2.1. Research Methodology
The purpose of the literature review was to identify studies related to thermal management, corridor insulation systems, CFD, and BIM in data centres. These studies confirm that corridor isolation is the optimal cooling system option among many possibilities. However, despite the accuracy of the data, the use of CFD and BIM is still rare in studies.
Based on the area of use, several categories of articles could be identified: cold aisle containment (CAC) thermal control; hot aisle containment (HAC) thermal control; room cooling systems; energy consumption optimisation by using cooling systems. The most relevant papers were selected and reviewed within the framework of this article. Twenty-six articles were analysed and based on these, extensive analysis was carried out, which is reflected in
Table 1.
CFD modelling can be used to solve a wide range of problems. CFD contains special tools to model the thermal conditions of electronic devices. For example, if we are talking about data centres, the use of this technology makes it possible to optimise the air conditioning system and increase the energy efficiency. Therefore, CFD represents a logical and necessary extension of BIM design.
BIM incorporates details of the structure including its architectural and engineering systems. This model is subsequently imported into the AutoDesk CFD 2019 (CFD software), which analyses the room geometry, establishes appropriate boundary and initial conditions, and computes three-dimensional temperature, velocity, pressure, and additional physical parameters.
In reality, CFD provides a virtual testing facility to assess crucial and highly responsible engineering systems ahead of construction efforts. CFD computations assure system performance along with designing a room layout early on in the process, allowing for the placement of air conditioning units and IT equipment racks in a data centre. The final results display the temperature mode of the system operation.
Table 1 shows the extensive analysis for the main categories.
As described by Lee et al. [
3], a containerised data centre with CAC has been proposed as an effective thermal control strategy. The system design utilises a downflow system with a heat exchanger positioned directly above the data centre on the air side. Additionally, there is an evaporative water cooler on the water side that facilitates a cooling process. The return temperature index (RTI), rack cooling index (RCI), and supply heat index (SHI) were used to evaluate the design performance of the proposed test data centre.
As reported in an ageing data centre retrofit project by Cho and Woo [
4], the referenced data centre’s previous air conditioning system was replaced with a new, separate row-based cooling system, implemented in the same IT server room to meet the target. Quantitative evaluation metrics reflecting the performance of the data centre were used. The results of the performance analysis were validated by comparing them to the results obtained during an analysis using the same conditions as the experimental measurements. Air temperature, humidity, and air distribution efficiency in the IT environment were studied using in situ measurements of the performance indices. The differences and technical features of this experimental study were that two different forms of cooling systems were evaluated based on the same IT environment conditions, IT equipment location, and cooling load in the same server room. The results showed that data centre cooling systems using the “full cold and hot aisle containment” method are very effective in relieving the heat load from IT servers.
The article written by Rong et al. [
5] examined the energy-saving technologies accessible for data centres. Energy-saving technologies have varying impacts on energy conservation at diverse stages of data centre evolution from a global viewpoint. In data centre site selection, proper site selection through scientific and rational use of natural energy can reduce the consumption of CRAC by about 30%. CRAC typically accounts for 40–50% of the total energy consumed, so careful siting can reduce the total energy consumption by up to 12–15%. When building data centres, energy consumption can be effectively reduced by 25–30% by selecting low-power servers and auxiliary energy-saving devices. Data centre operators can reduce the overall energy consumption by around 10–15% by optimising their resource planning algorithm and management strategy. The rational utilisation of renewable energy sources can also effectively lower the total energy consumption of data centres during their construction.
The article written by Cho and Kim [
6] investigated the feasibility of available energy saving solutions for dedicated data centre cooling using a data centre energy consumption estimation tool. The research was carried out by employing energy or environmental norms together with a comprehensive building design technique. This approach guarantees that state-of-the-art knowledge and innovation are applied in all stages of planning and analysis. The integrated approach, also known as integrated building design, in the case of data centres includes data centre infrastructure management (DCIM). This study aimed to discover adaptive design strategies that can optimise energy consumption by continuously adjusting the environmental conditions and cooling systems of IT centres through simulation.
