ArticlePDF Available

Abstract

Energy efficiency is increasingly important for future information and communication technologies (ICT), because the increased usage of ICT, together with increasing energy costs and the need to reduce green house gas emissions call for energy-efficient technologies that decrease the overall energy consumption of computation, storage and communications. Cloud computing has recently received considerable attention, as a promising approach for delivering ICT services by improving the utilization of data centre resources. In principle, cloud computing can be an inherently energy-efficient technology for ICT provided that its potential for significant energy savings that have so far focused on hardware aspects, can be fully explored with respect to system operation and networking aspects. Thus this paper, in the context of cloud computing, reviews the usage of methods and technologies currently used for energy-efficient operation of computer hardware and network infrastructure. After surveying some of the current best practice and relevant literature in this area, this paper identifies some of the remaining key research challenges that arise when such energy-saving techniques are extended for use in cloud computing environments.
© The Author 2009. Published by Oxford University Press on behalf of The British Computer Society. All rights reserved.
For Permissions, please email: journals.permissions@oxfordjournals.org
doi:10.1093/comjnl/bxp080
Energy-Efficient Cloud Computing
Andreas Berl1,, Erol Gelenbe2, Marco di Girolamo3, Giovanni Giuliani3,
Hermann de Meer1, Minh Quan Dang4and Kostas Pentikousis5
1Fakultät für Informatik und Mathematik, University of Passau, Innstr. 43, 94032 Passau, Germany
2Electrical and Electronic Engineering Department, Imperial College London, South Kensington Campus,
London SW7 2AZ, UK
3HP-European Innovation Centre, HP IIC (Italy Innovation Centre), Italy
4School of Information Technology, International University in Germany, Bruchsal, Germany
5VTT Technical Research Centre of Finland, Kaitoväylä 1, FI-90571 Oulu, Finland
Corresponding author: berl@uni-passau.de
Energy efficiency is increasingly important for future information and communication technologies
(ICT), because the increased usage of ICT, together with increasing energy costs and the need to
reduce green house gas emissions call for energy-efficient technologies that decrease the overall
energy consumption of computation, storage and communications. Cloud computing has recently
received considerable attention, as a promising approach for delivering ICT services by improving the
utilization of data centre resources. In principle, cloud computing can be an inherentlyenergy-efficient
technology for ICT provided that its potential for significant energy savings that have so far focused
on hardware aspects, can be fully explored with respect to system operation and networking aspects.
Thus this paper, in the context of cloud computing, reviews the usage of methods and technologies
currently used for energy-efficient operation of computer hardware and network infrastructure.After
surveying some of the current best practice and relevant literature in this area, this paper identifies
some of the remaining key research challenges that arise when such energy-saving techniques are
extended for use in cloud computing environments.
Keywords: energy-efficient computing and networking; energy-aware data centres; cloud computing
Received 28 July 2009; revised 28 July 2009
Handling editor:
1. INTRODUCTION
Significant savings in the energy budget of a data centre,
without sacrificing service level agreements, are an excellent
economic incentive for data centre operators, and would also
make a significant contribution to greater environmental sus-
tainability. According to Amazon.com’s estimates [1], at its
data centres (as illustrated in figure 1), expenses related to
the cost and operation of the servers account for 53% of the
total budget (based on a 3-year amortization schedule), while
energy-related costs amount to 42% of the total, and include
both direct power consumption (19%) and the cooling
infrastructure (23%) amortized over a 15-year period.
Dennis Pamlin, the Global Policy Advisor of WWF, Sweden
[2] highlighted different IT solutions and their beneficial
impact on green house gases (GHG), which include CO2,
emissions. These opportunities include IT- based solutions:
e.g. smart buildings, smart transportation and communication,
smart commerce and services, and smart industrial production.
The colloquial term ‘smart’ in this case means ‘with low
carbon footprint’, showing that the adoption of such ‘smart’
IT solutions will enable a potentially large GHG reduction,
including information and communication technologies (ICT)
itself which is a large power consumer (and therefore a GHG
emitter), and IT solutions that have a huge potential impact in
reducing GHG emissions in many sectors (Fig. 1).
Based on a recent ‘Data Centre Energy Forecast Report’[3], it
can be expected that savings of the order of 20% can be achieved
in server and network energy consumption with respect to
current levels, and that these savings may induce an additional
30% saving in cooling needs as detailed in a study by HP and the
Uptime Institute [4]. It shows that ‘most of data centre power
is spent on cooling ICT equipment (between 60 and 70%)’.
Thus there are very significant economic and environmental
gains to be obtained from a serious research thrust on energy
efficiency in the general area of IT and computer networks.
In particular, cloud computing is an inherently energy-efficient
virtualization technique [5], in which services run remotely
The Computer Journal,2009
The Computer Journal Advance Access published August 19, 2009
at CERDI on January 11, 2011comjnl.oxfordjournals.orgDownloaded from
2A. Berl et al.
FIGURE 1. Energy distribution in the data centre.
in a ubiquitous computing cloud that provides scalable and
virtualized resources. Thus peak loads can be moved to other
parts of the cloud and the aggregation of a cloud’s resources can
provide higher hardware utilization.
The rest of this paper is organized as follows.After reviewing
the current state of energy efficiency in ICT in Section 2, we
discuss the salient aspects of energy-efficient cloud computing
in Section 3. In Section 4, we detail the main research challenges
that lie ahead, and provide concluding remarks in Section 5
2. CURRENT STATE OF ENERGY EFFICIENCY IN
ICT INFRASTRUCTURES
ICT consumes an increasing amount of energy, but is also
instrumental in increasing productivity and economic prosperity
and in reducing energy expenditure from other sources through
e-work, e-commerce and e-learning. Traditional network design
has sought to minimize infrastructure costs and maximize
quality of service (QoS). However, ICT also plays a complex
role in energy consumption via the ‘communicate more and
travel less’paradigm, as well as through the use of smart devices
in homes and offices to optimize energy management. Thus, ICT
can reduce energy consumption and carbon emissions, but this
potential reduction is partially offset by the power used by data
centres and computer networks [6] which runs into billions of
dollars or euros. Thus, a fraction of energy savings in ICT and
networks could lead to significant financial and carbon savings.
