Liquid cooling storage cluster branch level health assessment and early warning method

Abstract

The present application relates to liquid cooling storage cluster operation monitoring and fault early warning technical field, specifically discloses a kind of liquid cooling storage cluster branch level health assessment and early warning method.The method is for each independent liquid cooling branch in liquid cooling storage cluster, establishes the association between independent liquid cooling branch and corresponding storage node, collects liquid cooling operating parameter and storage operating parameter and carries out benchmark normalization processing;Based on the branch level correlation characteristics generated by normalized parameter sequence, the branch level health score, node level health score and cluster level health score are calculated;And according to the change state of branch level correlation characteristics in continuous statistical period, early warning judgment is carried out, and branch level early warning information is output when liquid cooling operating parameter and storage operating parameter do not exceed alarm threshold.The method can improve the identification ability of early-stage abnormality in branch level of liquid cooling storage cluster.

Description

Technical Field

[0001] This invention relates to the field of liquid-cooled storage cluster operation monitoring and fault early warning technology, and in particular to a method for branch-level health assessment and early warning of liquid-cooled storage clusters. Background Technology

[0002] With the rapid development of data centers, cloud storage platforms, and high-performance computing systems, the deployment density and operating load of storage clusters are constantly increasing, and storage nodes generate a large amount of heat during continuous operation. Liquid cooling, due to its high heat exchange efficiency, is increasingly being used in high-density storage servers, enterprise-level storage arrays, and distributed storage clusters to ensure the stable operation of storage devices under high load conditions.

[0003] In a liquid-cooled storage cluster, multiple storage nodes typically form a liquid-cooled circulation system via coolant piping, cold plates, manifolds, pump sets, and heat exchange units. The operating status of the liquid-cooled branches corresponding to different storage nodes, in terms of flow rate, pressure, and inlet / outlet temperature difference, affects the operating temperature and operational stability of the corresponding storage node or storage medium. Therefore, changes in the state of the liquid-cooled branches are closely related to the reliable operation of the storage cluster.

[0004] Existing liquid cooling storage systems typically employ monitoring methods at the system-wide, rack-level, or overall system level, primarily focusing on overall operating parameters such as total inlet temperature, total outlet temperature, total flow rate, and total pressure. While these methods can reflect the overall operating status of the liquid cooling system, they are insufficient to detect early branch-level anomalies due to the small scope of the abnormality and the fact that overall parameters may still be within the normal range.

[0005] Meanwhile, existing storage system health monitoring typically focuses on operational indicators such as the storage medium's own temperature, read / write performance, bit error rate, and number of bad blocks, while liquid cooling system monitoring focuses on the coolant circulation status. These two types of monitoring data are often judged independently, making it difficult to accurately reflect the impact of changes in the liquid cooling tributary status on the operational stability of the corresponding storage node, resulting in tributary-level degradation being difficult to identify in its early stages.

[0006] Furthermore, existing early warning methods mostly rely on parameter threshold triggers, meaning that alarms are only issued after temperature, flow rate, pressure, or storage operation indicators exceed preset thresholds. For liquid-cooled storage clusters, branch anomalies typically have the characteristics of slow accumulation and localized development. If alarms are only issued after parameters exceed limits, the lifespan of the corresponding storage media and data reliability may have already been adversely affected, making it difficult to meet the needs of preventative maintenance.

[0007] Therefore, there is an urgent need for a branch-level health assessment and early warning method suitable for liquid-cooled storage clusters to improve the ability to identify early anomalies in liquid-cooled branches and to support the stable operation and preventive maintenance of liquid-cooled storage clusters. Summary of the Invention

[0008] The purpose of this invention is to provide a method for branch-level health assessment and early warning of liquid-cooled storage clusters, which has the advantages of improving the ability to identify early anomalies in liquid-cooled branches and providing support for the stable operation and preventive maintenance of liquid-cooled storage clusters.

[0009] The above-mentioned technical objective of the present invention is achieved through the following technical solution: A method for branch-level health assessment and early warning of liquid-cooled storage clusters includes the following steps: S1. For each independent liquid cooling branch in the liquid cooling storage cluster, establish the association between the independent liquid cooling branch and the corresponding storage node, and synchronously collect the liquid cooling operation parameters of each independent liquid cooling branch and the storage operation parameters of the corresponding storage node. Perform benchmark normalization processing on the collected liquid cooling operation parameters and storage operation parameters to obtain the normalized parameter sequence corresponding to each independent liquid cooling branch. S2. Based on the normalized parameter sequence corresponding to each independent liquid cooling branch, generate branch-level correlation features. The branch-level correlation features include the cumulative correlation features between branch heat dissipation efficiency and medium bit error rate, the correlation features between branch pressure fluctuation and read / write latency jitter, the correlation features between branch temperature drift and medium temperature deviation, and the correlation features between branch flow attenuation and bad block growth rate. S3. Calculate the branch-level health score of each independent liquid cooling branch based on the normalized parameter sequence of each independent liquid cooling branch and the branch-level correlation characteristics, and generate the node-level health score of the corresponding storage node and the cluster-level health score of the liquid cooling storage cluster based on the branch-level health score. S4. Based on the changes in the branch-level correlation characteristics of each independent liquid cooling branch within a continuous statistical period, a warning judgment is made; when the branch-level correlation characteristics of the target independent liquid cooling branch meet the preset warning triggering conditions, and the liquid cooling operation parameters of the target independent liquid cooling branch and the storage operation parameters of the corresponding storage node do not exceed the preset alarm threshold, the branch-level warning information of the target independent liquid cooling branch is output.

