The document summarizes the strengths and weaknesses of using a utility computing model to understand cloud computing. While the utility model provides some useful insights, it is an overly simplistic analogy that risks missing key opportunities and challenges. Technically, cloud computing faces challenges around the rapid pace of innovation, scalability limits, and latency due to distance. Business-wise, cloud computing differs in its needs around complementarity with other innovations, issues of lock-in and interoperability between providers, and new security concerns. An accurate understanding of cloud computing requires moving beyond the utility analogy to address its technical and business complexities.

1. v
viewpoints
DOI:10.1145/1735223.1735234 Erik Brynjolfsson, Paul Hofmann, and John Jordan
economic and
Business dimensions
Cloud Computing and electricity:
Beyond the Utility Model
Assessing the strengths, weaknesses, and general applicability
of the computing-as-utility business model.
B
U s iN e s s e s re LY No less on cloud computing and use, paid for via subscription, and ac-
electricity than on IT. Yet the electricity model cessed over the Web.” From an aca-
corporations don’t need a Definitions for cloud computing demic perspective: “Cloud computing
“Chief Electricity Officer” vary. From a practitioner standpoint: refers to both the applications deliv-
and a staff of highly trained “Cloud computing is on-demand ac- ered as services over the Internet and
professionals to manage and integrate cess to virtualized IT resources that the hardware and systems software
electricity into their businesses. Does are housed outside of your own data in the data centers that provide those
the historical adoption of electricity of- center, shared by others, simple to services. … The data center hardware
fer a useful analogy for today’s innova- and software is what we will call a
tions in cloud computing? cloud. When a cloud is made available
While the utility model offers some an overly simplistic in a pay-as-you-go manner to the pub-
insights, we must go beyond this sim- lic, we call it a public cloud; the service
ple analogy to understand cloud com- reliance on the utility being sold is utility computing.”1
puting’s real challenges and opportu- model risks blinding Both definitions imply or explicitly
nities. Technical issues of innovation, use the “utility” model that embeds the
scale, and geography will confront us to the real logic of water supply, electrical grids, or
managers who attempt to take advan- opportunities and sewage systems. This model is ubiqui-
tage of offsite resources. In addition, tous. While it has important strengths,
business model challenges related to challenges of it also has major weaknesses.
complementarity, interoperability, cloud computing. Hardware providers introduced the
and security will make it difficult for language of “utility” computing into the
a stable cloud market to emerge. An market. But perhaps the most rigorous
overly simplistic reliance on the util- and vigorous assertion of the electric-
ity model risks blinding us to the real ity model comes from Nicholas Carr,
opportunities and challenges of cloud an independent blogger in his recent
computing. book, The Big Switch: “At a purely eco-
32 communications of th e ac m | m ay 2 0 1 0 | vo l . 5 3 | n o. 5

2. viewpoints
nomic level, the similarities between technical Weaknesses consistency and scalability at the same
electricity and information technology of the utility model time. The problem of scalable data stor-
are even more striking. Both are what The Pace of Innovation. The pace of in- age in the cloud with an API as rich as
economists call general-purpose tech- novation in electricity generation and SQL makes it difficult for high-volume,
nologies. … General-purpose technolo- distribution happens on the scale of mission-critical transaction systems to
gies, or GPTs, are best thought of not as decades or centuries.8 In contrast, run in cloud environments.
discrete tools but as platforms on which Moore’s Law is measured in months. Meanwhile, companies of a certain
many different tools, or applications, In 1976, the basic computational pow- size can get the best of both worlds by
can be constructed. … Once it becomes er of a $200 iPod would have cost one deploying private clouds. Intel, for ex-
possible to provide the technology cen- billion dollars, while the full set of ca- ample, is consolidating its data centers
trally, large-scale utility suppliers arise pabilities would have been impossible from more than 100 eventually down to
to displace the private providers. It may to replicate at any price, much less in a about 10. In 2008 the total fell to 75, with
take decades for companies to abandon shirt pocket. Managing innovative and cost savings of $95 million. According to
their proprietary supply operations and rapidly changing systems requires the Intel’s co-CIO Diane Bryant, 85% of In-
all the investment they represent. But attention of skilled, creative people, tel’s servers support engineering com-
in the end the savings offered by utili- even when the innovations are creat- putation, and those servers run at 90%
ties become too compelling to resist,
even for the largest enterprises. The
grid wins.”4
strengths of the utility model
Carr correctly highlights the concept
of a general-purpose technology. This
class of technology has historically
been the greatest driver of productivity
growth in modern economies. They not
only contribute directly, but also by cat-
alyzing myriad complementary innova-
tions.3 For electricity, this includes the
electric lighting, motors, and machin-
ery. For IT, this includes transaction
processing, ERP, online commerce and
myriad other applications and even
business model innovations.