Data centre ventilation configurations and how the airflow is geometrically influenced by the floor inlet were examined by Lu et al. [
7]. Experimental and numerical investigations of typical cases were also summarised, along with the relevant parameters. The advantages and limitations of different airflow control methods including CAC systems, HAC systems, and vertical exhaust duct system (VEDS) were analysed.
Wang et al. [
8] examined the impact of different cold aisle blocking and containment schemes on the overall efficiency of a containerised data centre. A comprehensive understanding of airflow and temperature distribution is essential for proper thermal management in a data centre to maintain the temperature limits and avoid excessive cooling utilisation. The arrangement and characteristics of all data centres were similar. The most popular was the hot aisle/cold aisle configuration, which serves to reduce the blending of hot and cold air. Either a suspended ceiling or raised floor can be used to supply cold air. Full hot or cold aisle insulation can be used to further minimise hot air recirculation and cold air bypass.
Nada and Elfeky [
9] investigated the thermal performance of data centres with different partitioning and cold aisle content. The objective of this study was to evaluate and compare three different cold aisle arrangements: (i) an open free cold aisle, (ii) a semi-closed cold aisle where the sides of the aisles were closed, and (iii) a completely closed cold aisle where the sides and top of the aisles were closed. The study explored the impact of distinct cold aisle arrangements on the distribution of temperature in the racks and servers as well as performance metrics (SHI and RHI) at varying power densities. For the purpose of this study, a physically simulated data centre with one row of three racks and four simulated servers in each rack was designed and built.
In the referenced article written by Khalaj et al. [
10], an operating data centre was considered. In order to investigate the thermal behaviour of the air distributed inside the data centre, numerical simulations were carried out using experimental data. The simulation results showed several undesirable hotspots near the racks. To minimise the associated thermal inefficiencies, this study examined the effectiveness of utilising cold aisles as a highly efficient and industrially available method. The efficiency of this technique was measured based on three dimensionless indices called SHI, RCI, and the coefficient of performance (COP). In this example, the application of the proposed method improved the SHI, RCI, and COP of the cooling system by more than 0.45%, 17%, and 19.5%, respectively. The results demonstrate the effectiveness of the cold aisle containment method in improving the energy efficiency of data centres.
It is common for a data centre to generate heat that can cause damage to the equipment. Therefore, an air conditioning system is required to remove it. Typically, data centres use a CRAC system to blow cold air to achieve the necessary cooling capacity. The main contributions in the article by Lin et al. [
11] are as follows:
In order to reduce energy consumption in the air conditioning system, the mass flow rate in the optical distribution frame (ODF) can be decreased or the temperature difference between the supply and return branches of the ODF can be increased. The best temperature difference is 10.86 °C.
Separating the hot and cold passages will prevent the mixing of hot and cold fluids, significantly reducing the rate of entropy generation and exergy destruction in the air system.
Compared to the worst design point, the total exergy destruction can be reduced by 12.6% at the optimal design point. The energy saving is 7.5% compared to the mean value of the total exergy loss at the chosen design points. This can be majorly attributed to the reduced energy losses in the air conditioning system.
At the optimal design point, the anticipated PUE is 1.30, which is lower than the average of 1.59 obtained in 2020.
At the optimal design point, the three thermal efficiency indicators, namely RCILo (the subscript Lo represents the temperature conditions at low end of temperature range), RTI, and SHI, are 100%, 101%, and 0%, respectively, and are all within the range of ideal conditions.
Lu and Zhang [
12] analysed how geometric configurations influence the thermal efficiency of data centres and offered extra insights into data centre design. This paper proposed a hypothesis of a significant relationship between the air distribution, geometry, and pressure field in a UFAD system, in conjunction with the existing knowledge. It provides additional insights into the design of retrofitted or newly built data centres. Furthermore, the paper experimentally examined the operating parameters of CRAC. The paper introduced and assessed improved methods to determine the thermal performance. Creating an effective cooling system for a data centre can dramatically impact its thermal performance. This paper discussed two types of factors affecting the thermal performance of a data centre. First, these were geometric configurations: plenum height (0.4, 0.5, 0.6, and 0.7 m), perforation percentage (10, 20, 30, and 40%), and the location of air conditioners in the computer room. Second, the operating parameters included air temperature and air volume. In order to investigate the effect of the location of the cooling system on the overall thermal performance of the data centre, 32 configurations were designed with varying percentages of perforation along the height of the plenum. The results showed that there is a critical range of plenum height (up to 0.6 m) and percentage of perforation (greater than 20%) in which the thermal performance of the system can be significantly improved by changing the geometric configurations. Outside this range, the performance is essentially unchanged.
