Introduction

The development of information technology (IT) has brought great convenience to mankind. A new level of transformation in data centres (DCs) is driven by the emergence and development of new technologies such as Internet of Things, artificial intelligence and 5th generation. DCs are usually physical spaces that house computer systems and associated devices (such as communication and storage systems) for processing, storing, transferring, exchanging, and managing data. They also contain data communication connections, environmental control equipment, monitoring equipment, and various security devices (Joshi and Joshi 2012; Ashrae 2011). During the past two decades, the development and application of emerging technologies have brought about the explosive growth of data. In response to related demands, the number of DCs has been rapidly growing (Dayarathna et al. 2016; Zhang et al. 2010). However, an increasingly serious energy crisis comes with this rapid growth because large DCs consume as much electricity as a small town industrial business (Mittal 2014). In addition, the heat dissipation per unit area significantly increases in the operation and maintenance processes of DCs (Dewan and Srivastava 2015). The global energy consumption of DCs is estimated at 416 TWh per year (Avgerinou et al. 2017), which is significantly higher than the total energy consumption of the UK (approximately 300 TWh). Koomey pointed out that the energy consumption of DCs doubles every 4 years, and DCs accounted for approximately 1.3% of the total global electricity consumption in 2010 (Ham et al. 2015a). Consequently, DCs have become more energy consumption than conventional office buildings. In addition, DC is the fastest growing carbon footprint sector in the entire IT industry, accounting for 2% of global carbon dioxide emissions (Avgerinou et al. 2017).

Because the thermal environment of the electronic equipment in DCs is critical to the operation safety of data processing, the appropriate temperature and humidity must be maintained. Currently, the air-conditioning system in DCs needs to operate continually 365 days a year (Kurkjian and Glass 2007). According to a survey conducted by www.researchandmarkets.com (Yuan et al. 2019), the global DC cooling market is expected to exceed $8 billion by 2023. The overall energy consumption of a DC is mainly composed of two parts as follows: IT equipment (e.g. servers, storage, and network) and infrastructure (e.g. cooling system). Analyses of the power consumption of DCs showed that a considerable portion of this usage (approximately 40%) is typically for powering the cooling system (Habibi Khalaj and Halgamuge 2017; Ham et al. 2015b). The major power consumers are the chillers, which supply chilled water to the cooling coils to maintain the indoor environment conditions by removing the heat dissipated by the servers (Joshi and Joshi 2012; Schmidt et al. 2005; Siriwardana et al. 2013; Siriwardana 2012; Siriwardana et al. 2012). Hence, it is necessary to reduce the power consumption of DC cooling systems through thermal management techniques.

However, also optimal cooling energy distribution inside the DCs is as important as powering the cooling system. The low cooling efficiency of DCs is mainly due to the uneven airflow distribution (Qian et al. 2013, 2015; Zhang et al. 2018; Sakanova et al. 2019), the resistance of piping systems (Boucher et al. 2006), and high energy consumption of the cold source (Cho and Kim 2016; Ma et al. 2016). At present, experts have proposed several methods for improving cooling efficiency and reducing energy consumption in DCs. Among them, free cooling and airflow management are the most widely used (Oró et al. 2015). The free cooling technology is an ideal method. When the outdoor temperature or sea water temperature is lower than the indoor temperature, the data center is cooled by natural cold source (Pawlish and Aparna 2010). The power usage effectiveness (the ratio of the total power consumed by the data centre to that consumed by the IT equipment) value approaches 1 and the green level of the DC is higher. Existing methods include the direct introduction of outdoor air or water into the DC and utilizing outdoor cold energy by heat exchangers, rotary heat wheel, or two-phase thermosyphon (Zhang et al. 2014; Le Masson et al. 2012; Daraghmeh and Wang 2017; Wang et al. 2019a, b; Phan et al. 2019; Zou et al. 2019; Chauhan and Kandlikar 2019).

