The Sacred Fire of a Data Center

Author: Donal Lally

How the combination of domesticated fire, the data center, and artificial intelligence is reshaping our communities, opening them up to a stealthier and more pervasive form of the Smart City.

In his 1851 text The Four Elements of Architecture, architectural theorist Gottfried Semper puts forth fire as “the first and most important and moral element of architecture.”

For Semper, the first human groups assembled, formed alliances, and developed religious concepts around the domesticated fire. “Throughout all phases of society, the hearth formed that sacred focus around which took order and shape,” he said.


Ignis Vestae

Fire is both metaorganic and metacultural. It is more than a natural phenomenon and more than a technology; fire is a primary agent in the shaping of human society. Fire, like the human animal, is a process: it breathes oxygen, emits heat, and needs to be fed, bred, trained, and maintained. Fire is an agent in its own right; an agent without consciousness, an actor possessing the power to shape the course of human and nonhuman habitats.

Ancient Europeans, having tamed the garden paving the way for the agricultural revolution, sought to subordinate fire to the rule of order. In Latin, the word “hearth” means focus; the hearth fire became the focus of social life. Houses were built around the hearth and villages and cities were constructed around communal hearths which often took the form of a perpetual flame. For Europeans, the orderly and obedient anthropogenic hearth fire became a sacred symbol for a disciplined and orderly society. According to environmental historian Stephen J. Pyne, “civilization would be impossible without fire.”

In ancient Rome, a sacred order, the Vestal Virgins, sustained the Ignis Vestae or vestal flame at the temple of Vesta, dedicated to the goddess of the hearth. The Pontifex Maximus demanded celibacy to ensure absolute dedication to the task. For Romans, the shrine at Vesta predated all others; it had no inauguration, for it had always been. Keeping the flame alive and under control was seen as vital to maintaining the shape and order of Roman society.

Triumph of Chastity: unicorns draw a carriage bearing a female figure seated next to a sphinx and holding a standard, Vestal virgins walk behind the carriage, round temple of Vesta at right, from the series 'The Triumphs of Petrarch',ca. 1539. Image courtesy the Metropolitan Museum of Art.

Domesticated fire belonged to the hearth and to the altar; natural fire belonged to the primitive fringes where society begins to break down. Pyne describes how over time, Europe came to resemble a fire whose core had burnt out; a smoldering glow at the center with flickering flames at the edges.


Ignis Machinae

For historian Lewis Mumford, infrastructural technologies are container technologies. A container technology, such as the hearth or data center, foreground the figure while making invisible the ground—the former foregrounding the flame, and the latter the cloud.

Simply put, a data center is a container which organizes a company’s ICT (Information and Communication Technology) equipment such as servers, switches, and storage facilities. The data center manages the environmental conditions of the space (temperature, humidity, and dust) to ensure that the ICT systems operate reliably, safely, and efficiently.

The global data center industry consumes vast amounts of energy and generates vast amounts of heat. A proposed (now canceled) Apple facility to be located on the west coast of Ireland was projected to consume as much energy as a medium-sized Irish city, or roughly 230,000 homes. When fully operational, the new Amazon development in the Dublin suburb of Mulhuddart is forecast to consume as much as 4 percent of Ireland’s total energy demand.

Proposed Amazon data center facility, Mullhuddart, Dublin. Image courtesy baxtel.com

Ireland’s state energy distribution agency, Eirgrid, has predicted that by 2027, the total data center industry will consume as much as 31 percent of Ireland’s overall energy demand. Data centers are currently responsible for about 3 percent of global electricity consumption. Anders Andrae, who works on sustainable ICT at Huawei Technologies Sweden in Kista, predicts that number will increase to 15 percent by 2030.

The increase in the power consumption of data centers is not just a function of increasing size, but also of increasing server density. Demand continues to move faster than supply; 90 percent of the world’s data has been created in the last two years. Following the Jevons paradox, efficiency does not reduce consumption but increases it.

