Green Cloud Computing_ Balancing Energy in Processing, Storage, and Transport

CONTRIBUTED P A P E R Green Cloud Computing: Balancing Energy in Processing, Storage, and Transport For processing large amounts of data, management and switching of communications may contribute significantly to energy consumption and cloud computing seems to be an alternative to office-based computing. By Jayant Baliga, Robert W. A. Ayre, Kerry Hinton, and Rodney S. Tucker, Fellow IEEE ABSTRACT | Network-based cloud computing is rapidly expanding as an alternative to conventional office-based computing. As cloud computing becomes more widespread, the energy consumption of the network and computing resources that underpin the cloud will grow. This is happening at a time when there is increasing attention being paid to the need to manage energy consumption across the entire information and communications technology (ICT) sector. While data center energy use has received much attention recently, there has been less attention paid to the energy consumption of the transmission and switching networks that are key to connecting users to the cloud. In this paper, we present an analysis of energy consumption in cloud computing. The analysis considers both public and private clouds, and includes energy consumption in switching and transmission as well as data processing and data storage. We show that energy consumption in transport and switching can be a significant percentage of total energy consumption in cloud computing. Cloud computing can enable more energy-efficient use of computing power, especially when the computing tasks are of low intensity or infrequent. However, under some circum- stances cloud computing can consume more energy than conventional computing where each user performs all com- puting on their own personal computer (PC). KEYWORDS | Cloud computing; core networks; data centers; energy consumption I . INTRODUCTION The increasing availability of high-speed Internet and corporate IP connections is enabling the delivery of new network-based services [1]. While Internet-based mail services have been operating for many years, service offerings have recently expanded to include network-based storage and network-based computing. These new services are being offered both to corporate and individual end users [2], [3]. Services of this type have been generically called Bcloud computing[ services [2]–[7]. The cloud computing service model involves the provision, by a service provider, of large pools of high- performance computing resources and high-capacity stor- age devices that are shared among end users as required [8]–[10]. There are many cloud service models, but generally, end users subscribing to the service have their data hosted by the service, and have computing resources allocated on demand from the pool. The service provider’s offering may also extend to the software applications re- quired by the end user. To be successful, the cloud service model also requires a high-speed network to provide con- nection between the end user and the service provider’s infrastructure. Cloud computing potentially offers an overall financial benefit, in that end users share a large, centrally managed pool of storage and computing resources, rather than owning and managing their own systems [5]. Often using existing data centers as a basis, cloud service providers invest in the necessary infrastructure and management Manuscript received November 26, 2009; accepted July 3, 2010. Date of publication August 30, 2010; date of current version December 17, 2010. This work was supported by the Australian Research Council and by Cisco Systems. The authors are with the Department of Electrical and Electronic Engineering, University of Melbourne, Melbourne, Vic. 3010, Australia (e-mail: jbaliga@gmail.com; r.ayre@ee.unimelb.edu.au; k.hinton@unimelb.edu.au; r.tucker@unimelb.edu.au). Digital Object Identifier: 10.1109/JPROC.2010.2060451 Vol. 99, No. 1, January 2011 | Proceedings of the IEEE 1490018-9219/$26.00 �2010 IEEE systems, and in return receive a time-based or usage-based fee from end users [6], [8]. Since at any one time, sub- stantial numbers of end users are inactive, the service provider reaps the benefits of the economies of scale and from statistical multiplexing, and receives a regular in- come stream from the investment by means of service subscriptions [6]. The end user in turn sees convenience benefits from having data and services available from any location, from having data backups centrally managed, from the availability of increased capacity when needed, and from usage-based charging [2], [3]. The last point is impor- tant for many users in that it averts the need for a large one- off investment in hardware, sized to suit maximum demand, and requiring upgrading every few years [5]. There are many definitions of cloud computing, and discussion within the IT industry continues over the pos- sible services that will be offered in the future [8], [10]. The broad scope of cloud computing is succinctly sum- marized in [11]: Cloud computing is a model for enabling convenient, on-demand network access to a shared pool of con- figurable computing resources that can be rapidly pro- visioned and released with minimal management effort or service provider interaction. Cloud computing architectures can be either public or private [8], [9]. A private cloud is hosted within an enterprise, behind its firewall, and intended only to be used by that enterprise [8]. In such cases, the enterprise invests in and manages its own cloud infrastructure, but gains benefits from pooling a smaller number of centrally maintained high-performance computing and storage resources instead of deploying large numbers of lower performance systems. Further benefits flow from the centralized maintenance of software packages, data back- ups, and balancing the volume of user demands across multiple servers or multiple data center sites. In contrast, a public cloud is hosted on the Internet and designed to be used by any user with an Internet connection to provide a similar range of capabilities and services [8]. A number of organizations are already hosting and/or offering cloud computing services. Examples include Google Docs [12], Amazon’s Elastic Compute Cloud and Simple Storage services [13], Microsoft’s Windows Azure Platform [14], IBM’s Smart Business Services [15], Salesforce.com [16], and Webex [17]. But while its financial benefits have been widely dis- cussed, the shift in energy usage in a cloud computing