A steady-state performance test regulation method for a liquid-cooled storage system

Abstract

The present application relates to liquid cooling heat dissipation state regulation and storage performance test technical field, disclose a kind of steady-state performance test regulation method of liquid cooling storage system, this method in the process of liquid cooling storage system performance test, synchronous acquisition liquid cooling system operating parameter and storage system performance parameter, lock test steady-state window;Based on the heat dissipation carrying capacity index of liquid cooling branch operating parameter, storage unit core temperature and test load heat contribution parameter calculation;Accordingly determine the target test load level of next test steady-state window, and complete load adjustment in transition interval;Meanwhile generate performance benchmark compensation coefficient, carry out compensation processing to storage performance data, and according to heat dissipation carrying capacity index, steady-state effectiveness is graded.This method can improve the authenticity, steady-state consistency and comparability of liquid cooling storage system performance test data.

Description

Technical Field

[0001] This invention relates to the field of liquid cooling heat dissipation state control and storage performance testing technology, and particularly to a method for steady-state performance testing and control of a liquid cooling storage system. Background Technology

[0002] With the widespread application of high-density storage devices in data centers, edge computing, and high-performance computing scenarios, the heat generated by storage systems during high-concurrency read / write operations is constantly increasing. Liquid cooling, due to its high heat exchange efficiency and good temperature control capabilities, is increasingly being used in server storage arrays, distributed storage nodes, and other high-density storage systems. For liquid-cooled storage systems, performance test results typically need to reflect metrics such as throughput, IOPS, and read / write latency under stable operating conditions. Therefore, the stability of the load and the stability of the heat dissipation during testing directly affect the authenticity and comparability of the test data.

[0003] Existing storage performance testing methods often employ fixed loads or preset load sequences, typically failing to adequately account for fluctuations in flow rate, pressure, inlet temperature, and heat exchange status that may occur in the liquid cooling system during testing. When liquid cooling capacity decreases, a fixed high-intensity test load may cause the storage cell temperature to rise or even trigger frequency reduction protection, distorting performance data. Conversely, when cooling capacity is sufficient, a fixed low-intensity test load may fail to fully reflect the storage system's performance ceiling. While some solutions can adjust the device load based on cooling parameters, frequent changes in read / write loads during performance testing can disrupt the original test steady state, causing performance data collected at different times to correspond to different test conditions, thereby reducing the comparability of test results.

[0004] Furthermore, existing methods for evaluating heat dissipation typically focus on the temperature, flow rate, or pressure parameters of the liquid cooling system itself, rarely considering the contribution of the current storage read / write load to heat generation. This makes it difficult to accurately characterize the actual load-bearing capacity of the liquid cooling system under the current test load. Simultaneously, existing test results often lack grading labels for the steady-state validity of test data, making it difficult for testers to distinguish between data that can be used for final performance conclusions, data that can only be used as a reference, and heat-limited data that should be excluded. Therefore, there is an urgent need for a test control method that can balance adaptability to liquid cooling heat fluctuations with the steady-state consistency of storage performance testing. Summary of the Invention

[0005] The purpose of this invention is to provide a method for controlling the steady-state performance of a liquid-cooled storage system. This method has the advantage of maintaining the steady-state consistency of storage performance testing by adjusting the test load in a closed loop based on operating variables such as liquid cooling inlet temperature, liquid cooling branch flow rate, liquid cooling branch circulation pressure, and storage unit core temperature, thereby preventing fluctuations in liquid cooling heat dissipation.

[0006] The above-mentioned technical objective of the present invention is achieved through the following technical solution: A method for testing and controlling the steady-state performance of a liquid-cooled storage system includes the following steps: S1. During the performance test of the liquid cooling storage system, the operating parameters of the liquid cooling system and the performance parameters of the storage system are collected simultaneously. Based on the operating parameters of the liquid cooling system and the performance parameters of the storage system, it is determined whether the operating state during the test meets the preset steady-state judgment condition. The time interval that continuously meets the preset steady-state judgment condition is locked as the test steady-state window. S2. For the test steady-state window, obtain the liquid cooling branch operating parameters, storage unit core temperature and thermal contribution parameters corresponding to the current test load of the storage node under test, and calculate the heat dissipation capacity index corresponding to the test steady-state window based on the liquid cooling branch operating parameters, the storage unit core temperature and the thermal contribution parameters. S3. Based on the heat dissipation load capacity index corresponding to the current test steady-state window, determine the target test load level corresponding to the next test steady-state window, and adjust the current test load level to the target test load level within the transition interval between the current test steady-state window and the next test steady-state window. S4. Based on the difference between the current test load level and the target test load level, generate a performance benchmark compensation coefficient, and perform compensation processing on the storage performance data collected in the next test steady-state window based on the performance benchmark compensation coefficient. S5. Label the storage performance data collected within each test steady-state window with the corresponding heat dissipation capacity index, test load level and performance benchmark compensation coefficient, and classify the steady-state effectiveness of the storage performance data according to the heat dissipation capacity index.

[0007] Further configuration: In step S1, the operating parameters of the liquid cooling system include at least two of the following: liquid cooling inlet temperature, liquid cooling branch flow rate, and liquid cooling branch circulation pressure; The storage system performance parameters include at least two of the following: storage read / write load, IOPS, read / write latency, and storage cell core temperature. The synchronous acquisition refers to the corresponding acquisition of the operating parameters of the liquid cooling system and the performance parameters of the storage system according to the same sampling period or the sampling period aligned with the timestamp.

[0008] By adopting the above technical solution, and by limiting the operating parameters of the liquid cooling system, the performance parameters of the storage system, and the synchronous acquisition method, the source of test data is made clearer. This ensures that the liquid cooling status data and the storage performance data correspond consistently in the time dimension, avoiding the inaccurate correlation between heat dissipation status and performance due to inconsistent sampling times. This improves the reliability of subsequent steady-state judgment and heat dissipation capacity calculation.

[0009] Further settings: In step S1, the preset steady-state determination conditions include: Within N consecutive sampling periods, the fluctuation range of the read / write load of the storage system does not exceed the preset load fluctuation threshold, and the fluctuation range of the inlet temperature of the liquid cooling system does not exceed the preset temperature fluctuation threshold. When the above conditions are met simultaneously, the time interval corresponding to the N consecutive sampling periods is locked as the test steady-state window; The sampling period is 1s-10s, N is 30-100, the preset load fluctuation threshold is no greater than 5%, and the preset temperature fluctuation threshold is no greater than 2℃.

[0010] By adopting the above technical solution, by limiting the fluctuation range of read / write load and liquid cooling inlet temperature within N consecutive sampling periods, a clear test steady-state window judgment standard is formed. This makes performance testing no longer dependent on single instantaneous data or human experience judgment, but based on the stability within a continuous time interval to lock the window, which is beneficial for filtering out non-steady-state test data with large load fluctuations or inlet temperature disturbances.

[0011] Further setting: In step S2, the thermal contribution parameter corresponding to the current test load is the thermal contribution coefficient. The thermal contribution coefficient Calculated based on the read ratio, write ratio, rated power consumption under pure read load, rated power consumption under pure write load, and maximum rated power consumption of the storage unit in the current test load; The thermal contribution coefficient The formula for calculation is: ; in, This represents the rated power consumption of the storage unit under a pure write load. This represents the write ratio in the current test load. This represents the rated power consumption of the storage unit under a pure read load. This represents the read ratio in the current test load. This represents the maximum rated power consumption of the storage unit.

[0012] By adopting the above technical solution and introducing the thermal contribution coefficient of the current test load, the read / write ratio, pure read power consumption, pure write power consumption and maximum rated power consumption are included in the calculation. This makes the heat dissipation evaluation no longer based solely on the parameters of the liquid cooling system itself, but can reflect the actual impact of different read / write loads on the heat generation of the storage unit, thereby improving the matching between the heat dissipation capacity evaluation and the storage performance test scenario.