In this section, we review recent research in energy efficiency
for standalone hardware, and then review work that considers
energy consumption as part of the cost functions to be used
for scheduling in multiprocessor and grid architectures. Finally,
we briefly review energy consumption in cluster servers and
wired/wireless networks.
2.1. Energy-efficient hardware
One of the approaches to increase the energy efficiency is to
develop more energy-efficient hardware. This effort is fostered
by labels such as the US Energy Star [7] or the European
TCO Certification [8] which rate IT products (mostly monitors)
according to their environmental impact. Novel emerging
technologies such as solid-state discs are content with much
less energy than the currently used hard disc drives. Computer
power can be saved by means of various well-known techniques.
First, the processor can be powered down by mechanisms like
SpeedStep [9], PowerNow [10], Cool’nQuiet [11] or Demand-
Based Switching [12]. These measures enable slowing down
CPU clock speeds (clock gating), or powering off parts of the
chips (power gating), if they are idle [13,14]. By sensing the
lack of user–machine interaction, different redundant hardware
parts can incrementally be turned off or put in hibernating mode
(display, disc etc.).
The advanced configuration and power interface (ACPI)
specification [15] defines four different power states that an
ACPI-compliant computer system can be in. These states range
from G0-working to G3-mechanical-off. The states G1 and G2
are subdivided into further substates that describe which com-
ponents are switched off in the particular state. For devices and
the CPU, separate power states (D0–D3 for devices and C0–C3
for CPUs) are defined, which are similar to the global power
states. Some of the mentioned techniques are usually applied
to mobile devices but can be used for desktop PCs as well.
2.2. Energy-aware scheduling in multiprocessor
and grid systems
Energy-aware scheduling in multiprocessor and grid systems
is a widely discussed problem in the literature as the following
overview shows. In [16] the authors present an energy-aware
method to schedule multiple real-time tasks in multiprocessor
systems that support dynamic voltage scaling (DVS). The key
in their approach is to consider the probabilistic distribution
of the tasks’ execution time in order to partition the workload
and reduce energy consumption. Memory energy consumption
[17] can also be reduced by scheduling techniques that impact
the effectiveness of frequency scaling by combining the effect
of tasks running on a multicore system, including memory
contention and the technical constraint of chip-wide frequency
and voltage settings. The DVS capability [18] addresses energy
minimization for periodic pre-emptive hard real-time tasks
that are scheduled on an identical multiprocessor platform.
AlEnawy and Aydin [18] suggest partitioned scheduling and
assume that the tasks are assigned rate-monotonic priorities.
To solve this problem, they proposed an integrated approach,
including rate monotonic scheduling, an admission control
test, a partitioning heuristic and a speed assignment algorithm.
All the above work proposes to control the energy consumption
of hardware by adjusting voltage levels.
Cong et al. [19] consider energy-efficient scheduling for data
grids supporting real-time and data-intensive applications. They
use both, the location of data and application properties to
design a novel distributed energy-efficient scheduler that aims
The Computer Journal,2009
at CERDI on January 11, 2011comjnl.oxfordjournals.orgDownloaded from
Energy-Efficient Cloud Computing 3
to seamlessly integrate scheduling tasks with data placement
strategies to provide energy savings. The main energy savings
are obtained by reducing the amount of data replication and task
transfers. In [20] an energy-constrained scheduling scheme for a
grid environment is investigated both for energy minimization in
mobile devices and for grid utility optimization, by formalizing
energy-aware scheduling using nonlinear optimization theory
under the constraints of energy budget and the job deadline. In
addition, [20] proposes a distributed pricing scheme that makes
trade-offs between energy and deadlines to achieve a system-
wide optimization based on the preference of the grid user.
Another approach that increases energy efficiency in data
centres is based on server consolidation by service virtualization
[2125]. Virtualization partitions computational resources and
allows the sharing of hardware. Many services often need
only a small fraction of the available computational resources
[26] of a data centre server. However, even when run at a
low utilization, servers typically need up to 70% of their
maximum power consumption [27]. Such services can be
virtualized and run within a virtual machine (VM) resulting
in significant increases in overall energy efficiency. Depending
on their utilization, many VMs can run on a single hardware
unit (server consolidation). Therefore, less hardware is needed
overall, thus reducing energy wasted for cooling, while the
deployed hardware utilization increases. This consolidation of
shared hardware fosters energy efficiency, measured as work
accomplished per unit of consumed energy [28].
Resources can be virtualized on different layers and imple-
ment different forms of virtualization: full virtualization, hosted
virtualization or operating system (OS) layer virtualization [29].
When system virtualization [30] (e.g. full virtualization) is sup-
ported, the virtualization software emulates full-featured hard-
ware and runs on top of the local OS. In the paravirtualization
approach, guest VMs are modified in order to perform so-called
‘hyper calls’ instead of system calls, leading to higher perfor-
mance of VMs [31], as used, for instance, in the XEN systems.
In XEN 3.0 [32] guests can be virtualized without being mod-
ified by using virtualization support of X86 CPUs. OS layer
virtualization has been proposed in the Linux-VServer [33,34],
a kernel-based virtualization.
It is important to point out that virtualization comes at a
cost which needs to be managed efficiently. When resources are
virtualized, additional management of VMs is needed to create,
terminate, clone or move VMs from host to host. Migration of
VMs can be done off-line (the guest in the VM is powered off)
or on-line (live migration of a running VM to another host).
The management solution infrastructure 3 [35] of VMWare, for
instance, supports live migration.
