HAL Id: hal-01160234https://hal.inria.fr/hal-01160234

Submitted on 21 Feb 2016

HAL is a multi-disciplinary open accessarchive for the deposit and dissemination of sci-entific research documents, whether they are pub-lished or not. The documents may come fromteaching and research institutions in France orabroad, or from public or private research centers.

L’archive ouverte pluridisciplinaire HAL, estdestinée au dépôt et à la diffusion de documentsscientifiques de niveau recherche, publiés ou non,émanant des établissements d’enseignement et derecherche français ou étrangers, des laboratoirespublics ou privés.

Interconnecting Smart Grids and Clouds to save EnergyAnne-Cécile Orgerie

To cite this version:Anne-Cécile Orgerie. Interconnecting Smart Grids and Clouds to save Energy. International Confer-ence on Smart Cities and Green ICT Systems (SMARTGREENS), May 2015, Lisbon, Portugal. pp.6.�hal-01160234�

Interconnecting Smart Grids and Clouds to Save Energy

Anne-Cecile OrgerieCNRS, IRISA, Rennes, Franceanne-cecile.orgerie@irisa.fr

Keywords: Energy Efficiency, Cloud Computing, Smart Grids

Abstract: Cloud computing is becoming an essential component for Internet services. However, its energy consump-tion has become a key environmental and economic concern. The distributed nature of Cloud infrastructuresinvolves that their components are spread across wide areas, interconnected through different networks, andpowered by diverse energy sources and providers, making overall energy monitoring and optimizations chal-lenging. In this paper, we present the opportunity brought by the Smart Grids to exploit renewable energyavailability and to optimize energy management in distributed Clouds. The presence of smart sensors whichare both integrated into the electricity Grid and connected to the Internet, indeed offers for the first time thepossibility of exploiting the availability of various energy sources, and of making complete energy measure-ments of all the Cloud resources – computing, storage and especially networking resources – problems whichhave previously been intractable.

1 INTRODUCTION

Global warming, oil reserve limits and ever-growingelectricity consumption are all powerful motivationsfor designing more energy-efficient systems. The en-ergy consumed to power Information and Communi-cation Technologies (ICT) has become a major con-cern. Data centers and network infrastructures repre-sent an ever-growing part of this consumption (Mills,2013). As an example, for 2010, Google used900,000 servers and consumed 1.9 billion kWh ofelectricity (Koomey, 2011). Electricity becomes thenew limiting factor for deploying data center equip-ment. Indeed, the energy consumption of state-of-the-art servers grows inexorably as they embed more andmore powerful cores and advanced features and tech-nologies. So, electricity consumption is hamperingtechnological advances. It also raises major environ-mental, economic and societal concerns. Curbing thisconsumption is thus crucial and urgent.

In this paper, we mainly aim at investigating howto exploit the capabilities of Smart Grids technologiesin the context of Cloud computing in order to save en-ergy. This has the potential for reducing global energyconsumption and thus, inducing real economic, envi-ronmental, and societal benefits. On the Cloud’s side,we also aim at increasing the user’s energy-awarenessand the Cloud’s energy-efficiency which are the mainlevers needed to trigger substantial involvement from

Cloud providers and designers in order to reduce theelectricity demand, and contribute to a more sustain-able ICT industry.

On the generation side, it is estimated that 10%of electric energy produced by power plants is cur-rently lost during transmission and distribution toconsumers, with 40% of these losses occurring on thedistribution network (Feng et al., 2009). As an ex-ample, in 2006 in the United-States, the total energylosses and distribution losses were about 1,638 billionand 655 billion kWh, respectively (Feng et al., 2009).Smart Grids provide the means to control the energysupply more efficiently and to dynamically managepeak load. We argue that their utilization will greatlyimprove understanding of the energy consumption ofClouds and will help to reduce it significantly.

The remainder of the paper is organized as fol-lows. Section 2 presents the context and motivation,which is further discussed over a canonical exampleon Section 3. Section 4 illustrates the problem com-plexity. The related work is presented in Section 5.Section 6 concludes and presents our future work.

2 CONTEXT AND MOTIVATION

Recently, the maturity of virtualization techniques hasled to the emergence of Cloud Computing, whichis a new paradigm becoming increasingly essential

to Information Technology (IT) services. It aimsto provide dynamic, reliable, customized and QoS-guaranteed environments to end-users. By putting to-gether individual requirements and by gathering theneeded equipment, Cloud providers can efficientlymanage and offer virtually unlimited resources, mini-mizing the costs incurred by organizations when pro-viding Internet services.