model has received little attention. Through the use of large shared servers and storage units, cloud computing can offer energy savings in the provision of computing and storage services, particularly if the end user migrates toward the use of a computer or a terminal of lower capa- bility and lower energy consumption. At the same time, cloud computing leads to increases in network traffic and the associated network energy consumption. In this paper, we explore the balance between server energy consump- tion, network energy consumption, and end-user energy consumption, to present a fuller assessment of the benefits of cloud computing. The issue of energy consumption in information tech- nology equipment has been receiving increasing attention in recent years and there is growing recognition of the need to manage energy consumption across the entire information and communications technology (ICT) sector [18]–[20]. It is estimated that data centers accounted for approximately 1.2% of total United States electricity consumption in 2005 [20]. The transmission and switch- ing networks in the Internet account for another 0.4% of total electricity consumption in broadband-enabled countries [21]. In addition to the obvious need to reduce the greenhouse impact of the ICT sector [4], [19]–[22], this need to reduce energy consumption is also driven by the engineering challenges and cost of managing the power consumption of large data centers and associated cooling [23], [24]. Against this, cloud computing will involve in- creasing size and capacity of data centers and of networks, but if properly managed, cloud computing can potentially lead to overall energy savings. The management of power consumption in data centers has led to a number of substantial improvements in energy efficiency [25], [26]. Cloud computing infra- structure is housed in data centers and has benefited significantly from these advances. Techniques such as, for example, sleep scheduling and virtualization of computing resources in cloud computing data centers improve the energy efficiency of cloud computing [24]. While it is important to understand how to minimize energy consumption in data centers that host cloud com- puting services, it is also important to consider the energy required to transport data to and from the end user and the energy consumed by the end-user interface. Previous studies of energy consumption in cloud computing [4], [24], [27] have focused only on the energy consumed in the data center. However, to obtain a clear picture of the total energy consumption of a cloud computing service, and understand the potential role of cloud computing to provide energy savings, a more comprehensive analysis is required. In this paper, we present an overview of energy con- sumption in cloud computing and compare this to energy consumption in conventional computing. For this com- parison, the energy consumption of conventional comput- ing is the energy consumed when the same task is carried out on a standard consumer personal computer (PC) that is connected to the Internet but does not utilize cloud com- puting. We consider both public and private clouds and include energy consumption in switching and transmis- sion, as well as data processing and data storage. Speci- fically, we present a network-based model of the switching and transmission network [21], [28], [29], a model of user Baliga et al. : Green Cloud Computing: Balancing Energy in Processing, Storage, and Transport 150 Proceedings of the IEEE | Vol. 99, No. 1, January 2011 computing equipment, and a model of the processing and storage functions in data centers [7], [30], [31]. We exa- mine a variety of cloud computing service scenarios in terms of energy efficiency. In essence, our approach is to view cloud computing as an analog of a classical supply chain logistics problem, which considers the energy con- sumption or cost of processing, storing, and transporting physical items. The difference is that in our case, the items are bits of data. As with classical logistics modeling, our analysis allows a variety of scenarios to be analyzed and optimized according to specified objectives. We explore a number of practical examples in which users/customers outsource their computing and storage needs to a public cloud or private cloud [8], [9]. Three cloud computing services are considered, including storage as a service [3]–[6], [8], processing as a service [2]–[6], [8], and software as a service [2]–[4], [6], [8]. As the name implies, storage as a service allows users to store data in the cloud. Processing as a service gives users the ability to outsource selected computationally intensive tasks to the cloud. Software as a service combines these two services and allows users to outsource all their computing to the cloud and use only a very-low-processing-power terminal at home. We show that energy consumption in transport and switching can be a significant percentage of total energy consumption in cloud computing. Cloud computing can enable more energy-efficient use of computing power, especially when the users’ predominant computing tasks are of low intensity or arise infrequently. However, we show that under some circumstances cloud computing can consume more energy than conventional computing on a local PC. Our broad conclusion is that cloud computing can offer significant energy savings through techniques such as virtualization and consolidation of servers [25], [32] and advanced cooling systems [26], [33]. However, cloud computing is not always the greenest computing technology. II . CLOUD SERVICE MODELS We focus our attention on three cloud servicesVstorage as a service, processing as a service and software as a service. In the following sections, we outline the functionality of each of the three cloud services. Note that we use the terms Bclient,[ Buser,[ and Bcustomer[ interchangeably. A. Software as a Service Consumer software is traditionally purchased with a fixed upfront payment for a license and a copy of the software on appropriate media. This software license typi- cally only permits the user to install the software on one computer. When a major update is applied to the software and a new version is released, users are required to make a further payment to use the new version of the software. Users can continue to use an older version, but once a new version of software has been released, support for older versions is often significantly reduced and updates are infrequent. With the ubiquitous availability of broadband