2.3. Power minimization in clusters of servers
Recent research has considered power minimization in server
clusters, with guaranteed throughput and response time [36].
Energy consumption depends primarily on CPU utilization, but
components, such as discs, memory and network devices, also
use energy so that a server that seems to remain idle may still use
up to 60% of its peak power. In [37] policies are developed that
use economic criteria and energy as criteria to despatch jobs to a
small set of active servers, while other servers are down to a low-
power state. Similar dynamic provisioning algorithms [38] are
studied for long-lived TCP connections as in instant messaging
and gaming. A queuing approach to [39] dynamic provisioning
technique has also been studied to obtain the minimum number
of servers required to respect the required QoS and reactive
provisioning can be used to compensate for sudden surges
in load.
2.4. Power minimization in wireless and wired networks
According to some estimates, ‘the Internet’ may by consuming
more than 860 TWh annually [40], but such figures can only
be considered as educated guesses due to the number of
assumptions one has to make. Traditionally fixed network
operators have not considered energy consumption as a major
cost factor. Lately, however, as sustainability is becoming a
key business objective, fixed network operators are looking
for ways to decrease their energy footprint. On the contrary,
wireless network operators due to regulatory requirements and
operational considerations regarding base station deployments
have been trying to minimize energy consumption for over a
decade. In fact, it is reported that [41] the radio access network
(rather than the core network) is the most energy consuming part
of the infrastructure, and in many cases the associated energy
bills are comparable to the total costs for the personnel who
work on network operations and maintenance. The ICT energy
estimates in [41] report that the Vodafone Group radio access
network alone consumed nearly 3 TWh in 2006.
Surprisingly enough, energy savings for infrastructure
networks have not received much attention until very recently,
while energy-saving routing protocols in wireless sensor
networks have already been studied in detail [4244] because of
the specific needs of battery powered networks and the related
research has included the use of topology control [45,46]
that modify the network graph to optimize properties such as
network capacity and QoS. Since processing and transmission
power in nodes are the essential consumers of energy, it is
also necessary to optimize the number of hops traversed by
packets. An interesting trade-off then arises between high
transmission power which can reduce the number of hops,
low transmission power which can lead to more hops being
necessary due to shorter ranges and transmission interference
which can be affected by power in a complex manner. Related
work can be found in [42,43] where the idea of turning
nodes on and off is also considered. In a wired node, power
consumption depends and influences other factors, such as the
node’s throughput; furthermore, up to 60% of a node’s energy
consumption can originate with peripheral devices such as link
drivers. Turning wired network nodes on and off may be very
The Computer Journal,2009
at CERDI on January 11, 2011comjnl.oxfordjournals.orgDownloaded from
4A. Berl et al.
difficult in a wired context because of the high volumes, rate of
traffic and the stringent QoS constraints. Routing for wireless
ad hoc networks with battery-powered nodes [47] have also
been considered in the selection paths so as to satisfy QoS
constraints and minimize power. Generally speaking, we feel
that the research community has begun to seriously consider
energy consumption in infrastructure networks, and the IEEE
has now focused on developing a standard for energy-efficient
Ethernet (IEEE 802.3 az).
In a future Internet where cloud computing may become
a mainstay for economic growth, businesses and individuals
will require energy-efficient operation that involves not only
computation and storage facilities but networking as well. It is
further anticipated that the majority of users will access cloud
computing resources from mobile, battery-powered devices,
which impose stringent limitations on power consumption.
Clearly, one therefore needs to address not just the issues arising
from individual components (such as storage and processing
elements), resource utilization algorithms (job scheduling,
virtualization, migration) and topology considerations, but the
entire chain of services and infrastructure enablers.
3. TOWARDS ENERGY-EFFICIENT CLOUD
COMPUTING
The previous discussion highlights the need to develop a
comprehensive approach for energy efficiency that involves all
system layers and aspects, including physical nodes, cooling
of nodes, networking hardware, communication protocols and
finally the servers and services themselves. The conceptual
framework of cloud computing may therefore be a way forward
to analyze, identify and implement overall energy savings in a
system to attain truly ‘green’ computing services.
In contrast to hardware-oriented optimization, software
systems can potentially be optimized at development time by
specifying their energy characteristics and by adapting the
implementation. However, this requires individual adaptation
of each component and it also requires understanding the
interaction between individual components when they operate
as a system. A major challenge is therefore to explore the
relations among system components and the trade-offs that can
result in an optimal balance between performance,QoS and
energy consumption and include self-aware runtime adaptation
[40,4850]. Thus in this section we briefly discuss some
areas of energy-efficiency research based on a cloud computing
perspective.
Significant energy savings can result from using energy-
aware scheduling mechanisms pervasively throughout a system.
Our survey in the previous sections shows that progress has
been made in this area, but that much more needs to be done
in holistically examining where and how such mechanisms are
needed, in applying the available mechanisms, and inventing
and evaluating new ones as needed. In the case of any one
service provider site, algorithms to multiplex and de-multiplex
workload in order to save energy are needed, and they
should incorporate the trade-off between performance and the
reduction in service cost due to energy savings. In addition to
scheduling and the mapping of workflows, the improvement
of energy-aware cloud applications themselves can also benefit
from software optimization.
In a business environment based on cloud computing,
workflows that run over many sites will tend to be popular.
Thus, developing methods that map the workflowonto resources
under the constraint of energy optimization becomes a central
problem of great value and novelty. Furthermore, in order
to comprehensively raise the energy efficiency of a system,
all of its layers have to be considered, including application
layer services. Services have different needs concerning the
environment they are running on or have special properties that
support the energy efficiency of the underlying system (e.g.
certain usage patterns). A service, for instance, might only be
used weekdays, say, from 8 to 18h or have peak usage at a
certain time of the day. A user also may also consider a trade-
off between a more energy-efficient service and a more reliable
or faster service, and compose the service in a way that fits its
needs. Thus it should also be possible to develop accounting
mechanisms that track and depend on the energy that has been
used by a service.