[0010] Further, in step S1, the liquid cooling operating parameters include the branch coolant inlet temperature, the branch coolant outlet temperature, the branch real-time flow rate, the branch circulation pressure, and the coolant conductivity. The storage operating parameters include storage medium core temperature, read / write IOPS, read / write latency, bit error rate, number of bad blocks, and bad block growth rate.

[0011] Further, in step S1, the benchmark normalization process includes: Using the reference value of each independent liquid cooling branch as the normalization reference, the ratio of the real-time parameters collected at each sampling time of the same independent liquid cooling branch to the corresponding reference value is calculated to obtain the normalized parameters of the independent liquid cooling branch. The reference value is the design rating of the independent liquid cooling branch; if no design rating exists, the reference value is the initial reference value obtained by the independent liquid cooling branch during the initial stable operation phase.

[0012] Further, in step S2, the branch heat dissipation efficiency is calculated according to the following formula: ; in, Let be the branch heat dissipation efficiency at time t, c be the specific heat capacity of the coolant at constant pressure, and ρ be the density of the coolant. Let be the real-time flow of the branch at time t. Let be the outlet temperature of the branch coolant at time t. Let be the branch coolant inlet temperature at time t. Let t be the real-time operating power consumption of the storage node at time t.

[0013] Further, in step S2, the cumulative correlation feature between the branch heat dissipation efficiency and the medium bit error rate is denoted as M1, and its calculation method is as follows: ; in, Let be the branch heat dissipation efficiency at time t. This is the baseline heat dissipation efficiency for this independent liquid cooling branch. Let be the real-time bit error rate of the storage medium at time t. The reference bit error rate is the corresponding storage medium, and N is the number of sampling points in the statistical period.

[0014] Further, in step S2, the correlation feature between the branch pressure fluctuation and read / write latency jitter is denoted as M2, and its calculation method is as follows: ; in, This is the pressure fluctuation sequence of the target independent liquid cooling branch within the statistical period. This is a sequence of read / write latency jitter for the corresponding storage node within the statistical period. This is the function for calculating covariance.

[0015] Further, in step S2, the correlation feature between the branch temperature difference drift and the medium temperature deviation is denoted as M3, and its calculation method is as follows: ; in, Let be the temperature difference between the outlet and inlet temperatures of the coolant in the branch at time t. This serves as the reference temperature difference for the independent liquid cooling branch. Let t be the core temperature of the storage medium at time t. This is the reference core temperature for the corresponding storage medium.

[0016] Further, in step S2, the correlation feature between branch flow attenuation and bad block growth rate is denoted as M4, and its calculation method is as follows: ; in, The reference flow rate for the target independent liquid cooling branch, Let be the real-time flow rate of the target independent liquid-cooled branch at time t. This represents the bad block growth rate of the corresponding storage medium per unit time.

[0017] Further, in step S3, the branch-level health score is calculated according to the following formula: ; in, For branch-level health scores, This is the normalized value of the branch cooling efficiency. , , , These are the anomaly thresholds for the corresponding branch-level association features. , , , , The preset weighting coefficients satisfy the following conditions: ; When calculated When the value is less than 0, the branch-level health score is set to 0.

[0018] Further, in step S4, the preset warning trigger condition is: Within M consecutive statistical periods, any two or more of the branch-level correlation features of the target independent liquid cooling branch continuously exceed the corresponding preset warning thresholds, and the liquid cooling operation parameters of the target independent liquid cooling branch and the storage operation parameters of the corresponding storage node do not exceed the preset alarm thresholds; where M ranges from 3 to 10. In step S4, after outputting the branch-level early warning information of the target independent liquid cooling branch, the adjacent series branches and adjacent parallel branches corresponding to the target independent liquid cooling branch are determined based on the liquid cooling pipeline topology of the liquid cooling storage cluster, and cluster-level operation and maintenance suggestions are generated according to the early warning type of the target independent liquid cooling branch.

[0019] Compared with the prior art, the present invention has at least the following beneficial effects: This invention establishes a correlation between each independent liquid cooling branch in a liquid-cooled storage cluster and the storage nodes, and simultaneously collects liquid cooling operating parameters and the corresponding storage node operating parameters. This refines the health assessment from the overall system or cluster level down to the individual liquid cooling branch level. Consequently, the operating status of a single liquid cooling branch, such as flow rate, pressure, and temperature difference, can be independently assessed, avoiding the problem of relying solely on overall liquid cooling parameters and failing to promptly detect anomalies in local branches.

[0020] This invention performs benchmark normalization on the collected liquid cooling and storage operating parameters, thereby reducing parameter differences between different liquid cooling branches caused by variations in design rated flow rate, design rated pressure, benchmark temperature difference, or initial state of the storage medium. As a result, subsequent branch-level correlation feature generation and health score calculation can be performed based on a relatively uniform data scale, improving the comparability of health assessment results between different branches.

[0021] This invention generates branch-level correlation features based on normalized parameter sequences. These features include the cumulative correlation between branch heat dissipation efficiency and media bit error rate, the correlation between branch pressure fluctuation and read / write latency jitter, the correlation between branch temperature drift and media temperature deviation, and the correlation between branch flow rate attenuation and bad block growth rate. Therefore, it can reflect the correlation between changes in the state of liquid-cooled branches and changes in the operating state of corresponding storage nodes from different perspectives such as heat dissipation efficiency, pressure fluctuation, temperature drift, and flow rate attenuation. This avoids the problem of fragmented monitoring data between the liquid cooling system and the storage system, which makes it difficult to identify branch-level anomalies.