Internet, software developers are increasingly moving towards providing software as a service [2]–[4], [6]. In this service, clients are charged a monthly or yearly fee for access to the latest version of software [2], [3]. Additionally, the soft- ware is hosted in the cloud and all computation is per- formed in the cloud. The client’s PC is only used to transmit commands and receive results. Typically, users are free to use any computer connected to the Internet. However, at any time, only a fixed number of instances of the software are permitted to be running per user. One example of software as a service is Google Docs [12]. When a user exclusively uses network- or Internet- based software services, the concept is similar to a Bthin client[ model, where each user’s client computer functions primarily as a network terminal, performing input, output, and display tasks, while data are stored and processed on a central server. Thin clients were popular in office environments prior to the widespread use of PCs. In Section IV, we explore the opportunity for reduced energy consumption in the client’s PC when we only use software services. In this scenario, data storage and pro- cessing is always performed in the cloud and we are thus able to significantly reduce the functionality, and conse- quently, the power consumption, of the client’s PC. B. Storage as a Service Through storage as a service, users can outsource their data storage requirements to the cloud [3]–[6]. All pro- cessing is performed on the user’s PC, which may have only a solid state drive (e.g., flash-based solid-state stor- age), and the user’s primary data storage is in the cloud. Data files may include documents, photographs, or videos. Files stored in the cloud can be accessed from any com- puter with an Internet connection at any time [5]. How- ever, to make a modification to a file, it must first be downloaded, edited using the user’s PC and then the modified file uploaded back to the cloud. The cloud service provider ensures there is sufficient free space in the cloud and also manages the backup of data [5]. In addition, after a user uploads a file to the cloud, the user can grant read and/or modification privileges to other users. One example of storage as a service is the Amazon Simple Storage service [13]. C. Processing as a Service Processing as a service provides users with the re- sources of a powerful server for specific large computa- tional tasks [2]–[6]. The majority of tasks, which are not computationally demanding, are carried out on the user’s PC. More demanding computing tasks are uploaded to the cloud, processed in the cloud, and the results are Baliga et al. : Green Cloud Computing: Balancing Energy in Processing, Storage, and Transport Vol. 99, No. 1, January 2011 | Proceedings of the IEEE 151 returned to the user [6]. Similar to the storage service, the processing service can be accessed from any computer connected to the Internet. One example of pro- cessing as a service is the Amazon Elastic Compute Cloud service [13]. When utilizing a processing service, the user’s PC still performs many small tasks and is consequently required to be more powerful than the Bthin client[ considered in the software service (Section II-A). However, the user’s computer is not used for large computationally intensive tasks and so there is scope to reduce its cost and energy consumption, relative to a standard consumer PC, by using a less powerful computer. D. Summary of Models Table 1 provides a summary of the location of pro- cessing, location of storage, and function of transport for each of these cloud services. In a storage service, the majority of processing occurs at the user’s PC (the client) and the majority of storage is in the cloud. The trans- mission and switching network transports the user’s files between the data center and the user. With a processing service, the user’s computer processes only short tasks and the cloud processes large computationally intensive tasks. Long-term storage of data is on the user’s computer and transport is required to transfer the files relevant to each large task. In a software service, processing and storage are performed in the cloud. Transport is required for all tasks to enable transmission of commands to the cloud and to return the results. III . MODELS OF ENERGY CONSUMPTION In this section, we describe the functionality and energy consumption of the transport and computing equipment on which current cloud computing services typically operate. We consider energy consumption models of the transport network, the data center, plus a range of customer-owned terminals and computers. The models described are based on power consumption measurements and published specifications of representative equipment [7], [21], [22], [30]. Those models include descriptions of the common energy-saving techniques employed by cloud computing service providers. The models are used to calculate the energy consump- tion per bit for transport and processing, and the power consumption per bit for storage. The energy per bit and power per bit are fundamental measures of energy con- sumption, and the energy efficiency of cloud computing is the energy consumed per bit of data processed through cloud computing. Performing calculations in terms of energy per bit also allows the results to be easily scaled to any usage level. We consider both public and private clouds. Fig. 1 shows schematics of a public cloud computing network [Fig. 1(a)] and a private cloud computing network [Fig. 1(b)]. For the public cloud, the schematic includes the data center as well as access, metro and edge, and core networks. The private cloud schematic includes the data center as well as a corporate network. These schematics form the basis for the analysis in the following sections of this paper. From a hardware perspective, the key difference between public cloud computing and private cloud computing is the network connecting the users to the respective data center. As described earlier, a data center for a public cloud is hosted on the Internet and designed to be used by anyone with an Internet connection. Public cloud users are typically residential users and connect to the public Internet through an Internet service provider’s (ISP) network. Looking forward, it is expected that the access portion of such networks will increasingly use passive optical network (PON) technologies, which are particularly energy efficient [34]. Within the ISP’s net- work, Ethernet switches aggregate user traffic, broadband net