1. Introduction
Due to the rapid development of the big data era, not only the growth has been in the number of the data centers, but also in the scale and thermal density of the data centers. The increasing demand for communication and computing for IT services has led to a significant increase in power demand (kW) and energy consumption (kWh) in data centers. The power density of a single server rack is approximate 5–10 kW, and the heat dissipation of the microelectronic chip alone has reached 60–90 W/cm
2. More specifically, the power density of the single server rack is expected to exceed 50 kW by 2025 and Information and Communications Technology infrastructure could use up to 50% of the world’s electricity in 2030, requiring greater efforts to preserves the power grid’s stability. Beyond that, the total power consumption of the data center has accounted for 1.3% of global power consumption and continues to increase [
1,
2,
3,
4,
5,
6,
7].
In terms of the thermal management strategy, the data center commonly uses two thermal management strategies: room-level and cabinet-level [
8]. The room-level thermal management strategy is to achieve the purpose of heat dissipation for each cabinet through the cooling of the whole data room. Although the room-level cooling has the advantages of low investment cost and easy layout, due to the expanding size of the data center; if the room-level thermal management strategy is continuously adopted, the total energy consumption of the cooling system will further increase.
The researchers proposed the cabinet-level thermal management strategy. This strategy designs a one-to-one cooling system for each cabinet with different heat dissipation, and places it in different forms inside the cabinet. This approach has higher energy efficiency and better energy saving potential.
With regard to the cooling technology commonly used in data centers, it can be categorized into three categories: air-type, water-type and heat-pipe type. Air-type is more widely used in room-level thermal management data centers. In a data center equipped with high-temperature IT equipment, it is necessary to run a huge cooling system throughout the year. Because of the low heat dissipation performance of air, many scholars have actively studied how to reduce the energy consumption of air cooling systems. Many scholars have proposed to reduce the cooling energy consumed by introducing air-side economizers, thereby improving heat source efficiency [
9,
10]. Park et al. [
11] analyzed the energy-saving effects of direct and indirect air-side economizers in Korean data centers; the results showed that its annual cooling-energy consumption increased by approximately 6.1% compared to the case with no recirculation, and the indirect air-side economizer also exhibited an approximately 9% increase in cooling-energy consumption. Cho et al. [
12] proposed that the air cooling system will have problems such as hot air recirculation and cold air bypass, which reduces cooling efficiency and generates local hot spots. Wang et al. [
13] pointed out that inefficient cooling systems and immature control methods are the most significant problems in cooling telecommunications base stations. Samadiani [
14] proposed that blade servers can dissipate more than 60 kW of heat, while the current CRAC systems are designed for 10–15 kW racks. At present, the heat dissipation of the data center is nearly 100 W/cm
2, and the heat dissipation capacity of air cooling is only 37 W/cm
2 [
15]. Although air cooling methods have the characteristics of a simple system, low failure rate and mature technology, its limitations are increasingly obvious, and its heat dissipation capacity is low, which makes it difficult to meet the heat dissipation requirements of high-density equipment rooms.
The working principle of heat pipe cooling is to drive the heat transfer medium to undergo phase change and flow by using the temperature difference between the evaporation end and the condensation end to finally achieve heat dissipation [
16]. Heat pipe cooling is an indirect liquid cooling without direct contact between the heat source and liquid coolant, contrasting the traditional indirect liquid cooling method. Many scholars usually focus on microchannel heat sinks and their efficiency enhancement as they have superior augmented heat transfer characteristics compared to the traditional indirect liquid-cooled methods [
15,
17]. Wherein, the heat pipe coolant may be water, methanol, acetone, ammonia, R141b, NF or SiO2-H2O, and the condenser coolant may be air or water [
18]. The heat-pipe type is more widely used in cabinet-level thermal management data centers. Jouhara and Wang [
19,
20] both proposed combining heat pipe cooling with conventional compression refrigeration and applying them to the data center cooling system. It was found that the heat pipe can be used for natural cooling, reducing the operating energy consumption of the air conditioning unit and the PUE value of the data center. Ding et al. [
21] used a separate heat pipe for cabinet cooling, and the experimental results showed that the PUE value of the data center could reach 1.20. Dang et al. [
22] proposed a new rack structure with inner air channel and pulsating heat pipe; the simulation found that the temperature distribution in the rack was relatively even, effectively eliminating local hot spots. Although the heat pipe has strong heat transfer capacity, it has the maximum heat transfer limit, and its engineering application stability and energy conservation still need further in-depth research.