Typically, the cooling energy required is transferred to computer room air-conditioner (CRAC) with air. The airflow management strategies are dedicated to changing the DC configuration and CRAC air supply status to improve the cooling efficiency and thermal environment. There are two ventilation methods in conventional DCs as follows: one is up-supply and down-return, while the other is up-return and down-supply. Wang et al. (Wang et al. 2015) found that the height of a raised floor causes significant impacts on the airflow distribution. They proposed a drawer-type rack, which increases the effective hot aisle space and reduces the cold aisle space. With this design, hot air recirculation and cold air bypass will be significantly reduced. Ni et al. (Ni et al. 2017) pointed out that the airflow distribution through the tiles has an important influence on the cooling of computer equipment, and the airflow distribution is more uniform when the air outlet is at an angle of 60° from the tiles. In addition, Nada et al. (Nada and Elfeky 2016) conducted experimental studies on airflow and thermal management of DCs for different arrangements of cold aisle containments, which indicated that the thermal performance of fully enclosed cold aisles is optimal. Subsequently, Zhang et al. (Zhang et al. 2017) investigated the airflow optimization for a small-scale DC with a cold aisle closure. They concluded that the cold aisle closure can optimize the thermal environment in addition to eliminating the bypass airflow.

Many researchers have shown that the application of baffles can significantly improve the airflow effect; therefore, such techniques have been widely investigated. Schmidt et al. (Schmidt et al. 2007) claimed that the installation of baffles can reduce the mixing of hot and cold airflow and improve the airflow distribution in racks. The fan power consumption in IT equipment can be greatly reduced by the efficient use of baffles, and the likelihood of hotspots in DCs can be reduced simultaneously.

The rack baffles can effectively improve the heat distribution in the DC, and the wider the rack baffles, the better the heat distribution in the DC. In recent years, some new methods of applying flexible baffles (FBs) have been studied. The installation of FBs can improve the airflow configuration in DCs without changing their overall structure. FBs are suitable for both new and renovated DCs. In addition, this method requires only a partial modification of the rack level, and there is no need to adjust the air supply parameters of air conditioning. Furthermore, no additional energy consumption is required.

The optimal configuration of FBs has been proposed by Yuan et al. (Yuan et al. 2018, 2019). The application of 20-cm rack baffles tilted to 75° resulted in the most uniform temperature distribution in the rack air outlet, and the hotspot temperature of the rack was reduced by 1.5 °C. In addition, they also applied baffles in the lower terminal of servers and concluded that 8 cm × 45° server terminal baffles can make the rack air outlet temperature distribution more uniform, with the temperature of the rack hotspot reduced by 2.5 °C. The server terminal baffles can also effectively improve the heat distribution in the rack air outlet. Separately, the effect of rack and server terminal baffles on the airflow pattern and heat distribution have been studied by Yuan et al. (Yuan et al. 2018, 2019). However, the integration simultaneously of both rack and server baffles has not received much attention. As the density of racks in DCs continues to increase, the optimization effect of a single baffle system on airflow organization is no longer sufficient. Therefore, the novelty of this study is to combine rack baffle and server baffle system and analyse the optimization effect of the combined system. This study is an extension of the separate previous studies on rack baffles (Yuan et al. 2018, 2019) and server baffles (Yuan et al. 2019). Existing studies on FBs in DCs are verified and analysed. Besides, the optimization mechanisms of these two baffle systems are analysed and validated.

This paper is organized as follows. The first section describes the current state of DC cooling and energy conservation. Then, a combined baffle system using rack baffles and server terminal baffles is proposed. In the second section, a new experimental DC and a new numerical model are introduced. The third section presents the experimental validation on the optimal combination of the rack and server baffle systems in DCs. The velocity fields are analysed to investigate the optimization mechanism of the combined baffle system and numerical simulations are used to verify the results. The optimized combined system is established and analysed, and the simulation results are verified by experiments in the fourth part. Finally, the conclusions are drawn in the last section.

Methodology

Description of the test DC

The DC used for the experiment and simulation is located in a university in Nanjing, China, which operates all year round. This DC has three data rooms, namely, the core DC, super DC, and managed DC. The experiment was carried out in the managed DC. This DC has one air-cooled CRAC (Model TDAR1822, Schneider Electric), and the air supply ducts are installed under the plenum. Figure 1 shows a photo of the DC, which employs a cold aisle containment with the air supplied from an underfloor plenum. The layout of the racks in the test DC is shown in Fig. 2. Perforated tiles are located in the cold aisle to supply cold air to the servers. The hot air discharged by the server fans is extracted by the CRAC unit to re-cool and supplied to the DC plenum. This cooling strategy prevents recirculation of the hot air discharged from the servers and mixing with the supplied cold air from the perforated tiles. The server location in each rack is different, and the number and specification of the servers configured in the racks vary; therefore, the power requirement of each rack differs. Table 1 lists the geometry and technical parameters of the DC for this experiment.