3.1. Energy-aware data centres
The key current technology for energy-efficient operation of
servers in data centres is virtualization. VMs that encapsulate
virtualized services can be moved, copied, created and deleted
depending on management decisions. Consolidating hardware
and reducing redundancy can achieve energy efficiency. Unused
servers can be turned off (or hibernated) to save energy.
Some hardware gets higher load, which reduces the number
of physical servers needed. However, the degree of energy-
efficient self-management in data centres is still limited today.
Services should not only be virtualized and managed within
a data centre site but they should be moved to other sites if
necessary. Not only the aspect of load has to be considered,
also the ‘heat’ generated by a service has to be measured and
accounted for before migrating operations. Every operational
physical node produces heat. When a particular node is
excessively used or is near other high-loaded nodes, hotspots
can appear in a given data centre. To avoid such hotspots,
heat can be distributed across sites. Furthermore, services can
be moved from sites with high load or high temperature to
sites with smaller loads and lower temperatures. Generally,
services should be moved to those locations, where they
can operate in the most energy-efficient way. This kind of
energy-efficient management of resources has to be realized
by an autonomous energy management that is as transparent
as possible to the user of a service. Energy-related problems
have to be solved according to defined policies without needing
The Computer Journal,2009
at CERDI on January 11, 2011comjnl.oxfordjournals.orgDownloaded from
Energy-Efficient Cloud Computing 5
human interaction. Machine-readable descriptions of the needs
and features of services, servers, networks and even whole sites
have to be available to enable energy efficiency in the highly
autonomous and adaptive [48] systems of the future.
3.2. Energy savings in networks and protocols
Research has shown that communications, in particular, is one of
the largest consumers of energy, however, energy optimization
for communications must deal with the trade-offs between
performance, energy savings and QoS [51].
Some hardware already offers features that create an
opportunity for energy-efficient operation such as turning
off network interfaces and throttling of processors. Network
protocols could also be optimized, or even be redeveloped in a
way that enhances the energy-efficient operation of the network
elements. Network devices could be enabled to delegate services
to other devices so as to transfer services from energy inefficient
to more energy-efficient devices or to devices that need to
be always on, while certain other devices are turned off. The
delegating device can then become dormant and be turned
off. Currently, many basic network services have to remain
active to periodically confirm their availability even when no
communication is taking place. These ‘soft states’ make it
impossible to turn off certain system components; therefore,
new protocols need to be designed to work around such soft
states so as to increase the energy efficiency of the network.
Signalling can also be revisited in this context; whereas data
and signalling traffic vary widely, the same technology and
mechanisms are used for both (in so-called in-band signalling).
While signalling needs only low bandwidth but can occur
anytime, data traffic occurs after signalling has taken place,
usually requires high bandwidth and traverses all network layers
up to the application layer, and uses processing power through
multiple layers of the network. Therefore the use of out-of-
band signalling should also be evaluated to design and improve
energy-aware communication protocols.
3.3. The effect of Internet applications
Thus far, we have considered the opportunities offered by cloud
computing as a possible foundation for energy-efficient ICT
infrastructures but have not discussed the nature of the applica-
tions themselves. We note that one large application area for the
Internet is in information dissemination. From digital cameras
embedded in mobile phones to environmental sensors to Web
2.0, end users are generating and interconnecting unprece-
dented amounts of information and this trend is expected to
continue unabated. However, the professional, expedited and
reliable distribution of content requires increasing investments
in infrastructure build-out and maintenance, and a matching
electricity bill to run the underlying ICT. Web, peer-to-peer
and web-based video-on-demand services currently dominate
Internet traffic and, taken together, consistently comprise
85% or more of the Internet traffic mix for several years. In
practice, dissemination networks operate using methods and
paradigms based on remote-access, replicating functionality
in several parts of the protocol stack, and fail to benefit from
recent advances in wired and wireless communications, storage
technologies and Moore’s law. If cloud computing becomes a
significant platform for producing and accessing information,
the amount of data that will be transferred over the Internet will
increase significantly. Content replication and dissemination
algorithms will then need to consider energy as a key param-
eter of optimal operation, and therefore cloud computing calls
for a thorough re-examination of the fundamentals of major
computation/communication/storage and energy/performance
trade-offs.
4. CONCLUSIONS
This paper has reviewed the potential impact of energy-
saving strategies for the management of integrated systems that
include computer systems and networks. We have surveyed the
contributions that are available in this area from recent research.
We propose that cloud computing with virtualization as a way
forward to (i) identify the main sources of energy consumption,
and the significant trade-offs between performance, QoS and
energy efficiency and (ii) offer insight into the manner in which
energy savings can be achieved in large-scale computer services
that integrate communication needs. Based on the approaches
that we have identified, we think that specific plug-ins and
energy-control centres for networked large-scale hardware and
software can be implemented and that they can have significant
impact, including:
(i) reducing the software and hardware related energy cost
of single or federated data centres that execute ‘cloud’
applications;
(ii) improving load balancing and hence QoS and
performance of single and federated data centres;
(iii) reducing energy consumption due to communications;
(iv) saving GHG and CO2emissions resulting from data
centres and networks so as to offer computing power
that is ‘environment protecting/conserving’.
Such improvements can have additional impact by reducing
energy utilization for transportation and work by encouraging
‘green’ ICT-based smart solutions for e-work, e-learning and
smart climate control for homes.
REFERENCES
[1] Hamilton, J. (2009) Cooperative Expendable Micro-Slice Servers
(CEMS): Low Cost, Low Power Servers for Internet-Scale
Services. Proc. 4th Biennial Conf. Innovative Data Systems
Research (CIDR), Asilomar, CA, USA, January.
The Computer Journal,2009
at CERDI on January 11, 2011comjnl.oxfordjournals.orgDownloaded from
6A. Berl et al.
