Problem Definition
The escalating growth in data center infrastructure and the concurrent surge in IT equipment utilization have e led to an increase in electricity consumption.
Since servers, integral components of the data centers, convert electricity into heat while operating, we face the problem of dealing with high temperatures and cool data center premises and equipment.
Let us briskly remember the basics of school physics: following the fundamental principles of thermodynamics, the energy does not disappear but is transformed. Thus, if a data center consumes 1 MW of electricity – this entire energy quantum is transformed into an equivalent amount of heat. Consequently, the more electricity is consumed, the greater the challenge of managing the resulting heat within the data center.
The situation becomes even more complicated, as IT equipment can have varying energy consumption levels while having disparate physical sizes. For instance, equipment with high energy consumption may have a small size, posing issues in efficiently cooling the concentrated heat. On the other hand, larger IT equipment with a relatively moderate electricity consumption rate is easier to cool due to its larger surface area. Data centers typically house a mix of equipment sizes and consumption levels, presenting the challenge of not only cooling diverse IT equipment but also doing so at different speeds, dictated by the temperature requirements of each equipment type. Needless to say, to cool the DC we need substantial amounts of electricity, which adds to the operational costs.
The problem of efficient power use in data centers becomes especially acute with the global rise in electricity prices. According to
Data center electricity consumption is growing despite increasing performance efficiency.
In other words, though performance per watt is improving, the demand for resources grows even faster, so overall consumption inevitably increases as well as the costs. However, substantial cost savings can be achieved by optimizing the cooling systems. This made me want to have a deeper look into the efficient cooling methods in general and free cooling in particular.
Assessment of energy consumption levels in data centers usually relies on the Power Usage Effectiveness (PUE) metric. PUE gauges a data center's efficiency by evaluating the total energy consumption against that used exclusively for IT equipment. We'll talk about it in more detail a bit later. What we need to know now is that a lower PUE signifies a more efficient data center, indicating reduced reliance on non-computing power. In the face of growing infrastructure and escalating electricity consumption, optimizing PUE with more efficient cooling systems provides financial prudence and sustainable operations.
In this article, we'll explore traditional and innovative cooling methods and discover which of them offers maximum efficiency.
Cooling Methods Overview
Air and non-air
In a simplified classification, cooling techniques can be delineated into two primary categories: air-based and non-air-based methods. To elaborate, air cooling encompasses conventional approaches, while the non-air category includes diverse methods employing substances like water, oil, or solid materials. It is noteworthy that the overwhelming majority, constituting 99%, of cooling methods fall under the air cooling umbrella.
Air Conditioners
Air conditioning systems are the most prevalent means of air cooling in professional data centre setups. Their fundamental principle mirrors that of residential air conditioners: the air flowing through the servers is circulated through an air conditioner, cooled via a radiator grille, and then recirculated back into the servers. This cyclical process ensures a continuous cooling mechanism.
Chillers
Following air conditioners, chillers represent the second most widely adopted cooling system. Unlike air conditioners, chillers employ water (or a water-based solution) to transfer heat away from spaces requiring climate control. Though air conditioning is simpler and generally more affordable, its higher energy costs can sometimes be a deterrent for businesses. On the other hand, chilled water systems are more energy-efficient, but require more components and complexities in their installation and maintenance.
Adiabatic Chambers and Mats
Adiabatic cooling involves the use of chambers or mats where water is poured and evaporated. As the water evaporates, the chambers and mats cool along with the air inside. While adiabatic cooling represents a third viable option, it is considered somewhat exotic and is not as commonly employed in data center cooling when compared to air conditioners and chillers.
Water Cooling
In water cooling systems, water or water-containing liquids are used for heat dissipation. Water pipes are strategically placed in server rooms, with each server connected to two pipes — one for hot water outflow and the other for cold water inflow. Radiators on CPUs, GPUs, and on other equipment are linked directly to this water supply system. This approach not only cools the data center equipment and premises but also generates a supply of warm water for additional uses.
Heat Exchangers
This method enhances cooling efficiency by leveraging external cold environments.
When a nearby cold source, such as a lake, sea, or cold ground, is available, water pipes can be deployed right into it to transmit large amounts of heat from IT equipment.
Exotics
There are also unconventional methods. One of them is based on Peltier elements or thermoelectric coolers (TECs). This approach relies on semiconductor effects and involves supplying electricity to a special plate that is heated on one side and cooled on the other.
Another avant-garde approach is the deployment of underwater data centers. In
Free Cooling
This technique is particularly aimed at maximizing cooling efficiency. Free cooling refreshes the air within a data center without relying on traditional cooling systems. It uses natural outside air as it is. Typically the outside air is just subject to humidity control and then natural thermodynamic processes regulate temperatures within the data rooms.
