A server water cooling heat dissipation online control method, system and device

Abstract

The present application relates to the technical field of water cooling heat dissipation, in particular to a server water cooling heat dissipation online control method, system and device. The method analyzes the response correlation between the flow data and load monitoring in the cooling branch of a single server node, combines the change trend between the cooling liquid flow data and the temperature data of the server node, and evaluates the thermal response matching degree. According to the relationship between the thermal response matching degree of the server nodes and the temperature difference change of the inlet and outlet pipes, the heat transfer balance deviation is analyzed. According to the synchronous response of the loop pressure data change in each server node branch, the control deviation influence coefficient is obtained, and the cooling liquid distribution is regulated and controlled comprehensively. The present application adjusts the cooling liquid distribution control of different nodes by periodically monitoring the heat dissipation balance and execution response between nodes, improves the reasonable balance of the cooling liquid distribution in the water cooling heat dissipation process, and reduces the system energy consumption.

Description

Technical Field

[0001] This invention relates to the field of water cooling technology, specifically to an online control method, system, and device for server water cooling. Background Technology

[0002] With the rapid development of information technology and the continuous growth of high-performance computing services such as big data, artificial intelligence and cloud computing, the computing density of servers is constantly increasing, and the heat generated per unit volume is increasing dramatically. Traditional air-cooled heat dissipation systems are limited by factors such as airflow guidance, wind resistance loss and noise control, making it difficult to meet the heat dissipation performance requirements of high-power servers. At this time, water-cooled heat dissipation technology has gradually become the mainstream heat dissipation method for data centers and high-performance computing equipment.

[0003] In typical water-cooling systems, a cold plate heat dissipation structure is often used. This involves placing a cold plate with excellent thermal conductivity in close contact with core components such as the server CPU, GPU, and memory, allowing the coolant to quickly dissipate the heat generated by the chips. However, in real-world data center environments with multiple racks operating in parallel, the load on each server node changes frequently, resulting in significant differences in cooling requirements. Because traditional CDU control logic is mostly based on fixed setpoints or local temperature difference feedback, it lacks global perception and dynamic balancing capabilities regarding coolant distribution. This can easily lead to uneven distribution of coolant in different branch loops. This imbalance manifests as excessive or insufficient flow in some cooling loops, resulting in cooling redundancy or insufficient heat dissipation. This can easily cause hotspots to accumulate on servers, reduce the heat transfer efficiency of the cold plate, and increase system energy consumption. Summary of the Invention

[0004] To address the shortcomings of existing technologies, such as the lack of global perception and dynamic balancing of coolant distribution, which easily leads to uneven coolant distribution in various branch circuits, causing hotspot accumulation in servers, decreased heat transfer efficiency of cold plates, and increased system energy consumption, this invention aims to provide an online control method, system, and device for server water cooling. The specific technical solution adopted is as follows:

[0005] This invention provides an online control method for server water cooling, the method comprising:

[0006] Collect coolant status data for each server node's corresponding cooling branch in the water-cooled heat dissipation system, as well as load and temperature data for each server node; coolant status data includes: inlet and outlet temperature difference, pressure data, and flow rate data;

[0007] Based on the monitoring period, the dynamic response between the load change of the server node and the flow data change in the corresponding cooling branch is coupled and analyzed. Combined with the changing trends of the server node temperature data and flow data, the thermal response matching degree of each server node is obtained.

[0008] For each server node, the correlation between its thermal response matching degree and the temperature difference between the inlet and outlet of the corresponding cooling branch is analyzed during the monitoring period to obtain the heat transfer balance deviation.

[0009] Based on the timing linkage of the pressure data change response of each server node and other server nodes in the corresponding cooling branch, as well as the control execution delay of the server node temperature data change, the control deviation influence coefficient of the server node is obtained.

[0010] Adjust the cooling branch valve control based on the heat transfer balance deviation and control deviation influence coefficient of each server node.

[0011] Furthermore, the method for obtaining the thermal response matching degree includes:

[0012] For any given server node, the coolant distribution coupling degree of that server node is obtained based on the degree of synchronization between the high fluctuation of flow data and the high fluctuation of load in the corresponding cooling branch during the monitoring period.

[0013] During the monitoring period, the expected cooling trend of the server node is obtained based on the correlation between the temperature data curve and the traffic data curve in the time series.

[0014] By combining the coolant distribution coupling degree and the expected cooling trend of the server node, the thermal response matching degree of the server node is obtained.

[0015] Furthermore, the method for obtaining the coolant distribution coupling degree includes:

[0016] During the monitoring period, the absolute value of the load slope of the server node at each moment is analyzed, and the set of moments with high changes in the absolute value of the load slope is taken as the load mutation set; the absolute value of the flow data slope of the cooling branch corresponding to the server node at each moment is analyzed, and the set of moments with high changes in the absolute value of the flow data slope is taken as the flow mutation set.

[0017] By combining the absolute difference in slope and time difference between the same sequence number of the load mutation set and the flow mutation set, the coolant distribution coupling degree is obtained. The absolute difference in slope and time difference between the sequence number of the corresponding time are negatively correlated with the coolant distribution coupling degree.

[0018] Furthermore, the method for obtaining the heat transfer equilibrium deviation includes:

[0019] For any given server node, each other server node is taken as the analysis node in turn. Based on the degree to which the thermal response matching degree of the server node is higher than that of the analysis node, and the degree of similarity between the inlet and outlet temperature differences of the corresponding cooling branches of the server node and the analysis node during the monitoring period, the heat transfer efficiency comparison between the server node and the analysis node is obtained.