Since the heat transfer coefficient of liquid is much larger than air, researchers have proposed to apply liquid cooling technology in high heat density data centers. Zimmermann et al. [
23] studied the energy efficiency of Aquasar, the first hot water cooled supercomputer prototype and found that the water cooling system can dissipate the server with a shorter and more efficient heat transfer path and save 40% energy consumption. Kim et al. [
24] integrated a hot water cooling system with a desiccant assisted evaporative cooling system for cooling the data center, the heat removed by the hot water is used for the regeneration of the desiccant. The operating energy consumption of this system is 95% lower than that of the conventional CRAH system. Furthermore, domestic and foreign scholars have proposed hybrid cooling. Gao et al. [
25] proposed a hybrid cooling system model realized by the rear door heat exchanger. Based on the analysis of two CFD simulation methods, it was found that the model has better heat dissipation ability and is suitable for high heat density data center. Chi et al. [
26] compared the full energy consumption of a hybrid air-water cooling system and enclosed liquid cooling system, and found that in the partial PUE of 1.14, hybrid air-water cooling system improved efficiency by 34% compared to enclosed direct liquid cooling systems. As a commonly used liquid cooling medium, water has certain defects, namely liquid leakage risk and water pollution. Therefore, the relevant design specifications of the data center stipulate that water is not allowed to enter the machine room and its engineering application is limited.
In addition, some scholars [
27,
28] discussed the current status of data center cooling technology, including air side energy savers, water side energy savers and heat pipe technology. They proposed that the use of free cooling on the airside or water side depends to a large extent on environmental conditions. On the other hand, although heat pipes have unique characteristics, they only transfer heat at a small temperature difference, and there are still some limitations.
Furthermore, electrical transformers are one of the essential and expensive devices in supplying electrical energy for industries and buildings. Nevertheless, high temperature is known as one of the major sources of failure and aging of a transformer, which must be removed by efficient cooling techniques [
29]. In view of availability and low cost, precisely with the beginning of energy evolution, insulating liquids have been used for the insulation and cooling of electrical devices, such as transformers, cables, switches and capacitors [
30]. Kim et al. [
31] introduced prediction and experimental study on the cooling performance of radiators used in oil-filled power transformer applications and found that the maximum cooling capacity of direct-oil-forced flow increased by 20.1% compared with non-direct flow.
In view of the above-mentioned two thermal management strategies and various characteristics of three cooling technologies, it is also considered that oil cooling technology has been widely used in transformer cooling for a long time, and insulating oil is already ideal working fluid in liquid cooling technology. Under these conditions thereupon, in order to solve the heat dissipation problem of high heat density data center, this paper proposed a rear door oil-cooling heat exchanger device that could be applied to cabinet level cooling, and combined the thermal management strategy of cabinet level with oil cooling technology. More specifically, compared with the traditional CRAH system, the cabinet-level RDHE system of this paper was more targeted; it could effectively eliminate local hotspots in the data center, and avoided waste of cooling as much as possible. Likewise, compared with the water cooling system, the insulating cooling oil can effectively reduce the damage caused to the server due to the leakage of the working fluid and reduce the limitation of engineering application. Finally, the high-temperature cooling oil after cooling the cabinet could be utilized for the preparation of hot water, or regenerative desiccant and other uses to achieve waste heat recovery and reuse.
4. Analysis and Discussion of the Testing Results
In this part, preliminary experiments were conducted on the heat dissipation performance test of the air-cooled cabinet and the liquid-cooled cabinet, and the experimental results were analyzed and discussed. Furthermore, the heat dissipation performance of the RDHE of the liquid-cooled cabinet was studied. Based on the test results, the heat dissipation performance of the RDHE will be analyzed from four evaluation indicators for the system operating under different working conditions, namely the average heat dissipation efficiency, the cumulative heat dissipation, energy saving potential and the coefficient of thermal performance.
4.1. Comparison between Traditional Air-Cooled Cabinets and Liquid-Cooled Cabinets
This section compared the heat dissipation of traditional air-cooled cabinets with liquid-cooled cabinets. Taking the working conditions 1–4 as an example, the test was divided into four groups, and the heat dissipation of the two cabinets was analyzed from the average heat dissipation efficiency, the cumulative heat dissipation and the operating temperature of the cabinet. It could be seen from
Figure 7 that, under the same working conditions, the average heat dissipation efficiency of the RDHE in the liquid-cooled cabinet was higher than that of the air-cooled cabinet Among them, compared with the traditional air-cooled cabinet, under the same heat source power, the average heat dissipation efficiency of RDHE could be increased by up to 66.6%. The statistics of the cumulative heat dissipation of the two cabinets are shown in
Table 4; the cumulative heat dissipation of the liquid-cooled cabinet was much higher than that of the air-cooled cabinet. In addition, this section also compared the operating temperatures of the two cabinets. It can be seen in
Table 5 that the operating temperature of the cabinet of the traditional air-cooled cabinet was higher than the temperature of the liquid-cooled cabinet, and the operating temperature of the liquid-cooled cabinet did not exceed 50.5 °C. There was no doubt that the heat dissipation capability of the liquid-cooled cabinet was much higher than that of the air-cooled cabinet, and the communication cabinet could be maintained stably and safely.