Fig. 1
figure 1

Analysed data centre

Full size image
Fig. 2
figure 2

Schematic of experimental DC

Full size image
Table 1 Specifications of DC components
Full size table

Experimental setup

Experimental validations were conducted in 2019 in an operating DC during June to verify separately the effectiveness of rack baffles and server terminal baffles in optimizing the thermal environment of the DC. In the experiment, we chose a different rack from previous studies. The experimental procedure is as follows: (1) select a rack with higher power density which is suitable for installing both the rack baffles and server terminal baffles, (2) measure the thermal distribution and possible rack hotspots of the rack airflow outlet, and (3) apply the rack baffles and server terminal baffles to the rack to optimize the airflow distribution and reduce the temperature of the rack outlet. Yuan et al. (Schmidt et al. 2007) defined the rack hotspot as the point where the highest exhaust temperature of the rack is located.

This experiment selected rack B4 for the validation of the combined baffles. The total power of the rack B4 is 16.98 kW. Its rack power density is high, which satisfies the requirement of the experiment, and the layout of the server is relatively uniform. The layout and rated power of the servers in the rack B4 are shown in Fig. 3. All servers have a size of 0.09 m (H) × 0.46 m (W) × 0.8 (L) m. In this experiment, nine temperature test points were set in the rack B4, where each measuring point was located at the central line of the rack rear door to ensure that the temperature sensor measures a more uniform air temperature at the air outlet of the rack. The arrangement of the measuring points is presented in Fig. 4.

Fig. 3
figure 3

The layout of servers in the rack B4 (three-dimensional view)

Full size image
Fig. 4
figure 4

Arrangement of measuring points in the rack B4

Full size image

K-type thermocouples were used as temperature sensors to measure the supply air temperature (SAT) of the rack B4, and Agilent 34,970 Data Acquisition was used for recording data. The temperature sensors were arranged in one-to-one correspondence with the measuring points. The first temperature sensor was located 30 cm away from the plenum, and the distance between each adjacent temperature sensor was 20 cm. The exhaust air thermal distribution in the rack B4 was obtained, and the rack hotspot was located. The supply air velocity (SAV) of the rack B4 through the plenum was measured by a handheld airflow anemometer (KIMO VT200/FC300). Ranges and uncertainties of the measuring equipment were shown in Table 2. In the experiment, the data acquisition logger recorded the air temperature measured by the temperature sensor every 10 s, and finally, all the temperature data were transmitted by the data acquisition logger to the personal computer. Each measurement included an adaptive time of 0.5 h and an actual test time of 2 h, i.e. a total of 2.5 h. The adaptive time consists of three parts as follows: the time to change the baffles, time to adjust the angle of baffles, and time which the DC environment recovers to a steady state. Each sensor recorded approximately 1080 temperature values per experiment. To avoid interference from transient temperature fluctuations, the temperature of each measurement point was taken as the arithmetic mean of the measured 1080 temperature values. The reading of the anemometer was between 24 °C and 24.2 °C. Thus, the SAT of the rack B4 was assumed as 24 °C. Meanwhile, the SAV of the rack B4 was measured to be 5.33 m/s.

Table 2 Ranges and uncertainties of the measuring equipment
Full size table

Model setup

The computational fluid dynamics (CFD) modelling study was performed based on the validated model of the same experimental DC. The model was constructed by a commercial CFD software, Fluent Airpak3.0, as shown in Fig. 5. Taking into account such factors as DC scale, racks, servers, and airflow conditions, Abanto et al. (Abanto et al. 2004) considered that the turbulent mixing mode should be chosen as the flow state of the airflow model for the room-level DC. The standard k–ε turbulence model was utilized in the process of numerical simulation due to its satisfactory performance for DC modelling studies (Yuan et al. 2018; Song 2016; Priyadumkol and Kittichaikarn 2014; Almoli et al. 2012). In the calculation of CFD model, the finite volume method was used to transform the differential equation into a discrete equation (Song 2016). This simulation study adopts a steady-state approach, so the phenomena associated with transient flow instability in the DC are not considered, and previous numerical studies (Abanto et al. 2004; Zametaev et al. 2019; Sheth and Saha 2019; Liu et al. 2014; Najjar et al. 1995) have shown the feasibility and benefits of such an approach. Several assumptions were set for the model as follows: (1) the effects of air leakage and radiation were not considered, (2) air was considered as an incompressible fluid, and (3) the heat dissipation caused by fluid viscosity was ignored.

