Data Center Efficiency at the Hyperscale

Introduction

Fig. 1: This is a data center of BalticServers. (Source: Wikimedia Commons)

According to some estimates, in 2011 the world produced around 1.8 zettabytes of data (where a zettabyte is an eye-watering 10²¹ bytes). This has grown to around 44 zettabytes in less than 10 years! [1] It is easy to imagine digital data as something intangible and ephemeral. This is not true; even in the ultra-light and compact world of modern electronics, data must be physically stored somewhere. Data centers - brick and mortar building hosting servers that process and store the data produced by various digital services - are the information hubs of the modern world. Fig. 1 shows an example interior of a data center.

Data Center Energy Efficiency

In the United States alone, there are over 2600 data centers (representing 33% of the global total), used by leading tech companies such as Google, Meta and Amazon. [2] These data centers utilize a great deal of energy; in the US, around 1.8% of total energy consumed was used by data centers in 2014, which was around 70 billion kWh. [3] It is possible that this number is quite a bit higher now; as data center usage increased by nearly sixfold from 2010 to 2018. [4] However, numbers related to data center energy usage should be taken with a grain of salt. Studies of data center energy usage are quite complicated, with very large variances depending on the sources and/or models used. [5] Additionally, estimates (including the one used previously in this article) regarding total energy consumption usually rely on commercial/private sources, with all of the raw data not being available via either open access or journal subscriptions. [5]

Despite these complexities, it remains important to understand the efficiency of data centers. One of the ways to do this is by the power usage efficiency (PUE), defined as

The ideal PUE is 1.0, from the fact that in a perfectly efficient scenario, all the energy required by the data center would go directly to powering its computing equipment. However, the need for cooling, lighting etc. makes reaching values close to 1.0 extremely difficult.

Hyperscale Data Centers

Not all data centers are created equal (from an efficiency perspective). According to data from the Lawrence Berkeley National Laboratory, closet-sized data centers (which cover <100 ft²) have PUEs of around 2.0, high-end service provider data centers (typically >20,000 ft²) have PUEs of 1.7, while hyperscale data centers (up to over 400,000 ft²) have PUEs of 1.2! [3] The same paper estimates that if 80% of small-scale US data shifted to hyperscale centers, there could be a 25% decrease in energy usage. [3]

Hyperscale data centers are more efficient by design. For example, hyperscale data centers may house ultra-utilitarian servers that do not have superfluous power draws such as blinking lights or video connectors. [6] Their servers are also used more productively; in conventional data centers around 25% of servers soak up power but do not perform any productive work often due to corporations over- budgeting for peak loads. [3,6] Indeed, one server in a hyperscale center can replace 3.75 servers in more conventional, smaller centers. [6] Furthermore, huge energy savings are often made by locating hyperscale data centers in cooler climates, as conventional data centers use around 40% of their energy on cooling. [6]

Conclusion

As the reliance on the internet and related digital infrastructure continues to grow, so too does the need for more data centers. There is already a shift towards using hyperscale data centers. By December 2021, 700 hyperscale facilities were in use, with over 60 new centers being set up every year. [5] However, further advancements in energy efficient computing are still necessary to keep up with humanity's ballooning demand for digital services.

© Hannah Ashai. The author warrants that the work is the author's own and that Stanford University provided no input other than typesetting and referencing guidelines. The author grants permission to copy, distribute and display this work in unaltered form, with attribution to the author, for noncommercial purposes only. All other rights, including commercial rights, are reserved to the author.

References

[1] Y. Shi, Advances in Big Data Analytics (Springer, 2022), Ch. 1.

[2] B. Daigle, "Data Centers Around the World: A Quick Look," United States International Trade Commission, May 2021.

[3] A. Shehabi et al., "United States Data Center Energy Usage Report," Lawrence Berkeley National Laboratory, LBNL-1005770, June 2016.

[4] E. Masanet et al., "Recalibrating Global Data Center Energy-Use Estimates," Science 367, 6481 (2020).

[5] D. Mytton and M. Ashtine, "Sources of Data Center Eenergy Estimates: A Comprehensive Review," Joule 9, 2032 (2022).

[6] N. Jones, "How to Stop Data Centres From Gobbling Up the World's Electricity," Nature 561, 163 (2018).