Modern technologies do not reveal the sources of their power, but transform them unrecognizably. John Durham Peters, a philosopher of elemental media, sees electricity as “repressed fire.” The vestal virgins maintained the Vestal fire’s purity through a strict diet of oak tree called materia. The oak tree is statistically the most likely to be hit by lightning, and thus was a pan-cultural symbol of the gods of lightning, such as Zeus and Thor. Lightning purifies the oak; the oak feeds the vestal flame.

The data center consumes a much dirtier materia, produced at the edges to maintain its perpetual warm glow. In only 10 years of global data center proliferation, a new industry has emerged whose carbon emissions equal that of the airline industry. With 90 percent of all data produced in the last two years, one must assume this unprecedented growth cannot be supported with a parallel surge in clean energy production. Ireland relies on fossil fuels to produce 70 percent of its electricity. Amazon, Microsoft, and Google offer their AI and cloud services to the fossil fuel industry to aid in the streamlining of extraction of oil and gas. While all three make significant efforts to eliminate fossil fuels from their in-house company operations, they appear intent on maintaining the global supply of fossil-derived energy.

For Durham Peters, “our bodies are fire containers, each cell an image of the vestal hearth. Heat control is one of the classic cybernetic processes that unite humans and machines, and it remains the central design problem for the chief medium of our time, the computer.” Too much heat causes server failure; this makes the design and manufacture of thermal management systems one of the most challenging aspects of the data center design. A data center radically transforms the quality of energy it consumes; it takes in high-grade electricity and expels large quantities of low-grade heat. As media theorist Julia Velkova points out, “intense computation processes can make server rooms reach temperatures of 35-45°C that could result in server failure. Surprisingly, the effort that the industry exerts to offset this threat has largely evaded scholarly attention, which has instead predominantly discussed the input problem, the electricity.”

A computer server takes electricity and uses it in three ways. The majority is converted to heat through computational effort. Another portion is used to drive fans to push the heat away from the server. And a very small amount is used to drive electrons or photons on the network interface. According to the large Finnish DC operator Telia, the operation of its new DC generates 200 GWh of heat annually, with 1 GWh being roughly the equivalent of 412 wind turbines.

Vast amounts of energy are used to cool the server room floors to an optimal industry standard—from around 40°C down to somewhere between 10°C and 22°C. Google’s most efficient data center in St. Ghislain, Belgium, is designed to rise above 35°C during the summer months. The server hall becomes a data furnace not safely habitable by workers. For Durham Peters, “container technologies show media at their most environmental.”

Saint-Ghislain data center in Belgium. Image courtesy Google.

A feral electrical spark can be extraordinarily destructive—the temple of Vesta burned down at least four times. In 2014, the world witnessed a contemporary mining accident. A five megawatt server farm in Bangkok, exclusively tasked to mine bitcoin, burned to the ground as a result of an electrical fire. The press reported that hundreds of bitcoin mining rigs were destroyed, with estimates of their value running into the hundreds of thousands and even millions of dollars.

The Bangkok fire illustrates the risks of running high-density computing equipment in buildings operating without advanced electrical infrastructure and fire suppression systems. The safest and most energy-efficient data centers compete for global locations which are the most politically stable, have colder climates, and have invested in the most advanced network infrastructure, enabling abundant access to fast and reliable network connections.

While speaking of ancient Rome, Pyne stated that “desired fire belonged in the hearth and altar; unwanted fire appeared along the rough fringes of an unravelling society; in the cracks of disintegrating cities.” In an affirmation of the global digital divide, unsafe server farms such as the Bangkok mine will find their homes where building regulation is weakest.

The data center is the hearth around which gathers the most pervasive of contemporary metacultural megastructures—the data economy. From the landing of the world’s first transatlantic telegraph cable at Valentia Island over 150 years ago to its present status as a strategic location for hosting data technology infrastructure and services, Ireland has played a unique role in the historical development of global network communications and technologies. Ireland’s position as a significant node in the global data economy is a result of its geographic location and its interconnecting relations with other nations and entities. Ireland plays host to the European headquarters of Amazon, Facebook, Google, and Microsoft, as well as many other well-known technology companies. Ireland is attractive for its geographic location, favorable tax conditions, and English-speaking workforce.