Zhan et al. [
13] proposed and experimentally verified a ducted hot aisle containment for refrigeration racks. A new type of hot aisle confinement system for refrigerated cabinets was proposed and experimentally validated. Experiments were conducted under different hot aisle conditions, door condition, diffusers, percentage of installed blanking panels, and airflow scenarios. The main results can be summarised as follows: for the overall improvement in the cooling effect of the racks, the sealed envelope was slightly better than the duct envelope. Sealing is obviously better than a collar alone. However, due to the high investment cost of airtight enclosures under realistic conditions, it was ultimately recommended that duct enclosures be used as the most effective method of improving the average rack inlet temperature.
The paper written by Chu et al. [
14] addressed airflow management issues that have a significant impact on cooling efficiency in data centres. Based on the available airflow management techniques, long and short airflow paths were identified. Long paths can result in significant losses due to recirculation, bypass, misdistribution, and leakage, but are relatively inexpensive to install. In contrast, a short path cooling system can effectively reduce or eliminate the above disadvantages, but can be expensive to install. In addition, it may have the problem of improper airflow distribution. Typically, CFD is widely used to study thermal management and airflow control in data centres. However, to further improve the CFD model, it needs to be pre-calibrated with experimental data. Airflow control techniques applicable to raised floor data centres were discussed for long distance cooling systems. This paper reviewed and discussed airflow control techniques applicable to the most common raised floor data centres. The significant airflow control difficulties were hot air recirculation, cold air bypass, leakage, over- and under-supply, airflow, and temperature non-uniformity. Nonetheless, such effects tend to interact with the geometric arrangement of the data centre. Therefore, studies on the effects of plenum depth, perforated tiles, improved means such as forced bypass fans, aisle, and leakage containment, infrastructure layout were discussed and compared, and some air-flow manipulation rules based on previous studies were proposed. Furthermore, studies on the suspended airflow method were examined and contrasted with the raised floor method. Research was conducted on a closed cooling system utilising a heat exchanger and heat pipe for short-distance cooling.
Sundaralingam et al. [
15] concentrated on experimentally characterising the following cold aisle configurations: open aisle, partially open aisle, and fully closed aisle. The presentation of the experimental data aimed to evaluate various configurations including rack inlet contour plots, tile and rack airflow, and measurements of pressure and temperature within the central processing unit (CPU) of the server to assess their respective performances. One of the aims of this paper was to present experimental data that could be used in the future, either for numerical modelling or to illustrate trends. The paper presented a significant amount of data obtained using a variety of measurement instruments. In addition, recommendations for cold aisle sealing can be made based on the experimental results. The following recommendations were proposed, having considered various factors relating to data centre conditions similar to those described in this paper: (1) Where possible, a fully sealed aisle with redundant provision is preferred; (2) Where there are geometric or cost constraints and partial sealing is considered, either an under-provisioned (UP) or over-provisioned (OP) top only arrangement is recommended.
Attia et al. [
16] described the effects of three air distribution systems on the rack temperature profile and thermal management using two performance indices to analyse data centre air management systems. Two indices, SHI and RHI, were calculated experimentally to present the thermal performance of a scaled data centre model. Based on the size analysis, a scaled data centre with a four server rack was designed and built with a scaling factor of 1/6. The temperature distributions of the front and rear racks, server temperatures, and air inlet and outlet heat indices were measured to examine, compare, and assess the performance of three air distribution system configurations in the scaled data centre and to evaluate their effectiveness in simulating actual data centres. The results indicated that both methods of insulation, namely aisle partitioning and aisle enclosure, significantly reduced rack inlet temperatures at critical locations such as the tops of server racks. Insulation using aisle partitioning and aisle containment could reduce rack inlet temperatures up to approximately 3% to 13% for partitioning and around 13% to 15.5% for aisle containment at various power densities. Therefore, CAC provides the best thermal performance because aisle containment completely blocks the top airflow.