This method significantly reduces power consumption (75% to 92% less compared to other CRAH systems), lowers carbon dioxide emissions, and eliminates the need for water in the cooling system.
Free cooling is one of the most eco-friendly choices that requires less energy. Besides, it can help save costs as 40% of the power used by data centers goes into cooling. This system enhances the performance of all air-cooled equipment, even in tough conditions. Here is a simple visual representation of the free-cooling process:
As you can see the system operates in a very straightforward way, channeling outside air through filters, IT equipment, and expelling it. This reduction in complexity, with only fans as potential vulnerabilities, bolsters the overall reliability of a DC on free cooling.
Unlike systems with complex equipment, the absence of intricate components also reduces both initial setup costs and ongoing maintenance expenses. So the financial advantages begin already at the building stage, where the streamlined design of free cooling translates into tangible savings.
The Agony of Choice
During conferences and meetings, I often get numerous questions that revolve around the paradox: if free cooling is as advantageous in terms of cost savings and simplicity, why isn't it universally adopted in the industry?
This prompts the broader question of why, despite its benefits, only a limited number of companies have embraced free cooling, while others persist with conventional methods. The answer to this lies in a multifaceted examination of the prevailing industry dynamics.
Industry Dictatorship
In the data centre industry, where reliability is of primary importance, the adoption of innovative solutions often faces resistance. I attribute this, first of all, to the DC industry's conservative nature, where decision-makers prioritize proven concepts over innovative solutions.
While new technologies like free cooling promise cost-effectiveness and efficiency, industry representatives would rather prefer traditional yet reliable approaches to ensure the seamless operation of the servers.
Marketing obstacles
Another point here is that commercial DC providers, constituting about 80% of the industry, rely on certifications from independent bodies like the
Global Warming Concerns
Critics often raise concerns about the impact of global warming on the viability of free cooling. However, the argument is debunked by recognizing the gradual nature of global warming, with an approximate increase of 1.5 degrees over a decade. This modest temperature change is unlikely to compromise the stability of free cooling solutions in the near term.
"Just in Case" Argument
One more common practice for companies choosing a DC cooling method is the inclusion of backup air conditioning units as a precautionary measure in addition to free cooling. This "just in case" argument undermines the core concept of free cooling, introducing unnecessary complexity and compromising financial and operational efficiency.
Even for a small air conditioner, the need to provide various components such as freon, wires, liquids, systems, and controls arises. Instead of embracing the idea of having a backup air conditioner, the industry should focus on adapting its free cooling system to a wide range of conditions without relying on false hopes.
Real Risks and Considerations
When contemplating a free cooling solution, some tangible risks and considerations demand attention. One primary consideration is the geographical region, as deploying free cooling in a region like the Arab Emirates might not be justified.
Accessibility is another aspect to bear in mind. The selected region must possess the requisite infrastructure and be easily reachable by specialized personnel tasked with data center maintenance. Connectivity, including the availability of optical lines, is also important. For instance, establishing a free-cooling data center beyond the Arctic Circle becomes impractical due to the absence of communication lines and the challenge of maintaining a skilled workforce.
Beyond these logistical considerations, the only restrictions for free cooling pertain to the region's maximum temperature (app. 38-40 degrees) and air quality. Areas with excessive pollution, such as those near busy highways or intense agricultural activities, may pose problems. While there is no outright prohibition, filters in such locations will need frequent replacement. Unlike conventional air-conditioned data centers that circulate internal air, free-cooling centers draw in outside air, demanding more diligent filter maintenance. Other location parameters align with those applicable to traditional data centers.
Pioneers and Future Trends
Despite the industry's conservative nature, some forward-thinking corporate companies evaluate the tangible benefits of alternatives. Calculating the cost-effectiveness of free cooling through numerical analysis, they realize the potential cost savings it offers.
Several prominent companies, such as Facebook (now Meta), Google, Amazon, Yandex, and Wildberries, are pioneers in adopting free cooling technology. Their trailblazing status stems from their willingness to assess risks and recognize the financial advantages inherent in this technology. The choice for these companies was clear – either go for conventional schemes and incur higher costs or take the risks and benefits of becoming pioneers in data center cooling.
The industry's evolving landscape indicates a growing trend among corporate hyper-scalers towards implementing free cooling solutions. As more companies recognize the cost-effectiveness and operational advantages of this technology, it is anticipated that an increasing number of corporate-free cooling data centers will emerge in the future.
If you are interested in learning more about the physics of free cooling, explore the topic in my new article Free Cooling: Technology Deep Dive.