[0013] Further settings: In step S2, the heat dissipation load capacity index Φ is calculated as follows: ; in, The core temperature limit is preset for the storage unit. To test the average core temperature of the memory cells within the steady-state window, This represents the real-time coolant flow rate of the liquid cooling branch corresponding to the tested storage node. The design rated flow rate of this liquid cooling branch is... This is the real-time circulating pressure of the liquid cooling branch. The design rated pressure of this liquid cooling branch is... The thermal contribution coefficient of the current test load is α, β, γ, and δ, which are preset weighting coefficients, and α+β+γ+δ=1.

[0014] By adopting the above technical solution and constructing a heat dissipation capacity index, the core temperature margin, liquid cooling branch flow rate, liquid cooling branch circulation pressure, and test load heat contribution coefficient are comprehensively calculated. This allows for the evaluation of the liquid cooling system's capacity to withstand the current test load from multiple dimensions, including storage temperature status, liquid cooling branch cooling status, circulation drive status, and load heat intensity.

[0015] Further settings: The rules for determining the preset weighting coefficients are as follows: The values ​​of α range from 0.4 to 0.6, β range from 0.15 to 0.25, γ range from 0.1 to 0.2, and δ range from 0.05 to 0.15. Wherein, α is used to characterize the influence weight of the core temperature margin of the storage unit on the heat dissipation capacity index, β is used to characterize the influence weight of the liquid cooling branch flow rate on the heat dissipation capacity index, γ is used to characterize the influence weight of the liquid cooling branch circulation pressure on the heat dissipation capacity index, and δ is used to characterize the influence weight of the current test load thermal contribution on the heat dissipation capacity index.

[0016] By adopting the above technical solution and limiting the value range of each weight coefficient and its corresponding meaning, the influencing factors in the heat dissipation capacity index have a clear contribution relationship. This not only highlights the main impact of the core temperature margin of the storage unit on the test safety and stability, but also takes into account the contribution factors of flow rate, pressure and load heat, thereby improving the interpretability and feasibility of the index calculation process.

[0017] Further settings: In step S3, the test load is pre-divided into multiple consecutive test load levels, and each test load level corresponds to a set of read and write strength parameter combinations; The read / write intensity parameter combination includes read / write ratio, data block size, queue depth, and number of concurrent threads; A mapping relationship is established between the numerical range of the heat dissipation load capacity index and the test load level, and the target test load level corresponding to the next test steady state window is determined based on the numerical range of the heat dissipation load capacity index corresponding to the current test steady state window.

[0018] By adopting the above technical solution, the test load is divided into multiple continuous levels, and a mapping relationship is established between the heat dissipation capacity index range and the test load level. This transforms the load adjustment from a vague manual adjustment to a graded control based on the capacity index. This is beneficial for selecting a matching test load under different heat dissipation conditions, avoiding maintaining an excessively high load when the heat dissipation capacity is insufficient or having an excessively low test load when the heat dissipation capacity is sufficient.

[0019] Further settings: In step S3, when the change in the heat dissipation load capacity index corresponding to two consecutive test steady-state windows exceeds the preset fluctuation threshold, the test load level adjustment is triggered. During the test load level adjustment process, the read and write strength parameter combination corresponding to the current test load level is smoothly adjusted to the read and write strength parameter combination corresponding to the target test load level; Within a single adjustment cycle, the parameter adjustment range shall not exceed 20% of the difference between the corresponding parameters between the current test load level and the target test load level.

[0020] By adopting the above technical solution, and by setting the trigger condition for the change in heat dissipation capacity index and the limit on the adjustment range of single-cycle parameters, the load adjustment is triggered only when there is a substantial change in heat dissipation capacity, and the load level switching is completed in a smooth manner. This reduces the damage to the test steady state caused by frequent and sudden adjustments, and ensures a smoother transition process between adjacent test steady state windows.

[0021] Further setting: In step S4, the performance benchmark compensation coefficient Calculate as follows: ; in, Test load level for target Rated IOPS of the lower storage unit This represents the rated IOPS of the storage unit under the current test load level L0. The rated read / write latency of the storage unit is tested under the target test load level Lx. This represents the rated read / write latency of the storage unit under the current test load level L0. The compensation process includes: normalizing the IOPS and read / write latency collected in the next test steady-state window based on the performance benchmark compensation coefficient.

[0022] By adopting the above technical solution, setting a performance benchmark compensation coefficient, and normalizing the rated IOPS and rated read / write latency under different load levels, the impact of load level changes on the horizontal comparison of performance data can be reduced, making the performance data collected under different test load levels comparable, thereby improving the fairness and consistency of the final test conclusions.

[0023] Further configuration: In step S5, the steady-state effectiveness classification includes: When the heat dissipation capacity index Φ≥0.8, the storage performance data within the corresponding steady-state test window will be marked as the steady-state valid core data; When 0.6≤Φ<0.8, the storage performance data within the corresponding steady-state test window will be marked as the valid steady-state reference data; When Φ < 0.6, the storage performance data within the corresponding steady-state test window will be marked as unsteady-state heat-limited data. The final performance test conclusion is generated based on the steady-state effective core data and the steady-state effective reference data, and the non-steady-state heat dissipation-limited data is not included in the calculation of the final performance test conclusion.

[0024] By adopting the above technical solution, and by classifying and labeling the storage performance data within the test steady-state window according to the heat dissipation capacity index, the test data is divided into steady-state effective core data, steady-state effective reference data, and non-steady-state heat-limited data. This clarifies which data can be used for the final performance conclusion, which data is only for reference, and which data should be excluded, thereby improving the credibility of the test report conclusions and the accuracy of data management.

[0025] Compared with the prior art, the present invention has the following beneficial effects: This invention simultaneously collects the operating parameters of the liquid cooling system and the performance parameters of the storage system during the performance testing of the liquid cooling storage system, and locks the time interval that continuously meets the preset steady-state judgment conditions as the test steady-state window. This ensures that the performance data collection is established within a relatively stable test interval, avoiding the direct use of data with large load fluctuations or large fluctuations in liquid cooling inlet temperature as performance test results, thereby improving the steady-state consistency and authenticity of the liquid cooling storage system performance test data.

[0026] This invention targets the steady-state test window and calculates the heat dissipation capacity index by combining the operating parameters of the liquid cooling branch corresponding to the tested storage node, the core temperature of the storage unit, and the thermal contribution parameters corresponding to the current test load. This makes the evaluation of heat dissipation capacity no longer rely solely on the temperature, flow rate, or pressure parameters of the liquid cooling system itself, but further considers the impact of the current read / write load on the heat generation of the storage unit. This allows for a more accurate characterization of the actual load-bearing capacity of the liquid cooling system for the current storage test load, and improves the matching degree between the heat dissipation status evaluation and the storage performance test scenario.

[0027] This invention determines the target test load level for the next test steady-state window based on the heat dissipation capacity index corresponding to the current test steady-state window, and completes the test load adjustment within the transition interval between adjacent test steady-state windows. This allows the load adjustment to avoid the already locked test steady-state window, reducing the interference of load changes on the steady-state performance data acquisition process. Therefore, this invention can adapt to changes in liquid cooling heat dissipation conditions while reducing the disruption of performance test steady-state conditions caused by frequent or abrupt adjustments.

[0028] This invention generates a performance benchmark compensation coefficient based on the difference between the current test load level and the target test load level, and performs compensation processing on the storage performance data collected in the next test steady-state window based on the performance benchmark compensation coefficient, so that the performance data obtained under different test load levels have a unified comparison basis, reduce the performance data deviation caused by changes in test load level, and thus improve the comparability and evaluation fairness of the liquid-cooled storage system performance test results.

[0029] This invention simultaneously labels the storage performance data collected within each test steady-state window with a heat dissipation capacity index, test load level, and performance benchmark compensation coefficient. It also classifies the steady-state validity based on the heat dissipation capacity index, enabling the test report to distinguish between steady-state valid core data, steady-state valid reference data, and non-steady-state heat-limited data. This avoids including distorted data under heat-limited or unstable conditions in the final performance conclusion, thereby improving the data management accuracy and reliability of the test results. Attached Figure Description

[0030] Figure 1 This is a schematic diagram of the overall process of a steady-state performance testing and control method for a liquid-cooled storage system according to the present invention; Figure 2 This is a schematic diagram of the liquid cooling branch and parameter acquisition location of the liquid cooling storage system in an embodiment of the present invention. Detailed Implementation

[0031] The present invention will be further described below with reference to embodiments. It should be understood that the following embodiments are only used to explain the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention. Without departing from the technical concept of the present invention, any parameter adjustments, equivalent substitutions of steps, or adaptive adjustments to the execution method made by those skilled in the art based on the type of liquid-cooled storage system, the specifications of the storage medium, the structure of the liquid-cooled branch, and the performance testing standards should fall within the scope of protection of the present invention.