[0022] This invention calculates branch-level health scores based on normalized parameter sequences and branch-level correlation characteristics, and further generates node-level and cluster-level health scores. This allows health assessment results to be aggregated from individual liquid-cooled branches to the corresponding storage nodes and the entire liquid-cooled storage cluster. Therefore, it can both locate specific abnormal branches and reflect the health status of the corresponding storage nodes and the cluster as a whole, facilitating maintenance personnel to assess the operational risks of the liquid-cooled storage cluster from both local and overall perspectives.

[0023] This invention uses the changes in branch-level correlation characteristics over a continuous statistical period to make early warning judgments. When the liquid cooling operating parameters of the target independent liquid cooling branch and the storage operating parameters of the corresponding storage node do not exceed preset alarm thresholds, branch-level early warning information is output. Therefore, early abnormal trends at the branch level can be identified before conventional threshold alarms are triggered. This allows maintenance personnel to proactively check and handle suspected branch anomalies such as blockages, scaling, and micro-leakage, reducing the risk of branch anomalies further affecting the reliability of the storage medium.

[0024] After outputting branch-level early warning information, this invention can also analyze the degradation impact range of a target independent liquid-cooled branch based on the liquid-cooled piping topology of the liquid-cooled storage cluster, and generate cluster-level operation and maintenance suggestions. Therefore, after detecting an anomaly in a single branch, it is possible to simultaneously monitor adjacent series branches, adjacent parallel branches, or other branches on the same manifold path, reducing the impact of a single branch anomaly on the overall heat dissipation stability of the liquid-cooled storage cluster. Attached Figure Description

[0025] Figure 1 This is a flowchart illustrating the tributary-level health assessment and early warning method for liquid-cooled storage clusters according to the present invention. Figure 2 This is a schematic diagram of the associated structure between the liquid cooling branch and the storage node in an embodiment of the present invention. Detailed Implementation

[0026] The present invention will be further described in detail below with reference to the accompanying drawings.

[0027] Example: This embodiment provides a branch-level health assessment and early warning method for liquid-cooled storage clusters, applied to liquid-cooled storage clusters. The liquid-cooled storage cluster includes multiple storage nodes and independent liquid-cooling branches corresponding to each storage node. Each independent liquid-cooling branch is used to perform liquid cooling heat dissipation on the corresponding storage node or storage medium. The operating status of the independent liquid-cooling branch affects the temperature status, read / write stability, and storage medium reliability of the corresponding storage node. Therefore, before performing a health assessment, it is necessary to establish the correlation between each independent liquid-cooling branch and its corresponding storage node.

[0028] This embodiment uses a distributed liquid-cooled storage cluster deployed in a data center as the evaluation object. For example... Figure 2 As shown, this liquid-cooled storage cluster includes 8 storage nodes, each corresponding to one independent liquid-cooling branch. The 8 independent liquid-cooling branches are connected in parallel via manifolds to form a cluster liquid-cooling system. For ease of explanation, the 8 storage nodes can be referred to as the first storage node to the eighth storage node, and the corresponding 8 independent liquid-cooling branches can be referred to as the first liquid-cooling branch to the eighth liquid-cooling branch. The first liquid-cooling branch corresponds to the first storage node, the second liquid-cooling branch corresponds to the second storage node, and so on, with the eighth liquid-cooling branch corresponding to the eighth storage node.

[0029] In this embodiment, the rated flow rate of a single independent liquid cooling branch is 1.5 L / min, the rated pressure is 0.3 MPa, and the reference inlet and outlet temperature difference is 6°C. Each storage node includes 12 16TB NVMe SSDs, and the reference bit error rate of the storage medium is 1×10⁻⁶. -13 The bit error rate alarm threshold is 1×10 -12The core temperature alarm threshold for the storage medium is 70℃. The above-mentioned design rated flow rate, design rated pressure, reference inlet and outlet temperature difference, reference bit error rate, and alarm threshold serve as the basic parameters for subsequent parameter normalization processing, branch-level correlation feature generation, health score calculation, and branch-level early warning judgment.

[0030] With the above settings, before performing branch-level health assessment and early warning, this embodiment first clarifies the correspondence between each independent liquid-cooled branch and the corresponding storage node. This allows the subsequently collected liquid-cooled operating parameters, such as branch coolant inlet temperature, branch coolant outlet temperature, branch real-time flow rate, branch circulation pressure, and coolant conductivity, to be matched with the storage operating parameters of the corresponding storage node, such as storage medium core temperature, read / write IOPS, read / write latency, bit error rate, number of bad blocks, and bad block growth rate. This provides a data foundation for subsequent branch-level health assessment and early warning judgment.

[0031] Under the basic environment of the above embodiments, the method of the present invention sequentially performs branch-level parameter acquisition and normalization processing, branch-level correlation feature generation, health score calculation, and branch-level early warning judgment. Specifically, in step S1, parameters are acquired and normalized for each independent liquid-cooled branch and its corresponding storage node; in step S2, branch-level correlation features are generated based on the normalized parameter sequence; in step S3, branch-level health scores, node-level health scores, and cluster-level health scores are calculated based on the branch-level correlation features; in step S4, early warning judgment is made based on the change status of the branch-level correlation features within a continuous statistical period, and branch-level early warning information is output.