Fig. 5
figure 5

CFD model of the studied DC

Full size image

In actual operation, the server is not always fully loaded. However, running at the full or partial load will have no significant effect on this experiment. Therefore, in the simulation, it was assumed that the server is running at full capacity. In addition, the setting of parameters in the simulation will inevitably deviate from the actual situation, resulting in certain differences between the simulation results and experimental results.

The governing equations for an impressible fluid are expressed as Eqs. (1)–(3), which represents the continuity, momentum, and energy conservation equations, respectively.

$$ \nabla \cdotp \overrightarrow{u}=0 $$
(1)
$$ \frac{\partial }{\mathrm{\partial t}}\left(\overrightarrow{u}\right)+\overrightarrow{u}\cdotp \nabla \overrightarrow{u}=-\nabla p+\nabla \cdotp \left(\overset{=}{T}\right)+\overrightarrow{g} $$
(2)
$$ \uprho {c}_p\left[\frac{\partial T}{\partial t}+\left(\overrightarrow{u}\cdotp \nabla \right)T\right]=\nabla \cdotp \left({k}_{eff}\nabla T\right)+S $$
(3)

where \( \overrightarrow{u} \) represents the average velocity vector, p represents the static pressure, T represents the static temperature, \( \overset{\rightharpoonup }{g\ } \)represents the gravitational acceleration vector, Veff and keff are, respectively, the effective fluid viscosity and thermal conductivity, ρ is the density, S denotes the volumetric heat source, and cp is the specific heat capacity.

According to Silva-Llanca et al. (Silva-Llanca et al. 2018), the mesh generation is important for the simulation period and simulation accuracy and the success or failure of the simulation depends directly on it. Thus, the following criterion was formulated for the mesh generation as follows: smooth element changes and changes in density in the entire solution domain should be based on the variables (Ni et al. 2017). The calculation reached convergence when the residuals of the velocity in the X, Y, and Z directions reached 10−3, and that for energy was 10−6 (Ni et al. 2017; Alkharabsheh et al. 2015). The boundary conditions for this model are given in Table 3. Only racks B3, B4, and B5 are modelled with actual heat power to simplify the numerical model and reduce the simulation time.

Table 3 Boundary conditions of the CFD model
Full size table

The grid number can not only affect the simulation period but also determine the accuracy of the simulation (Ni et al. 2017). A suitable number of meshes can reduce the numerical errors caused by a coarse grid or rounding-off errors caused by dramatic increases in the number of meshes. The grid independence test is widely used in CFD numerical investigations to ensure the feasibility of simulation calculations. Four grid independence tests were performed in this study for four scenarios as follows: with no baffles, with rack baffles, with server terminal baffles, and with the combined baffles. The number of grids was different. The optimum mesh numbers for the scenarios are shown in Fig. 6, in which the hexa-unstructured mesh for this model achieved the balance between good accuracy and high efficiency of the numerical calculation. The mesh quality had been checked automatically by the Grid Quality Inspection Tool of the CFD Airpak3.0 software, and more than 95% of the meshes had a quality of 1 for the four scenarios. The results showed no significant difference when the number of nodes increased, and the minimum gap and maximum side length of all grids were, respectively, 0.001 m and 0.1 m.

Fig. 6
figure 6

Object maximum temperature variation with grid number in (a) no baffles, (b) rack baffles, (c) server terminal baffles, and (d) combined baffles

Full size image

Experimental results and mechanism analysis

The validation experiments considered three scenarios (scenarios E0–E2) as presented in Table 4. In scenario E0, there are no baffles introduced. In scenario E1, the angle between the baffle and the vertical downward direction of the rack door is defined as the baffle angle. In scenario E2, the baffle angle is the angle between the baffle and underside of the server. The baffle sizes and angles are selected for the validation from the earlier studies (Qian et al. 2013, 2015). Figure 7 shows the actual mounting positions of the rack baffles and server terminal baffles. Figure 7 (right panel) shows that all servers are equipped with server terminal baffles of the same size and angle except for the lowest server. No baffles were applied in the lowest server because the flow of cold air in the lower channel will be blocked by the baffle.