[0020] By comparing the heat transfer efficiency of this server node with all other server nodes, the heat transfer balance deviation of this server node is obtained.

[0021] Furthermore, the method for obtaining the heat transfer efficiency comparison includes:

[0022] The ratio of the thermal response matching degree of the server node to that of the analysis node is used as the relative response coefficient between the server node and the analysis node.

[0023] The differences in the temperature difference between the server node and the analysis node at each moment during the monitoring period were analyzed. By combining the differences in the temperature difference between the server node and the analysis node at all moments during the monitoring period, the heat dissipation similarity between the server node and the analysis node was obtained. The difference in the temperature difference between the server node and the analysis node was negatively correlated with the heat dissipation similarity.

[0024] By combining the heat dissipation similarity and relative response coefficients of the server node and the analysis node, a comparison of the heat transfer efficiency of the server node and the analysis node is obtained.

[0025] Furthermore, the method for obtaining the control deviation influence coefficient includes:

[0026] During the monitoring period, the absolute value of the pressure data slope of the server node at each time point is analyzed, and the set of times with high changes in the absolute value of the pressure data slope is taken as the pressure mutation set.

[0027] For any given server node, the stress response coupling degree between that server node and other server nodes is obtained based on the minimum time difference between stress mutations and the difference in data volume in the stress mutation set. Both the minimum time difference between stress mutations and the difference in data volume are negatively correlated with the stress response coupling degree.

[0028] Obtain the control time of the CDU control change of the server node during the monitoring period; determine the execution response latency of the server node based on the time difference between each control time and the preceding temperature change time.

[0029] By combining the pressure response coupling degree and execution response latency of the server node, the control deviation influence coefficient of the server node is obtained.

[0030] Furthermore, the method for obtaining the execution response latency includes:

[0031] During a preset reference period before each control moment, the difference between the temperature data of the server node at each moment and the temperature data at the previous moment is analyzed, and the moment corresponding to the largest difference is taken as the temperature change moment of each control moment.

[0032] The time difference between each control moment and the temperature change moment is used as the delay of each control moment; the execution response delay of the server node is obtained by combining the delays of all control moments of the server node; there are no control moments in the preset reference time period.

[0033] Furthermore, the adjustment of cooling branch valve control based on the heat transfer balance deviation and control deviation influence coefficient of each server node includes:

[0034] For any given server node, obtain the target opening degree of the cooling branch valves of each server node; use the difference between the target opening degree and the current opening degree as the adjustment amount.

[0035] By combining the heat transfer equilibrium deviation and control deviation influence coefficient of the server node, the current monitoring control coefficient is obtained; by combining the control coefficient and the adjustment opening amount, the corrected adjustment opening amount is obtained.

[0036] When the adjusted opening value is greater than the preset adjustment threshold, the current opening of the server node is adjusted to the target opening value by adjusting the adjusted opening value; otherwise, no adjustment is made.

[0037] The present invention also provides an online control system for server water cooling, the system comprising:

[0038] The data acquisition module is used to collect coolant status data of the corresponding cooling branches of each server node in the water cooling system, as well as load and temperature data of each server node; the coolant status data includes: inlet and outlet temperature difference, pressure data and flow data.

[0039] The node heat transfer equalization analysis module is used to couple and analyze the dynamic response between the load change of the server node and the flow data change in the corresponding cooling branch based on the monitoring period. Combined with the changing trends of the server node temperature data and flow data, the thermal response matching degree of each server node is obtained.

[0040] For each server node, the correlation between its thermal response matching degree and the temperature difference between the inlet and outlet of the corresponding cooling branch is analyzed during the monitoring period to obtain the heat transfer balance deviation.

[0041] The node execution response analysis module is used to obtain the control deviation influence coefficient of the server node based on the timing linkage between each server node and other server nodes in the corresponding cooling branch pressure data change response, as well as the control execution delay of the server node temperature data change.

[0042] The control module is used to control the cooling branch valves based on the heat transfer balance deviation and control deviation influence coefficient of each server node.

[0043] The present invention also provides an online control device for server water cooling, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of any of the above-described online control methods for server water cooling.

[0044] The present invention has the following beneficial effects:

[0045] This invention analyzes the response correlation between flow data and load monitoring in the cooling branch of a single server node, and comprehensively evaluates the thermal response matching degree of the current server node by combining the changing trends of coolant flow data and temperature data of the current node. It preliminarily analyzes the coolant distribution response capability to heat dissipation demands based on the adjustment of individual nodes during the monitoring period. Then, based on the relationship between the thermal response matching degree between server nodes and the temperature difference changes in the inlet and outlet pipes, it analyzes the balance of actual heat dissipation conversion between server nodes and other nodes, and measures the adverse effects of uneven coolant distribution on the heat dissipation capacity and thermal response at this time. Simultaneously, based on the synchronous response of loop pressure data changes in each server node branch, considering the transmission effect of pressure fluctuations, and combining the delayed response of distribution execution, a control deviation influence coefficient is obtained, reflecting the degree of influence generated during control response. This comprehensive approach regulates the coolant distribution of the current node to improve the flow response level of coolant in each server node during water cooling. This invention adjusts the coolant distribution control of different nodes by periodically monitoring the heat dissipation balance and execution response between nodes, improving the rational balance of coolant distribution during water cooling and reducing system energy consumption. Attached Figure Description

[0046] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1This is a flowchart of an online control method for server water cooling provided in one embodiment of the present invention;

[0048] Figure 2 This is a structural diagram of an online water-cooling control system for servers, provided in one embodiment of the present invention. Detailed Implementation

[0049] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a server water-cooling heat dissipation online control method, system, and apparatus proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0050] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0051] The following description, in conjunction with the accompanying drawings, details the specific solution of the online control method, system, and device for server water cooling provided by this invention.