[0032] like Figure 1 As shown, in the basic environment of the above embodiment, step S1 is executed: for each independent liquid cooling branch in the liquid cooling storage cluster, establish the association relationship between each independent liquid cooling branch and the corresponding storage node, and synchronously collect the liquid cooling operation parameters of each independent liquid cooling branch and the storage operation parameters of the corresponding storage node.

[0033] Specifically, the first liquid cooling branch corresponds to the first storage node, the second liquid cooling branch corresponds to the second storage node, and so on, with the eighth liquid cooling branch corresponding to the eighth storage node. When the system collects parameters, it uses the independent liquid cooling branch as the basic data unit, matching and recording the liquid cooling operation parameters collected at the same sampling time with the storage operation parameters of the corresponding storage node, so that each independent liquid cooling branch forms a corresponding branch-level data record.

[0034] The preset acquisition frequency can be set to 10s-60s, and the statistical period can be set to 1 hour-24 hours. In this specific implementation, the parameter acquisition frequency is set to 30s / time, and the statistical period is set to 12 hours. That is, the system synchronously acquires data from each independent liquid-cooled branch and its corresponding storage node every 30s, and uses a 12-hour statistical period to perform subsequent normalization processing, branch-level correlation feature generation, and health score calculation on the acquired data.

[0035] In this embodiment, the liquid cooling operating parameters include the branch coolant inlet temperature, branch coolant outlet temperature, branch real-time flow rate, branch circulation pressure, and coolant conductivity. Specifically, the branch coolant inlet and outlet temperatures reflect the heat exchange status of the independent liquid cooling branch; the branch real-time flow rate reflects the coolant delivery status of the independent liquid cooling branch; the branch circulation pressure reflects the pressure fluctuation status of the independent liquid cooling branch; and the coolant conductivity helps determine whether the coolant status is abnormal.

[0036] Storage operating parameters include storage medium core temperature, read / write IOPS, read / write latency, bit error rate, number of bad blocks, and bad block growth rate. Specifically, storage medium core temperature reflects the thermal state of the corresponding storage medium; read / write IOPS, combined with read / write latency, determines the load status of the storage node; read / write latency reflects the read / write response stability of the storage node; bit error rate reflects the data reliability of the storage medium; number of bad blocks records basic data on storage medium degradation; and the bad block growth rate reflects the change in the number of bad blocks over time. These parameters are derived from the definitions of liquid cooling operating parameters and storage operating parameters in the documentation.

[0037] After data acquisition, the liquid cooling operating parameters and storage operating parameters are normalized to a benchmark. Specifically, the design rating of each independent liquid cooling branch is used as the normalization benchmark. The ratio of the real-time parameters collected at each sampling time for the same independent liquid cooling branch to the corresponding benchmark value is calculated to obtain the normalized parameters for that independent liquid cooling branch. For parameters without a design rating, the initial benchmark value obtained by the independent liquid cooling branch or the corresponding storage node during the initial stable operation phase is used as the normalization benchmark.

[0038] For example, for the real-time flow rate of a branch, the ratio of the real-time flow rate of the branch collected at time t to the design rated flow rate of the independent liquid-cooled branch can be calculated to obtain the normalized parameter of the real-time flow rate of the branch; for the circulation pressure of a branch, the ratio of the circulation pressure of the branch collected at time t to the design rated pressure of the independent liquid-cooled branch can be calculated to obtain the normalized parameter of the circulation pressure of the branch; for the bit error rate, the ratio of the real-time bit error rate collected at time t to the reference bit error rate of the corresponding storage medium can be calculated to obtain the normalized parameter of the bit error rate.

[0039] Through the above-mentioned benchmark normalization process, the parameter differences between different independent liquid cooling branches caused by differences in design rated flow, design rated pressure, benchmark temperature difference or initial state of storage medium can be weakened, so that the subsequent branch-level correlation feature generation and health score calculation can be carried out based on a relatively uniform data scale.

[0040] After completing the parameter acquisition and baseline normalization processing in step S1, step S2 is executed: based on the normalized parameter sequence corresponding to each independent liquid cooling branch, branch-level correlation features are generated. These branch-level correlation features reflect the correlation between the operating state changes of the independent liquid cooling branch and the corresponding storage node's operating state changes.

[0041] In this embodiment, the branch-level correlation features include the cumulative correlation features between branch heat dissipation efficiency and media bit error rate, the correlation features between branch pressure fluctuation and read / write latency jitter, the correlation features between branch temperature drift and media temperature deviation, and the correlation features between branch flow rate attenuation and bad block growth rate. All of these correlation features are generated based on the collected data of the same independent liquid-cooled branch and its corresponding storage node within the same statistical period.

[0042] First, calculate the heat dissipation efficiency of each independent liquid-cooled branch at the sampling time. The branch heat dissipation efficiency is calculated according to the following formula: ; in, Let be the branch heat dissipation efficiency at time t, c be the specific heat capacity of the coolant at constant pressure, and ρ be the density of the coolant. Let be the real-time flow of the branch at time t. Let be the outlet temperature of the branch coolant at time t. Let be the branch coolant inlet temperature at time t. Let t be the real-time operating power consumption of the storage node at time t.