To offer better QoS and support more users,Clouds are distributed: they consist in several datacenters geographically distributed. Computer net-works are the crucial elements that interconnect thedata centers providing the computing and storage ca-pacities of these distributed Clouds. As the size ofClouds increases and their traffic demands diversify,computer network resources, inside and in-betweenthe data centers, are often stretched to their limits and,in many cases, become a performance bottleneck. Be-sides, they represent a non negligible part of the en-ergy consumption of these distributed systems (Org-erie et al., 2014).

Advances in distributed systems have historicallybeen related to improving their performance, scala-bility and quality of service. Yet, it is now urgentto drastically increase the efficiency of large-scaledistributed systems in order to curb the rising en-ergy consumption of ICT (Information and Commu-nications Technologies). Indeed, a recent study fo-cused on Clouds (Mills, 2013) estimates that the ICTecosystem now approaches 10% of world electric-ity generation. The authors promise a faster growthin ICT energy use, pointing out the information ap-petite of Big Data, big networks and big infrastructurewhich unavoidably leads to big power.

Smart Grid sensors are the monitoring devices ofSmart power Grids. They provide high fidelity controlthrough high-speed, two-way communication, sens-ing, and real-time coordination of all assets down tothe customer meter and the end-user devices (Prattet al., 2010). As they are literally located at the in-tersection between the electrical and communicationnetworks, smart sensors enable improvement in thestability, efficiency, flexibility, and reliability of elec-tricity services. Through their communicating sen-sors, they offer better support of renewable energyand more comprehensive views of the consumptionof infrastructures. As for the Smart Grid actuators,the operating counterpart of smart sensors, they canact directly on the devices to enforce energy-efficientpolicies (Hledik, 2009). For instance in smart homes,they can adjust air conditioning level according to theparameters detected by the smart sensors: the temper-ature, the operation of appliances producing heat andhuman presence.

While the design of smart actuators is under study,smart sensors starts to be deployed at the scale of en-tire cities (like in Lyon, France, with the smart cityproject for instance). This deployment comes alongwith a Big Data challenge since the collected data rep-resent huge amounts, coming from distributed places,and requiring cross-computations. While Cloudscome naturally to the rescue of Smart Grids for deal-ing with this big data issue, little attention has beenpaid to the benefits that Smart Grids could bring toClouds.

We propose exploiting Smart Grid technologies tocome to the rescue of energy-hungry Clouds. Unlikein traditional electrical distribution networks, wherepower can only be moved and scheduled in very lim-ited ways, Smart Grids dynamically and effectivelyadapt supply to demand. This recent technology of-fers the unique chance to monitor the energy con-sumption in real-time of entire and distributed infras-tructures through their smart sensors, to reduce en-ergy use through their smart actuators, and to act asthe bridge enabling hitherto impossible joint energyproduction/consumption synergies.

3 A CANONICAL EXAMPLE

When a user submits a Google request (such as asearch for instance), it is routed to a data center thatprocesses it, computes the answer and sends it backto the user. Google owns several data centers spreadacross the world. For performance reasons, the cen-ter answering the user’s request is more likely to bethe one closest to the user. However, this data cen-ter may be supplied by a far away energy source. Amore distant data center may be using a renewable en-ergy source which happens to be close by and avail-able at the time of the request. This request may haveconsumed less energy, or a different kind of energy(renewable or not), if it had been sent to this furtherdata center. In this case, the response time would havebeen increased but maybe not noticeably: a differenttrade-off between performance and energy-efficiencycould have been adopted.

However, two issues make this trade-off impos-sible to determine: 1) the cloud (hosting Google’sservices here) is not able to estimate the energy con-sumption of a user’s request because it implies vari-ous and distributed resources (several servers to pro-cess the request and communication networks to carrythe request and the answer), and 2) the cloud has nomeans of knowing where the energy for its data cen-ters comes from, and whether it is sufficiently avail-able at a given time. Such data could be provided by

CommunicationNetwork

ElectricalNetwork

User Router Power stations Data center

HardwareResources

Smart Gridlayer

Cloudlayer

Figure 1: Representation of the involved resources and layers on the example of a worldwide Cloud

the smart sensors included in Smart Grid infrastruc-tures. Smart sensors and Cloud’s knowledge about itsutilization could be combined to take near real-timeallocation decisions for the Cloud’s resources, and en-ergy management optimizations through the smart ac-tuators for the electrical Grid.