Some of the economies of scale and
cost savings of cloud computing are
also akin to those in electricity genera-
tion. Through statistical multiplexing,
centralized infrastructure can run at
higher utilization than many forms of
distributed server deployment. One
system administrator, for example, can
tend over 1,000 servers in a very large
data center, while his or her equivalent
in a medium-sized data center typical-
ly manages approximately 140.7 ed by others, unlike managing stable utilization—a combination of strategic
By moving data centers closer to technologies. importance and operational perfor-
energy production, cloud computing The Limits of Scale. The rapid avail- mance that would negate any arguments
creates additional cost savings. It is far ability of additional server instances is a for shifting that load to a cloud vendor.
cheaper to move photons over the fiber- central benefit of cloud computing, but Ironically, even as the utility model is
optic backbone of the Internet than it it has its limits. In the first place, paral- being touted for computing, the highly
is to transmit electrons over our power lel problems are only a subset of diffi- centralized approach is becoming less
grid. These savings are captured when cult computing tasks: some problems effective for electricity itself: an emerg-
illustration by st ua rt bra dford
data centers are located near low-cost and processes must be attacked with ing distributed power generation sys-
power sources like the hydroelectric other architectures of processing, mem- tem features smaller nodes running
dams of the northwest U.S. ory, and storage, so simply renting more micro-hydro, wind, micro-turbines and
Along with its strengths, however, nodes will not help. Secondly, many fuel cells. What’s more, many enterpris-
the electric utility analogy also has business applications rely on consis- es do in fact generate their own electric-
three technical weaknesses and three tent transactions supported by RDBMS. ity or steam, for the same reasons they
business model weaknesses. The CAP Theorem says one cannot have will continue to keep certain classes of
m ay 2 0 1 0 | vo l . 5 3 | n o. 5 | c o m m u n i c at i o n s o f t he acm 33

3. viewpoints
IT in house: reliability, strategic advan- the user and pose severe new security
tage, or cost visibility. issues not encountered by on-premise
Latency: Distance is Not Dead. One of if the utility model computing behind firewalls.
the few immutable laws of physics is were adequate,
the speed of light. As a result, latency conclusion
remains a formidable challenge. In the the challenges to If the utility model were adequate, the
network realm, the demands for nearly cloud computing challenges to cloud computing could
instantaneous execution of machine-to- be solved with electricity-like solu-
machine stock trades has led financial could be solved tions—but they cannot. The reality is
services firms to locate their data cen- with electricity-like that cloud computing cannot achieve
ters as physically close to stock exchang- the plug-and-play simplicity of electric-
es as possible. The read/write limits of solutions—but ity, at least, not as long as the pace of
magnetic disks can only drop so far, but they cannot. innovation, both within cloud comput-
increased speed comes at the cost of ca- ing itself, and in the myriad applica-
pacity: big disks are slow, and fast disks tions and business models it enables,
are small. For many classes of applica- continues at such a rapid pace. While
tions, performance, convenience, and electric utilities are held up as models
security considerations will dictate that of simplicity and stability, even this in-
computing be local. Moving data cen- iTunes store helped it quadruple reve- dustry is not immune from the trans-
ters away from their customers may save nues in four years. The tight integration formative power of IT.8,9 Innovations
on electricity costs, but those savings are between Apple’s ERP system and the like the “smart grid” are triggering fun-
often outweighed by the costs of latency. billing engine handling some 10 mil- damental changes at a pace not seen
lion sales per day would have been dif- since the early days of electrification.