[0043] After obtaining the branch heat dissipation efficiency, calculate the cumulative correlation feature M1 between the branch heat dissipation efficiency and the medium bit error rate. M1 is calculated according to the following formula: ; in, Let be the branch heat dissipation efficiency at time t. This is the baseline heat dissipation efficiency for this independent liquid cooling branch. Let be the real-time bit error rate of the storage medium at time t. The reference bit error rate is the corresponding storage medium, and N is the number of sampling points in the statistical period.

[0044] Then, the correlation characteristic M2 between branch pressure fluctuation and read / write latency jitter is calculated. M2 is calculated according to the following formula: ; in, This is the pressure fluctuation sequence of the target independent liquid cooling branch within the statistical period. This is a sequence of read / write latency jitter for the corresponding storage node within the statistical period. This is the function for calculating covariance.

[0045] Next, the correlation characteristic M3 between branch temperature difference drift and medium temperature deviation is calculated. M3 is calculated according to the following formula: ; in, Let be the temperature difference between the outlet and inlet temperatures of the coolant in the branch at time t. This serves as the reference temperature difference for the independent liquid cooling branch. Let t be the core temperature of the storage medium at time t. This is the reference core temperature for the corresponding storage medium.

[0046] Finally, the correlation characteristic M4 between branch flow attenuation and bad block growth rate is calculated. M4 is calculated according to the following formula: ; in, The reference flow rate for the target independent liquid cooling branch, Let be the real-time flow rate of the target independent liquid-cooled branch at time t. This represents the bad block growth rate of the corresponding storage medium per unit time.

[0047] Through the above processing, each independent liquid-cooled branch can form corresponding M1, M2, M3, and M4 within a statistical period. The above four branch-level correlation features characterize the correlation between the operating status changes of the independent liquid-cooled branch and the corresponding storage node operating status changes from four aspects: changes in heat dissipation efficiency and changes in bit error rate, pressure fluctuations and read / write latency jitter, inlet / outlet temperature difference drift and core temperature deviation of storage medium, and flow rate decay and bad block growth rate changes. They serve as the data basis for subsequent health score calculations and branch-level early warning judgments.

[0048] After generating branch-level association features in step S2, in order to facilitate subsequent health score calculation and branch-level early warning judgment, this embodiment sets corresponding threshold parameters for branch-level association features and sets weight coefficients for each evaluation item in the branch-level health score.

[0049] In this embodiment, the threshold parameters include a warning threshold, an alarm threshold, and an anomaly threshold. The warning threshold is used in step S4 to determine whether the target independent liquid cooling branch meets the branch-level warning triggering condition; the alarm threshold is used to determine whether the liquid cooling operating parameters or storage operating parameters have reached a normal alarm state; and the anomaly threshold is used in the normalization calculation of the branch-level health scoring formula in step S3. As a specific implementation, the anomaly threshold in the branch-level health scoring formula can be the alarm threshold corresponding to the branch-level associated characteristics.

[0050] In this embodiment, the warning thresholds and alarm thresholds for branch-level associated features M1, M2, M3, and M4 are configured as follows: The aforementioned warning thresholds are used to identify early abnormal changes in branch-level correlation features before reaching a normal alarm state; the aforementioned alarm thresholds are used to determine whether the corresponding branch-level correlation features have entered a high degree of abnormality. For liquid cooling operating parameters and storage operating parameters, the bit error rate alarm threshold of 1×10 in the basic environment of the aforementioned embodiments can also be used. -12 Routine alarm judgments are made based on parameters such as the core temperature alarm threshold of 70℃ for the storage medium and the design rated flow rate and design rated pressure of the liquid cooling branch.

[0051] In the branch-level health rating, the weighting coefficients are set as follows in this embodiment: =0.2、 =0.3、 =0.2、 =0.15、 =0.15. Wherein, The weight of the normalization deviation of the corresponding branch heat dissipation efficiency. The weights corresponding to M1, The weights corresponding to M2, The weight corresponding to M3, The corresponding weights for M4. The above weight coefficients satisfy: ; By configuring the thresholds and weights as described above, the subsequent branch-level health score calculation can simultaneously consider the branch heat dissipation efficiency deviation and the influence of the four branch-level correlation features M1, M2, M3, and M4. Furthermore, it enables branch-level early warning judgment to distinguish between early abnormal states and regular alarm states based on early warning thresholds and alarm thresholds. The aforementioned threshold and weight parameters are derived from the specific implementation configuration in the disclosure document.

[0052] After completing the branch-level association feature generation and the aforementioned threshold and weight configuration in step S2, step S3 is executed: based on the normalized parameter sequence and branch-level association features of each independent liquid cooling branch, the branch-level health score of each independent liquid cooling branch is calculated, and the node-level health score of the corresponding storage node and the cluster-level health score of the liquid cooling storage cluster are further generated.

[0053] Specifically, for any target independent liquid-cooled branch, the system calculates the branch-level health score based on the normalized value of the branch's heat dissipation efficiency, M1, M2, M3, and M4. The branch-level health score is calculated according to the following formula: ; in, For branch-level health scores, This is the normalized value of the branch cooling efficiency. , , , These are the anomaly thresholds for the corresponding branch-level association features. , , , , These are preset weighting coefficients.

[0054] When calculated When the value is less than 0, the branch-level health score of the target's independent liquid-cooled branch is set to 0. This avoids negative scores when the branch-level correlation features are highly abnormal, keeping the score results within a range that can be used to express the health status.