[0032] Example: This embodiment provides a method for steady-state performance testing and control of a liquid-cooled storage system. This method is applied to the performance benchmark testing process of liquid-cooled storage systems, and is particularly suitable for liquid-cooled server storage arrays, liquid-cooled distributed storage nodes, liquid-cooled edge storage devices, and other information storage systems that adopt liquid-cooled heat dissipation structures. During the performance testing process, the liquid-cooled storage system applies read and write loads to the storage units using a test load generation tool, and collects performance or operating parameters such as IOPS, throughput, read and write latency, error rate, and core temperature of the storage units under different load conditions. Simultaneously, the liquid-cooling system provides heat dissipation capacity to the storage units through coolant circulation, and can provide corresponding heat dissipation to different storage nodes through liquid-cooled branches.

[0033] In this embodiment, the steady-state performance testing and control method for the liquid-cooled storage system can be executed by a test control platform. The test control platform can be test management software deployed on a test server, an independent test control device connected to the liquid-cooled storage system, or a functional module integrated into the liquid-cooled storage system management platform. The test control platform is communicatively connected to the liquid-cooled system sensors, storage performance testing tools, load generation module, and test data recording module, respectively, for collecting the operating parameters of the liquid-cooled system and the performance parameters of the storage system, and for performing steady-state window locking, heat dissipation capacity index calculation, test load level adjustment, performance benchmark compensation, and steady-state validity grading of test data based on the collected data.

[0034] Specifically, the operating parameters of the liquid cooling system include at least one or more of the following: liquid cooling inlet temperature, liquid cooling branch flow rate, and liquid cooling branch circulation pressure. The liquid cooling inlet temperature reflects the temperature state of the coolant entering the tested liquid cooling branch or liquid cooling distribution unit; the liquid cooling branch flow rate reflects the coolant supply capacity of the liquid cooling branch corresponding to the tested storage node; and the liquid cooling branch circulation pressure reflects the coolant circulation drive state within the liquid cooling branch.

[0035] The storage system performance parameters include at least one or more of the following: storage read / write load, IOPS, read / write latency, and storage cell core temperature. The storage read / write load characterizes the load intensity and read / write structure applied to the storage system during the current test; the IOPS characterizes the storage system's ability to complete input / output operations per unit time; the read / write latency characterizes the latency level of the storage system in responding to read / write requests; and the storage cell core temperature characterizes the thermal state of the storage cell under the current test load and current liquid cooling conditions.

[0036] The method in this embodiment generally includes the following execution flow: like Figure 1 As shown, the method in this embodiment generally includes the following steps: S1. Steady-State Window Determination and Locking: During the performance testing of the liquid-cooled storage system, the operating parameters of the liquid-cooling system and the performance parameters of the storage system are collected simultaneously. Based on preset steady-state determination conditions, it is determined whether the operating state during the test meets the steady-state requirements. When the fluctuation range of the storage read / write load and the fluctuation range of the liquid-cooling inlet temperature within a continuous time interval both meet the preset steady-state determination conditions, that continuous time interval is locked as the test steady-state window. This step ensures that the data used for subsequent performance evaluation comes from a relatively stable test interval, avoiding the direct inclusion of data significantly affected by load fluctuations or heat dissipation disturbances in the performance test results.

[0037] S2. Calculation of Heat Dissipation Capacity Index: For the test steady-state window locked in step S1, the test control platform acquires the operating parameters of the liquid cooling branch corresponding to the tested storage node, the core temperature of the storage unit, and the thermal contribution parameters corresponding to the current test load. Based on these parameters, the platform calculates the heat dissipation capacity index corresponding to the test steady-state window. This index characterizes the actual heat dissipation capacity of the current liquid cooling system under the current storage test load. It is not solely determined by the liquid cooling system's own temperature, flow rate, or pressure parameters, but also by the impact of the current read / write load on the heat generated by the storage unit.

[0038] S3. Determination and Adjustment of Target Test Load Level: Based on the heat dissipation capacity index corresponding to the current test steady-state window, determine the target test load level corresponding to the next test steady-state window. When the heat dissipation capacity is high, a higher test load level can be matched; when the heat dissipation capacity is low, a lower test load level can be matched. To avoid the test load adjustment disrupting the locked test steady-state window, this embodiment arranges the adjustment of the test load level within the transition range between the current test steady-state window and the next test steady-state window, thus distinguishing the load adjustment process from the steady-state performance data acquisition process.

[0039] S4. Performance Benchmark Compensation Processing: Based on the difference between the current test load level and the target test load level, a performance benchmark compensation coefficient is generated, and the storage performance data collected in the next test steady-state window is compensated based on the performance benchmark compensation coefficient. Through this step, performance data collected under different test load levels can obtain a unified comparison basis, reducing the problem of horizontal incomparability of performance data caused by changes in test load levels.

[0040] S5. Steady-State Validity Classification of Test Data: Storage performance data collected within each test steady-state window is simultaneously labeled with the corresponding heat dissipation capacity index, test load level, and performance benchmark compensation coefficient. The steady-state validity of the storage performance data is then classified according to the heat dissipation capacity index. This allows for the differentiation of test data into steady-state valid data used for the final performance conclusions, data used only as a reference, and heat-limited data not included in the final conclusions, thereby improving the authenticity, steady-state consistency, and comparability of liquid-cooled storage system performance test results.

[0041] Through the above steps S1 to S5, this embodiment forms a complete test control chain consisting of test steady-state window locking, heat dissipation capacity index calculation, target test load level determination, performance benchmark compensation, and steady-state effectiveness grading. This test control chain can maintain the match between the test load and the liquid cooling heat dissipation capacity when the liquid cooling heat dissipation state fluctuates, while reducing the interference of load adjustment on the steady-state performance data acquisition process, thus making it suitable for standardized performance benchmark testing of liquid-cooled storage systems.

[0042] This embodiment uses a liquid-cooled server storage array as the test object to illustrate the steady-state performance testing and control method of the liquid-cooled storage system. The test object can be a rack-mounted liquid-cooled server storage array, a liquid-cooled distributed storage node, or a liquid-cooled edge storage device. In this embodiment, the test object is a 2U rack-mounted liquid-cooled server storage array, which includes multiple storage units. The storage units can be U.2 NVMe SSDs, enterprise-grade solid-state drives, liquid-cooled storage modules, or other storage devices suitable for liquid cooling. As a specific example, the liquid-cooled server storage array includes 16 U.2 NVMe SSD storage units.

[0043] like Figure 2 As shown, the liquid-cooled server storage array adopts a manifold-type parallel liquid-cooling branch structure. Each storage node under test is equipped with an independent liquid-cooling branch. The coolant enters each liquid-cooling branch through the liquid-cooling distribution structure and exchanges heat with the corresponding storage node through liquid-cooling plates, cooling channels, or heat exchange contact structures. By associating the storage node under test with the corresponding liquid-cooling branch, branch-level liquid-cooling operating parameters can be obtained during performance testing, rather than just the overall system-level liquid-cooling operating parameters. This facilitates subsequent evaluation of the heat dissipation capacity of a single storage node or a local storage area.

[0044] In this embodiment, the liquid cooling branch has a preset design rated flow rate and design rated pressure. As a specific example, the design rated flow rate of the liquid cooling branch corresponding to the tested storage node is: ; The design rated pressure of the liquid cooling branch corresponding to the tested storage node is: ; in, This indicates the rated coolant flow rate of the liquid-cooled branch under design conditions. This indicates the rated circulating pressure of the liquid cooling branch under design conditions. The above-mentioned rated flow rate and rated pressure are used as a normalization benchmark in the subsequent calculation of the heat dissipation capacity index.