[0052] Example 1:

[0053] A typical server room water cooling system includes a chilled water system, distributed water cooling plates, coolant distribution units, and cooling towers or dry coolers. The basic principle of water cooling is that the heat generated by the server processor, graphics card, and memory chips is carried away by the coolant through circulation. Finally, the heat is exchanged with the server room water circulation system through the CDU to achieve heat dissipation for the server cluster.

[0054] Please see Figure 1 The diagram illustrates a flowchart of an online control method for server water cooling according to an embodiment of the present invention. The method includes:

[0055] S1: Collect coolant status data for each server node's corresponding cooling branch in the water-cooled heat dissipation system, as well as load and temperature data for each server node; coolant status data includes: inlet and outlet temperature difference, pressure data, and flow rate data.

[0056] In this embodiment of the invention, a server node refers to an independent computing entity within a data center that includes heat-generating units such as CPUs, GPUs, and memory chips. Temperature sensors and performance counters deployed on the chip surface are used to collect real-time temperature data and load from each node, thereby reflecting its instantaneous heat load and heat dissipation requirements.

[0057] At each server node, an independent cooling branch forms a local loop for coolant flow, precisely delivering coolant to the cold plate in contact with that node for direct heat dissipation of core heat-generating components. Sensors are deployed on the inlet and outlet pipes of this branch to collect key coolant status data, including: inlet and outlet temperature difference (measuring the temperature rise of the coolant after flowing through the node's cold plate, characterizing the node's actual heat dissipation); pressure data (reflecting the flow resistance within the branch and the system's pressure distribution); and flow rate data (the volume of coolant flowing through the branch per unit time).

[0058] When regulating coolant, CDU control typically triggers control commands based on a set fixed temperature threshold or simple temperature difference feedback from local loops to uniformly adjust valves and distribute coolant across different server nodes. However, in high-density computing environments, frequent and asynchronous load changes make this control method, based on local and static strategies, highly susceptible to low actual heat dissipation efficiency and uneven coolant distribution due to dynamic response lag and a lack of global awareness.

[0059] Therefore, in this embodiment of the invention, a periodic monitoring period is set, such as 5 minutes per monitoring period. The specific value can be adjusted by the implementer according to the specific implementation scenario, and no limitation is made here. By comprehensively analyzing the distribution thermal response and execution capability between nodes in each period, the control situation is adaptively adjusted to improve the dynamic optimization capability and balanced control capability of global coolant distribution.

[0060] It is understandable that monitoring data undergoes preprocessing, which may include data standardization and time-scale normalization to facilitate unified data analysis and remove the influence of units. It should be noted that data preprocessing is a well-known technique in the field, and the implementation of the data collection frequency setting can be adjusted by the implementer, and will not be elaborated upon or restricted here.

[0061] S2: Based on the monitoring period, the dynamic response between the load change of the server node and the flow data change in the corresponding cooling branch is coupled and analyzed. Combined with the changing trends of the server node temperature data and flow data, the thermal response matching degree of each server node is obtained.

[0062] Because server nodes in multi-rack environments experience frequent load fluctuations and varying heat dissipation requirements, the actual effectiveness of traffic allocation can easily be overlooked when analyzing individual nodes. Mismatches may occur, such as load changes without corresponding traffic responses, or stable loads with fluctuating traffic volumes. Therefore, analyzing the dynamic response relationship between traffic data and load on a single server node is crucial. Furthermore, the response of traffic allocation alone cannot fully represent the rationality of the allocation, and inconsistencies in the actual heat dissipation effect can exist across the entire node network.

[0063] Therefore, by analyzing the dynamic response of a single server node to load and traffic, and combining this with the server node's temperature change trend, the degree of actual thermal response matching of the single server node is evaluated. In this embodiment of the invention, the method for obtaining the thermal response matching degree includes:

[0064] First, for any given server node, during the monitoring period, the coolant distribution coupling degree of the server node is obtained based on the degree of synchronization between the high fluctuation of flow data and the high fluctuation of load in the corresponding cooling branch of the server node. This measures the dynamic response capability of the coolant supply to significant load changes.

[0065] In this embodiment of the invention, during the monitoring period, the absolute value of the load slope of the server node at each moment is analyzed. The set of moments with high changes in the absolute value of the load slope is taken as the load mutation set. After obtaining the absolute value of the load slope at each moment, the absolute values ​​of the load slope are arranged in ascending order. The difference between the sorted data is calculated. Between the two absolute values ​​of the slope corresponding to the largest difference, the smaller absolute value of the slope is taken as the dividing absolute value of the slope. All moments corresponding to the absolute values ​​of the load slope that are greater than the dividing absolute value of the slope are combined into the load mutation set. The load mutation set represents the moments when the load changes drastically.