A CAC isolation method was considered, which is one of the most effective and industrially available methods to reduce thermal inefficiency. The effectiveness of this method was evaluated using three dimensionless indices: the SHI, RCI, and COP of the cooling system. In this example, Scherer et al. [
10] presented an application of the proposed method that improved the SHI, RCI, and COP of the cooling system. The results obtained demonstrated the effectiveness of the cold aisle containment method in improving the energy efficiency of data centre systems.
The development of verified predictive models for CRAC in data centres was the focus of research written by Alkharabsheh et al. [
17]. These models were based on measured fan curves for the CRAC unit as well as load banks used for server modelling. Validation of the models was conducted by measuring airflow and temperature at critical points within the data centre. The modelling process also included a thorough assessment of the effects of cold aisle leaks that affect the temperature and flow rates. A detailed approach to CAC system modelling was derived and the results were checked with experimental data. The results indicate a strong agreement between the flow and temperature measurements and the CFD results. A detailed model of the rack including the rails for its attachment was also modelled in CFD. This study found that under the conditions considered, rack leaks can raise the temperature by 4 degrees Celsius at the inlet to the upper third of all racks in the cold aisle. This suggests that the upper sections of the racks are highly vulnerable to recirculation. Therefore, they should be considered as crucial areas for characterising leakage in the CAC system. In this study, a certain leakage threshold value was also observed beyond which the inlet temperature of the protection system became irregular and exceeded the supply temperature. This limit was set to about 15% of the total surface area of the vessel. When the leakage level reaches 15%, the containment system permits too much hot air to recirculate, resulting in the loss of the benefits provided by the cold aisle containment.
Gondipalli et al. [
18] explored the effect of CAC insulation on the rack inlet air temperature. Two data centre layouts with raised floor—simple and complex models—were considered in the study. Cold aisles were insulated in three ways: roof, door, roof and door. The analysis was performed for 60% and 80% CRAC airflow on a basic model comprised of 12 server racks. Both methods of insulation (i.e., roof and roof and doors) significantly reduced the rack inlet temperature at critical locations (tops of server racks). The improvement in thermoregulation was mainly due to the obstruction of the upper circulation cells. When the flow rate was at 60%, the CRAC experienced notable infiltration from both the top and edges. Therefore, an insulation technique that incorporates a roof and doors is the most effective. Conversely, when the flow rate increased to 80%, the optimal insulation method was just a roof, as it surpassed the efficiency of other insulation techniques. The door-only insulation technique has minor advantages and can improve the thermal performance of edge server racks by blocking edge recirculation chambers. However, blocking edge recirculation cells contributes to the deepening of the upper recirculation cells, causing the thermal performance at the top of the rack to deteriorate. For a complex scheme with 20 server racks and 80% CRAC flow rate, both insulation methods (i.e., roof and roof and doors) significantly reduced the rack inlet temperature at critical locations (tops of server racks). However, only the roof showed the best results.
The data centre studied in the article by Gao et al. [
19] had an improvement in energy efficiency. The CRAC air conditioners provided plenty of cold air to make the thermal environment ideal. By optimising the airflow pattern with three measures, the energy efficiency was improved. By partially enclosing the cold aisles, the supply air temperature could be increased by 3 °C while still maintaining the thermal environment within the recommended range. This will lead to a corresponding improvement in the CRAC system efficiency, and hence the energy efficiency of the system. Regarding the cooling efficiency evaluation parameters, in some complex cases, the RTI is not suitable for evaluating the bypass or recirculation condition because a recirculation or bypass problem may exist even when the RTI is equal to 1, which is assumed to mean that there is no recirculation or bypass problem. Similarly, an RTI value less than or greater than one may not accurately indicate the presence of bypass or recirculation. Although some metrics have been developed to assess cooling efficiency and airflow patterns in data centres, there is still a need for a comprehensive metric that combines cooling efficiency, energy costs, and equipment failure rates that can also be easily applied.