At the end of 2018, Dublin became Europe’s largest data hosting cluster, surpassing London with 25 percent of the European market. As a result, Ireland has experienced a rapid proliferation of large-scale data centers across its urban and rural landscape. There are currently 53 data center sites in Ireland. With new developments by Facebook and large-scale expansions by Amazon and Microsoft, the market is expected to double over the next five years.

Data centers, once hidden from sight, are beginning to play a direct role in shaping our urban environments. Currently, a bill is waiting to be ratified by the Irish parliament that would re-designate data centers as strategic infrastructure—meaning they would become categorically identical to major roads and bridges.

Facebook's Clonee data center. Image courtesy Facebook.

If the data center fails to maintain its perpetual fire, the order and shape of society breaks down. The business model for cloud computing is based on volume, stability, and efficiency of delivery. In May 2017, over 75,000 people had their three-day weekend plans thrown into chaos when British Airways suffered a massive data center failure. Passengers suffered lost luggage and immense frustration from canceled and delayed flights, and the airline lost $112 million.

This downtime was caused by a single engineer who disconnected and incorrectly reconnected the power supply, triggering a power surge that disrupted BA’s operations infrastructure. According to a 2018 report by the Uptime Institute’s Global Data Center Survey, the rate and severity of these incidents has grown significantly. As we rely more and more on the cloud to manage our society, we rely more and more on the data center to maintain the cloud’s order and shape.

The global data center trend is moving towards much larger hyperscale data centers as many organizations with small- and medium-scaled data centers continue to migrate their workload to the cloud. A typical data center may support hundreds of physical servers and thousands of virtual machines. A hyperscale facility needs to support thousands of physical servers and millions of virtual machines. An individual hyperscale data center would typically have a footprint of more than 10,000m2—and often much more. The latest Amazon facility in Dublin to be approved for construction will grow to eight buildings totaling 165,000m2, the combined equivalent of about 24 soccer pitches.

Hyperscale data centers are forecast to host a 60 percent share of the total global data center server space by 2021. In Ireland, these hyperscale data centers are typically designed and operated by large technology firms such as Amazon, Google, Microsoft, and Facebook.

District Heating System in Tallaght, Dublin. Image courtesy of South Dublin County Council.

In 2018, South Dublin County Council and Amazon Inc. announced they were partnering on a scheme to reuse exhausted heat from the fifth, and newest, of their data centers in the Dublin suburb of Tallaght. As part of the arrangement, Amazon would build a new energy center to collect and distribute exhausted heat into the local area. Phase one would involve plugging government buildings, an arts center, a university, a hospital, and about 1,700 residential units totalling around 44,000m2 into the data center. This would reportedly reduce CO2 emissions in Tallaght by nearly 1,900 tonnes per year.

For a data center to be “green,” two key areas must be addressed. Firstly, the input electricity needs to be from a renewable source. Secondly, the output heat must be managed in as efficient a way as possible. As data centers typically operate 24 hours a day all year round, they are a stable source of exhausted heat. The heat exhausted is of a low grade—less than 85°C. This heat then degrades further when converted to the hot water required by the district heating network.

In Denmark and Germany, where architecture has historically been built and insulated to a higher standard, Low-Temperature District Heating Systems (LTDH) have been successfully implemented. The LTDH, on average, requires an input average of 50°C—as opposed to 80°C in Ireland. In other words, if this heat recycling practice is to be used efficiently, the receiving buildings must be constructed and insulated to the highest of standards to prevent additional heat boosting technology from being required. In Tallaght, Amazon will boost the quality of the exhausted heat through a centralized large-scale heat pump housed in an on-site pump house.