Table 4 Three working scenarios in validation experiments
Full size table
Fig. 7
figure 7

Installation of rack baffles (left) and server terminal baffles (right)

Full size image

Reference case without baffles

The experiment was carried out for approximately 2 weeks to obtain the outlet airflow distribution in the rack B4 and examine the influence of different types of baffles on the heat distribution in the rack. Then, the experimental raw data were processed and analysed. In this paper, the average temperature of the nine measurement points was used to describe the overall temperature of the rack B4, and this temperature was used as one of the indicators for determining whether the DC airflow distribution was optimized.

Figure 8 shows the exhaust air temperature of the nine measuring points of the rack B4 with the reference scenario E0 without FBs. The average exhaust temperature of the rack B4 is 28.5 °C. Based on Fig. 8, the temperature of the measuring point at 110 cm distance from the plenum is the highest; therefore, this point is considered as the rack hotspot of the rack B and the temperature of the rack hotspot is 29.1 °C. The maximum temperature difference between the supply temperature and exhaust temperature of the rack reaches 5.1 °C. Furthermore, it can be observed from Fig. 8 that the maximum temperature difference among the nine measurement points reaches 1.9 °C, which indicates that the exhaust temperature of the rack B4 is not uniform enough and there is local overheating.

Fig. 8
figure 8

The exhaust air temperature of nine measuring points of the rack B4 without FBs

Full size image

In the succeeding validation work, the experiments involving scenarios E1–E2 were carried out, and the optimization schemes for the two different baffle systems (rack baffles and server terminal baffles) were respectively set in the rack B4 to investigate their effects on airflow optimization and energy conservation.

Rack baffle system

Figure 9 shows a comparison of the exhaust air temperature for the nine measuring points of the rack B4 in scenarios E0 and E1. In case E0, the temperature of the rack hotspot is approximately 29.1 °C. By applying the optimal rack baffles, the temperature of the rack hotspot drops to 28.8 °C, which is 0.3 °C lower than that without baffles. In addition, the overall temperature is reduced from 28.5 °C to 28.2 °C; thus, the heat accumulation phenomenon of the rack B4 is alleviated.

Fig. 9
figure 9

The exhaust air temperature of nine measuring points of the rack B4 in E0 and E1

Full size image

Figure 10 compares the airflow in the rack B4 with no baffles and with optimal rack baffles. In the case of no baffles, the cold air enters the rack obliquely upwards, and the angle of inclination is large so that the cold air entering the upper part of the rack is more than that below, and the airflow at different heights of the rack is uneven. Therefore, the cold air is tilted up into the rack, causing the bottom of the rack to be insufficiently cooled, resulting in heat accumulation. However, when the optimal rack baffles (20 cm × 75°) are applied, the direction of cold air entering the rack changes from upward tilt to horizontal. With this condition, the airflow at different heights of the rack is more evenly distributed, and the amount of cold air entering the lower part of the rack is guaranteed. Therefore, the optimization mechanism of the rack baffles is changing the direction of the cold air entering the rack, so that the cold air entering different heights of the rack is relatively uniform, and the amount of cold air entering the rack is increased to some extent.

Fig. 10
figure 10

Airflow in the rack B4 without baffles (left) and with optimal rack baffles (right)

Full size image

Server terminal baffle system

The exhaust air temperatures of the nine measuring points of the rack B4 in scenarios E0 and E2 are compared in Fig. 11. The overall temperature is reduced from 28.5 °C to 28.1 °C. In the case of no baffles, the temperature of the rack hotspot is approximately 29.1 °C. The temperature of the rack hotspot is reduced by 0.3 °C relative to the case without baffles by applying the optimal server terminal baffles.

Fig. 11
figure 11

The exhaust air temperature of nine measuring points of the rack B4 in E0 and E2

Full size image

Figure 12 shows a comparison of the airflow in the rack B4 with no baffles and that with optimal server terminal baffles. In the case of no baffles, the cold air enters the rack from the cold aisle containment and mostly flows through the path between the servers, forming a bypass airflow and flowing directly out of the rack rear door. However, by applying the server terminal baffles, the direction of the cold air coming out of the channel between the servers is changed, allowing more cold air to participate in cooling the server terminal. Therefore, the optimization mechanism of the server terminal baffles is changing the direction of the bypass airflow between adjacent servers, thereby taking away more heat at the terminal of the servers and improving the cooling efficiency.