4 ILLUSTRATION OF THEPROBLEM COMPLEXITY

Figure 1 shows an example of a large-scale Cloud in-frastructure. Its hardware layer is constituted of twophysical components: data centers to process the userrequests and communication network resources (i.e.routers) to transport the requests and to link the datacenters together. Electricity, which is also a physicalresource, is distributed through the electrical networkfrom power stations to data centers and network re-sources.

But, as shown in Figure 1, the electrical networkand the communications network of a Cloud infras-tructure are not directly interconnected. So, the Cloudmanager has no access to the electrical information.This implies two inherent problems for the Cloudmanager which aims to optimize the energy consump-tion of the entire Cloud infrastructure: 1) it has noway to know the consumption of the networking re-sources as they belong to other (numerous) compa-nies; 2) it has no access to the electricity availabilityin the power stations used by its resources.

As illustrated in Figure 1, Smart Grids constitutethe autonomic management layer that enables cooper-ation among the electrical network and the communi-cation network. Our goal is to make the Cloud layerand the Smart Grid layer cooperate in order to save

energy in the Cloud infrastructures (with its underly-ing physical resources) and in the electrical network.As pointed out in the figure, the communication net-work will play a key role in this cooperation as itphysically stands in between.

Smart Grids have been designed to be deployed atlarge scale in order to improve the efficiency of theproduction and distribution of electricity in a compre-hensive context. Although some smart sensors (homesmart meters for instance) are already available, smartactuators are still under development. This currentlyleaves Smart Grids with a measurement role only (ex-ploited for a smarter accounting for instance), andconsiderably limits the power of Smart Grids, whichare here simply able to collect more detailed monitor-ing information than regular power grids. Thus, thecurrent status of deployed Smart Grid technologiesis not allowing them yet to dynamically adapt sup-ply to demand. Nonetheless, we believe that they willstrengthen in the forthcoming years in order to answerto the urgent need of smarter infrastructures for sav-ing energy.

In this paper, we limit the investigation of the po-tential impact of Smart Grids to distributed Cloud in-frastructures, impact which has not been estimated yetto the best of our knowledge. Interconnecting Cloudcomputing infrastructures and Smart Grid technolo-gies is not straightforward due to the complexity ofthese two systems. It requires to face multi-parametermodeling and multi-objective optimization challengeswith distributed algorithmic and real-time constraints.A fine-grained modelization of the considered Cloudapplications is also needed in order to be able to ac-comodate users’ expectations in terms of QoS and inorder to anticipate the impact of infrastructure recon-figurations and optimization algorithms.

5 RELATED WORK

Large-scale distributed systems – such as computingdata centers, Grids, Clouds and dedicated networks –consist of vast collections of computing and storageresources interconnected through a network. Thesesystems, with different levels of scalability, interoper-ability and respect to user constraints, have becomethe building blocks for numerous applications andservices ranging from weather forecast to web search.

Cloud infrastructures consist in geographicallydistributed data centers which are linked throughcommunication networks (Wang et al., 2008a). Withthe emergence of the Future Internet and the dawningof new IT models such as cloud computing, the usageof data centers (DC), and consequently their powerconsumption, increase dramatically. As an example,for 2010, Google used 900,000 servers and consumed1.9 billion kWh of electricity (Koomey, 2011). Otherbig Cloud companies present similar figures and sim-ilar issues (Katz, 2009).

Electricity becomes the new limiting factor for de-ploying data center equipment. Indeed, the energyconsumption of state-of-the-art servers grows inex-orably as they embed more and more powerful coresand advanced features and technologies. So, elec-tricity consumption is hampering technological ad-vances. It also raises major environmental, economicand social concerns. Curbing this consumption is thuscrucial and urgent.

The first way to save energy at a data center levelconsist in locating it close to where the electricity isgenerated, hence minimizing transmission losses. Forexample, Western North Carolina, USA, attracts datacenters with its low electricity prices due to abundantcapacity of coal and nuclear power following the de-parture of the region’s textile and furniture manufac-turing (Greenpeace, 2011). As of writing, this regionhas three super-size data centers from Google, Appleand Facebook with respective power demands of 60 to100 MW, 100 MW and 40 MW (Greenpeace, 2011).