[0066] Similarly, the absolute value of the flow rate slope for the corresponding cooling branch of the server node at each time point is analyzed. The set of times with high changes in the absolute value of the flow rate slope is taken as the flow rate mutation set. After obtaining the absolute value of the flow rate slope at each time point, the absolute values ​​of the flow rate slope are arranged in ascending order. The difference between the sorted data is calculated. Between the two absolute values ​​of the slope corresponding to the largest difference, the smaller absolute value of the slope is taken as the dividing slope absolute value. All time points corresponding to the absolute values ​​of the flow rate slope greater than the dividing slope absolute value are combined to form the flow rate mutation set. The flow rate mutation set represents the time points where the coolant flow rate changes drastically.

[0067] By combining the absolute difference in slope and time difference between the same sequence number of the load mutation set and the flow mutation set at corresponding times, the coolant distribution coupling degree is obtained. Both the absolute difference in slope and the time difference between sequence number-corresponding times are negatively correlated with the coolant distribution coupling degree. Coupling is achieved from two aspects: the synchronicity of response time and the degree of matching of response intensity, reflecting the supply-side response distribution status.

[0068] In one specific embodiment of the present invention, the times in the load mutation set and the flow mutation set are arranged in chronological order. For the load mutation set, the time difference at the same index and the absolute value difference of the slope corresponding to each time are calculated. The product of the time difference and the absolute value difference of the slope at each index is used as the coupling deviation degree for each index. The smaller the time difference and the smaller the absolute value difference of the slope, the closer the response time and the more similar the amplitude of the high fluctuation situation, and the smaller the deviation degree. The mean of the coupling deviation degrees of all indices with the same index in the load mutation set and the flow mutation set is negatively correlated and normalized to obtain the coolant distribution coupling degree. The existence of times with the same index represents the coexistence of high fluctuation situations. Only the coexisting high fluctuation situations are analyzed. The smaller the deviation, the more matched the time response synchronicity and intensity, i.e., the higher the coupling degree. In this embodiment of the present invention, the difference can be obtained by the absolute value of the numerical difference.

[0069] It should be noted that normalization and negative correlation mapping are both techniques well-known to those skilled in the art. Normalization can be linear or standard normalization, etc., and negative correlation mapping can take the form of a negative exponential power with the natural constant as the base or an inverse proportional function, etc. The negative exponential function is negatively correlated and normalized to obtain the coolant distribution coupling degree, where x is the independent variable. In the process of obtaining the coolant distribution coupling degree, x represents the mean of the coupling deviation degree with the same index in the load change set and the flow change set. The specific calculation is not elaborated or restricted here.

[0070] The greater the coupling degree of coolant distribution, the more sensitive the supply side is to changes in the distribution state. However, the rationality of flow distribution also needs to consider the actual heat dissipation capacity. If the heat transfer of the cold plate is uneven or the local heat load is concentrated, the temperature may still be hot spots, and the high flow rate does not match the heat dissipation response.

[0071] Therefore, considering that the increased temperature due to load necessitates a larger flow rate of coolant for heat dissipation and timely cooling, the response relationship between temperature and flow rate should exhibit a synchronous trend change. Furthermore, during the monitoring period, the cooling trend expectation of the server node is obtained based on the correlation between the temperature data curve and the flow rate data curve over time. In one specific embodiment of this invention, curve fitting is performed on the temperature and flow rate data distributions during the monitoring period to obtain temperature and flow rate curves. The Pearson correlation coefficient between the temperature and flow rate curves is calculated and normalized to obtain the cooling trend expectation of the server node. A larger Pearson correlation coefficient indicates a higher degree of similarity in the trends between the curves, and the cooling flow rate change is more in line with expectations. The normalization process can employ Z-Score standardization.

[0072] It should be noted that the methods for curve fitting and Pearson correlation coefficient calculation are well-known techniques in the art, such as least squares fitting, and will not be limited or elaborated here.

[0073] Finally, by combining the coolant distribution coupling degree and the expected cooling trend of the server node, the thermal response matching degree of the server node is obtained. In this embodiment of the invention, the product of the coolant distribution coupling degree and the expected cooling trend of the server node is normalized to obtain the thermal response matching degree of the server node, which reflects the rationality of the coolant distribution response at a single server node. The larger the thermal response matching degree, the stronger the cooling supply response capability brought about by changes at the node. The normalization process can be performed using a hyperbolic tangent function.

[0074] S3: For each server node, analyze the correlation between its thermal response matching degree and the temperature difference between the inlet and outlet of the corresponding cooling branch with all other server nodes during the monitoring period to obtain the heat transfer balance deviation.

[0075] Based on the analysis of node thermal response, the overall cooling measurement can be optimized by comparing the thermal response matching degree of all nodes horizontally. By analyzing the consistency of the inlet and outlet water temperature difference of each server node, the heat dissipation uniformity of the entire server cluster can be evaluated. If the thermal response matching degree of a server node is high, meaning that the flow can respond to the load in a timely manner, but the inlet and outlet temperature difference is similar to that of other server nodes, it means that the heat dissipation efficiency per unit flow is not high, indicating an imbalance of excessive flow supply.