[0045] In this embodiment, the storage unit has a preset core temperature limit, rated power consumption for pure write load, rated power consumption for pure read load, and maximum rated power consumption.

[0046] As a specific example, the upper limit of the core temperature of the storage unit is: , The rated power consumption of the storage unit under pure write load is: , The rated power consumption of the storage unit under pure read load is: , The maximum rated power consumption of the storage unit is: , in, This indicates the upper limit of the core temperature allowed for the storage unit. This indicates the rated power consumption of the storage cell under pure write load conditions. This indicates the rated power consumption of the storage cell under read-only load conditions. This indicates the maximum rated power consumption of the storage unit. These parameters can be obtained from the storage unit's product specification sheet, test standard documents, or pre-calibrated results.

[0047] Before performance testing, a standard test environment is set up, and the test environment parameters are preset. As a specific example, in this embodiment, the ambient temperature is controlled at 25℃±2℃, the relative humidity is controlled at 60%±10%, and the coolant is an insulating electronic fluorinated liquid. The test control platform is communicatively connected to temperature sensors, flow sensors, pressure sensors, storage performance testing tools, and load generation modules to synchronously collect the operating parameters of the liquid cooling system and the performance parameters of the storage system during the test.

[0048] To ensure the correspondence of the collected data over time, this embodiment sets a uniform parameter acquisition cycle. As a specific example, the parameter acquisition cycle is set to 5 seconds per acquisition, meaning that the liquid cooling system operating parameters and storage system performance parameters are collected every 5 seconds. The collected liquid cooling system operating parameters include at least the liquid cooling inlet temperature, the real-time coolant flow rate of the liquid cooling branch corresponding to the tested storage node, and the real-time circulation pressure; the collected storage system performance parameters include at least the current read / write load, IOPS, read / write latency, and storage cell core temperature. For data from different sampling sources, the test control platform can synchronize the data using a unified sampling cycle or timestamp alignment.

[0049] In this embodiment, the determination of the test steady-state window is based on the parameter fluctuations over multiple consecutive sampling periods. As a specific example, 60 consecutive sampling periods are taken as a steady-state window determination period. Since a single sampling period is 5 seconds, the time length corresponding to 60 consecutive sampling periods is 5 minutes. If, within this continuous time interval, the fluctuation amplitude of the storage read / write load does not exceed a preset load fluctuation threshold, and the fluctuation amplitude of the liquid cooling inlet temperature does not exceed a preset temperature fluctuation threshold, then this continuous time interval is locked as the test steady-state window.

[0050] As a specific example, the preset load fluctuation threshold is set to 3%, and the preset temperature fluctuation threshold is set to 1℃. That is, when the storage read / write load fluctuation amplitude is no greater than 3% and the liquid cooling inlet temperature fluctuation amplitude is no greater than 1℃ within 5 consecutive minutes, the test control platform determines that the continuous time interval meets the test steady-state requirements and uses it as the test steady-state window; if either of the above conditions is not met, the continuous time interval is not used as the test steady-state window, or it is only recorded as non-steady-state process data.

[0051] In this embodiment, a preset fluctuation threshold for the heat dissipation capacity index is also included to determine whether to trigger test load level adjustment. As a specific example, the heat dissipation capacity index fluctuation threshold is set to 0.15. When the difference between the heat dissipation capacity index corresponding to two consecutive test steady-state windows exceeds 0.15, the test control platform determines that the liquid cooling heat dissipation load state has changed and triggers subsequent test load level adjustment; when the difference between the heat dissipation capacity index corresponding to two consecutive test steady-state windows does not exceed 0.15, the test control platform can maintain the current test load level to avoid frequent adjustments due to slight fluctuations.

[0052] In this embodiment, the calculation of the heat dissipation load capacity index involves multiple influencing factors. In order to reflect the contribution of different factors to the heat dissipation load capacity evaluation results, weight coefficients are pre-configured.

[0053] As a specific example, the weighting coefficient corresponding to the core temperature margin is: , The weighting coefficient corresponding to the flow rate of the liquid cooling branch is: , The weighting coefficient corresponding to the circulating pressure of the liquid cooling branch is: , The weighting coefficient corresponding to the thermal contribution of the current test load is: , The above weighting coefficients satisfy: .

[0054] Wherein, α is used to characterize the influence weight of the core temperature margin of the storage unit on the heat dissipation capacity index, β is used to characterize the influence weight of the liquid cooling branch flow rate on the heat dissipation capacity index, γ is used to characterize the influence weight of the liquid cooling branch circulation pressure on the heat dissipation capacity index, and δ is used to characterize the influence weight of the current test load thermal contribution on the heat dissipation capacity index.

[0055] This embodiment also pre-defines multiple test load levels. Each test load level corresponds to a set of read / write intensity parameter combinations, including read / write ratio, data block size, queue depth, and number of concurrent threads. As a specific example, the test load is divided into five consecutive levels, L1 to L5, where L1 is an extremely low load level, L2 is a low load level, L3 is a medium load level, L4 is a high load level, and L5 is an extremely high load level. The higher the load level, the higher the corresponding read / write intensity, and the greater the heat generated by the storage unit.

[0056] Specifically, in this embodiment, the L1 load level corresponds to a read / write ratio of 100 / 0, a data block size of 4KB, a queue depth of 1, and a concurrent thread count of 1; the L2 load level corresponds to a read / write ratio of 70 / 30, a data block size of 8KB, a queue depth of 4, and a concurrent thread count of 2; the L3 load level corresponds to a read / write ratio of 50 / 50, a data block size of 16KB, a queue depth of 8, and a concurrent thread count of 4; the L4 load level corresponds to a read / write ratio of 30 / 70, a data block size of 32KB, a queue depth of 16, and a concurrent thread count of 8; and the L5 load level corresponds to a read / write ratio of 0 / 100, a data block size of 64KB, a queue depth of 32, and a concurrent thread count of 16.

[0057] Accordingly, each test load level can also be pre-configured with rated IOPS and rated read / write latency for subsequent calculation of performance benchmark compensation coefficients. As a specific example, the L1 load level corresponds to a rated IOPS of 120,000 and a rated read / write latency of 80μs; ​​the L2 load level corresponds to a rated IOPS of 350,000 and a rated read / write latency of 150μs; the L3 load level corresponds to a rated IOPS of 600,000 and a rated read / write latency of 220μs; the L4 load level corresponds to a rated IOPS of 950,000 and a rated read / write latency of 280μs; ​​and the L5 load level corresponds to a rated IOPS of 1,300,000 and a rated read / write latency of 350μs.

[0058] In this embodiment, a mapping relationship between the heat dissipation capacity index and the test load level is also pre-established. As a specific example, when the heat dissipation capacity index satisfies: At that time, it corresponds to the L5 load level; when the heat dissipation capacity index meets: At that time, it corresponds to the L4 load level; when the heat dissipation capacity index meets: At that time, it corresponds to the L3 load level; when the heat dissipation capacity index meets: At that time, it corresponds to the L2 load level; when the heat dissipation capacity index meets: At that time, it corresponds to the L1 load level.

[0059] Through the above-described test objects, test environment, and basic parameter settings, this embodiment provides a parameter basis for subsequent execution of steady-state window determination, heat dissipation capacity index calculation, test load level adjustment, performance benchmark compensation, and steady-state validity classification of test data. The specific values ​​described above are merely illustrative of the execution method of this embodiment and do not constitute a limitation on the scope of protection of this invention. In other embodiments, the number of storage units, rated parameters of the liquid cooling branch, sampling period, steady-state determination threshold, weighting coefficient, number of test load levels, and their corresponding parameters can all be adaptively adjusted according to the liquid-cooled storage system structure, storage medium specifications, and performance testing standards.

[0060] In this embodiment, the determination and locking process of the test steady-state window corresponds to step S1. After the performance test of the liquid-cooled storage system begins, the test control platform synchronously collects the operating parameters of the liquid-cooled system and the performance parameters of the storage system according to a preset sampling period, and determines whether the test process meets the preset steady-state determination conditions based on the parameter fluctuations within multiple consecutive sampling periods. When the fluctuation range of the storage read / write load and the fluctuation range of the liquid-cooled inlet temperature are both within the allowable range within a continuous time interval, the test control platform locks that continuous time interval as the test steady-state window.

