While hearing presentations at the NYC Oracle Enterprise Cloud Computing Summit earlier this year, I found myself wondering about the impact cloud computing would have on the electrical needs of large data centers. Cloud computing, for the purposes of this column, means front-end computer users outsourcing activities such as processing, storage, applications, and services provided by data centers to back-end providers.
With that issue in mind, I asked several Summit attendees the following questions, “What effect do you believe rapidly growing cloud computing will have upon demand for data center resources, and what will be the impact upon data center electricity requirements?” The answers evidenced a strong belief that cloud computing, by spreading the demand for computing resources from limited front end-users to extensive back-end capabilities, enables greatly enhanced performance and makes much more efficient use of available resources. Everyone I asked felt that efficiency gains on the computation side would translate into comparable reductions in data center demand for electricity as compared to demand without cloud computing.
Despite these answers, my experience in the electric industry suggests to me that cloud computing will not reduce the growth of electrical demand. In fact, it is more likely to lead to increases.
An electric utility rate case considers the elasticity of demand relative to cost of service. A low cost signal promotes greater usage. Cloud services, at least initially, are likely to be offered at low metered cost based upon usage. One reason for expected initial low costs is that, so long as available resources can serve the load, the anticipated costs would likely be more OPEX than CAPEX. Also, services that would be included within cloud computing such as data storage are often packaged with other services and not provided separately. Gmail storage, which is limited to 8 gigabytes per account, is one example. The costs of these services on the cloud may be partially or completely defrayed by website advertising revenues.
Assuming that the cost is low, cloud computing that offers comprehensive, readily available, and vastly enhanced computer capabilities could result in rapidly escalating demand that may be difficult to limit since front-end, intermediary, and back-end participants would be ubiquitous, varied, and multi-national. Internet access to cloud computing by cell phones using operating systems such as those provided by Apple or Google are likely to exponentially increase use.
Other such developments might overwhelm existing computing capacity and increase rather than decrease the upward curve of electric power requirements. For that reason, adopting the assumption that cloud computing will provide economies of scale and fuller use of existing computational resources that will, in turn, decrease the amount of electric power that would otherwise be required, and to base electrical supply planning on that assumption, would present a risk of insufficient future supply.
A leading software provider recently emphasized the relationship between electric power and data centers. CNN Money.Com’s Steve Hargreaves recently quoted Microsoft CEO Steve Ballmer as saying at CERAWeek 2011 (IHS Cambridge Energy Research Associates, Inc.’s annual energy conference), “Cloud computing basically outsources the hard drive functions on individual machines to giant, centrally located server farms, boosting efficiency and power…the size of these server farms isn’t measured in square feet or dollars, but rather in kilowatt hours of energy consumption.” Ballmer also mentioned that Microsoft’s energy use is huge and that it is constantly seeking to make its use more efficient. Microsoft’s commitment to efficient energy use is consistent with a general overall trend towards greater data center energy efficiency-although the level of efficiency achieved by individual data centers varies.
It is vital to realistically project the impact of cloud computing when evaluating future electrical requirements. Large data centers are proliferating, and these can require transmission level (50 to 100+ MW) electric supplies on a 24x7 basis since data centers generally have substantially level demand curves. Looking ahead to meet such needs should include a review of the possible need to build new transmission lines and even new power plants. Regulatory siting issues and other requirements, together with the potential of having to condemn land, can require lengthy public processes that dictate that factors such as the impact of cloud computing should be considered long in advance.
A first-level electricity use analysis suggests:
Personal and organizational individual users who outsource to giant server farms to achieve greater speed and storage will use electricity to power individual computers during the outsourcing process
Additional electricity will be necessary to power intermediary resources during the back and forth transmission process
- Outsourcing will place a processing load upon the server farm causing it to use more electric power while providing cloud-computing services than would be the case without such a service. This buttresses the concern that, while cloud computing may yield significant computing benefits to individual machine users and provide business opportunities to data center operators, it is unlikely to lead to reduced data center growth or to reduce in the upward growth curve of data center energy use.
For these reasons, depending upon the cost of cloud services, it is reasonable to expect increased rather than reduced overall demands being imposed upon data centers as a result of cloud computing. On an even more fundamental level, it is common knowledge that the majority of computer users desire greater speed and more capability and will use it when it is available. As a corollary, assuming all other things are equal, those data centers with the greatest energy efficiency will be best positioned to serve the cloud-computing community in a cost-efficient manner.
The expansion of the cloud is also likely to see computational pricing models that bear similarity to electric system pricing. For example, these cloud pricing models might even include time-of-day pricing and demand response (reducing or shifting demand or firing up a spare computer) that seek to mitigate peak demand to make better use of resources and reduce prices. This analogy can take us only so far since these are separate systems with different types of components, which brings to mind a similar difference between electric power and telephone service since, unlike telephones, there is no busy signal (or “server unavailable” response) in electric service while the grid is in operation.
From a legal perspective, there are also factors that suggest that cloud computing may result in increased use. For example, discovery of electronically stored information (ESI) requires searches of information that are generally stored within a litigant’s system. Courts have also ordered seizure of computers, hard drives, and CD-ROMs, but such seizures could not be enforced upon cloud data. Discovery located in the cloud would likely require exhaustive searches of all information stored by that litigant. In light of the many lawsuits filed each year that involve discovery of ESI, conducting extensive searches of cloud stored information would impose additional demands upon the use of cloud systems.
Efficiencies and economies of scale in a cloud-computing network, therefore, do not necessarily mean that there will be a like effect upon the electric system. In fact, it appears that cloud computing might result in greater demands for electric power.
If so, this would put a premium upon energy efficiency. It would also mean that for certain locations and circumstances, data center owners might implement self-generation. Cogeneration would appear to be a particularly attractive choice, assuming a supply of natural or biogas, since the data center could become a thermal host for steam or hot water produced by the cogeneration system by installing absorption chillers to provide cooling. Data center owners have not embraced on-site economic generation such as cogeneration, but tight power supplies may mandate a closer look at alternatives such a cogeneration in the future.