[2] Pamlin, D. (2008) The Potential Global CO2Reductions from
ICT Use: Identifying and Assessing the Opportunities to Reduce
the First Billion Tonnes of CO2, Vol. May. WWF, Sweden.
[3] Accenture (2008) Data Centre Energy Forecast Report. Final
Report, Silicon Valley Leadership Group, July.
[4] Malone, C. and Belady, C. (2006) Metrics to Characterise Data
Centre & IT Equipment Energy Use. Proc. Digital Power Forum,
Richardson, TX, USA, September.
[5] Hewitt C. (2008) ORGs for scalable, robust, privacy-friendly
client cloud computing. IEEE Internet Comput., September,
96–99.
[6] Fan, X., Weber, W.-D. and Barroso, L.A. (2007) Power
provisioning for a warehouse-sized computer, Proc. 34th Annual
Int. Symp. Computer Architecture, San Diego, CA, USA, June
9–13, 2007. pp. 13–23. ACM, New York.
[7] Energy Star, http://www.energystar.gov,http://www.eu-
energystar.org. (Last accessed August 12, 2009).
[8] European TCO Certification, http://www.tcodevelopment.com.
(Last accessed August 12, 2009).
[9] Intel whitepaper 30057701 (2004) Wireless Intel SpeedStep
Power Manager: optimizing power consumption for the Intel
PXA27x processor family. http://sunsite.rediris.es/pub/mirror/
intel/pca/applicationsprocessors/whitepapers/30057701.pdf.
(Last accessed August 12, 2009).
[10] AMD PowerNow! Technology, http://www.amd.com/de-de/
Processors/ProductInformation/0„30_118_10220_10221ˆ964,00.
html. (Last accessed August 12, 2009).
[11] AMD Cool’n’Quiet Technology, http://www.amd.com/de-de/
Processors/ProductInformation/0„30_118_9485_9487ˆ10272,00.
html. (Last accessed August 12, 2009).
[12] Intel Software Network (2008) Enhanced Intel Speed-
Step® Technology and Demand-Based Switching on Linux.
http://softwarecommunity.intel.com/articles/eng/1611.htm.
(Last accessed August 12, 2009).
[13] Whitepaper Revision-001 (2007) ENERGY STAR* system
implementation. Intel with technical collaboration from the U.S.
Environmental ProtectionAgency. http://www.energystar.gov/ia/
partners/prod_development/revisions/downloads/316478-
001.pdf. (Last accessed August 12, 2009).
[14] Windeck, C. (2007) Energy Star 4.0. c’t Magazin für Computer
Technik,14, 52–53.
[15] Hewlett-Packard/Intel/Microsoft/Phoenix/Toshiba (2004)
Advanced configuration and power interface specification.
http://www.acpi.info/DOWNLOADS/ACPIspec30.pdf. (Last
accessed August 12, 2009).
[16] Changjiu, X., Yung-Hsiang, L. and Zhiyuan, L. (2007) Energy-
Aware Scheduling for Real-Time Multiprocessor Systems with
Uncertain Task Execution Time. Proc. 44th Annual Conf. Design
Automation, San Diego, CA, USA, June, pp. 664–669. ACM.
[17] Merkel, A. and Bellosa, F. (2008) Memory-Aware Scheduling
for Energy Efficiency on Multicore Processors. Proc. Workshop
on Power Aware Computing and Systems (HotPower’08), San
Diego, CA, USA, December, pp. 123–130. USENIX online.
[18] AlEnawy, A.T. and Aydin, H. (2005) Energy-Aware Task
Allocation for Rate Monotonic Scheduling. 11th IEEE Real Time
and Embedded Technology and Applications Symp. (RTAS’05),
San Francisco, CA, USA, March, pp. 213–223. IEEE.
[19] Cong, L., Xiao, Q., Kulkarni, S., Chengjun, W., Shuang, L.,
Manzanares, A. and Baskiyar, S. (2008) Distributed Energy-
Efficient Scheduling for Data-Intensive Applications with
Deadline Constraints on Data Grids. Proc. IEEE Int. Conf.
Performance, Computing and Communications, IPCCC 2008,
Austin, TX, USA, December, pp. 26–33. IEEE.
[20] Chunlin, L. and Layuan, L. (2008) Utility-based scheduling for
grid computing under constraints of energy budget and deadline.
Comput. Stand. Interfaces.
[21] Koomey, J. (2006) Server Energy Measurement Protocol, Ver-
sion 1.0. Following Energy Efficiency Server Benchmark Tech-
nical Workshop, Santa Clara, CA. http://www.energystar.gov/
ia/products /downloads/Finalserverenergyprotocol-v1.pdf. (Last
accessed August 12, 2009).
[22] IBM Project Big Green, http://www-03.ibm.com/press/us/
en/presskit/21440.wss. (Last accessed August 12, 2009).
[23] Berl, A. and de Meer, H. (2009) An Energy-Efficient
Distributed Office Environment. Proc. European Conf. Universal
Multiservice Networks (ECUMN 2009), Sliema, Malta, October,
IEEE Computer Society Press.
[24] Berl, A., Hlavacs, H., Weidlich, R., Schrank, M. and de Meer, H.
(2009) Network Virtualization in Future Home Environments.
Proc. Int. Workshop on Distributed Systems: Operations and
Management (DSOM09), Venice, Italy, October. Lecture Notes
in Computer Science. Springer, Berlin, Germany.
[25] Hlavacs, H., Hummel, K.A., Weidlich, R., Houyou, A., Berl, A.
and de Meer, H. (2009) Distributed energy efficiency in future
home environments. Ann. Telecommun.,63, 473-485.
[26] IBM (2007) Virtualization can help power efficiency. http://
www.03.ibm.com/systems/virtualization/view/011607.html.
(Last accessed August 12, 2009).