Although the liquid cooling system had the above benefits over other air-cooled base systems, the main disadvantage was to ensure liquid sealing and thus avoid leakage problems. The system was a rear door liquid-cooled cabinet, compared with the general liquid-cooled cabinet, the leakage of electronic components was reduced as much as possible, but there were still hidden dangers and the system needs to be sealed.
4.2. Average Heat Dissipation Efficiency of the RDHE
The average heat dissipation efficiency is the ratio of the cumulative heat dissipation to the total heat generation of the simulated heat source during the experimental test conditions. It is one of the main indicators reflecting the heat dissipation performance of the heat dissipation device of the communication cabinet.
The average heat dissipation efficiency is one of the main indicators reflecting the heat dissipation performance of the RDHE. Under the same test conditions, the larger average heat dissipation efficiency means greater heat dissipation capability. The average heat dissipation efficiency of the RDHE was tested by changing the cooling oil flow rate and the fan wind speed under the same simulated heat source power, as shown in
Table 6. When the simulated heat source power in the range of 800–1600 W, the average heat dissipation efficiency of the heat dissipation performance of the RDHE was 44.1%. The RDHE has a larger heat dissipation capacity when the heat source power is 800 W, and the average heat dissipation efficiency could reach 58.8%.
According to the test results, it could be concluded that under the same cooling oil flow rate, the average heat dissipation efficiency of the RDHE under the influence of different wind speeds showed a linear growth trend with the increase of wind speed, as shown in
Figure 8 and
Figure 9. It showed that the RDHE has better heat dissipation effect in the case of good airflow organization.
4.3. The Cumulative Heat Dissipation
The cumulative heat dissipation of the RDHE was mainly calculated by the temperature rise of the oil in the heat storage oil drum, and the cumulative heat dissipation of the device was calculated according to Equation (1).
Table 7 showed the cumulative heat dissipation of the heat storage oil drum under different working conditions. When the simulated heat source power in the range of 800–1600 W, the average cumulative heat dissipation of the RDHE could reach 5226 kJ, and when the heat source power was 1600 W, the maximum could reach 6009 kJ.
4.3.1. The Trend of Heat Dissipation
The entire experimental test time was 3 h. In order to study the trend of heat dissipation of the RDHE with time, the experimental results of the test 5–16 were taken as an example. It can be seen in
Figure 10 and
Figure 11 that in the early stage of the test, the heat dissipation rate of the RDHE increased rapidly, and the growth gradually became flat during the test period, and finally the heat dissipation amount tended to be stable. Choi et al. [
33] pointed out that the operating temperature of the CPU is generally lower than 70 °C. Bar-Cohen [
34] found that when the CPU operating temperature exceeded the allowable operating temperature, the reliability of the chip was reduced by 10% for every 2 °C. Through the data acquisition system, it was found that the final temperature in the cabinet is as shown in
Table 8 after three hours of operation. Under the heat dissipation work of the RDHE, the temperature of the cabinet could be lower than 70 °C, and the maximum was 50.9 °C, which indicated that the heat dissipation of the RDHE is more effective and can maintain good operation inside the cabinet.
4.3.2. Cooling Oil Flow
In order to explore the heat dissipation of the RDHE compared the heat dissipation of the system under different flow rates. In this experiment, two flow rates were set, 7.8 m
3/h and 8.8 m
3/h. In the case of the same heat source power and the same air flow rate, a comparative test was carried out. In the case of the condition of 800 W heat source power, the experimental results of tests 1 and 7 were taken as an example. In the case of the heat source power of 1600 W, the experimental results of tests 4 and 10 were taken as an example for analysis. The experimental results are shown in
Figure 12 and
Figure 13. It can be seen from
Figure 8 that the RDHE has better heat dissipation capability and could carry away higher heat when operating at a small flow rate of 7.8 m
3/h. Due to the reduced flow rate, the cooling oil could be better exchanged in the cabinet. Through comparative analysis, it could find when the cooling oil is operated at a flow rate of 7.7 m
3/h, the heat dissipation capacity can be increased by 6%.