[0076] Preferably, in this embodiment of the invention, the method for obtaining the heat transfer equilibrium deviation includes:

[0077] For any given server node, each other server node is sequentially used as the analysis node. Based on the degree to which the thermal response matching degree of the server node is higher than that of the analysis node, and the approximation of the inlet and outlet temperature differences of the corresponding cooling branches of the server node and the analysis node during the monitoring period, the heat transfer efficiency comparison between the server node and the analysis node is obtained. Through a comprehensive evaluation of response time and heat dissipation capacity, the degree of heat dissipation comparison between the two nodes is reflected.

[0078] In this embodiment of the invention, the ratio of the thermal response matching degree of the server node to that of the analysis node is used as the relative response coefficient between the server node and the analysis node. The larger the relative response coefficient, the higher the response capability of the server relative to the analysis node.

[0079] Furthermore, the differences in the temperature difference between the server node and the analysis node at each moment during the monitoring period are analyzed. Combining the differences in the temperature difference between the server node and the analysis node at all moments during the monitoring period, the heat dissipation similarity between the two nodes is obtained. The difference in the temperature difference between the server node and the analysis node is negatively correlated with the heat dissipation similarity. By calculating the difference in the temperature difference between the server node and the analysis node at each moment, and then negatively mapping the mean of the differences in the temperature difference between the server node and the analysis node, the heat dissipation similarity between the server node and the analysis node is obtained. The smaller the overall distribution of the difference in the temperature difference between the server node and the analysis node over the time period, the more consistent the heat dissipation level is over time, and therefore the greater the heat dissipation similarity. In a specific embodiment of the present invention, using... The negative exponential function is used to perform a negative correlation mapping to obtain the heat dissipation approximation. The independent variable is the mean of the difference between the inlet and outlet temperature at all times. The smaller the mean, the greater the heat dissipation approximation.

[0080] Further combining the heat dissipation similarity and relative response coefficient between the server node and the analysis node, a comparative analysis of their heat transfer efficiency is obtained. A larger relative response coefficient aligns with a larger heat dissipation similarity, indicating that while the server node has a higher response capability than the analysis server node, its heat dissipation level has not increased, suggesting a higher possibility of excessive imbalance in local heat distribution. In one specific embodiment of this invention, the product of the heat dissipation similarity and relative response coefficient between the server node and the analysis node is used as the comparative analysis of their heat transfer efficiency.

[0081] By combining the heat transfer efficiency of this server node with that of all other server nodes, the heat transfer balance deviation of this server node is obtained. In one specific embodiment of the invention, the mean value of the heat transfer efficiency comparison between this server node and all other server nodes is normalized to obtain the heat transfer balance deviation of this server node, which characterizes the degree of significance of local imbalance in this server node. The normalization process can be performed using Z-Score standardization.

[0082] S4: Based on the timing linkage of the pressure data change response of each server node and other server nodes in the corresponding cooling branch, and the control execution delay of the server node temperature data change, the control deviation influence coefficient of the server node is obtained.

[0083] During the operation of the current water cooling system, when the heat load distribution of each server node is uneven or the valve adjustment response is not synchronized, the local flow resistance will change dynamically, resulting in uneven pressure distribution between loops. At this time, because the pressure wave propagates at a very fast speed in the liquid, the rapid disturbance of the loop at a single node will be instantly transmitted to other loops. The linkage effect will cause control mismatch of some nodes, resulting in local coolant distribution imbalance.

[0084] Meanwhile, the imbalance of loop pressure in the water cooling system may be caused by differences in the flow resistance of the loop itself, or by deviations in pump speed regulation or valve control execution. Further analysis of the delay in control execution is needed to analyze the actual effect of control parameter changes on different loops and comprehensively evaluate the degree of imbalance at the nodes.

[0085] Preferably, in this embodiment of the invention, the method for obtaining the control deviation influence coefficient includes:

[0086] During the monitoring period, the absolute value of the pressure data slope of the server node at each moment is analyzed, and the set of moments with high changes in the absolute value of the pressure data slope is taken as the pressure mutation set. In this embodiment of the invention, after obtaining the absolute value of the pressure data slope at each moment, the absolute values ​​of the pressure data slope are arranged in ascending order. The difference between the sorted data is calculated. Between the two absolute values ​​of the slope corresponding to the largest difference, the smaller absolute value of the slope is taken as the dividing absolute value of the slope. All moments corresponding to the absolute values ​​of the pressure data slope greater than the dividing absolute value of the slope are combined to form the pressure mutation set. The pressure mutation set represents the moments when significant pressure changes occur.

[0087] For any given server node, the pressure response coupling degree between that server node and other server nodes is obtained based on the minimum time difference between the moments in the pressure mutation set and the difference in data volume within the pressure mutation set. Both the minimum time difference and the difference in data volume are negatively correlated with the pressure response coupling degree. In one specific embodiment of the invention, the time difference between the server node and each other server node in the minimum moment in the pressure mutation set is calculated. The initial time difference of the occurrence of significant pressure change events reflects the degree of correlation of the impact of the linked response when a disturbance occurs. The difference between the server node and each other server node in the total data volume within the pressure mutation set is calculated to reflect the correlation of the number of significant fluctuations after a disturbance.

[0088] The smaller the time difference and the closer the total data volume, the higher the probability of a stress-related response. Therefore, the product of the minimum time difference and data volume difference between the server node and each other server node is calculated and a negative correlation mapping is performed to obtain the correlation coupling between the server node and each other server node. The mean of the correlation coupling between the server node and all other server nodes is used as the stress response coupling degree of the server node. In a specific embodiment of the present invention, using... The negative exponential function is used to perform negative correlation mapping to obtain the correlation coupling. The product result is used as the independent variable. The smaller the product, the greater the correlation coupling.