[27] Bundesverband Informationswirtschaft (2008) Energieeffizienz
im Rechenzentrum. Telekommunikation und neue Medien e.V.,
2, 10.
[28] Rivoire, S., Shah, M.A., Ranganathan, P. and Kozyrakis, C.
JouleSort: A Balanced Energy-Efficiency Benchmark. Proc. 2007
ACM SIGMOD International Conference on Management of
Data, Beijing, China, 365–376, 2007, ACM NewYork, NY, USA.
[29] Siering P. (2006) Realitätsverschiebung Virtualisierungstech-
niken im Vergleich. c’t Magazin für Computer Technik,13, 182–
189.
[30] Berl, A., Fischer, A. and de Meer, H. (2009) Using System
Virtualization to Create Virtualized Networks. Workshops der
Wissenschaftlichen Konferenz Kommunikation in Verteilten
Systemen (WowKiVS2009), Vol. 17, Kassel, Germany, March,
EASST.
[31] Barham, P., Dragovic, B., Fraser, K., Hand, S. Harris, T., Ho, A.,
Neugebauer, R., Pratt, I. and Warfield, A. (2003) Xen and the Art
of Virtualization. 19th ACM Symp. Operating Systems Principles
(SOSP’03), New York, October, pp. 164–177. ACM, New York.
[32] Pratt, I. (2005) Xen 3.0 and the art of virtualization. XEN—
Computer Laboratory Architecture, University of Cambridge,
http://www.cl.cam.ac.uk/netos/papers/2005-xen-ols.ppt.
[33] Linux-VServer, http://linux-vserver.org.
[34] Des Ligneris, B. (2005) Virtualization of Linux Based
Computers: The Linux-VServer Project. 19th Int. Symp. High
PerformanceComputing Systems and Applications (HPCS 2005),
Guelph, ON, Canada, May, pp. 340–346. IEEE.
The Computer Journal,2009
at CERDI on January 11, 2011comjnl.oxfordjournals.orgDownloaded from
Energy-Efficient Cloud Computing 7
[35] Brochure of VMWare Infrastructure 3, VMWare.http://www.
vmware.com/pdf/vi_brochure.pdf. (Last accessed August 12,
2009).
[36] Economou, D., Rivoire, S., Kozyrakis, C. and Ranganathan, P.
(2006) Full-System Power Analysis and Modeling for Server
Environments. Workshop on Modeling, Benchmarking, and
Simulation (MoBS), June, Boston, USA.
[37] Chase, J.S., Anderson, D.C., Thakar, P.N., Vahdat, A.M. and
Doyle, R.P. (2001) Managing Energy and Server Resources
in Hosting Centres. Proc. 18th ACM Symp. Operating System
Principles, Banff, Canada, October, pp. 103–116. ACM.
[38] Chen, G., He, W., Liu, J., Nath, S., Rigas, L., Xiao, L. and Zhao, F.
(2008) Energy-Aware Server Provisioning and Load Dispatching
for Connection-Intensive Internet Services. NSDI’08: Proc. 5th
USENIX Symp. Networked Systems Design and Implementation,
San Francisco, CA, USA, April, pp. 337–350.
[39] Urgaonkar, B., Shenoy, P., Chandra, A., Goyal, P. and Wood,
T. (2008) Agile dynamic provisioning of multi-tier Internet
applications. ACM Trans. Auton. Adap. Syst.,3, 1–39.
[40] Gelenbe, E. (2009) Steps towards self-aware networks. Commun.
ACM,52, 66–75.
[41] Sarokin, D. (2007) Question: energy use of Internet.
http://uclue.com/?xq=724. (Last accessed August 12, 2009).
[42] Jia, X., Li, D. and Du, D. (2004) QoS Topology Control in
Ad hoc Wireless Networks. Proc. IEEE INFOCOM’04, Hong
Kong, China, March, IEEE.
[43] Boukerche, A., Cheng, X. and Linus, J. (2005) A performance
evaluation of a novelenergy-aware data-centric routing algorithm
in wireless sensor networks. Wirel. Netw.,11, 619–635.
[44] Xu, Y., Heidemann, J. and Estrin, D. (2000) Adaptive
Energy-Conserving Routing for Multihop Ad hoc Networks.
Research Report 527, USC/Information Sciences Insti-
tute, http://www.isi.edu/johnh/PAPERS/Xu00a.html. (Last
accessed August 12, 2009).
[45] Rajaraman, R. (2002) Topology control and routing in ad hoc
networks: a survey. SIGACT News,33, 60–73.
[46] Santi, P. (2005) Topology control in wireless ad hoc and sensor
networks. ACM Comput. Surv.,37, 164–194.
[47] Gelenbe, E. and Lent, R. (2004) Power-aware ad hoc cognitive
packet networks, Ad Hoc Netw.,2, 205–216.
[48] Gelenbe, E., Lent, R. and Nunez, A. (2004) Self-aware networks
and QoS. Proc. IEEE,92, 1478–1489.
[49] Dobson, S., Denazis, S.G., Fernández, A., Gaïti, D., Gelenbe, E.,
Massacci, F., Nixon, P., Saffre, F., Schmidt, N. and Zambonelli,
F. (2006) A survey of autonomic communications. ACM Trans.
Auton. Adap. Syst.,1, 223–259.
[50] Gelenbe, E. (2006) Users and services in intelligent networks.
Proc. IEE (ITS), 153, 213–220.
[51] Gelenbe, E. and Silvestri S. (2009) Reducing Power Consumption
in Wired Networks. Proc. 24th Annual Int. Symp. Computer
and Information Sciences (ISCIS 09), Cyprus, September 14–16,
to appear.
The Computer Journal,2009
at CERDI on January 11, 2011comjnl.oxfordjournals.orgDownloaded from
... Management and orchestration of network services on top of a distributed multi-cloud environment raise many challenges. Efficient resource management and energy consumption are the main factors [11,118] in operating such large-scale cloud environments, and this is also a key topic addressed in 5G networks. Motivated by the understanding of performance issues [P3, P4, P5] ...
