Over the weekend, a two-part article in the New York Times took issue with power consumption by data centers. There has been substantial reaction from the industry, including a response from GigaOM’s very own Katie Fehrenbacher. She assessed the argument and its rebuttals, pointing out that while the data center industry can perhaps do more, it has in fact been focused on energy efficiency for quite some time.
Supporting this view, Gregory Mone argues in the new Communications of the ACM that “leading companies have begun revising the way they design, maintain, and monitor data centers, from the physical building all the way down to the hardware…” While the industry attempts to balance inexorable economic forces with social and environmental responsibility, I’d like to view the debate from a broader perspective: perhaps we’re focusing on the wrong question.
Rather than focusing on the percentage of global energy that data centers use, suppose we focused on whether there is a net benefit in the use of IT, expressed and mediated through computing in data centers and the cloud. After all, if one knew the winning lottery numbers, one wouldn’t begrudge the dollar for the ticket purchase versus the millions in winnings. A variety of mechanisms would seem to generate net benefits: substitution, optimization, innovation, location independence, virtualization and multitenancy, and dynamic resource allocation.
For example, IT can deliver a net reduction in energy use via substitution. Examples abound. Conducting an audio or video conference, mediated by cloud data centers and cloud-resident multipoint conference bridges uses energy, but substantially less than the gasoline and jet fuel that would be required to drive and fly the meeting participants to their destination.
By one estimate, each meeting held this way can generate an average net benefit of nearly a half metric ton of greenhouse gases. Streaming a video out of a content delivery network takes energy, but less than the amount required to drive to and from the neighborhood video store, in the worst case for both pickup and return. Perhaps this is why data centers are responsible for less than 1.5 percent of global energy use, whereas transportation uses a reported 25 percent.
IT can also deliver a net reduction in energy use via optimization. UPS for example, has reportedly reduced both miles of delivery routes and tons of CO2 emissions by tens of millions using “package flow” software. In one recent study, mere feedback on energy consumption decreased energy use by several percent on average, but up to 25 percent among the “cybernetically sensitive.” Smart grids have been projected to reduce peak power requirements and thus achieve 4-to-6 percent efficiency improvements.
A related benefit is innovation. Compute-intensive simulation modeling is required to improve jet engines, automotive aerodynamics, molecular dynamics for drug discovery, and so forth, in virtually every domain of technology. And, open contests and innovation markets such as Innocentive leverage what Don Tapscott calls “networked intelligence” to accelerate collaborative innovation and crowdsourcing. These innovations, in turn, generate net benefits.
Location independence, a key attribute of cloud services, means that processing can be done where power is cheap. Restated, this means that relative prices, which express relative costs of generation plus costs associated with transmission and distribution losses, can signal globally optimal locations for processing and storage of information.
Virtualization and multitenancy drive additional efficiencies. Cloud providers don’t maintain a dedicated server for you personally on the off chance that you might want to send or receive an email. Rather, in the same way that thousands of people can enjoy Central Park, thousands of people can send and receive email via a single server, each using a small fraction of the available resources, generating orders of magnitude of efficiency. In the case of at least one SaaS provider, it’s been calculated that there can be a 65-fold savings Stanford’s Jonathan Koomey concludes that lower than projected growth rates in data center electricity use over the past few years are due in large part to adoption of virtualization.
A related efficiency gain in the cloud is due to dynamic resource allocation. In the same way that a hotel allocates and reallocates its rooms over time, a cloud provider can allocate and reallocate compute, storage, and network resources. When individual customer demands are independent, utilization can increase to nearly 100 percent, especially when given a boost by dynamic pricing. This is how the MGM Grand can achieve a reported average occupancy rate for over 5,000 rooms of 96.8 percent. Daily usage fluctuations impact utilization driven solely by interactive user demand, but this is where yield management and dynamic pricing through mechanisms such as spot instances can enhance overall cloud utilization.
Andreesen-Horowitz’s Mark Andreesen has famously argued that “software is eating the world,” and Kleiner Perkins Caufield Byers’ Mary Meeker has chronicled the “re-imagination of nearly everything”—news, photography, diaries, communication, etc.— to an Internet / cloud based model. The increasing transformation of the global economy from atoms to bits means that global productivity growth and therefore human welfare increasingly depend on harnessing IT, whether for the common good or for competitive advantage. And, yes, IT uses electricity. However, like preventative healthcare or education, in addition to optimizing the investment, perhaps we also focus on the return.
Joe Weinman is a senior vice president at Telx, the author of Cloudonomics: The Business Value of Cloud Computing, and a regular guest contributor to GigaOM. You can find him on Twitter @joeweinman and he’ll also be speaking at our upcoming Structure:Europe event in October.