[0089] Further considering the execution of allocation changes, when abnormal temperature changes occur, the CDU needs to control the coolant in the current branch to respond promptly to the temperature change. The control moment of the CDU control change on the server node during the monitoring period is obtained, representing each event where coolant allocation is performed based on control parameters. Understandably, if there is no control moment during the monitoring period, the most recent control moment before the current moment is determined for execution deviation analysis.

[0090] Based on the time difference between each control moment and the preceding temperature abrupt change moment, the execution response latency of the server node is determined. Since the timing of the control parameters is based on the response to temperature changes, the abnormal moment of the temperature abrupt change is located before the control moment. In this embodiment of the invention, during a preset reference period before each control moment, the difference between the temperature data of the server node at each moment and the temperature data at the previous moment is analyzed. The moment corresponding to the largest difference is taken as the temperature abrupt change moment of each control moment. By analyzing the temperature data changes between moments in the preceding time series, the moment of highest change is determined as the temperature abrupt change moment.

[0091] The time difference between each control moment and the moment of temperature change is used as the delay of each control moment. A larger delay indicates a slower execution response of the CDU control system. When multiple control moments exist within a monitoring period, the execution response delay of the server node is obtained by combining the delays of all control moments. In one specific embodiment of this invention, the average delay of all control moments is normalized to obtain the execution response delay of the server node. Because slow control is more likely to cause global flow or pressure shifts, affecting other loops, the required adjustment is higher. The normalization process can employ Z-Score standardization.

[0092] In this embodiment of the invention, the preset reference time period for each control moment is set to the time period within 1 minute before each control moment, and it must satisfy that there is no control moment in the preset reference time period to prevent the previous control response from affecting the current analysis. If there is a control moment in the preset reference time period, the time period between each control moment and the previous control moment is used as the preset reference time period to analyze the execution response.

[0093] Finally, by combining the pressure response coupling degree and execution response latency of the server node, the control deviation influence coefficient of the server node is obtained. In this embodiment of the invention, the product of the pressure response coupling degree and execution response latency of the server node is normalized and used as the control deviation influence coefficient of the server node. The larger the control deviation influence coefficient, the more significant the impact of uneven pressure distribution and control delay at the node, requiring a higher degree of adjustment to balance the imbalance. The normalization process can employ Z-Score standardization.

[0094] S5: Adjust the cooling branch valve control based on the heat transfer balance deviation and control deviation influence coefficient of each server node.

[0095] By analyzing the global balance of the thermal response of the integrated nodes and the control reliability affected by the control execution deviation, the control situation of each server node is analyzed, the adjustment intensity of each server node is quantified, and the adjustment intensity is greater for nodes with more severe heat transfer unevenness and execution deviation.

[0096] In this embodiment of the invention, for any server node, the target opening degree of the cooling branch valve of each server node is obtained. The target opening degree is the adjustment target of the CDU control system in direct response to temperature changes. It can be obtained by model construction through prior experiments on the temperature and opening degree requirements of the server node, which will not be elaborated here.

[0097] The difference between the target opening degree and the current opening degree is then used as the adjustment amount, representing the opening degree adjustment requirement under ideal conditions. However, considering the system imbalance, adjustment and regulation are performed. Combining the heat transfer balance deviation and control deviation influence coefficient of the server node, the currently monitored control coefficient is obtained. In this embodiment of the invention, the product of the heat transfer balance deviation and control deviation influence coefficient of the server node is normalized to obtain the currently monitored control coefficient, and the control coefficient is mapped to the range [0.5, 1.5]. This control coefficient is used to adjust the control of this server node in the next monitoring period. In a specific embodiment of the invention, it is possible to utilize... Normalization mapping process yields control coefficients. The function represents the normalization process, such as the hyperbolic tangent function. x represents the independent variable, which means that in the process of obtaining the control coefficient, the product of the heat transfer equilibrium deviation and the influence coefficient of the control deviation is used as the independent variable. The larger the product, the larger the control coefficient.

[0098] By combining the control coefficient and the adjustment opening amount, a corrected adjustment opening amount is obtained. In a specific embodiment of the present invention, the product of the control coefficient and the adjustment opening amount is used as the corrected adjustment opening amount, which is used as the adjusted control adjustment amount.

[0099] Furthermore, when the corrected adjustment opening value is greater than the preset adjustment threshold, it indicates that the correction is meaningful. The corrected adjustment opening amount is then adjusted from the current opening to the target opening. That is, when the target opening is greater than the current opening, the corrected adjustment opening amount is increased; when the target opening is less than the current opening, the corrected adjustment opening amount is decreased. It should be noted that the adjusted opening must be within the valve opening range to ensure that the opening value is controlled within a physically feasible range. If the adjusted value exceeds the opening range, the closest range boundary value will be used.

[0100] In this embodiment of the invention, the preset adjustment threshold can be set to 5%. If the adjustment force is too small, it will cause the valve to frequently perform small movements, increasing energy consumption and valve wear. Therefore, no adjustment is performed when the corrected adjustment opening value is less than or equal to the preset adjustment threshold. The control is adjusted in the monitoring period of the subsequent new cycle by using the current control coefficient, and the overall system balance is re-monitored. The valve opening is adjusted by iterative analysis of the control coefficient, which improves the dynamic optimization control of multiple nodes and effectively improves the dynamic distribution balance of water cooling heat dissipation.