Other companies opt for greener sources of en-ergy. For example, Quincy (Washington, USA) sup-plies electricity to data facilities from Yahoo, Mi-crosoft, Dell and Amazon with its low-cost hydro-electrics left behind following the shutting down ofthe region’s aluminum industry (Greenpeace, 2011).Several renewable energy sources like wind power,solar energy, hydro-power, bio-energy, geothermalpower and marine power can be considered to powerup super-sized facilities.

In spite of these approaches, numerous data facil-ities have already been built and cannot be moved.Cloud infrastructures, on the other hand, can still take

advantage of multiple locations to use green sourcesof energy with approaches such as follow-the-sun andfollow-the-wind (Figuerola et al., 2009). As sun andwind provide renewable sources of energy whose ca-pacity fluctuates over time, the rationale is to placecomputing jobs on resources using renewable energy,and migrate jobs as renewable energy becomes avail-able on resources in other locations.

Within the data center itself, a range of tech-nologies can be utilized to make cloud computinginfrastructures more energy efficient, including bet-ter cooling technologies, temperature-aware schedul-ing (Fan et al., 2007), Dynamic Voltage and Fre-quency Scaling (DVFS) (Snowdon et al., 2005), andresource virtualization (Talaber et al., 2009). Theuse of VMs (Barham et al., 2003) brings severalbenefits including environment and performance iso-lation; improved resource utilization by enablingworkload consolidation; and resource provisioningon demand. Nevertheless, such technologies shouldbe analyzed and used carefully for really improv-ing the energy-efficiency of computing infrastruc-tures (Miyoshi et al., 2002).

Concerning wired networking resources, whichconstitute an important part of Cloud infrastructures,the energy-efficient management techniques can bedivided into two primary categories (Orgerie et al.,2014; Bolla et al., 2011): sleeping techniques mak-ing use of future foreseen on/off capabilities of net-work devices, and rate adaptation techniques. Themost emblematic solution of the sleeping techniquesis named Low-Power Idle (Christensen et al., 2010),standardized under the name 802.03az, and startingto be implemented on Network Interface Cards andswitches available on the market. The precursor ofrate adaptation techniques is named ALR (adaptiveLink Rate) (Gunaratne et al., 2006), and proposes toadapt the negotiated link rates (between the two linkend devices) to the actual load.

As detailed in a survey (Orgerie et al., 2014),numerous works target resource allocation and taskscheduling with energy constraints for both comput-ing (Lefevre and Orgerie, 2010; Goiri et al., 2011),and networking (Baliga et al., 2011; Bolla et al., 2011;Betti et al., 2013) resources separately. However, forboth of these disjointed propositions, they either fo-cus on energy-efficient goals (ie. reducing the amountof consumed energy), or on energy-aware objectives(ie. maximizing the amount of consumed renewableenergy). Only few works consider both aims at thesame time, like it is the case in this position paper;and none considered both aims for both kind of re-sources: computing and networking.

The Cloud workload varies widely on a short time

scale (Elson and Howell, 2008). To quickly and de-centrally adapt Cloud resource allocation to renew-able energy availability, efficient real-time autonomicreconfiguration mechanisms are required. Centralizedsolutions have been proposed to this issue (Orgerieet al., 2014). Yet, their scalability is limited, and thus,their applicability to Cloud infrastructures is arguable.In order to solve this technical barrier, we argue forthe use of Software-Defined Networking techniques.These techniques allow for virtual networks decou-pling the control (routing messages for instance) anddata planes(users’ messages) (Jain and Paul, 2013).OpenFlow is currently the mostly used SDN technol-ogy (Lara et al., 2014).

Network virtualization is also a promising solu-tion to enable energy savings (Orgerie et al., 2014).Virtual ROuters On the Move (VROOM), a networkmanagement primitive, pioneering research work, ispresented in (Wang et al., 2008b). It simplifies themanagement of routers by virtualizing them and al-lowing them to freely move from one physical node toanother, thus avoiding logical topology changes. Themigration of virtual routers can also be used to saveenergy (Chen and Phillips, 2013).

On the electricity side, it is estimated that 10%of electric energy produced by power plants is cur-rently lost during transmission and distribution toconsumers, with 40% of these losses occurring on thedistribution network (Feng et al., 2009). As an ex-ample, in 2006 in the United-States, the total energylosses and distribution losses were about 1,638 billionand 655 billion kWh, respectively (Feng et al., 2009).Smart Grids provide the means to control the energysupply more efficiently and to dynamically managepeak load. Their utilization has the potential to greatlyimprove understanding of the energy consumption ofClouds and to help to reduce it significantly.

