enders, J., and Leisyte, L., 2007. Public
Sector Reform in Dutch Higher Education: The
Organizational Transformation of the University.
Public Administration, 85(1), pp. 27-46.
Drucker, P. F., 1999. Beyond the Information Revolution.
Atlantic Monthly, 284, pp. 47-57.
Drucker, P. F., 2002. They're not employees, they're
people. Harvard Business Review, 80(2), pp. 70-77.
Eurich, M., Calleja, P., and Boutellier, R., 2013. Business
Models of High Performance Computing Centres in
Higher Education in Europe. Journal of Computing in
Higher Education, 25(3), pp. 166-181.
Eurich, M., Giessmann, A., Mettler, T., and Stanoevska-
Slabeva, K., 2011. Revenue Streams of Cloud-based
Platforms: Current State and Future Directions.
Proceedings of the 17th Americas Conference on
Information Systems (AMCIS 2011). Detroit, MI,
August 4-7, 2011.
Eurich, M., Tahar, S., and Boutellier, R., 2011.
Effizienzdruck und technologische Innovation im
Hochschul-IT Management: Strukturwandel der ETH-
Informatikdienste. Hochschulmanagement, 2/2011, pp.
36-41.
Hedman, J. and Kalling, T., 2003. The Business Model
Concept: Theoretical Underpinnings And Empirical
Illustrations. European Journal of Information
Systems, 12(1), pp. 49-59.
Johnson, M. W., Christensen, C. M., and Kagermann, H.,
2008. Reinventing Your Business Model. Harvard
Business Review, 86(12), pp. 57-68.
Kundra, V. State of Public Sector Cloud Computing, CIO
Council Indonesia, May 20, 2010, http://dosen.
narotama.ac.id/wp-content/uploads/2012/01/State-of-
Public-Sector-Cloud-Computing.pdf, accessed June
2014.
BusinessModelsforCloud-basedHighPerfomanceComputingServiceProvision-InsightsfromtheSwissHigher
EducationSector
109
Kunszt, P., Maffioletti, S., Flanders, D., Eurich, M.,
Schiller, E., Bohnert, T. M., Haug, S., et al., 2014.
Towards a Swiss National Research Infrastructure.
Proceedings of the Euro-Par 2013: Parallel Processing
Workshops, Aachen, Germany, August 26-30, 2013,
Springer. Heidelberg, pp. 157-166.
Kunszt, P., Maffioletti, S., Messina, A., Flanders, D.,
Mathys, S., and Murri, R., 2013. Academic Cloud
Provisioning and Usage Project,
https://wiki.systemsx.ch/display/cloudresult, accessed
June 2014.
Lifka, D., Foster, I., Mehringer, S., Parashar, M., Redfern,
P., Stewart, C., and Tuecke, S. XSEDE Cloud Survey
Report. Extreme Science and Engineering Discovery
Environment, September 2013, https://www.cac.
cornell.edu/technologies/XSEDECloudSurveyReport.
pdf, accessed June 2014.
Macias, F. and Greg, T. Cloud Computing Advantages in
the Public Sector. Cisco, 2011,
https://www.cisco.com/web/strategy/docs/c11-
687784_cloud_omputing_wp.pdf, accessed June 2014.
McGrath, R., 2010. Business Models: A Discovery Driven
Approach. Long Range Planning, 43(2-3), pp. 247-
261.
Mell, P. and Grance, T., 2011. The NIST Definition of
Cloud Computing. US Department of Commerce,
National Institute of Standards and Technology, NIST
Special Publication 800-145, September 2011,
http://csrc.nist.gov/publications/nistpubs/800-
145/SP800-145.pdf, accessed June 2014.
Norwich University. A Pulse on Virtualization & Cloud
Computing, Norwich University, April 2011,
http://www.thecre.com/fisma/wp-content/uploads/
2011/05/ Norwich_Survey_Findings1.pdf, accessed
June 2014.
Osterwalder, A. and Pigneur, Y., 2010. Business Model
Generation: A Handbook for Visionaries, Game
Changers, and Challengers. Wiley. Hoboken, NJ.
Pring, B. Cloud Computing: The Next Generation of
Outsourcing, Gartner, November 10, 2010,
http://www.gartner.com/DisplayDocument?id=146041
6, accessed June 2014.
Red Shift Research. Adoption, Approaches & Attitudes:
The Future of Cloud Computing in the Public and
Private Sectors, Red Shift Research, June 2011,
http://whitepaper.techweekeurope.co.uk/adoption-
approaches-attitudes-the-future-of-cloud-computing-
in-the-public-and-private-sectors-340.html, accessed
June 2014.
Reuther, A. and Tichenor, S., 2006. Making the Business
Case for High Performance Computing: A Benefit-
Cost Analysis Methodology. CTWatch Quarterly,
2(4A), pp. 2-9.
Schubert, T., 2009. Empirical Observations on New Public
Management to Increase Efficiency in Public Research
- Boon or Bane?. Research Policy, 38(8), pp. 1225-
1234.
SGI. SGI Cyclone: Results On Demand, Silicon Graphics
International Corp., 2010, https://www.sgi.com/
pdfs/4205.pdf, accessed June 2014.
Sotomayor, B., Montero, R. S., Llorente, I. M., and Foster,
I., 2009. Virtual infrastructure management in private
and hybrid clouds. Internet Computing, IEEE, 13 (5),
pp. 14-22.
Teece, D. J., 2010. Business Models, Business Strategy
and Innovation. Long Range Planning, 43(2-3), pp.
172-194.
Timmers, P., 1998. Business Models for Electronic
Markets. Electronic Markets, 8(2), pp. 3-8.
Urquhart, C., 2001. An Encounter with Grounded Theory:
Tackling the Practical and Philosophical Issues. In
Trauth, E. (ed.), Qualitative Research in IS: Issues and
Trends. Idea Group Publishing. London, UK, pp. 104-
140.
Vaquero, L. M., Rodero-Merino, L., Caceres, J., and
Lindner, M., 2009. A Break in the Clouds: Towards a
Cloud Definition. ACM SIGCOMM Computer
Communication Review, 39(1), pp. 50-55.
Weinhardt, C., Anandasivam, A., Blau, B., Borissov, N.,
Meinl, T., Michalk, W., and Stößer, J., 2009. Cloud
Computing: A Classification, Business Models, and
Research Directions. Business & Information Systems
Engineering, 1(5), pp. 391-399.
ICE-B2014-InternationalConferenceone-Business
110