[0061] Specifically, the test control platform acquires the operating parameters of the liquid cooling system and the performance parameters of the storage system within each sampling period. The operating parameters of the liquid cooling system include at least the liquid cooling inlet temperature; the performance parameters of the storage system include at least the storage read / write load, and may also include parameters such as IOPS, read / write latency, throughput, error rate, and storage cell core temperature. For data from different sensors or different testing tools, the test control platform can use a unified sampling period for synchronous acquisition, or it can perform data alignment based on timestamps after acquisition to ensure that the liquid cooling status data and storage performance data at the same sampling moment have a corresponding relationship.

[0062] In this embodiment, the test control platform uses N consecutive sampling periods as a candidate judgment interval. For any candidate judgment interval, the test control platform determines whether the fluctuation amplitude of the storage read / write load within the interval exceeds a preset load fluctuation threshold, and whether the fluctuation amplitude of the liquid cooling inlet temperature within the interval exceeds a preset temperature fluctuation threshold. If the fluctuation amplitude of the storage read / write load does not exceed the preset load fluctuation threshold, and the fluctuation amplitude of the liquid cooling inlet temperature does not exceed the preset temperature fluctuation threshold, then the candidate judgment interval is determined to meet the preset steady-state judgment condition, and the candidate judgment interval is locked as the test steady-state window; if any fluctuation amplitude exceeds the corresponding threshold, then the candidate judgment interval is determined not to meet the preset steady-state judgment condition, and the data collected within the interval is not used as data within the test steady-state window, or is only recorded as non-steady-state process data.

[0063] In one implementation, the fluctuation range of storage read / write load can be determined based on the maximum, minimum, and average values ​​of storage read / write load within a candidate decision interval. For example, the difference between the maximum and minimum values ​​of storage read / write load within the candidate decision interval, relative to the average value of storage read / write load within that interval, can be used as the fluctuation range of storage read / write load. The storage read / write load can be characterized by the number of read / write requests, read / write bandwidth, IOPS, queue depth, number of concurrent threads, or overall load intensity. In this embodiment, overall load intensity or IOPS is preferably used as the calculation object for load fluctuation range.

[0064] In one implementation, the fluctuation range of the liquid cooling inlet temperature can be determined based on the maximum and minimum values ​​of the liquid cooling inlet temperature within the candidate determination interval. For example, the difference between the maximum and minimum values ​​of the liquid cooling inlet temperature within the candidate determination interval can be used as the fluctuation range of the liquid cooling inlet temperature. In this way, the stability of the liquid cooling inlet temperature within the candidate determination interval can be directly reflected.

[0065] In this embodiment, the preset steady-state determination condition can be specifically set as follows: within N consecutive sampling periods, the fluctuation amplitude of the storage read / write load does not exceed a preset load fluctuation threshold, and the fluctuation amplitude of the liquid cooling inlet temperature does not exceed a preset temperature fluctuation threshold. When both of the above conditions are met simultaneously, the test control platform determines the corresponding candidate determination interval as the test steady-state window.

[0066] As a specific example, with a sampling period of 5 seconds and N set to 60, the length of the candidate decision interval corresponding to 60 consecutive sampling periods is 5 minutes. The preset load fluctuation threshold is set to 3%, and the preset temperature fluctuation threshold is set to 1°C. Within a certain candidate decision interval, if the test control platform determines that the storage read / write load fluctuation is 2.4% and the liquid cooling inlet temperature fluctuation is 0.8°C, then since neither exceeds the corresponding threshold, the test control platform locks this consecutive 5-minute time interval as the test steady-state window. Conversely, if the storage read / write load fluctuation is 6% or the liquid cooling inlet temperature fluctuation is 1.5°C within the candidate decision interval, the test control platform determines that the candidate decision interval does not meet the test steady-state requirements and does not use it as the test steady-state window.

[0067] To avoid data overlap between test steady-state windows or misjudgment of test steady-state windows due to short-term disturbances, in one implementation, the test control platform can use a sliding window method to determine the continuously sampled data. Specifically, the test control platform uses N consecutive sampling periods as the initial candidate determination interval; when the candidate determination interval does not meet the preset steady-state determination conditions, the test control platform moves the starting point of the window backward by at least one sampling period and re-forms a new candidate determination interval; when the candidate determination interval meets the preset steady-state determination conditions, the test control platform locks the interval as the test steady-state window and associates and stores the liquid cooling system operating parameters and storage system performance parameters collected within the window.

[0068] In another implementation, to further improve the reliability of the test steady-state window, the test control platform can, in addition to meeting the conditions of storage read / write load fluctuation and liquid cooling inlet temperature fluctuation, also assess the changing trend of the storage cell core temperature, the fluctuation of the liquid cooling branch flow rate, or the fluctuation of the liquid cooling branch circulation pressure. For example, when the storage cell core temperature continues to rise rapidly within the candidate judgment interval, or when there are significant abnormal fluctuations in the liquid cooling branch flow rate or liquid cooling branch circulation pressure, even if the read / write load and inlet temperature meet the basic threshold conditions, the test control platform can still mark the candidate judgment interval as an interval to be confirmed, rather than directly using it as the test steady-state window. This additional judgment method can avoid interference from local anomalies in the liquid cooling branch or the heat accumulation process of the storage cell on the steady-state window determination.

[0069] Once the test steady-state window is locked, the test control platform generates a corresponding window identifier for that window and records data such as the start time, end time, window duration, average read / write load, average liquid cooling inlet temperature, average liquid cooling branch flow rate, average liquid cooling branch circulation pressure, average memory cell core temperature, average IOPS, and average read / write latency. This data is used for two purposes: firstly, to calculate the thermal load capacity index, and secondly, for subsequent performance benchmark compensation and steady-state effectiveness grading.

[0070] Through the above-described steady-state window determination and locking process, this embodiment can select a continuous test interval that simultaneously satisfies load stability and liquid cooling inlet temperature stability during the performance testing of the liquid cooling storage system. This allows subsequent heat dissipation capacity calculations and performance data evaluations to be based on stable test conditions, reducing the impact of instantaneous fluctuations, sudden load changes, or liquid cooling inlet temperature disturbances on the performance test results.

[0071] In this embodiment, the calculation process of the thermal contribution coefficient and the heat dissipation capacity index corresponds to step S2. After locking the test steady-state window, the test control platform calculates the heat dissipation capacity index corresponding to the test steady-state window based on the liquid cooling branch operating parameters, storage unit core temperature, and read / write structure corresponding to the current test load collected within the test steady-state window. The heat dissipation capacity index is used to characterize the actual heat dissipation capacity of the liquid cooling system to the storage test load under the current test state, and serves as the basis for subsequently determining the target test load level.

[0072] Specifically, after the test steady-state window is locked, the test control platform acquires the real-time coolant flow rate, real-time circulating pressure, and core temperature of the liquid cooling branch corresponding to the tested storage node within the test steady-state window, and also acquires the read and write ratios in the current test load. Since different read and write loads have different effects on the power consumption and heat generation of the storage unit, this embodiment first calculates the thermal contribution coefficient of the current test load, and then uses this thermal contribution coefficient together with the operating status of the liquid cooling branch and the temperature status of the storage unit to calculate the heat dissipation capacity index.

[0073] The thermal contribution coefficient is denoted as This is used to characterize the contribution of the current read / write load structure to the heat generation of the storage unit. A higher write ratio in the current test load generally corresponds to a greater contribution of power consumption and heat generation to the storage unit; a higher read ratio in the current test load allows the corresponding heat generation contribution to be calculated based on the rated power consumption of the pure read load. Therefore, this embodiment calculates the heat contribution coefficient of the current test load based on the rated power consumption of the pure write load, the current write ratio, the rated power consumption of the pure read load, the current read ratio, and the maximum rated power consumption of the storage unit.