To take advantage from this new technology, weneed to deal with three major issues: 1) complexity in-curred by the proliferation of energy sensors, the factthat different systems are interconnected (communi-cation networks and distribution networks), these sys-tems are highly multi-scale by nature, different tech-nologies are employed (different manufacturers), anda lot of parameters are involved (energy sources andavailability, location, etc.); 2) interdependability be-cause our solution relies on the collaboration betweenelectricity networks and communication networks; 3)performance of the system: it is mandatory for themonitoring infrastructure to provide real-time anal-ysis in spite of the large data volumes to process,and for the Cloud management system to integratedistributed management and failure recovery mech-anisms.

Current works on Smart Grids are dealing withsmoothing energy peak demand (Tang et al., 2013),improving the effective utilization of intermittent andfluctuating renewables, and improving the scalabilityof the proposed solutions (Fang et al., 2012). Our goalhere is not to contribute to the development and tech-nical advances of Smart Grids. It only consists in uti-lizing this emerging and innovative technology.

6 CONCLUSION AND FUTUREWORK

Most of the work on energy-efficient managementof Clouds does not consider networking resources asthey go beyond the limits of the data centers. Thus,they are not entirely managed by one provider andtheir consumption is not accessible to the cloud man-ager. Yet, the intrinsic energy consumption of theGoogle request includes transporting the request toa data center, computing the request and transport-ing the answer back to the user. Taking into accountonly the computing costs leads to proposing optimiza-tion algorithms that do not optimize the overall energyconsumption, but only the consumption of one datacenter. Indeed, network resources account for a non-negligible part in the energy equation (Orgerie et al.,2014; Orgerie et al., 2011). Making Smart Grids andCloud managers collaborate could solve these issuesand lead to important energy savings.

Our future work will focus on exploiting this en-ergy saving potential resulting from the interconnec-tion of Smart Grids and Clouds. This challenging goalwill require to propose implementable distributed re-configuration mechanisms and task scheduling algo-rithms for the allocation of Cloud resources (bothnetworking and computing resources) taking into ac-count energy efficiency and utilization of renewableenergies. It will also led us to dsign an efficient man-agement of electricity distribution and utilization inthe context of distributed energy sources linked bySmart Grids and used to power a distributed Cloudinfrastructure. Working towards this objective shouldboth increase the knowledge about these two complexsystems – namely Cloud infrastructures and SmartGrids – and remove technological barriers towardstheir interconnection for coping with the big powerissue.

REFERENCES

Baliga, J., Ayre, R., Hinton, K., and Tucker, R. (2011).Green Cloud Computing: Balancing Energy in Pro-

cessing, Storage, and Transport. Proceedings of theIEEE, 99(1):149–167.

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. In ACMSymposium on Operating Systems Principles (SOSP),pages 164–177.

Betti, G., Amaldi, E., Capone, A., and Ercolani, G. (2013).Cost-aware optimization models for communicationnetworks with renewable energy sources. In IEEEConference on Computer Communications Workshops(INFOCOM Workshops), pages 25–30.

Bolla, R., Bruschi, R., Davoli, F., and Cucchietti, F. (2011).Energy Efficiency in the Future Internet: A Surveyof Existing Approaches and Trends in Energy-AwareFixed Network Infrastructures. IEEE Communica-tions Surveys and Tutorials, 13(2):223–244.

Chen, X. and Phillips, C. (2013). An evolutionary baseddynamic energy management framework for ip-over-dwdm core networks. In Pierson, J.-M., Da Costa, G.,and Dittmann, L., editors, Energy Efficiency in LargeScale Distributed Systems, Lecture Notes in ComputerScience, pages 233–247. Springer Berlin Heidelberg.

Christensen, K., Reviriego, P., Nordman, B., Bennett, M.,Mostowfi, M., and Maestro, J. (2010). IEEE 802.3az:the road to Energy Efficient Ethernet. IEEE Commu-nications Magazine, 48(11):50–56.

Elson, J. and Howell, J. (2008). Handling Flash Crowdsfrom Your Garage. In USENIX Annual Technical Con-ference, ATC’08, pages 171–184.

Fan, X., Weber, W.-D., and Barroso, L. A. (2007). Powerprovisioning for a warehouse-sized computer. In ACMInternational symposium on Computer architecture(ISCA), pages 13–23.

