The advent of digital transformation across Africa has catalyzed the proliferation of data centres critical for managing the exponential growth in data generation.
As Africa strides towards a more digitized future, the operational efficiency of these facilities, particularly in their cooling strategies, will play a pivotal role. Among the several cooling strategies available, liquid cooling technologies present a promising solution to enhance thermal management and operational efficacy.
Key considerations
When transitioning an air-cooled data centre to liquid cooling, several important considerations must be taken into account. Modifications are required for both racks and servers, even when utilizing direct-to-chip liquid cooling. The racks need a manifold designed for the pipes connected to individual servers, and it might be necessary to create additional openings for the pipes to exit the racks.
The situation is similar for the servers themselves, which require holes for the pipes. Since these pipes are relatively thick and need to be integrated with cold plates in the housing, space constraints may arise, necessitating more adjustments. Another crucial aspect is adhering to stricter safety standards when routing pipes for the various liquid cooling circuits, especially if they contain a water/glycol mixture. Ideally, these pipes should be installed in the existing raised floor, which is generally already established in air-cooled facilities. Additionally, regular inspections for corrosion and leaks are essential. Conversely, pipes that carry dielectric fluids tend to be less of a concern, as leaks will not harm other components.
The structure of the building must also be considered. Generally speaking, no action needs to be taken with direct-to-chip liquid cooling, because the weight per square meter does not change significantly compared with air cooling. However, if immersion cooling is used, in larger plants there is a risk of dramatically increased point load on the ceiling due to the higher server density and large quantities of fluid. Therefore, it’s essential that structural engineers check potential ceiling loads before the conversion and that any necessary structural measures to increase load bearing capacity are taken.
If the cost of a complete conversion is too high, it is possible to convert just part of the data centre surface area to liquid cooling. In some cases, perhaps it even makes more sense to construct a new building or extension, which meets the requirements for liquid cooling right from the start. Existing air conditioning cabinets for room cooling can continue to be used for cooling residual heat that is still emitted into the room by non-liquid-cooled components, such as power supply units.
Coolant distribution unit (CDU) – a key component
The basic design of a liquid cooling system in a data centre is as follows: with direct-to-chip liquid cooling, the cold plates on the server boards are connected via flexible lines to a manifold in the rack. This rack manifold is in turn connected via pipes to a coolant distribution unit (CDU). If immersion cooling is used instead of the direct-to-chip method, the tank, not the rack, is connected to the CDU. The CDU separates the facility water system (FWS – the building’s cooling circuit) from the technology cooling system (TCS – the liquid cooling system for the servers) via a heat exchanger. In addition, the CDU uses valves and pumps to control the flow and temperature of the cooling liquid.
It’s important to separate the circuits, because different purity requirements apply to the cooling circuit and to the facility water. The necessary liquid quality in a cooling circuit of this kind depends on what materials the heat exchangers and cold plates are made of, and how large the integrated microchannels are.
Standards and specifications
Most data centre standards do not currently include specifications or recommendations for designing suitable systems for liquid cooling. The sole exception is ASHRAE Guideline TC 9.9, which deals explicitly with liquid cooling. It names two recommended designs, which differ in the location and number of CDUs.
Version 1: The first version has a central CDU to which several racks can be connected. This saves space in the rack but occupies additional space in the server room. If this central CDU design is used, 2N redundancy should ideally be established, to compensate for failures at any time.
Version 2: In the second version, a dedicated CDU is incorporated in each rack. This means that the racks cannot accommodate as many servers, but no additional space is needed outside the racks. If a CDU fails, this only affects a single rack; all other servers can continue to function unimpaired. Scaling is also easier to achieve, because each time another rack is added, a suitable CDU adapted to the required cooling capacity is added as well.
Alternative methods of implementation
Some plants do without a CDU completely. This is only possible with immersion cooling, however, because a heat exchanger to separate the circuits is already in place on the tank. The temperature needs to be regulated via the facility water, offering less flexibility than when a CDU is used. A design without CDU also has advantages like higher efficiency, because no heat exchangers need to be used to link to the facility water, and therefore no transfer losses are produced at this point.
As a rule, the advantages mentioned at the start of this article — and the greater flexibility of the CDU version — outweigh the merits of the non-CDU version. This is also because, without system separation by a CDU, glycol must be added to the facility water, which lowers efficiency. Compared to a centralized plant control system, a CDU enables autonomous, adaptive, and precise regulation of individual cooling circuits, thereby reducing overall control complexity while significantly enhancing operational stability.
There is another method for the use of liquid cooling involving an “active rear door.” This represents an economically attractive bridge between air and liquid cooling in the 30-50 kW performance range, as it offers a powerful and significantly less expensive air cooling solution despite the option of converting to liquid cooling. The active rear door at the back of the rack can also be connected to a CDU and contains an air-to-water heat exchanger, which transfers the residual heat out of the air and into the cooling circuit.
Cooling facility water
All liquid cooling technologies are designed to set the temperature of the facility water in such a way that sufficient reserves are available for the cooling circuits. To prevent condensation, however, these temperature windows must not get too close to the dew point of the room. In all liquid cooling versions, the facility water is cooled in a similar way to air cooling, but generally at higher temperatures. Depending on the required temperature, dry coolers, cooling towers, and chillers may be used.