[0055] After obtaining the branch-level health scores for each independent liquid-cooled branch, the system generates a node-level health score based on the correspondence between the independent liquid-cooled branches and the storage nodes. In this embodiment, each storage node corresponds to one independent liquid-cooled branch; therefore, the node-level health score of the target storage node can be determined based on the branch-level health scores of its corresponding independent liquid-cooled branch. In other embodiments, when a storage node corresponds to multiple independent liquid-cooled branches, the node-level health score can be a weighted average of the branch-level health scores of the multiple independent liquid-cooled branches corresponding to that storage node. The weighting coefficients can be determined according to the capacity ratio or access load ratio of the storage media corresponding to each branch.

[0056] Furthermore, the system generates a cluster-level health score based on the node-level health scores of each storage node within the liquid-cooled storage cluster. This cluster-level health score can be a weighted average of the node-level health scores of each storage node, with the weighting coefficients determined according to the storage capacity percentage or business importance level of each storage node. In this way, the health score results can be aggregated level by level from a single independent liquid-cooled branch to the corresponding storage node and the entire liquid-cooled storage cluster.

[0057] For example, after the cluster is running stably, taking the third liquid-cooled branch as an example, its baseline heat dissipation efficiency is 0.92. Within a 12-hour statistical period, the system calculates four branch-level correlation characteristics of the third liquid-cooled branch based on the collected data: M1 = 0.2 × 10 -13 M2=42, M3=1.2℃, M4=0.01. All four branch-level correlation characteristics are below their corresponding warning thresholds, indicating that the third liquid-cooled branch did not show any obvious early abnormal trend during this statistical period.

[0058] Furthermore, the system substitutes the normalized value of the branch heat dissipation efficiency of the third liquid-cooled branch, M1, M2, M3, M4, and their corresponding weighting coefficients into the branch-level health scoring formula to obtain the branch-level health score of the third liquid-cooled branch. Since the third storage node corresponds to the third liquid-cooled branch in this embodiment, the node-level health score of the third storage node can be determined based on the branch-level health score of the third liquid-cooled branch. Combining this with the node-level health scores of other storage nodes, the cluster-level health score of the liquid-cooled storage cluster can be obtained.

[0059] Through the above processing, the system can generate health status evaluation results at three levels: branch level, node level, and cluster level. Specifically, the branch-level health score is used to locate the operating status of a specific independent liquid-cooled branch, the node-level health score is used to characterize the local health status of the corresponding storage node, and the cluster-level health score is used to characterize the overall health status of the entire liquid-cooled storage cluster, thus providing an evaluation basis for subsequent branch-level early warning judgments and operational maintenance.

[0060] After completing the health score calculation in step S3, step S4 is executed: Early warning judgment is made based on the changes in the branch-level correlation characteristics of each independent liquid cooling branch within a continuous statistical period. When the branch-level correlation characteristics of the target independent liquid cooling branch meet the preset early warning triggering conditions, and the liquid cooling operation parameters of the target independent liquid cooling branch and the storage operation parameters of the corresponding storage node do not exceed the preset alarm threshold, the branch-level early warning information of the target independent liquid cooling branch is output.

[0061] Specifically, the system uses the statistical period as the judgment unit to continuously track M1, M2, M3, and M4 of each independent liquid-cooled branch. Within the continuous statistical period, if the branch-level correlation characteristics of an independent liquid-cooled branch show a continuous increase, a continuous deviation from the baseline value, or a continuous exceedance of the warning threshold, then this change is used as the basis for branch-level warning judgment.

[0062] In this embodiment, the preset early warning trigger condition is: within M consecutive statistical periods, any two or more of the branch-level correlation features of the target independent liquid cooling branch continuously exceed the corresponding preset early warning threshold, and the liquid cooling operation parameters of the target independent liquid cooling branch and the storage operation parameters of the corresponding storage node do not exceed the preset alarm threshold. The value of M can range from 3 to 10; as a specific implementation, M is set to 5 in this embodiment.

[0063] When a target independent liquid-cooled branch meets the aforementioned preset early warning trigger conditions, the system outputs branch-level early warning information. This branch-level early warning information may include the target independent liquid-cooled branch number, branch-level correlation characteristics exceeding a preset early warning threshold, corresponding statistical period, abnormal change trend, and preventative maintenance recommendations. The preventative maintenance recommendations can be generated based on the branch-level correlation characteristic types exceeding the preset early warning threshold. For example, when the correlation characteristic M4 between branch flow attenuation and bad block growth rate continuously increases, it prompts an inspection of the branch filter, pipe blockage, or scaling; when the correlation characteristic M2 between branch pressure fluctuation and read / write latency jitter continuously increases, it prompts an inspection of the branch pressure stability and the corresponding storage node load status.

[0064] For example, on the 38th day of cluster operation, the system detected a continuous abnormal trend in the branch-level correlation characteristics of the fifth liquid-cooled branch. Over five consecutive statistical periods, the M1 of the fifth liquid-cooled branch increased from 0.3 × 10⁻⁶. -13 It continued to rise to 0.7×10 -13 M4 continuously increased from 0.03 to 0.06, while both M1 and M4 exceeded their corresponding warning thresholds. Simultaneously, the real-time flow rate of the fifth liquid cooling branch slowly decreased from 1.5 L / min to 1.42 L / min, corresponding to a slow increase in the core temperature of the storage medium from 42°C to 48°C. However, the liquid cooling operating parameters of the fifth liquid cooling branch and the storage operating parameters of the corresponding storage node did not exceed the preset alarm thresholds.