[0101] In summary, this invention analyzes the response correlation between flow data and load monitoring in the cooling branch of a single server node, and comprehensively evaluates the thermal response matching degree of the current server node by combining the changing trends between the coolant flow data and the temperature data of the server node. It also preliminarily analyzes the coolant distribution response capability to heat dissipation demands based on the adjustment of a single node during the monitoring period. Then, based on the thermal response matching degree between server nodes and the relationship between the temperature difference changes in the inlet and outlet pipes, it analyzes the balance of actual heat dissipation conversion between the server node and other nodes, and measures the adverse effects of uneven coolant distribution on the heat dissipation capacity and thermal response at this time. Simultaneously, based on the synchronous response of loop pressure data changes in each server node branch, considering the transmission effect of pressure fluctuations, and combining the delayed response of distribution execution, a control deviation influence coefficient is obtained, reflecting the degree of influence generated during control response. This comprehensive approach regulates the coolant distribution of the current node under the overall imbalance situation, improving the flow response level of coolant in each server node during water cooling heat dissipation. This invention adjusts the coolant distribution control of different nodes by periodically monitoring the heat dissipation balance and execution response between nodes, thereby improving the rational balance of coolant distribution during water cooling and reducing system energy consumption.

[0102] Example 2:

[0103] This invention also provides an online control system for server water cooling; please refer to [link / reference]. Figure 2 The diagram shows a structural diagram of an online water-cooling control system for a server according to an embodiment of the present invention. The system includes: a data acquisition module 201, a node heat transfer equalization analysis module 202, a node execution response analysis module 203, and a control module 204.

[0104] The data acquisition module 201 is used to collect the coolant status data of the corresponding cooling branch of each server node in the water cooling heat dissipation system, as well as the load and temperature data of each server node; the coolant status data includes: inlet and outlet temperature difference, pressure data and flow data.

[0105] The node heat transfer equalization analysis module 202 is used to couple and analyze the dynamic response between the load change of the server node and the flow data change in the corresponding cooling branch based on the monitoring period. It combines the changing trends of the server node temperature data and flow data to obtain the thermal response matching degree of each server node.

[0106] For each server node, the correlation between its thermal response matching degree and the temperature difference between the inlet and outlet of the corresponding cooling branch is analyzed during the monitoring period to obtain the heat transfer balance deviation.

[0107] The node execution response analysis module 203 is used to obtain the control deviation influence coefficient of the server node based on the timing linkage of the pressure data change response of each server node and other server nodes in the corresponding cooling branch, as well as the control execution delay of the server node temperature data change.

[0108] The control module 204 is used to control the cooling branch valves based on the heat transfer balance deviation and control deviation influence coefficient of each server node.

[0109] It should be noted that the system provided in the above embodiments is only an example of the division of the above functional modules. In practical applications, the above functions can be assigned to different functional module units as needed, that is, the internal structure of the system can be divided into different functional module units to complete all or part of the functions described above. Since the specific implementation process of the server water cooling heat dissipation online control system in this embodiment is the same as the specific implementation process of the server water cooling heat dissipation online control method described above, it will not be described in detail here.

[0110] Example 3:

[0111] The present invention also provides an online control device for server water cooling, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the above-described online control method for server water cooling.

[0112] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0113] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A method for online control of server water cooling, characterized in that, The method includes: Collect coolant status data for each server node's corresponding cooling branch in the water-cooled heat dissipation system, as well as load and temperature data for each server node; coolant status data includes: inlet and outlet temperature difference, pressure data, and flow rate data; Based on the monitoring period, the dynamic response between the load change of the server node and the flow data change in the corresponding cooling branch is coupled and analyzed. Combined with the changing trends of the server node temperature data and flow data, the thermal response matching degree of each server node is obtained. For each server node, the correlation between its thermal response matching degree and the temperature difference between the inlet and outlet of the corresponding cooling branch is analyzed during the monitoring period to obtain the heat transfer balance deviation. Based on the timing linkage of the pressure data change response of each server node and other server nodes in the corresponding cooling branch, as well as the control execution delay of the server node temperature data change, the control deviation influence coefficient of the server node is obtained. Adjust the cooling branch valve control based on the heat transfer balance deviation and control deviation influence coefficient of each server node; The method for obtaining the control deviation influence coefficient includes: During the monitoring period, the slope of the pressure data of the server node at the specified time is analyzed, and the set of times when the pressure data slope changes significantly is taken as the set of pressure abrupt changes. For any given server node, the stress response coupling degree between that server node and other server nodes is obtained based on the minimum time difference between the stress mutation sets and the difference in the amount of data in the stress mutation sets. Both the minimum time difference between the stress mutation sets and the difference in the amount of data are negatively correlated with the stress response coupling degree. Obtain the control time of the CDU control change of the server node during the monitoring period; determine the execution response latency of the server node based on the time difference between each control time and the preceding temperature change time. By combining the pressure response coupling degree and execution response latency of the server node, the control deviation influence coefficient of the server node is obtained.

2. The online control method for server water cooling according to claim 1, characterized in that, The method for obtaining the thermal response matching degree includes: For any given server node, the coolant distribution coupling degree of that server node is obtained based on the degree of synchronization between the high fluctuation of flow data and the high fluctuation of load in the corresponding cooling branch during the monitoring period. During the monitoring period, the expected cooling trend of the server node is obtained based on the correlation between the temperature data curve and the traffic data curve in the time series. By combining the coolant distribution coupling degree and the expected cooling trend of the server node, the thermal response matching degree of the server node is obtained.