Beyond electricity: the ficult, if not impossible, in the cloud. The real strength of cloud computing is
Business model of the cloud Lock-in and Interoperability. Lock-in that it is a catalyst for more innovation. In
Important as the technical differences issues with electricity were addressed fact, as cloud computing continues to be-
are between electricity and cloud com- long ago by regulation of monopolies, come cheaper and more ubiquitous, the
puting, the business model differences then later by legal separation of gen- opportunities for combinatorial innova-
are even more profound. eration from transmission and the tion will only grow. It is true that this inev-
Complementarities and Co-invention. creation of market structures. Markets itably requires more creativity and skill
Like electricity, IT is a general-purpose work because electrons are fungible. from IT and business executives. In the
technology. This means that critical The rotary converter that enabled in- end, this not something to be avoided. It
benefits come from the co-inventions terconnection of different generating should be welcomed and embraced.
that the basic technology makes pos- technologies in the 1890s has no ana-
sible. It took 30 to 40 years for the full log for the customer of multiple cloud References
1. armbrust, m. et al. a view of cloud computing.
benefits of electricity to redound to vendors, and won’t anytime soon. For Commun. ACM 53, 4 (apr. 2010), 50–58.
America’s factories.5 Initially, assembly enterprise computing to behave like 2. bresnahan, t., greenstein, s., brownstone, d. and flamm,
k. technical progress and co-invention in computing and
lines and production processes were line voltage will require radically differ- in the uses of computers. Brookings Papers on Economic
not redesigned to take advantages of ent management of data than what is Activity—Microeconomics (1996), 1–83.
3. brynjolfsson, e. and saunders, a. Wired for Innovation:
electricity: large central steam engines on anyone’s technology roadmap. How IT is Reshaping the Economy. mit Press,
were simply replaced with large elec- Perhaps most critically, bits of infor- cambridge, ma, 2010.
4. carr, n. The Big Switch: Rewiring the World, from
tric motors, and then hooked up to the mation are not electrons. Depending on Edison to Google. norton, new york, 2008.
same old crankshafts and cogs. Only the application, its engineering, and its 5. david, P. the dynamo and the computer: an historical
perspective on the modern productivity paradox.
with the reinvention of the production intended use, cloud offerings will not American Economic Review 80, 2 (1990), 355–361.
process was the potential of electrifica- be interchangeable across cloud pro- 6. foley, J. Plug into the cloud. InformationWeek (sept.
28, 2008).
tion realized. Today, electricity has ma- viders. Put more simply, the business 7. hamilton, J. internet-scale service efficiency. in
Proceedings of the Large-Scale Distributed Systems
tured to become a relative commodity. processes supported by enterprise com- and Middleware (LADIS) Workshop, (sept. 2008).
In contrast, computing is still in the puting are not motors or light bulbs. 8. hughes, t. Networks of Power: Electrification
in Western Society, 1880–1930. Johns hopkins
midst of an explosion of innovation Security. The security concerns with university Press, baltimore, md, 1983.
and co-invention.2 Firms that simply cloud computing have no electricity 9. Waltz, d. and king, J. Information Technology and
America’s Energy Future. computing research
replace corporate resources with cloud analog. No regulatory or law enforce- association White Paper, Washington, d.c., 2009.
computing, while changing nothing ment body will audit a company’s
else, are doomed to miss the full ben- electrons, but processes related to Erik Brynjolfsson (erikb@mit.edu) is a professor at the
efits of the new technology. customer data, trade secrets, and clas- mit sloan school and the director of the mit center for
digital business in cambridge, ma.
The opportunities, and risks, from sified government information are all
IT-enabled business model innova- subject to stringent requirements and Paul hofmann (paul.hofmann@sap.com) is a vice
president at saP labs in Palo alto, ca.
tion and organizational redesigns are standards of auditability. The typically
reshaping entire industries.3 For in- shared and dynamic resources of cloud John Jordan (jmj13@smeal.psu.edu) is a senior lecturer
in the smeal college of business at Penn state university.
stance, Apple’s transition from a per- computing (including CPU, network-
petual license model to the pay-per-use ing, and so forth) reduce control for copyright held by author.
34 communications of th e ac m | m ay 2 0 1 0 | vo l . 5 3 | n o. 5