Fig. 12
figure 12

Airflow without baffles (left) and with optimal server terminal baffles (right)

Full size image

The temperatures of the nine different measuring points in all scenarios are compared in Fig. 13. If the temperature is lower than that of E0, it means that the thermal performance is better than the reference case. As shown in Fig. 13, the changing trend of the overall temperature for the rack B4 is consistent in E0–E2. The temperature at the measuring point 110 cm from the plenum is the highest, and the temperature at the measuring point 190 cm from the plenum is the lowest. The average temperature is 28.5 °C in scenario E0, and in scenarios E1 and E2, where rack baffles and server terminal baffles are respectively installed, the average temperatures are reduced to 28.2 °C and 28.1 °C, respectively. In addition, the temperature drops of each measuring point are given in Table 5. There are significant temperature drops in scenarios E1 and E2, and the maximum temperature drops of rack hotspot are 0.5 °C and 0.7 °C, respectively.

Fig. 13
figure 13

Variation of overall exhaust air temperature of the rack B4 in E0–E2

Full size image
Table 5 Temperature drop (°C) at each measuring point
Full size table

Considering the overall exhaust temperature, the temperature drops of the rack hotspot and the temperature of the nine measuring points, the rack baffles and server terminal baffles are both able to optimize the airflow distribution and significantly reduce the risk of local hotspot damages to the servers.

The reliability of the model was verified by optimization experiments of the rack baffles and server terminal baffles, and both systems can improve the thermal environment of the DC to a certain extent. According to the optimization mechanism research, the application of rack baffles increases the amount of cold air entering the rack and makes the distribution of cold air entering the cabinet more uniform. In contrast, the application of server terminal baffles improves the utilization of bypass airflow in the channels between adjacent servers, taking away more heat at the terminal of servers. The above researches have shown that the optimization mechanisms of the two systems are different and independent of each other. Through the analysis of the airflow velocity of these two baffle systems, a guess is proposed that their optimization mechanisms are different and will not affect each other. Thus, the two baffle systems can be easily combined to further optimize the heat distribution and airflow configuration in DCs.

Verification

To validate the accuracy of the simulation results, an experiment of the optimal baffle system combination was carried out. The experimental temperature values ​​in the nine measurement points of the rack B4 in two different situations (without baffles vs. with the optimal combination system) are compared in Fig. 14. The temperature trends in both cases are basically consistent. In the case of installing the optimal baffle combination system, the temperature of each measurement point is slightly lower than the original condition in the experimental data, and the largest temperature drop is 0.9 °C. The combined baffle system can effectively reduce the temperature of the hotspot and make the heat distribution in the DC more uniform.

Fig. 14
figure 14

Comparison of experimental values under two different conditions

Full size image

Overall, the exhaust air temperature distribution and thermal environment were optimized when the combined baffle system was applied. The application could realize the following optimization as follows: (i) the temperatures of the test points were all below 29 °C, (ii) the maximum temperature drop achieved was approximately 0.9 °C, (iii) the temperature of the hotspot in the rack B4 reduced by 0.7 °C from 29.1 °C to 28.4 °C, and (iv) the heat accumulation was eliminated and the thermal environment was improved.

Figure 15 illustrates the deviation of the nine measuring points between the experimental and numerical results. As shown in the figure, the maximum deviations are all below 10%. This proves that the simulations are very close to the experimental values. The deviation between the experimental data and numerical results of each measurement point is listed in Table 6. It can be observed that the numerical results agree well with the experimental results.

Fig. 15
figure 15

The deviation between measured and simulated results of nine measuring points in the rack B4

Full size image
Table 6 Deviation of each measuring point
Full size table

Combined rack and server baffle system

To further demonstrate the independence of these two baffle systems, 11 sets of simulations were performed, namely, scenario 1 (5 cases with 20 cm × 75° rack baffles and 8 cm × 46 cm SLTBs with five baffle angles of 15°, 30°, 45°, 60°, and 75°) and scenario 2 (6 cases with 8 cm × 45° SLTBs and 20 cm × 60 cm rack baffles with six baffle angles of 15°, 30°, 45°, 60°, 75°, and 90°). The optimal size and angle of the combined baffle system were obtained. In addition, the optimal airflow configuration and heat distribution characteristics were analysed. Figure 16 showed a sketch of the baffles’ installation of the combined system. Rack baffles are installed at the entrance of the front door of the rack B4 to change the direction of the cold air entering the servers, so that the cold air entering the rack at different heights is relatively uniform, and the amount of cold air entering is increased. The SLTBs is installed at the server end to change the direction of the bypass airflow at the outlet, thereby taking away more heat from the server terminal and improving the cooling efficiency.