4.4. Potential Heat Recovery Characterization
Based on the rapid development of data centers and the large amount of heat in the data center, waste heat recovery has become an important research direction. Regarding the waste heat recovery in data centers, Ebrahimi et al. [
1] proposed several directions that could be applied to waste heat utilization in data centers, including the following uses: (a) domestic space and water heating; (b) power plant co-location; (c) absorption cooling; (d) organic Rankine cycle; (e) desalination; (f) biomass processing; (g) piezoelectrics; (h) thermoelectrics.
The allowable temperatures for waste heat recovery in different cooling systems varied from a higher range of 50–60 °C and a lower range of 25–35 °C [
34]. Through experimental tests, it was found that the circulating cooling oil of this experiment and the exhaust at the top of the cabinet has a certain heat recovery value.
4.4.1. The Temperature of Cooling Oil
Due to the uncontrollable ambient temperature, there were small variations in the ambient temperature and humidity of each set of experiments and the initial temperature of the heat storage oil drum, but the initial conditions of each test in this experiment started at a dry bulb temperature of 26 °C ± 1 °C and a humidity of 60% ± 10%. The initial temperature of the oil was 28.5 ± 0.5 °C. The quality of the oil in the heat storage oil drum was always 170 kg before each test.
Table 9 showed the temperature changes of the heat storage oil drum during the experimental test under the different working conditions. Under the different working conditions, the final temperature of the heat storage oil drum could be higher than 42.8 °C, and the highest temperature could reach 46.9 °C. In the London district heating network program, for secondary heat sources, heat from the data center at temperatures ranging from 32–40 °C was considered to be the most cost-effective and carbon-efficient heat source [
1]. It could be seen that the cooling oil becomes high-temperature oil after the experiment, and could be used as a heat source at this time to prepare domestic water, which can improve energy utilization and achieve the purpose of utilizing waste heat of the data center. After the heat recovery is fully realized, the high temperature oil is returned to the low temperature oil and the data room is continuously re-cooled.
4.4.2. The Temperature of the Exhaust Air of the Cabinet
When the liquid cooling device performed heat dissipation, the top exhaust unit of the cabinet also performed the heat dissipation work at the same time. Under the test of different heat source power, by adjusting the wind speed of different fans, it was found that the exhaust air temperature of the cabinet also had a certain utilization value.
Table 10 shows the variation of exhaust air temperature during the experimental test under the different working conditions. It can be found that under the same oil flow rate and the same heat source power, the larger the air flow rate, the greater the temperature rise of the exhaust air. In the case of different working conditions, the final temperature of the exhaust air could be higher than 40 °C, and the highest could reach 42.8 °C. The temperature range of the captured waste heat from air cooled servers (35–45 °C) is more than sufficient for reuse in heating needs [
35]. The high temperature exhaust air temperature embodies two aspects. On the one hand, it effectively maintains the temperature stability inside the cabinet and contributed to work normally. On the other hand, the high-temperature exhaust air temperature could have sufficient heat recovery value, and could be used as a heat source to regenerate the desiccant, and to improve energy utilization. After the heat exchange, the high temperature air could re-enter into the machine room for cooling work to achieve the purpose of rational use of resources.
4.5. The Coefficient of Thermal Performance
The coefficient of thermal performance (COP) is the ratio of the heat dissipation of the heat sink to the input shaft power under the test conditions. It is another important indicator of the heat dissipation performance of the reaction liquid cold cabinet test system. It can also reflect the energy conversion rate of the device.
The coefficient of performance (COP) calculation results are shown in
Table 11 below. It can be seen that with the increase of wind speed and cooling oil flow, COP showed an increasing trend, indicating that in a good airflow organization environment, the heat dissipation environment in the cabinet is good, and the heat dissipation performance of RDHE is more effective. At different flow rates, smaller flows also exhibited better heat dissipation. In general, the average COP of the entire system could reach 0.85, indicating that the system has good heat dissipation performance.
4.6. The Discussion of the Model
In order to further illustrate the heat dissipation performance of this model, this part verified the experimental results and the energy balance calculation model. In addition, based on the experimental results, a calculation model was established to predict the increase trend in heat dissipation of RDHE with the test time.
4.6.1. The Verification of the Energy Balance Model
This part verified the experimental results and the energy balance calculation model. After verification, it was found that the system and the energy balance model can be well matched with an error of 8%. Thus, this energy balance model could be applied to the heat calculation of this system. The distribution of heat in the energy balance is shown in the
Figure 14 below.