... Based on these basic capabilities, operators can offer services like IaaS (where infrastructure or the actual hardware is provisioned to customers who are responsible for installing operating systems and necessary software), PaaS (the computing infrastructure and the installed platforms/operating systems) and SaaS (PaaS and the necessary software are provisioned). Among many problems [11] related to the management of services in data centres, the topic of application placement has intensively been researched in some earlier works [26,72,88,97,115,117,122]. Hyser et al. [97] reported the design of a controller that automatically assigns virtual resources to physical hosts. Later, Nguyen et al. [117] proposed a detailed architecture to control the dynamic placement of virtual resources. ...
... 11 The pmf of launched instances versus booting time of UPF instances3.6 Conclusions ...
Thesis
Full-text available
The fourth industrial revolution focuses on automation and digitalisation based on the knowledge gathered by connecting the physical and digital worlds, one of the most significant transformations in humankind. The increased number of vertical industries require high bandwidth and low latency services. IT and the telecommunication industry are progressing through a colossal transformation with several enablers (cloud computing technology, virtualisation technology and software development and delivery) to respond to the demands of vertical industries. Further opportunities arise for service realisation with the emergence of new players (cloud operators). For example, telecommunication operators can utilise the "Anything as a service" (XaaS) and Infrastructure as a service (IaaS) offered by cloud operators to provide service for customers without the necessity of severe HW investment for new customers. In this thesis, we evaluate the performance of two algorithms that can be used for the resource management of network functions in cloud environments. First, we consider the interaction aspects between on-demand requests and the allocation of virtual machines in a server farm operated by a specific infrastructure owner (cloud operator). We formulated an analytic performance model of the server farm, taking into account the Quality of Service (QoS) guaranteed to users and the operational energy consumption in the server farm. We compared several scheduling algorithms based on servers’ average energy consumption and heat emission and the blocking probabilities of on-demand requests. Based on the comparison of different allocation schemes, we can conclude that saving energy is possible while the blocking probability of on-demand requests is kept below a certain threshold. Furthermore, we have identified that the energy consumption savings can be achieved by the policies LF and PR depending on the load conditions of the server system. Second, we investigate the performance analysis of a threshold-based horizontal scaling algorithm for User Plane Function (UPF) of 5G Core Network. We proposed a model where a threshold-based algorithm controls the number of UPF instances depending on users’ traffic. We presented a method to efficiently compute the steady-state probabilities and performance measures based on the model. We investigated the performance of the UPF scaling algorithm in various scenarios with the assumption of specific hardware, giving this method for benchmarking the performance of horizontal scaling. In the presented model, we proposed a two-dimensional Continuous Time Markov Chain (CTMC), and we described the states of the system with the instant number of sessions served and the number of UPF instances deployed. We also provided a method to recursively calculate the steady-state probabilities from the balance equations from each state. Our results can be applied to improve and design appropriate resource management solutions to reduce energy consumption and operational cost in the environments of information and communications technology (ICT) operators.
... Open to overloading of servers mostly for popular web server. (1) Load balancing satisfies two important needs, first to make use of servers efficiently and second to promote performance. In order for load evaluation to work it should also convince below major goals: a. Cost effective: System performance is improved at an economical cost. ...
Article
Full-text available
Today there is an increase in energy consumption which is generated around the world simultaneously day by day. The growth as well resource allocation, scheduling is depended on the Data centres for the cloud load balancing to hold the circulation of workload traffic and demands. Here the need for energy efficiency is emphasized by making cloud computing as a vital contributor to save the energy consumption and to reduce the energy wastage. The most important challenge is optimization of energy utilization in this paper the context of cloud computing reviews the usage of methods and technologies used in energy efficient operations. In search of under best practices in certain area, this paper identifies some of the key research challenges will come to light in certain energy saving techniques are outstretched for use in cloud computing domain.
... When studying energy efficiency in cloud computing, we usually focus on parameters such as CPU usage, memory usage, network traffic, power consumption, number of executed instructions and execution time [2]. There are complex correlations and influences between these parameters, and traditional optimisation methods often find it difficult to comprehensively consider the interactions between these factors. ...
Preprint
For this paper, a prediction study of cloud computing energy consumption was conducted by optimising the data regression algorithm based on the horned lizard optimisation algorithm for Convolutional Neural Networks-Bi-Directional Gated Recurrent Units. Firstly, through Spearman correlation analysis of CPU, usage, memory usage, network traffic, power consumption, number of instructions executed, execution time and energy efficiency, we found that power consumption has the highest degree of positive correlation with energy efficiency, while CPU usage has the highest degree of negative correlation with energy efficiency. In our experiments, we introduced a random forest model and an optimisation model based on the horned lizard optimisation algorithm for testing, and the results show that the optimisation algorithm has better prediction results compared to the random forest model. Specifically, the mean square error (MSE) of the optimisation algorithm is 0.01 smaller than that of the random forest model, and the mean absolute error (MAE) is 0.01 smaller than that of the random forest.3 The results of the combined metrics show that the optimisation algorithm performs more accurately and reliably in predicting energy efficiency. This research result provides new ideas and methods to improve the energy efficiency of cloud computing systems. This research not only expands the scope of application in the field of cloud computing, but also provides a strong support for improving the energy use efficiency of the system.