Fig. 16
figure 16

A sketch of the baffles’ installation of the combined system

Full size image

Simulation analysis of the combined system

The numerical simulation analysis considered three scenarios (scenarios S0–S2) and 12 cases, as listed in Table 7. The heat distribution characteristics of the DC in each combination case were analysed, including the possibility that the optimal scheme of different baffle systems can be simply superimposed.

Table 7 The analysed scenarios of the combined baffle system
Full size table

Simulation results of combined baffle systems

Based on the optimal rack baffle (20 cm × 75°), different angles of server terminal baffles were installed at the server to further optimize the thermal environment. Figure 17 shows the exhaust air temperature profiles (the plane of the back door) of the rack B4 in the scenarios S0 and S1. In these scenarios, there is all significant heat accumulation in the middle portions of the rack B4. When the 15° server terminal baffles are used, the heat accumulation in the lower part of the rack B4 does not improve, but instead deteriorates and becomes central heat accumulation. When the angle of server terminal baffles is 30°, the heat accumulation is relieved to some extent. When the baffle angles are 45°, 60°, and 75°, the heat accumulation range of the rack B4 is slightly reduced, the hotspot temperature of the rack is lowered, the heat accumulation of the rack B4 is greatly improved, and the thermal environment is improved. In these three cases, the overall temperature of the rack B4 has dropped. Although the temperature of the rack hotspot decreases when the baffle angles are 60° and 75°, the heat accumulation problem and thermal environment are not improved. Both the heat accumulation problem and thermal environment are improved when the baffle angle is 45°. The exhaust air temperature distribution in the upper and lower parts of the rack B4 is relatively uniform in this case. Therefore, when the rack baffles are fixed and the angle of the server terminal baffles is 45°, the combined system performs best in eliminating heat accumulation and improving the thermal environment.

Fig. 17
figure 17

Exhaust air temperature distribution in the rack B4 for scenarios S0 and S1

Full size image

Figure 18 depicts the exhaust air temperatures of the rack B4 in S0 and S1. Compared with those in scenario S0, the temperatures of the recording points in S1 have different degrees of declines. In particular, when the angle of the server terminal baffles is 45°, the temperatures of all recording points are below 31 °C and the temperature drop at measuring point NO.7 is the largest. In addition, the maximum temperature difference between the nine recording points in S1 is 1.8 °C, and the temperature distribution is relatively uniform. Therefore, it can be observed from the temperature changes of the nine points that the heat distribution performance of the air outlet of the rack B4 is optimal when the angle of the server terminal baffles in S1 is 45°. Table 8 shows the standard deviation of the temperatures recorded by rack B4 in S0 and S1. When the angle of the server terminal baffles in S1 is 45°, the standard deviation of the minimum of 1.18, thus the temperature of rack B4 is more stable than that in other cases.

Fig. 18
figure 18

Exhaust air temperatures of nine temperature recording points for the rack B4 in S0 and S1

Full size image
Table 8 The standard deviation of the exhaust temperature at rack B4 in S0 and S1
Full size table

Similarly, based on the optimal server terminal baffles (8 cm × 45°), baffles with different angles were installed on the front door of the rack to further optimize the thermal environment. Figure 19 shows the exhaust temperature profile (the plane of the back door) of the rack B4 in scenarios S0 and S2. In scenario S0, there is significant heat accumulation in the lower and middle portions of the rack B4. When the server terminal baffles and rack baffles are used at the same time, the heat accumulation in the middle and lower portions of the rack B4 is lower. As the angle increases, the heat accumulation improvement is more evident, and the temperature of the rack hotspot is reduced. However, when the angle is increased to 90°, the temperature of the rack hotspot rises again. The exhaust air temperature distribution in the upper and lower parts of the rack B4 is relatively uniform when the angle of the rack baffles is 75°. Therefore, when the server terminal baffles are fixed and the angle of the rack baffles is 75°, the hotspot temperature of the rack is the lowest, and the heat distribution in the rack is the most uniform.