To improve the cooling efficiency and reduce recirculation, an increasing number of data centres are turning to the hot aisle system. In accordance with heat transfer principles, Zhou et al. [
20] provided a dynamic model for regulating cooling in open and closed environments and suggested decentralised model predictive controllers (MPCs) to manage CRAC units. The study examined an approach to segment the data centre into overlapping CRAC influence zones. In each zone, the MPC controller regulated the rack inlet temperature whilst minimising the cooling power consumption by controlling the CRAC unit fan speed and supply air temperature. The proposed decentralised cooling control method was tested in a production data centre featuring hot aisles enclosed with plastic strips. Experimental results demonstrated its robustness and ability to withstand various perturbations.
In Wibron et al. [
21], CFD modelling was utilised to obtain precise information concerning airflow in a data centre. Experimental values were incorporated to validate the findings, ultimately enhancing the accuracy and reliability of the CFD modelling of the data centres. It is imperative to consider the buoyancy effect when modelling airflow. However, the simplified boundary conditions used in this study may not accurately reflect the results at the rack level (i.e., just upstream of the inlet surfaces located on the components). However, it is evident that the results at the room level were precise. The objective of this investigation was not only to certify the CFD model, but also to contrast the effectiveness of different turbulence models. It was observed that the fluid flow for the various turbulence models diverged at positions that were not adjacent to the main elements of the data centre. Of the three models compared, it was observed that the deviations from the experimental values were similar, with variations above the server racks. For future studies, it is advisable to conduct measurements at additional locations where the flow field is less influenced by boundary conditions than at many of the measurement locations in this research.
In the research by Onyiorah et al. [
22], the idea of utilising rack-level containment was evaluated as a potential airflow management strategy, viable for certain data centre layouts and applications. A parametric investigation was conducted in this paper on a 1008 m
2 fundamental data centre with an IT load of 161 kW. The study was based on a case analysis for CFD to compare the efficiency of independent rack-level containment and snorkels in eradicating hot spots by routing cold air from a lower sub-floor plenum to cabinet inlets, in turn, minimising the inlet temperatures. The article analysed over 40 CFD models and ascertained the comparative degrees of inlet temperature reduction and cooling enhancement achieved through this technique. Test cases such as simulations for comparison with traditional hot aisle/cold aisle containment systems, different combinations of containment strategies, and for different rack density distributions in a data centre model were also included. The results provide a preliminary understanding of the feasibility of this strategy and offer initial recommendations for implementing this concept based on the layout and design of the data centre.
The objective of [
23] as mentioned by Tsuda et al. was to explore the disparities between the CAC and HAC approaches regarding the temperature of the air taken in by the information and communication technology (ICT) equipment in the rack. To achieve this aim, a setup was established to compare the two capping methods under the equivalent load and the same conditioned atmosphere, and the thermal conditions were then measured. With the ICT equipment mounted vertically in the centre of the rack, the intake air temperature was lower when using CAC compared to HAC. The inlet air temperatures of the ICT equipment were adjusted to the same level for both types of aisle sealing by regulating the volume of air provided through the air conditioner. As a result, while maintaining the same inlet temperature of each piece of equipment during capping tests, the power consumption of the air conditioner fan was twice as high when using HAC as opposed to CAC. This suggests that the CAC system can save more energy consumed by air conditioning systems as it can effectively maintain low temperature conditions.
Nemati et al. [
24] examined three containment configurations: cold aisle, hot aisle, and duct (with suspended ceiling). In the first two cases, the CRAH flow rate was controlled based on the pressure differential measured by three sensors inside and outside the containment. In the third case, the CRAH flow rate was controlled based on the required IT airflow. To uphold the designated supply air temperature, the CRAH flow rate was modified by measuring the return air temperature and the overall heat load in the premises. Leakage was monitored by checking the temperature and pressure readings in the aisles and at the server inlets/outlets, respectively.