Fang, X., Misra, S., Xue, G., and Yang, D. (2012). SmartGrid: The New and Improved Power Grid: A Survey.IEEE Communications Surveys Tutorials, 14(4):944–980.

Feng, X., Peterson, W., Yang, F., Wickramasekara, G., andFinney, J. (2009). Smarter grids are more efficient.ABB review.

Figuerola, S., Lemay, M., Reijs, V., Savoie, M., and St. Ar-naud, B. (2009). Converged Optical Network Infras-tructures in Support of Future Internet and Grid Ser-vices Using IaaS to Reduce GHG Emissions. Journalof Lightwave Technology, 27(12):1941–1946.

Goiri, I., Le, K., Haque, M., Beauchea, R., Nguyen,T., Guitart, J., Torres, J., and Bianchini, R. (2011).GreenSlot: Scheduling energy consumption in greendatacenters. In International Conference for High Per-formance Computing, Networking, Storage and Anal-ysis (SC), pages 1–11.

Greenpeace (2011). How dirty is your data? Greenpeacereport.

Gunaratne, C., Christensen, K., and Suen, S. (2006). Ether-net Adaptive Link Rate (ALR): Analysis Of A BufferThreshold Policy. In IEEE Global Telecommunica-tions Conference (GLOBECOM), pages 1–6.

Hledik, R. (2009). How green is the smart grid? The Elec-tricity Journal, 22(3):29 – 41.

Jain, R. and Paul, S. (2013). Network virtualization andsoftware defined networking for cloud computing: asurvey. IEEE Communications Magazine, 51(11):24–31.

Katz, R. H. (2009). Tech Titans Building Boom. IEEESpectrum, 46(2):40–54.

Koomey, J. (2011). Growth in Data Center Electricity Use2005 to 2010. Analytics Press.

Lara, A., Kolasani, A., and Ramamurthy, B. (2014). Net-work Innovation using OpenFlow: A Survey. IEEECommunications Surveys Tutorials, 16(1):493–512.

Lefevre, L. and Orgerie, A.-C. (2010). Designing and Eval-uating an Energy Efficient Cloud. The Journal of Su-perComputing, 51(3):352–373.

Mills, M. (2013). The Cloud Begins With Coal - Big Data,Big Networks, Big Infrastructure, and Big Power.Technical report, Digital Power Group.

Miyoshi, A., Lefurgy, C., Van Hensbergen, E., Rajamony,R., and Rajkumar, R. (2002). Critical power slope:understanding the runtime effects of frequency scal-ing. In ACM International conference on Supercom-puting (ICS), pages 35–44.

Orgerie, A.-C., de Assuncao, M. D., and Lefevre, L. (2014).A Survey on Techniques for Improving the Energy Ef-ficiency of Large Scale Distributed Systems. ACMComputing Surveys, 46(4).

Orgerie, A.-C., Lefevre, L., and Guerin-Lassous, I. (2011).On the Energy Efficiency of Centralized and De-centralized Management for Reservation-Based Net-works. In IEEE GLOBECOM.

Pratt, R., Balducci, P., Gerkensmeyer, C., Katipamula, S.,Kintner-Meyer, M., Sanquist, T., Schneider, K., andSecrest, T. (2010). The Smart Grid: An Estimationof the Energy and CO2 Benefits. Technical report,Pacific Northwest National Laboratory (PNNL), U.S.Department of Energy.

Snowdon, D., Ruocco, S., and Heiser, G. (2005). PowerManagement and Dynamic Voltage Scaling: Mythsand Facts. In Workshop on Power Aware Real-timeComputing.

Talaber, R., Brey, T., and Lamers, L. (2009). Using Virtual-ization to Improve Data Center Efficiency. Technicalreport, The Green Grid.

Tang, S., Huang, Q., Li, X.-Y., and Wu, D. (2013). Smooth-ing the energy consumption: Peak demand reductionin smart grid. In IEEE Conference on Computer Com-munications (INFOCOM), pages 1133–1141.

Wang, L., Tao, J., Kunze, M., Castellanos, A., Kramer, D.,and Karl, W. (2008a). Scientific Cloud Computing:Early Definition and Experience. In IEEE Interna-tional Conference on High Performance Computingand Communications (HPCC), pages 825–830.

Wang, Y., Keller, E., Biskeborn, B., van der Merwe, J., andRexford, J. (2008b). Virtual routers on the move: liverouter migration as a network-management primitive.ACM SIGCOMM Computer Communication Review,38(4):231242.