[0065] Under the above circumstances, the system determines that the fifth liquid cooling branch meets the branch-level early warning conditions and outputs branch-level early warning information. The branch-level early warning information may be: the fifth liquid cooling branch shows an abnormal trend in the correlation between flow attenuation and bit error rate, suspected of branch scaling or slight blockage, and it is recommended to check the branch filter and pipeline status and perform preventive maintenance.

[0066] Maintenance personnel inspected the fifth liquid-cooled branch based on the branch-level early warning information and found that the inlet filter of the fifth liquid-cooled branch was slightly clogged, causing the branch flow to decrease slowly. After cleaning the filter, the liquid-cooling operating parameters and branch-level correlation characteristics of the fifth liquid-cooled branch gradually returned to normal.

[0067] Through the above branch-level early warning judgment process, the system can identify the early abnormal trend of the target independent liquid cooling branch based on the continuous change status of the branch-level correlation characteristics when the liquid cooling operating parameters and storage operating parameters have not yet reached the conventional alarm threshold, and output preventive maintenance suggestions to the operation and maintenance personnel.

[0068] After outputting branch-level early warning information for the target independent liquid cooling branch, the system can also analyze the scope of the degradation impact of the target independent liquid cooling branch based on the liquid cooling pipeline topology of the liquid cooling storage cluster, and generate cluster-level operation and maintenance suggestions.

[0069] Specifically, the system determines the position of the target independent liquid cooling branch in the liquid cooling pipeline topology based on the connection relationship between each independent liquid cooling branch and manifold, coolant distribution unit, and return pipeline in the liquid cooling storage cluster, and identifies other liquid cooling branches that are on the same distribution path, adjacent parallel path, or adjacent series path as potentially affected branches.

[0070] When a target independent liquid cooling branch triggers a branch-level warning, the system determines its potential propagational impact based on the warning type. For example, if a target independent liquid cooling branch has a suspected micro-leakage risk, it may cause a drop in the total manifold pressure, thereby affecting the flow stability of other parallel liquid cooling branches; if a target independent liquid cooling branch has a suspected blockage or scaling risk, it may cause a decrease in the flow rate of that branch and change the flow distribution of other parallel branches in the liquid cooling system; if a target independent liquid cooling branch has abnormal pressure fluctuations, it may cause pressure disturbances to other branches on the same liquid cooling distribution path.

[0071] After identifying potentially affected branches, the system generates cluster-level maintenance recommendations. These recommendations include inspecting and maintaining the target independent liquid-cooled branch, continuously monitoring the flow rate, pressure, inlet and outlet temperature difference of the potentially affected branch, and the temperature changes of the corresponding storage nodes. If necessary, the system can also prompt maintenance personnel to adjust the output pressure or flow distribution strategy of the coolant distribution unit.

[0072] For example, during cluster operation, a micro-leakage warning is triggered in the second liquid cooling branch. Based on the parallel topology of the cluster's liquid cooling pipelines, the system determines that the micro-leakage in the second liquid cooling branch may cause a drop in the total manifold pressure, further affecting the flow stability of the remaining parallel liquid cooling branches. Based on this, the system generates cluster-level operation and maintenance recommendations: The second liquid cooling branch has a micro-leakage risk; it is recommended to address the fault in this branch while continuously monitoring the flow changes of the remaining liquid cooling branches and adjusting the output pressure of the coolant distribution unit as needed.

[0073] Based on the above cluster-level maintenance recommendations, the maintenance personnel inspected and addressed the second liquid cooling branch, and simultaneously monitored the flow and pressure changes of other parallel liquid cooling branches, thereby reducing the impact of a single liquid cooling branch malfunction on the overall heat dissipation stability of the liquid-cooled storage cluster.

[0074] Through the above branch degradation propagation analysis, the system can not only locate and alert the target independent liquid-cooled branch that triggers the warning, but also judge its potential impact on the operation status of adjacent liquid-cooled branches or the overall cluster by combining the liquid-cooled pipeline topology, so that the branch-level warning results can further serve the cluster-level operation and maintenance decision-making.

[0075] Through the above implementation process, this embodiment first establishes the association between each independent liquid cooling branch and the corresponding storage node in the liquid cooling storage cluster, and simultaneously collects liquid cooling operation parameters and storage operation parameters; then, it performs normalization processing on the collected data and generates branch-level association features; subsequently, it calculates branch-level health scores, node-level health scores, and cluster-level health scores based on the branch-level association features; finally, it performs branch-level early warning judgment based on the change status of branch-level association features within a continuous statistical period, and generates cluster-level operation and maintenance suggestions in combination with the liquid cooling pipeline topology.

[0076] Therefore, this embodiment can realize the health status assessment from a single independent liquid cooling branch to the corresponding storage node and then to the entire cluster in the liquid-cooled storage cluster. It can also provide early abnormal trends of the target independent liquid cooling branch when the liquid cooling operation parameters and storage operation parameters have not yet reached the conventional alarm threshold, providing a basis for preventive operation and maintenance of the liquid-cooled storage cluster.

[0077] The embodiments described above do not constitute a limitation on the scope of protection of this technical solution. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the above embodiments should be included within the scope of protection of this technical solution.