Article
Full-text available
Internet hosting centers serve multiple service sites from a common hardware base. This paper presents the design and implementation of an architecture for resource management in a hosting center operating system, with an emphasis on energy as a driving resource management issue for large server clusters. The goals are to provision server resources for co-hosted services in a way that automatically adapts to offered load, improve the energy efficiency of server clusters by dynamically resizing the active server set, and respond to power supply disruptions or thermal events by degrading service in accordance with negotiated Service Level Agreements (SLAs).Our system is based on an economic approach to managing shared server resources, in which services "bid" for resources as a function of delivered performance. The system continuously monitors load and plans resource allotments by estimating the value of their effects on service performance. A greedy resource allocation algorithm adjusts resource prices to balance supply and demand, allocating resources to their most efficient use. A reconfigurable server switching infrastructure directs request traffic to the servers assigned to each service. Experimental results from a prototype confirm that the system adapts to offered load and resource availability, and can reduce server energy usage by 29% or more for a typical Web workload.
Article
Associate Professor & HOD (CSE),IGCE Punjab Technical univ. Punjab technical univ. Punjab technical Univ. Abstract— Cloud computing is current buzzword in the market. It is paradigm in which the resources can be leveraged on per use basis thus reducing the cost and complexity of service providers. Cloud computing promises to cut operational and capital costs and more importantly let IT departments focus on strategic projects instead of keeping datacenters running. It is much more than simple internet. It is a construct that allows user to access applications that actually reside at location other than user's own computer or other Internet-connected devices. There are numerous benefits of this construct. For instance other company hosts user application. This implies that they handle cost of servers, they manage software updates and depending on the contract user pays less i.e. for the service only. Confidentiality, Integrity, Availability, Authenticity, and Privacy are essential concerns for both Cloud providers and consumers as well. Infrastructure as a Service (IaaS) serves as the foundation layer for the other delivery models, and a lack of security in this layer will certainly affect the other delivery models, i.e., PaaS, and SaaS that are built upon IaaS layer. This paper presents an elaborated study of IaaS components' security and determines vulnerabilities and countermeasures. Service Level Agreement should be Considered very much importance.
Conference Paper
The article deals with how selected companies doing business in the Czech Republic are using information technology and business strategy Customer Relationship Management (CRM). If they use complete solutions of CRM as purchase and installation of software or if they use the CRM system only as part of services which are provided through Cloud Computing technology. Apparently also because of ending of the economic crisis is CRM in the Czech Republic beginning more use. CRM is implemented to help increase quality of relationship and communication with customers. In the article there are described types of Cloud Computing and most common definition of CRM. The analysis of using of CRM and Cloud Computing in selected companies, which are grouped by number of employees, is described, too. At the end there is summary of analysis results and contribution of CRM and Cloud Computing.
Conference Paper
Electricity consumption of data centers increases continuously. Beside of the IT industry which tries to reduce this consumption by improving efficiency of components in data centers, there are research solutions based on an optimized energy management of data centers by acting on the IT load placement, then on cooling, start-up and shut down. In this context, this paper focus on energetic modeling of servers in data centers. In the state of art, the IT load is usually presented as a whole unit by means of the percentage CPU, while in this work, the percentage CPU is separated in two parts. The first one is the percentage CPU due to server self applications (for example a virtual machine manager), while the second part is due to services turning on the server. This classification led to a new linear model which shows that electricity consumption of data centers can be modeled as accumulated layers depending on what kind of software is running on the servers. The model is developed and then validated with experimental measurements on actual server conduct with the help of industrial partners. This modeling presents the first step of further works aim to optimize the energy consumption of data centers by knowing the IT load that is held on its servers.
Article
In den stark automatisierten, arbeitsteiligen Wirtschaftsystemen der Industrienationen ist die eine effiziente und zuverlässige zentrale Informationstechnik (IT) der Unternehmen entscheidend für den Geschäftserfolg. Immer mehr Unternehmensabläufe werden durch die IT unterstützt. Vielfach ist es sogar nur noch durch die umfassende IT-Unterstützung der Geschäftsprozesse möglich, im globalen Wettbewerb erfolgreich zu sein. Die installierte Rechenleistung in modernen Unternehmen steigt dabei ständig an. Neue und verbesserte Anwendungen und Programm-Features erfordern leistungsfähige Server. In den Rechenzentren kleiner und mittlerer Unternehmen sind heute Rechenleistungen installiert, die vor wenigen Jahren ausschließlich einigen Großunternehmen vorbehalten waren. Diese im Grundsatz positive Entwicklung kann aber auch zu hohen Strom- und Kühlleistungen in Rechenzentren führen. Damit stellen sich neue Herausforderungen an die Planung, Ausführung und den Betrieb einer IT-Infrastruktur.
Article
The Xen Virtual Machine Monitor will soon be undergoing its third major release, and is ma- turing into a stable, secure, and full-featured virtualization solution for Linux and other op- erating systems. Xen has attracted consider- able development interest over the past year, and consequently the 3.0 release includes many exciting new features. This paper provides an overview of the major new features, includ- ing VM relocation, device driver isolation, sup- port for unmodified operating systems, and new hardware support for both x86/64 and IA-64 processors.
Article
Energy efficiency in the field of information and communication technology becomes increasingly important due to the increase in energy costs and the desire to reduce CO$_2$ emissions. Office environments of public administration and companies promise high potential in terms of energy saving. In such environments a high number of hosts are operating on a 24/7 basis. This paper suggests an Energy-Efficient Distributed Office Environment that reduces the number of always-on hosts and raises the energy efficiency within office environments. It describes the suggested architecture and its key technologies. In addition, it evaluates the benefits of the architecture analytically and illustrates experimental results to evaluate resulting overhead.
Article
The Cooperative Expendable Micro-Slice Servers (CEMS) project evaluates low cost, low power servers for high-scale internet-services using commodity, client-side components. It is a follow-on project to the 2007 CIDR paper Architecture for Modular Data Centers [11]. The goals of the CEMS project are to establish that low-cost, low-power servers produce better price/performance and better power/performance than current purpose-built servers. In addition, we aim to establish the viability and efficiency of a fail-in-place model. We use work done per dollar and work done per joule as measures of server efficiency and show that more, lower-power servers produce the same aggregate throughput much more cost effectively and we use measured performance results from a large, consumer internet service to argue this point.