Meeting unpredictability with flexibly: The modular approach to data centre cooling

Jacob Wolfe, Global Key Account Manager, Data Centres, at Armstrong Fluid Technology, discusses why the unpredictability of the data centre sector must be matched by flexibility in design methodology.

The task of achieving maximum energy efficiency for mission-critical cooling applications is never straightforward. In data centres, in particular, extreme scrutiny of environmental performance is constant, in a situation which demands maximum reliability and zero downtime. With energy requirements increasing steeply, driven largely by AI, availability of power from the grid is a major issue with the potential to limit expansion, as is the need to secure the goodwill of stakeholders such as local planning authorities. These issues are magnified by commercial and technological factors, such as the need for adaptability and flexibility to address short equipment update cycles.

In this complex landscape, the best way of setting up for future expansion is to build a system designed specifically for incremental expansion from the start. Rather than making a large front-loaded investment in a plant room designed for a hypothetical full load, expanding in line with increasing demand for processing capacity provides more control over critical factors for efficiency, while achieving faster deployment, and futureproofing the development within an unpredictable sector.

More follows …

Why not just design for 100% straight away?

The traditional approach of creating a cooling system capable of supplying 100% of demand from the outset is unsuitable for data centres for a number of reasons. Firstly, it involves unnecessary front-loading of investment in building services for the site, and does not provide the necessary flexibility to increase incrementally over time. Secondly, a cooling system designed for 100% demand at the outset (when actual demand is far lower during the early phases) involves significant risks of energy wastage due to the operation of over-sized equipment. Although the cooling technologies themselves are highly reliable, technical issues are more likely in situations with over-sized equipment. Simply put, it risks disappointing energy efficiency levels from the start, whilst complicating future efforts to change or expand the system.

Aligning cooling systems with data centre business models

The keys to effective incremental expansion of cooling systems in alignment with increasing processing capacity include modularity, repeatability, scalability, demand-based control and Active Performance Management.

• Modularity: The best cooling systems for data centres are those that are designed on modular principles, specifically for incremental expansion. This ensures that additions to the system can be integrated quickly and seamlessly, without time-consuming and costly installation and commissioning, or additional development in-situ. In general, increasing ease and speed of expansion efforts while avoiding oversizing at all times is a good strategy.
• Repeatability: To safeguard reliability and reduce risk, it is important that the modules have been carefully designed for the stringent demands of data centre applications, but are readily available as off-the-shelf products. Ad hoc equipment and customised systems are ineffective in these situations, as they introduce variables/unknowns that can impact reliability adversely due to unpredictability.

To create a cooling system which delivers in terms of both modularity and repeatability, it is helpful to think in terms of packaged plant or offsite-manufactured plantrooms. As the packaged cooling plant is fully assembled and tested before it leaves Armstrong’s factory, many potential project risks (such as poor system integration) are eliminated. Solutions such as this, which are capable of ‘bolting-on’ additional cooling in line with expansion of IT processing capacity, can avoid the energy wastage of an oversized plant, whilst assisting profitability by preventing front-loading of capital investment, and providing repeatability of performance. Increasingly we are finding that data centre operators with multiple sites are moving towards integrated offsite-manufactured plantrooms, which can replicate the same cooling systems at each site. This speeds introduction and expansion of new facilities in this highly competitive industry sector, and assists with day-to-day operation.

• Scalability: With proper forecasting, plant capacity can be expanded using repeatable designs in a scalable fashion as each data centre expands. As many of the companies establishing new data centres are international businesses, an equipment supply partner capable of manufacturing the same solution at multiple production facilities worldwide is also beneficial, as it helps to ensure continuity irrespective of the location of each data centre.
• Demand-based control: As data centre cooling systems need to be reliable and efficient over wider ranges of operating conditions as the site expands, it is crucial that system components and control technologies are designed for variable demand and ultra-efficient performance at part-load. This requires variable-speed components (such as chillers, pumps and fans) across the system, and a control strategy specific to the operating characteristics of variable speed devices. There are no exceptions to this, because constant-speed devices cannot solve the challenges of applications such as data centre cooling.

When a variable frequency drive (VFD) is added to a compressor, pump or fan to improve part-load efficiency, the energy saving potential is huge due to the pump fan laws which state that power is proportional to rotary speed cubed (PaN3). This would equate to a potential 400% increase in operating efficiencies. This is only possible, however, if the pump fan law relationship between pressure and rotary speed, along the Natural Curve, is maintained at the decreased speed.

Traditional control practices often fail to optimise this potential. Pumps, for example, are often set to maintain a fixed or minimum differential pressure across the pump supply and return headers. This means the pump will not have the freedom to operate along its Natural Curve and will consume more energy. Best practice is to utilise advanced integrated control across the system. In the case of variable speed chillers, integrated control ensures operation along the chiller’s Natural Curve for all operating scenarios, ensuring optimum efficiency at all loads.

Another important design principle is the employment of capacity-based (rather than demand-based) sequencing. With capacity based sequencing, each pump would be taken up to 90% loading, for example, before the next pump was introduced. Demand-based sequencing, however, balances the load across the system as a whole, unlocking additional energy efficiencies which might otherwise remain under-exploited.

• Active Performance Management: Advanced connectivity and visibility of system performance are also important throughout the lifetime of ultra-efficient critical cooling systems for data centres. Without information on fluid f low, across the system, it’s difficult to diagnose and optimise efficiency. With accurate flow information, the picture changes entirely.

The Active Performance Management developed by Armstrong Fluid Technology, for example, helps to optimise HVAC systems at any stage of a data centre’s life-cycle, responding to changing cooling requirements. Online trending and analysis across multiple parameters on single pumps, or on an aggregated basis for multiple pumps, assists in identifying performance degradation and facilitates a predictive and proactive approach. It can, for example, report issues such as excessive vibration, pump in hand, risk of cavitation or a dead head should they start to occur, enabling preventative maintenance to address the issue and prevent downtime.

Conclusion

It’s important not to copy and paste the kind of HVAC solutions common in other situations, including those suitable for other mission-critical situations such as hospitals or laboratories. With data centres, the unpredictability of the sector itself must be met by flexibility in design methodology.