[0074] The thermal contribution coefficient The calculation formula is: ; in, This represents the rated power consumption of the storage unit under a pure write load. This represents the write ratio in the current test load. This represents the rated power consumption of the storage unit under a pure read load. This represents the read ratio in the current test load. This represents the maximum rated power consumption of the storage unit. Through the above calculations, the read / write structure of the current test load is converted into thermal contribution parameters that can be used to evaluate the heat dissipation capacity, thereby avoiding the problem of ignoring the heat dissipation differences of the storage load while only evaluating the heat dissipation status based on the parameters of the liquid cooling system itself.

[0075] After obtaining the thermal contribution coefficient, the test control platform further calculates the heat dissipation capacity index corresponding to the test steady-state window. The heat dissipation capacity index is denoted as... It comprehensively considers the core temperature margin of the storage unit, the real-time coolant flow rate of the liquid cooling branch, the real-time circulation pressure of the liquid cooling branch, and the thermal contribution coefficient of the current test load.

[0076] The heat dissipation capacity index The calculation formula is: ; in, The core temperature limit is preset for the storage unit. To test the average core temperature of the memory cells within the steady-state window, This represents the real-time coolant flow rate of the liquid cooling branch corresponding to the tested storage node. The design rated flow rate of this liquid cooling branch is... This is the real-time circulating pressure of the liquid cooling branch. The design rated pressure of this liquid cooling branch is... The thermal contribution coefficient of the current test load is α, β, γ, and δ, which are preset weighting coefficients, and α+β+γ+δ=1.

[0077] In the above calculation of the heat dissipation capacity index This value represents the margin of the storage cell's core temperature relative to its upper temperature limit. The larger the value, the further the storage cell is from the upper temperature limit, and the higher the thermal safety margin under the current test conditions; the smaller the value, the closer the storage cell's core temperature is to the upper temperature limit, and the current heat dissipation load is approaching the limit.

[0078] This value is used to characterize how well the actual coolant flow rate of the liquid cooling branch corresponding to the tested storage node meets the design rated flow rate. The closer this value is to or higher than 1, the closer the coolant supply capacity of the liquid cooling branch is to the design state; when this value is significantly lower than 1, it indicates that there may be insufficient flow rate or a decrease in localized coolant supply capacity in the liquid cooling branch.

[0079] This value is used to characterize the degree to which the real-time circulating pressure of the liquid cooling branch corresponding to the tested storage node meets the design rated pressure. It reflects the coolant circulation drive status within the liquid cooling branch and helps determine whether the liquid cooling branch is operating under normal circulation conditions.

[0080] This is used to characterize the deduction effect of the current test load's thermal contribution on the heat dissipation capacity index. In other words, with other liquid cooling operating parameters remaining the same, the higher the thermal contribution coefficient of the current test load, the greater the heat generated by the storage unit, and the higher the actual load pressure on the liquid cooling system for that test load, thus reducing the heat dissipation capacity index accordingly.

[0081] In one specific implementation, α is used to characterize the influence weight of the core temperature margin of the storage cell on the heat dissipation capacity index, β is used to characterize the influence weight of the liquid cooling branch flow rate on the heat dissipation capacity index, γ is used to characterize the influence weight of the liquid cooling branch circulation pressure on the heat dissipation capacity index, and δ is used to characterize the influence weight of the current test load thermal contribution on the heat dissipation capacity index.

[0082] As a specific example, α=0.5, β=0.2, γ=0.15, δ=0.15. The above weighting coefficients ensure that the core temperature margin of the storage cell plays a major role in the evaluation of heat dissipation capacity, while also taking into account the impact of liquid cooling branch flow rate, liquid cooling branch circulation pressure, and the thermal contribution of the test load on heat dissipation capacity.

[0083] In specific calculations, if data from multiple sampling periods exists within the test steady-state window, the test control platform can determine the corresponding statistical values ​​based on the data collected within that test steady-state window. For example, the core temperature of the memory cell can be the average core temperature within the test steady-state window; the real-time coolant flow rate and real-time circulation pressure of the liquid cooling branch can be the average value, the representative value of the stable segment, or the statistical value after outlier removal within the test steady-state window. This process reduces the impact of single-point sampling noise on the calculation results of the heat dissipation capacity index.

[0084] As a specific example, within a certain steady-state test window, the current test load is L3 level, the read / write ratio is 50 / 50, the rated power consumption of the storage unit under pure write load is 12W, the rated power consumption of the storage unit under pure read load is 5W, and the maximum rated power consumption of the storage unit is 12W. Then, the thermal contribution coefficient of the current test load is: ; Within this test steady-state window, if the average core temperature of the storage cell is 42℃, the preset upper limit of the core temperature of the storage cell is 75℃, the real-time coolant flow rate of the liquid cooling branch corresponding to the tested storage node is 1.18L / min, the design rated flow rate of the liquid cooling branch is 1.2L / min, the real-time circulation pressure is 0.34MPa, the design rated pressure is 0.35MPa, and the weighting coefficients are α=0.5, β=0.2, γ=0.15, and δ=0.15, then the heat dissipation capacity index corresponding to this test steady-state window is: ; This heat dissipation capacity index indicates that, within the current test steady-state window, the liquid cooling system possesses the corresponding heat dissipation capacity to handle the current test load. The test control platform can then determine the target test load level for the next test steady-state window based on the numerical range of this heat dissipation capacity index.

[0085] In this embodiment, the process of dividing the test load level and determining the target test load level corresponds to the first half of step S3. After obtaining the heat dissipation capacity index corresponding to the current test steady-state window, the test control platform determines the target test load level corresponding to the next test steady-state window based on the pre-established mapping relationship between the heat dissipation capacity index and the test load level. This process is used to match the test load intensity with the current heat dissipation capacity of the liquid cooling system, avoiding the continued maintenance of a high-intensity test load when the heat dissipation capacity is insufficient, and also avoiding an excessively low test load when the heat dissipation capacity is sufficient, thus failing to fully reflect the performance of the storage system.

[0086] Specifically, in this embodiment, the test load is pre-divided into multiple consecutive test load levels, and each test load level corresponds to a set of read / write intensity parameter combinations. These read / write intensity parameter combinations include read / write ratio, data block size, queue depth, and number of concurrent threads. A higher load level indicates a higher corresponding test load intensity, which typically leads to higher power consumption and heat generation in the storage unit; a lower load level indicates a lower corresponding test load intensity, suitable for test phases where liquid cooling capacity is relatively insufficient or where thermal stress needs to be reduced.

[0087] As a specific example, this embodiment divides the test load into five consecutive levels, L1 to L5. L1 is the extremely low load level, L2 is the low load level, L3 is the medium load level, L4 is the high load level, and L5 is the extremely high load level. The parameters corresponding to each load level can be set as follows: In the above load levels, the read / write ratio represents the proportion of read and write operations in the current test load; the block size represents the amount of data in a single read / write operation; the queue depth represents the number of input / output requests that the storage system can process simultaneously; and the number of concurrent threads represents the number of execution threads that simultaneously initiate read / write requests. Defining the test load level using these multiple parameters avoids relying solely on a single read / write ratio or a single IOPS metric to classify load intensity, allowing the load level to more accurately reflect the actual load conditions in storage performance testing.

[0088] In this embodiment, the test control platform pre-establishes a mapping relationship between the heat dissipation capacity index and the test load level. As a specific example, when the heat dissipation capacity index satisfies Φ≥0.8, it corresponds to the L5 load level; when it satisfies 0.65≤Φ<0.8, it corresponds to the L4 load level; when it satisfies 0.5≤Φ<0.65, it corresponds to the L3 load level; when it satisfies 0.35≤Φ<0.5, it corresponds to the L2 load level; and when it satisfies Φ<0.35, it corresponds to the L1 load level.

[0089] During testing, the test control platform determines the numerical range of the heat dissipation capacity index calculated based on the current steady-state test window, and uses the test load level corresponding to this range as the target test load level for the next steady-state test window. For example, if the heat dissipation capacity index calculated for the current steady-state test window is 0.721, then this value falls within the range of 0.65 ≤ Φ < 0.8, corresponding to a target test load level of L4; if the calculated heat dissipation capacity index is 0.512, then this value falls within the range of 0.5 ≤ Φ < 0.65, corresponding to a target test load level of L3.