Shrivastava et al. [
25] compared the performance and cooling energy costs of various types of air containment systems. Furthermore, they offered several recommendations that should be considered when selecting a containment system. Based on this paper, it can be concluded that a containment system can decrease the cooling energy costs by limiting the mixing of hot and cold air flows. The level of insulation directly affects the savings. For instance, if vertical exhaust duct (VED) or CAC insulation is implemented on all cabinets, a hypothetical situation with 100% containment results in low cooling airflow and high supply and return air temperatures. A VED system provides an advantage when high ambient room air temperatures are unacceptable. In typical data centres such as those with storage network cabinets where some equipment in the room cannot be isolated, a VED system captures exhaust air and prevents it from mixing with the cold supply air in an uninsulated area. Regardless of the type of cooling system used, the containment system provides significant opportunities for cooling energy savings. Due to an elevation in the permitted supply air temperature and an upsurge in the return air temperature, the water-based cooling systems analysed in the study exhibited increased savings with a CAC system, followed by a VED system, and ultimately a 100% isolated version of either system. Thus, it is evident that even a slight progressive increase in the supply and return air temperature results in economic benefits with the aforementioned cooling systems. These findings could also apply to an air-side economiser or evaporative cooling system as OpEx performance alone does not determine the optimal containment strategy.
Alkharabsheh et al. [
26] interrogated the impact of distinct containment leakage coefficients on the flow rate and temperature in a typical data centre. The modular data centre represented a genuine cold aisle system in an experimental laboratory. The numerical model was verified through experimental measurements. The model included fan rotation curves in CRAC and IT equipment. Several parametric studies have been carried out, varying leakage rates from 0% (perfectly sealed protection system) to 100% (unsealed cold aisle), to investigate the impact of leakage factors on the airflow rate of the CRAC unit and IT equipment. Technical term abbreviations were explained and clear, concise language was used to maintain comprehensibility. Common academic sections and regular author and institution formatting were included in conventional structure. Precise word choice was used and the text was grammatically correct with no spelling or punctuation errors. Additionally, the effect on the inlet temperature was also studied. The register was formal with positions on subjects made clear through hedging. The results demonstrate that leakage in the containment system significantly affects the performance of the containment system. Objective evaluations were excluded unless clearly marked as such. The language was objective, avoided biased or ornamental language, and employed a passive tone. Clear causal connections were made between statements and a logical flow of information was maintained. A threshold leak rate was discovered beyond which a containment system offers no benefit over a traditional non-containerised cold aisle system. This methodology can be utilised to define leakage in localised systems.
The analysis by Tatchell-Evans et al. [
27] and the corresponding experimental findings confirmed that cold aisle containment in server racks, used in data centres, could lead to considerable levels of bypass. There are numerous HVAC models for thermal comfort that forecast energy consumption under different conditions or schemes. At the cold aisle, up to 20% of the cold air provided can bypass the servers, depending on the level of DpCH and the degree of blockage of the leakage paths in the server racks. It is predicted that improvements in rack design or the blocking of leakage paths can lead to reductions in ET ranging from 1 to 8.8%. The reduction is significantly influenced by factors such as DpCH, IT power density, and server fan speed response to cold aisle pressure. Additionally, ET was found to be dependent on DpCH, with an up to 16% reduction in ET achievable through DpCH optimisation as a function of IT power density, server fan speed response to cold aisle pressure, and degree of rack leakage path blockage. The optimum pressure drop between the hot and cold aisles (DpCH) varies depending on certain parameters. This system model, developed for the first time, examines the impact of bypass and DpCH on the data centre’s total power consumption (ET).
As can be seen from the present study, the number of papers addressing issues related to the use of BIM and CFD for aisle isolation systems is quite minimal. As evidenced by the limited number of sources, the utilisation of BIM has received little attention, despite the fact that precise object geometry significantly enhances the accuracy of subsequent CFD modelling. It is important to note that BIM allows for filling the data centre element models with the necessary details needed to operate the data centre. It is important to explore this area as it is under-researched.