4.6.2. The RDHE Calculation Model
Based on the experimental results, a calculation model was established to predict the increase trend in heat dissipation of RDHE with the test time. The study found that the cumulative heat dissipation of the RDHE cooling device was a logarithmic change. Based on the experimental test data, we could assume that the logarithmic model equation is as follows:
The coefficients and regression formula correlation are gained by the test results and shown in
Table 12.
Comparing the test results with the calculating value by the calculation model, the calculation model can perform well in predicting the increase trend in heat dissipation of RDHE with the test time under different test conditions. The comparison of the test results with the model calculating value are show from
Figure 15,
Figure 16,
Figure 17 and
Figure 18 under different conditions.
5. Conclusions
This paper proposed a novel rear door oil-cooling heat exchanger device for cabinet-level cooling. The heat dissipation performance of the RDHE was tested experimentally and compared with the heat dissipation performance of the conventional air-cooled cabinet. In order to explore the heat dissipation performance of the RDHE, the test rig was constructed and tested in the laboratory of Guangdong University of Technology under different simulated heat source conditions and cooling oil flow conditions. The parameters used to evaluate the performance of the RDHE, namely the average heat dissipation efficiency, the cumulative heat dissipation, the potential heat recovery characterization and the coefficient of thermal performance. Moreover, a calculation model based on energy balance was established and the accuracy of the experiment was verified. Furthermore, according to the variation law of the cumulative heat dissipation of the RDHE, the regression fitting could be used to predict the heat dissipation law. It was found that:
Compared with the traditional air-cooled cabinet, the RDHE’s heat dissipation capacity has increased by an average of 56.2%, and the maximum can be increased by 66.6%.
When the simulated heat source power is in the range of 800–1600 W, the average heat dissipation efficiency of the heat dissipation performance of the RDHE was 44.1%. When the cooling oil flow rate was 7.8 m3/h, the wind speed was 7 m/s and the RDHE was operated at a heat source of 800 W, the heat dissipation efficiency was up to 58.8%.
After running for three hours, under the cooling work of the RDHE, the cabinet temperature under different working conditions could be lower than 70 °C (this temperature was the specified allowable operating temperature of the cabinet). The experimental results showed that the maximum temperature of the cabinet temperature in all tests was only 50.9 °C, which indicated that the heat dissipation of the RDHE is more effective and can maintain good internal operation of the cabinet.
Both the circulating cooling oil of this experiment and the exhaust unit of the cabinet have potential heat recovery values. Among them, the final temperature of the oil drum could be higher than 42.8 °C; the highest could reach 46.9 °C. The final temperature of the exhaust air could be higher than 40 °C; the highest could reach 42.8 °C. After heat reuse, the temperature of the circulating oil could be lowered, and the next cooling could be performed better.
With the increase of wind speed and cooling oil flow, COP showed an increasing trend, indicated that in a good airflow organization environment, the heat dissipation environment in the cabinet is good and the heat dissipation performance of RDHE is more effective. At different flow rates, smaller flows exhibited better heat dissipation. In general, the average COP of the entire system could reach 0.85, which indicated that the system has good heat dissipation performance.
After calculation, the system and the established energy balance calculation model could be well matched with an error of 8.1%. According to the logarithmic variation law of the cumulative heat dissipation of the RDHE, the regression fitting was carried out. It was found that under different experimental conditions, the model could predict the heat dissipation law of the RDHE well and have a good agreement with the experimental values.
The study demonstrated that, compared with the traditional air-cooled cabinet, the RDHE could exhibit better heat dissipation and has a potential heat recovery value. The system is suitable for applying in high heat density data room, and it could be used in areas with district heating networks to achieve waste heat utilization in data centers. Typically, in practical applications, both liquid-cooled and air-cooled data centers would have completely different designs and hardware configurations, and it may be difficult to use them directly in practice. Although the liquid cooling system has the above advantages over other air-cooled base systems, the main disadvantage is to ensure liquid sealing and thus avoid leakage problems. In most liquid-cooled designs, coolant pumps are required for delivery. To ensure the safety of personnel and equipment, leak detection systems are required for practical applications. The system is a rear door liquid-cooled cabinet; compared with the general liquid-cooled cabinet, the system does not directly touch the electronic components inside the cabinet, the leakage of electronic components is reduced as much as possible, but there were still hidden dangers and the system needs to be sealed. The next step of this study is to study the thermal performance of the entire system, the economic benefits of heat recovery and the optimization of the experimental equipment, providing an effective theoretical basis for the use of the system in practical applications.