Claims

1. A method for branch-level health assessment and early warning of liquid-cooled storage clusters, characterized in that, Includes the following steps: S1. For each independent liquid cooling branch in the liquid cooling storage cluster, establish the association between the independent liquid cooling branch and the corresponding storage node, and synchronously collect the liquid cooling operation parameters of each independent liquid cooling branch and the storage operation parameters of the corresponding storage node. Perform benchmark normalization processing on the collected liquid cooling operation parameters and storage operation parameters to obtain the normalized parameter sequence corresponding to each independent liquid cooling branch. S2. Based on the normalized parameter sequence corresponding to each independent liquid cooling branch, generate branch-level correlation features. The branch-level correlation features include the cumulative correlation features between branch heat dissipation efficiency and medium bit error rate, the correlation features between branch pressure fluctuation and read / write latency jitter, the correlation features between branch temperature drift and medium temperature deviation, and the correlation features between branch flow attenuation and bad block growth rate. S3. Calculate the branch-level health score of each independent liquid cooling branch based on the normalized parameter sequence of each independent liquid cooling branch and the branch-level correlation characteristics, and generate the node-level health score of the corresponding storage node and the cluster-level health score of the liquid cooling storage cluster based on the branch-level health score. S4. Make early warning judgments based on the changes in the branch-level correlation characteristics of each independent liquid cooling branch within a continuous statistical period; When the branch-level association characteristics of the target independent liquid cooling branch meet the preset early warning triggering conditions, and the liquid cooling operation parameters of the target independent liquid cooling branch and the storage operation parameters of the corresponding storage node do not exceed the preset alarm threshold, the branch-level early warning information of the target independent liquid cooling branch is output. In step S2, the branch heat dissipation efficiency is calculated according to the following formula: ； in, Let be the branch heat dissipation efficiency at time t, c be the specific heat capacity of the coolant at constant pressure, and ρ be the density of the coolant. Let be the real-time flow of the branch at time t. Let be the branch coolant outlet temperature at time t. Let be the branch coolant inlet temperature at time t. Let t be the real-time operating power consumption of the storage node at time t; In step S2, the cumulative correlation feature between the branch heat dissipation efficiency and the medium bit error rate is denoted as M1, and its calculation method is as follows: ； in, Let be the branch heat dissipation efficiency at time t. This is the baseline heat dissipation efficiency for this independent liquid cooling branch. Let be the real-time bit error rate of the storage medium at time t. The reference bit error rate is the corresponding storage medium, and N is the number of sampling points in the statistical period; In step S2, the correlation feature between the branch pressure fluctuation and read / write latency jitter is denoted as M2, and its calculation method is as follows: ； in, This is the pressure fluctuation sequence of the target independent liquid cooling branch within the statistical period. This is a sequence of read / write latency jitter for the corresponding storage node within the statistical period. The function for calculating covariance; In step S2, the correlation feature between the branch temperature difference drift and the medium temperature deviation is denoted as M3, and its calculation method is as follows: ； in, Let be the temperature difference between the outlet and inlet temperatures of the coolant in the branch at time t. This serves as the reference temperature difference for the independent liquid cooling branch. Let t be the core temperature of the storage medium at time t. This is the reference core temperature for the corresponding storage medium; In step S2, the correlation feature between branch flow attenuation and bad block growth rate is denoted as M4, and its calculation method is as follows: ； in, The reference flow rate for the target independent liquid cooling branch, Let be the real-time flow rate of the target independent liquid-cooled branch at time t. This represents the bad block growth rate of the corresponding storage medium per unit time. In step S3, the branch-level health score is calculated according to the following formula: ； in, For branch-level health scores, This is the normalized value of the branch cooling efficiency. , , , These are the anomaly thresholds for the corresponding branch-level association features. , , , , The preset weighting coefficients satisfy the following conditions: ； When calculated When the value is less than 0, the branch-level health score is set to 0.

2. The method for branch-level health assessment and early warning of liquid-cooled storage clusters according to claim 1, characterized in that, In step S1, the liquid cooling operating parameters include branch coolant inlet temperature, branch coolant outlet temperature, branch real-time flow rate, branch circulation pressure, and coolant conductivity. The storage operating parameters include storage medium core temperature, read / write IOPS, read / write latency, bit error rate, number of bad blocks, and bad block growth rate.

3. The method for branch-level health assessment and early warning of liquid-cooled storage clusters according to claim 1, characterized in that, In step S1, the benchmark normalization process includes: Using the reference value of each independent liquid cooling branch as the normalization reference, the ratio of the real-time parameters collected at each sampling time of the same independent liquid cooling branch to the corresponding reference value is calculated to obtain the normalized parameters of the independent liquid cooling branch. The reference value is the design rating of the independent liquid cooling branch; if no design rating exists, the reference value is the initial reference value obtained by the independent liquid cooling branch during the initial stable operation phase.

4. The method for branch-level health assessment and early warning of liquid-cooled storage clusters according to claim 1, characterized in that, In step S4, the preset warning trigger condition is: Within M consecutive statistical periods, any two or more of the branch-level correlation features of the target independent liquid cooling branch continuously exceed the corresponding preset warning threshold, and the liquid cooling operation parameters of the target independent liquid cooling branch and the storage operation parameters of the corresponding storage node do not exceed the preset alarm threshold; wherein, the value of M ranges from 3 to 10. In step S4, after outputting the branch-level early warning information of the target independent liquid cooling branch, the adjacent series branches and adjacent parallel branches corresponding to the target independent liquid cooling branch are determined based on the liquid cooling pipeline topology of the liquid cooling storage cluster, and cluster-level operation and maintenance suggestions are generated according to the early warning type of the target independent liquid cooling branch.

Images