Fig. 19
figure 19

Exhaust air temperature distribution in the rack B4 for scenarios S0 and S2

Full size image

Figure 20 shows the exhaust temperature of the rack B4 in S0 and S2. Compared with those in scenario S0, the temperatures of the recording points in S2 have different degrees of declines. Especially when the angle of the rack baffles is 75°, the temperature of the recording point is lower than 31 °C, and the temperature at the measuring point NO.7 is the most optimized. In addition, the temperature distribution is relatively uniform. Therefore, considering the temperature change of the nine temperature recording points of the rack B4, when the angle of the rack baffles in S2 is 75°, the heat distribution performance of the rack B4 is optimal. Table 9 shows the standard deviation of the temperature recorded by frame B4 in S0 and S2. When the angle of the server terminal baffles in S2 is 75°, the standard deviation of the minimum of 1.22, thus the temperature of rack B4 is more stable than that in other cases.

Fig. 20
figure 20

Exhaust air temperatures of nine temperature recording points of the rack B4 in S0 and S2

Full size image
Table 9 The standard deviation of the exhaust temperature at rack B4 in S0 and S2
Full size table

The simulation validation shows that when the rack baffles are fixed at 20 cm × 75°, the angle of the server terminal baffles is changed, and the optimal heat distribution in the rack B4 is obtained at 8 cm × 45°. Conversely, when the server terminal baffles are fixed at 8 cm × 45°, the thermal environment of the rack is optimal when the rack baffles are 20 cm × 75°. The simulation results of the above two scenarios show that considering the rack hotspot temperature and heat distribution uniformity of the rack B4, the combined baffle system of 20 cm × 75° rack baffle and 8 cm × 45° server terminal baffles has the best performance in eliminating heat accumulation and improving heat distribution. In addition, the above analysis proves that the optimization mechanisms of the two methods are different and are not affected by each other. Therefore, the two baffle systems can be easily combined to further optimize the heat distribution and airflow configuration in DCs.

Optimal combination of baffle system

Figure 21 shows a comparison of the temperature profiles of the exhaust air temperature in four different situations (no baffles, only optimal rack baffles, optimal server terminal baffles, optimal combined baffles). When the optimal rack baffles (20 cm × 75°) are installed together with the optimal server terminal baffles, the temperature of the rack hotspot is significantly reduced and the temperature distribution in the rack is more uniform. Figure 22 shows the airflow without baffles and that with the optimum combined baffles. A more uniform air supply is achieved by applying the combined baffle system. With this condition, the temperature of the rack hotspot is reduced relative to the original state and the heat distribution in the rack B4 is the most uniform. Both the heat accumulation problem and thermal environment are improved when the combination of baffle systems is applied.

Fig. 21
figure 21

Comparison of exhaust air temperature cloud maps under different conditions

Full size image
Fig. 22
figure 22

Airflow without baffles (left) and with optimum combined baffles (right)

Full size image

The 12 sets of simulations show that the applications of the two systems are not affected by each other; thus, the two baffle systems can be simply combined to further optimize the heat distribution in the DC airflow configuration and the air outlet of racks. With this strategy, the temperature of each measuring point of the rack is lower than 29 °C, and the temperature of the rack hotspot is reduced by 0.7 °C.

Conclusions

A new method of combining rack and server baffle systems was proposed innovatively in this study, which can be applied to high-density DCs with uniform servers without changing the overall structure of DCS. It is suitable for both new and renovated DCs and the optimization effects of the rack baffles and server terminal baffles on the thermal distribution in the DC were validated by experiments. In addition, the optimization mechanisms of the different baffle systems and the effect on the performance of the best baffle combinations were investigated. The optimal combined simulation model was validated by field experiments. The experimental results were in good agreement with the numerical results and the main conclusions are as follows:

  1. (1)

    Based on the velocity field analysis and numerical simulation, the optimization mechanisms of the two baffle systems are different and do not affect each other; therefore, they can be easily combined to further optimize the heat distribution in the DC.

  2. (2)

    The combined system of the optimal server terminal baffles (8 cm × 45°) and optimal rack baffles (20 cm × 75°) achieved the most uniform heat distribution, which greatly reduces the risk of local hotspot damages to the servers.

  3. (3)

    The exhaust air uniformity in all server racks was significantly improved by applying the rack baffles and server terminal baffles. Both the rack baffles and server terminal baffles can reduce the rack hotspots by 0.3 °C, while the combined baffles system can reduce the rack hotspots by 0.7 °C