3. The online control method for server water cooling according to claim 2, characterized in that, The method for obtaining the coolant distribution coupling degree includes: During the monitoring period, the load slope of the server node at the specified time is analyzed, and the set of times with high changes in load slope is taken as the load mutation set; the flow data slope of the cooling branch corresponding to the server node at the specified time is analyzed, and the set of times with high changes in flow data slope is taken as the flow mutation set. By combining the slope difference and time difference between the same sequence number of the load mutation set and the flow mutation set, the coolant distribution coupling degree is obtained. The slope difference and time difference between the sequence number of the corresponding time are negatively correlated with the coolant distribution coupling degree.

4. The online control method for server water cooling according to claim 1, characterized in that, The method for obtaining the heat transfer equilibrium deviation includes: For any given server node, each other server node is taken as the analysis node in turn. Based on the degree to which the thermal response matching degree of the server node is higher than that of the analysis node, and the degree of similarity between the inlet and outlet temperature differences of the corresponding cooling branches of the server node and the analysis node during the monitoring period, the heat transfer efficiency comparison between the server node and the analysis node is obtained. By comparing the heat transfer efficiency of this server node with all other server nodes, the heat transfer balance deviation of this server node is obtained.

5. The online control method for server water cooling according to claim 4, characterized in that, The method for obtaining the heat transfer efficiency comparison includes: The ratio of the thermal response matching degree of the server node to that of the analysis node is used as the relative response coefficient between the server node and the analysis node. The differences in the temperature difference between the server node and the analysis node at each moment during the monitoring period were analyzed. By combining the differences in the temperature difference between the server node and the analysis node at all moments during the monitoring period, the heat dissipation similarity between the server node and the analysis node was obtained. The difference in the temperature difference between the server node and the analysis node was negatively correlated with the heat dissipation similarity. By combining the heat dissipation similarity and relative response coefficients of the server node and the analysis node, a comparison of the heat transfer efficiency of the server node and the analysis node is obtained.

6. The online control method for server water cooling according to claim 1, characterized in that, The method for obtaining the execution response latency includes: During a preset reference period before each control moment, the difference between the temperature data of the server node at each moment and the temperature data at the previous moment is analyzed, and the moment corresponding to the largest difference is taken as the temperature change moment of each control moment. The time difference between each control moment and the temperature change moment is used as the delay of each control moment; the execution response delay of the server node is obtained by combining the delays of all control moments of the server node; there are no control moments in the preset reference time period.

7. The online control method for server water cooling according to claim 1, characterized in that, The adjustment of cooling branch valve control based on the heat transfer balance deviation and control deviation influence coefficient of each server node includes: For any given server node, obtain the target opening degree of the cooling branch valves of each server node; use the difference between the target opening degree and the current opening degree as the adjustment amount. By combining the heat transfer equilibrium deviation and control deviation influence coefficient of the server node, the current monitoring control coefficient is obtained; by combining the control coefficient and the adjustment opening amount, the corrected adjustment opening amount is obtained. When the adjusted opening value is greater than the preset adjustment threshold, the current opening of the server node is adjusted to the target opening value by adjusting the adjusted opening value; otherwise, no adjustment is made.

8. A server water-cooling heat dissipation online control system, characterized in that, The system includes: The data acquisition module is used to collect coolant status data of each server node's corresponding cooling branch in the water-cooled heat dissipation system, as well as load and temperature data of each server node; coolant status data includes: inlet and outlet temperature difference, pressure data, and flow rate data. The node heat transfer equalization analysis module is used to couple and analyze the dynamic response between the load change of the server node and the flow data change in the corresponding cooling branch based on the monitoring period. Combined with the changing trends of the server node temperature data and flow data, the thermal response matching degree of each server node is obtained. For each server node, the correlation between its thermal response matching degree and the temperature difference between the inlet and outlet of the corresponding cooling branch is analyzed during the monitoring period to obtain the heat transfer balance deviation. The node execution response analysis module is used to obtain the control deviation influence coefficient of the server node based on the timing linkage between each server node and other server nodes in the corresponding cooling branch pressure data change response, as well as the control execution delay of the server node temperature data change. The control module is used to control the cooling branch valves based on the heat transfer balance deviation and control deviation influence coefficient of each server node. The method for obtaining the control deviation influence coefficient includes: During the monitoring period, the slope of the pressure data of the server node at the specified time is analyzed, and the set of times when the pressure data slope changes significantly is taken as the set of pressure abrupt changes. For any given server node, the stress response coupling degree between that server node and other server nodes is obtained based on the minimum time difference between the stress mutation sets and the difference in the amount of data in the stress mutation sets. Both the minimum time difference between the stress mutation sets and the difference in the amount of data are negatively correlated with the stress response coupling degree. Obtain the control time of the CDU control change of the server node during the monitoring period; determine the execution response latency of the server node based on the time difference between each control time and the preceding temperature change time. By combining the pressure response coupling degree and execution response latency of the server node, the control deviation influence coefficient of the server node is obtained.

9. A server water-cooling heat dissipation online control device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the online control method for server water cooling as described in any one of claims 1 to 7.

Images