Industrial Scale AI

In 2016, DeepMind, the AI arm of Alphabet, announced it was training several neural networks with a type of machine learning called reinforcement learning to regulate the management of Google’s data center cooling operations. Reinforcement learning, described by Richard S. Sutton and Andrew G. Barto, teaches a computational system what to do—how to map situations to actions—to maximize a numerical reward signal. The learner is not told which actions to take, but instead must discover which actions yield the most reward by trying them.

In a data center, cooling is typically carried out via large mechanical devices such as pumps, chillers, and cooling towers. Data centers are dynamic environments—both in the complexity of the operational systems and in how those systems interact with external environments, such as the weather.

Google Data Center in Dublin. Image courtesy Google.

Each data center has a unique architecture, context, and environment. The physical architecture and the systems architecture that is designed for one context may not be applicable to another. DeepsMind’s algorithm is trained to learn from existing data sets generated from Google’s global data center operations, in addition to simulating millions of other potential scenarios and developing a general intelligence based on trial and error.

In 2018, Google said that it had effectively handed control to the algorithm, which autonomously manages cooling at several of its data centers. The AI yields impressive results, saving, on average, 30 percent on energy usage per month.

According to Google, this was the first time AI was given autonomous industrial control at such a scale. “It was amazing to see the AI learn to take advantage of winter conditions and produce colder than normal water, which reduces the energy required for cooling within the data center,” Dan Fuenffinger, one of Google’s data center operators, said. “Rules don’t get better over time, but AI does.”

Google sees this system being rolled out beyond data centers, applying it to other settings and even using it to help tackle climate change at scale. In the feeding (oil and gas extraction) and maintenance (through heat distribution systems) of these contemporary perpetual hearths, are powerful machine learning algorithms assuming the mantle of the contemporary vestal virgins?


A data furnace

DeepMind’s data center management algorithm is the most high-profile example of an autonomous AI data center management system. One must assume that Amazon (if it has not already) will implement a similar system. It is plausible that the cooling operations at Amazon Tallaght will at some point be governed by AI. What does it mean when a private industrial-scale AI is plugged into the civic realm?

A municipal AI system would be trained, one would hope, to ensure a balanced human and nonhuman ecology. For example, a municipal district heating AI system could make explicit the cybernetic processes at play that unite fire, the human, and the machine. It could collectively lower the heat in a neighborhood’s homes in order to save an energy center from overloading and causing outages. Or, it may use real-time global weather models in combination with carbon footprint models, cross referenced with building use and occupancy data, to efficiently manage the collective energy sources of a neighborhood ecosystem—one inclusive of animal, vegetable, mineral, and machinic users.

A private AI would have different goals—above all else, security of the data.

On the face of it, the Amazon Tallaght data center recycling its waste heat into the community would appear a sensible Idea. And, if implemented correctly, it would act as a model to be used to make data centers the glowing hearth of communities across the globe. As communities gather around and plug-in to these new mega-hearths, are we witnessing the latest morphology of society shaped by fire? For the Romans, the vestal fire represented the health of civic society. These neo-hearths represent the interests of private capital. Is the Smart City’s most radical manifestation occurring in our hot water pipes?

As we outpace the Jevons paradox, producing and consuming exponentially more data, we are exhausting more and more heat; contemporary global society has become a data furnace, spewing heat into an atmosphere already burning up. To media theorist Nicole Starosielski, “the language of thermodynamics can help to describe the extensive transformations of technological modernity—one that is heating up, increasing in entropy, and ultimately moving toward a ‘heat death.’” Or perhaps the dilemma is more prophetically observed by poet John Donne:

“Fire ever doth aspire,

And makes all like itself, turns all to fire.”

Donal Lally

An architect and researcher whose work focuses on the data center. Donal is a founding member of Annex, the multidisciplinary curatorial team chosen to deliver Entanglement, the Irish Pavilion at the Venice Biennale 2020. Donal is the director of the architectural practice zero-degree machine and a lecturer in design theory and researcher at TU Dublin.

Cover image: Dedication of a New Vestal Virgin (1710) by Alessandro Marchesini

If you noticed a typo or mistake, highlight it and send to us by pressing Ctrl+Enter.