[0090] Through the above-described process of classifying test load levels and determining target test load levels, this embodiment can convert the continuously changing heat dissipation capacity index into discrete, executable test load levels, so that subsequent test load adjustments have a clear object and parameter basis, thereby improving the stability and feasibility of the liquid cooling storage system performance testing and control process.

[0091] In this embodiment, the load smoothing adjustment process between adjacent test steady-state windows corresponds to the latter half of step S3. After determining the target test load level corresponding to the next test steady-state window, the test control platform does not directly change the test load within the currently locked test steady-state window. Instead, it performs load adjustment within the transition interval between the current test steady-state window and the next test steady-state window. This method avoids load changes directly interfering with the performance data acquisition within the current steady-state window, thus distinguishing the test load adjustment process from the steady-state performance data acquisition process.

[0092] Specifically, after each steady-state test window ends, the test control platform acquires the heat dissipation load capacity index corresponding to that window and compares it with the heat dissipation load capacity index corresponding to the previous steady-state test window. When the change in the heat dissipation load capacity index corresponding to two consecutive steady-state test windows does not exceed the preset fluctuation threshold, it indicates that the liquid cooling heat dissipation load status has not changed significantly. The test control platform can maintain the current test load level without triggering load adjustment to avoid frequent changes in the test load due to slight fluctuations.

[0093] When the change in the heat dissipation load capacity index corresponding to two consecutive test steady-state windows exceeds the preset fluctuation threshold, it indicates that the liquid cooling heat dissipation load state has changed and requires a response. The test control platform determines the target test load level for the next test steady-state window based on the range of the heat dissipation load capacity index corresponding to the current test steady-state window, and gradually adjusts the current test load level to the target test load level within the transition range between the two test steady-state windows.

[0094] During load smoothing, the test control platform does not switch the read / write intensity parameter combination corresponding to the current test load level to the target test load level all at once. Instead, it adjusts parameters such as read / write ratio, data block size, queue depth, and number of concurrent threads step by step according to a preset adjustment cycle. Within each adjustment cycle, the parameter adjustment range does not exceed 20% of the difference between the corresponding parameters of the current and target test load levels. By setting a single-cycle adjustment range limit, the impact of sudden load changes on the storage system's operating status and liquid cooling performance can be reduced.

[0095] For example, when the current test load level is L3 and the target test load level is L2, the test control platform can gradually reduce the queue depth, the number of concurrent threads, and the data block size within the transition interval, and gradually adjust the read / write structure according to the read / write ratio corresponding to the target load level, instead of directly jumping from the L3 parameter combination to the L2 parameter combination. Correspondingly, when the current test load level is L3 and the target test load level is L4, the test control platform can also gradually increase the queue depth, the number of concurrent threads, and the data block size, allowing the test load intensity to increase smoothly.

[0096] As a specific example, the thermal load capacity index of the current test steady-state window is 0.721, and the thermal load capacity index of the next test steady-state window drops to 0.512. The change between the two exceeds the preset fluctuation threshold of 0.15, and the test control platform determines that load adjustment needs to be triggered. Since a thermal load capacity index of 0.512 corresponds to a lower target test load level, the test control platform gradually adjusts the test load from the original load level to the target test load level during the transition period after the current test steady-state window ends, and controls the single-cycle parameter adjustment range to not exceed 20% of the difference between the old and new load parameters. For example, a 15% single-cycle adjustment range can be used to complete the smooth transition.

[0097] In another specific example, after the liquid cooling system recovers, the thermal load capacity index corresponding to the new steady-state test window rises to 0.735, indicating that the liquid cooling system has the capacity to handle higher test loads. If the change in this thermal load capacity index relative to the previous steady-state test window exceeds a preset fluctuation threshold, the test control platform can smoothly revert the test load level to a higher level within the transition range between adjacent steady-state test windows. This allows for increased test load intensity once the cooling capacity is restored, further obtaining test data that better reflects the upper limit of the storage system's performance.

[0098] In this embodiment, the generation of the performance benchmark compensation coefficient and the performance data compensation process correspond to step S4. Since the test control platform adjusts the test load level based on the liquid cooling heat dissipation capacity index, the storage performance data collected within different test steady-state windows may correspond to different test load levels. Directly comparing performance data such as IOPS and read / write latency under different load levels is easily affected by the differences in load levels, resulting in a lack of a unified basis for comparison in the test results. Therefore, in this embodiment, after the test load level changes, a performance benchmark compensation coefficient is generated based on the difference between the current test load level and the target test load level, and the storage performance data collected in the next test steady-state window is compensated based on this performance benchmark compensation coefficient.

[0099] Specifically, after determining the target test load level, the test control platform obtains the rated IOPS and rated read / write latency corresponding to the current test load level, and also obtains the rated IOPS and rated read / write latency corresponding to the target test load level. The rated IOPS and rated read / write latency can be derived from storage unit specifications, pre-calibrated test results, or preset load level benchmark parameters in the test standard. By converting the rated performance parameters under the current test load level and the target test load level, a performance benchmark compensation coefficient for normalizing performance data between different load levels can be obtained.

[0100] The performance benchmark compensation coefficient Calculate as follows: ; in, Test load level for target Rated IOPS of the lower storage unit This represents the rated IOPS of the storage unit under the current test load level L0. The rated read / write latency of the storage unit is tested under the target test load level Lx. This represents the rated read / write latency of the storage unit under the current test load level L0. The aforementioned performance benchmark compensation coefficients consider both IOPS and read / write latency. IOPS characterizes the difference in throughput between the target load level and the current load level; read / write latency characterizes the variation in response latency across different load levels. By combining these two metrics, evaluation biases caused by using only IOPS or only read / write latency as compensation criteria can be avoided, making the compensated performance data more suitable for comparisons across load levels.

[0101] In one implementation, when the target test load level differs from the current test load level, and the test control platform has already completed load smoothing within the transition range, the test control platform associates the performance benchmark compensation coefficient with the next test steady-state window. Storage performance data such as IOPS, read / write latency, and throughput collected within the next test steady-state window can be normalized or annotated according to this performance benchmark compensation coefficient. The normalized performance data is used for subsequent test report statistics; the unnormalized raw performance data can be retained synchronously for test process traceability.

[0102] As a specific example, when the system is adjusted from L3 load level to L2 load level, the rated IOPS for L3 load level is 600,000, and the rated read / write latency is 220μs; the rated IOPS for L2 load level is 350,000, and the rated read / write latency is 150μs. The test control platform calculates the performance baseline compensation coefficient based on the above parameters: ; This performance benchmark compensation factor is used to characterize the difference in rated performance benchmarks between the two load levels after switching from L3 load level to L2 load level. The test control platform writes this performance benchmark compensation factor into the data annotation information of the next test steady-state window, and normalizes the storage performance data collected in the next test steady-state window based on this performance benchmark compensation factor, so that the performance data in this window can be compared with the test steady-state window data before adjustment.

[0103] In this embodiment, the steady-state validity classification and labeling process of test data corresponds to step S5. After completing the data acquisition for each test steady-state window, the test control platform associates and labels the storage performance data within that test steady-state window with the corresponding heat dissipation capacity index, test load level, and performance benchmark compensation coefficient, so that it can distinguish the usage attributes of different data when generating test reports later.

[0104] Specifically, for each test steady-state window, the test control platform records information such as the window identifier, start time, end time, thermal load capacity index, current test load level, performance benchmark compensation coefficient, average IOPS, average read / write latency, average throughput, and storage unit core temperature. This information, along with the raw performance data collected within the test steady-state window, is stored together, enabling testers to trace the corresponding thermal state, load state, and compensation benchmark for each set of performance data.

[0105] In this embodiment, the test control platform classifies the steady-state validity of storage performance data within the test steady-state window based on the heat dissipation capacity index. As a specific example, when the heat dissipation capacity index Φ ≥ 0.8, the storage performance data within the corresponding test steady-state window is marked as steady-state valid core data; when 0.6 ≤ Φ < 0.8, the storage performance data within the corresponding test steady-state window is marked as steady-state valid reference data; and when Φ < 0.6, the storage performance data within the corresponding test steady-state window is marked as non-steady-state heat-limited data.

[0106] Among them, steady-state effective core data indicates that the liquid cooling system has a high heat dissipation capacity for the current test load, and this type of data can be given priority in the calculation of the final performance conclusion; steady-state effective reference data indicates that the liquid cooling system has a basic heat dissipation capacity for the current test load, and this type of data can be used as a reference or auxiliary statistical basis for the final performance conclusion; non-steady-state heat-limited data indicates that the performance data within the steady-state window of the test may be affected by insufficient heat dissipation capacity, and is usually not included in the calculation of the final performance test conclusion, but can be retained as heat-limited state analysis or test process traceability data.

[0107] During the test report generation phase, the test control platform prioritizes the use of performance data marked as steady-state valid core data and steady-state valid reference data, and performs statistical analysis in conjunction with the corresponding performance benchmark compensation coefficients to form the steady-state performance test results of the liquid-cooled storage system. For data marked as non-steady-state heat-limited data, the test report can list its window number, heat dissipation capacity index, and the reason for the limitation separately, without including it in the calculation of the final core performance conclusions such as average IOPS and average read / write latency.

[0108] In this embodiment, the test control platform performs steady-state performance testing and regulation of the liquid-cooled storage system according to the aforementioned steps S1 to S5. The following describes the steady-state test window locking, heat dissipation capacity index calculation, load level adjustment, performance benchmark compensation, and test data output process using a specific set of test procedures.

[0109] After the test started, the test control platform synchronously collected the operating parameters of the liquid cooling system and the performance parameters of the storage system at a sampling cycle of 5 seconds. After 60 consecutive sampling cycles, it was determined that the fluctuation range of the storage read / write load was no more than 2% and the fluctuation range of the liquid cooling inlet temperature was no more than 0.8℃ within 5 minutes, both of which met the preset steady-state judgment conditions. Therefore, the test control platform locked this continuous 5-minute interval as the initial test steady-state window.

[0110] Within this initial steady-state test window, the average core temperature of the storage unit was 42℃, the real-time coolant flow rate of the liquid cooling branch corresponding to the tested storage node was 1.18L / min, and the real-time circulation pressure was 0.34MPa. The current test load is L3 load level, with a read / write ratio of 50 / 50. Based on the rated power consumption of 12W for pure write load, 5W for pure read load, and the maximum rated power consumption of 12W, the thermal contribution coefficient of the current test load is: ; Furthermore, based on the upper limit of the storage unit core temperature of 75℃, the rated flow rate of the liquid cooling branch of 1.2L / min, the rated pressure of the liquid cooling branch of 0.35MPa, and the weighting coefficients α=0.5, β=0.2, γ=0.15, and δ=0.15, the heat dissipation capacity index corresponding to this initial test steady-state window is calculated: ; Since the heat dissipation load capacity index of 0.721 falls within the range of 0.65 ≤ Φ < 0.8, it corresponds to the L4 load level. However, the change in the heat dissipation load capacity index between this test steady-state window and the previous test state did not exceed the preset fluctuation threshold. Therefore, the test control platform maintains the current L3 load level and does not immediately trigger load adjustment.

[0111] During the second hour of testing, the ambient temperature rose, causing the liquid cooling inlet temperature to increase. The test control platform then locked onto a new steady-state testing window. Within this window, the average core temperature of the storage unit rose to 58°C. The real-time coolant flow rate of the liquid cooling branch corresponding to the tested storage node was 1.15 L / min, and the real-time circulation pressure was 0.33 MPa. The current test load remained at L3 load level.

[0112] With the current read / write ratio unchanged, the thermal contribution coefficient of the current test load remains at 0.708. The test control platform calculates a heat dissipation capacity index of 0.512 based on the parameters within this test steady-state window. This value falls within the range of 0.5 ≤ Φ < 0.65, corresponding to an L3 load level. Simultaneously, the change in the heat dissipation capacity index between this test steady-state window and the previous test steady-state window is 0.209, exceeding the preset fluctuation threshold of 0.15. This indicates a significant change in the heat dissipation load state of the liquid cooling system, triggering the test control platform to adjust the test load level.

[0113] During the transition period between two steady-state test windows, the test control platform smoothly adjusts the test load from the original load level to the target load level. During the adjustment process, the parameter adjustment range within a single adjustment cycle is controlled to be 15% of the difference between the old and new load parameters, not exceeding the 20% limit, thereby avoiding interference from sudden changes in test load on the formation of subsequent steady-state test windows.

[0114] Simultaneously, the test control platform calculates the performance baseline compensation coefficient based on the rated performance parameters of the load level before and after adjustment. Taking the adjustment from L3 load level to L2 load level as an example, the rated IOPS corresponding to L3 load level is 600,000, and the rated read / write latency is 220μs; the rated IOPS corresponding to L2 load level is 350,000, and the rated read / write latency is 150μs. Therefore, the performance baseline compensation coefficient is: ; The test control platform associates the performance benchmark compensation coefficient with the next test steady-state window for subsequent performance data normalization and test report statistics.

[0115] After four hours of testing, the ambient temperature returned to normal, the liquid cooling inlet temperature decreased, and the heat dissipation capacity of the liquid cooling system improved. The test control platform re-locked a new steady-state test window and calculated the corresponding heat dissipation capacity index to be 0.735. This value falls within the range of 0.65 ≤ Φ < 0.8, indicating that the liquid cooling system can withstand a higher test load.

[0116] If the change in the heat dissipation capacity index of the current test steady-state window relative to the previous test steady-state window exceeds a preset fluctuation threshold, the test control platform will smoothly revert the test load level to a higher load level during the transition interval between adjacent test steady-state windows, and simultaneously update the performance benchmark compensation coefficient. Through this process, after the heat dissipation capacity is restored, the system can increase the test load intensity to collect steady-state test data that better reflects the performance ceiling of the liquid-cooled storage system.

[0117] Throughout the testing process, the test control platform locked 28 steady-state test windows, and simultaneously labeled the performance data within each steady-state test window with the heat dissipation capacity index, test load level, and performance benchmark compensation coefficient. According to the heat dissipation capacity index grading rules, there were 8 steady-state test windows with Φ ≥ 0.8, and the corresponding data was labeled as steady-state valid core data; 12 steady-state test windows with 0.6 ≤ Φ < 0.8, and the corresponding data was labeled as steady-state valid reference data; and 8 steady-state test windows with Φ < 0.6, and the corresponding data was labeled as non-steady-state heat-limited data.

[0118] Ultimately, the test control platform used steady-state effective core data and steady-state effective reference data as the basis for performance conclusion calculations, and performed normalization processing based on the corresponding performance benchmark compensation coefficients, obtaining a steady-state average IOPS of 920,000 and an average read / write latency of 275μs for the liquid-cooled server storage array. Data labeled as non-steady-state heat-limited data was retained by the test control platform as heat-limited process data but not included in the final performance conclusion calculations.

[0119] As can be seen from the above exemplary execution process, this embodiment can filter data based on the test steady-state window, adjust the load level based on the heat dissipation capacity index, and improve the authenticity, comparability and traceability of test data through performance benchmark compensation and steady-state effectiveness classification when the liquid cooling heat dissipation state fluctuates.

[0120] It should be noted that the parameters regarding the number of storage units, liquid cooling branch type, coolant type, sampling period, steady-state window length, load fluctuation threshold, temperature fluctuation threshold, weighting coefficient, number of test load levels, and the read / write ratio, data block size, queue depth, and number of concurrent threads corresponding to each test load level in the above embodiments are only used to illustrate specific implementations of the present invention and do not constitute a limitation on the scope of protection of the present invention. In other embodiments, the above parameters can be adaptively adjusted according to the structure of the liquid-cooled storage system, the specifications of the storage medium, the liquid cooling heat dissipation capacity, and the performance testing standards; as long as the steady-state performance testing and control of the liquid-cooled storage system is still based on test steady-state window locking, heat dissipation capacity index calculation, target test load level determination, performance benchmark compensation, and steady-state effectiveness grading, it should all fall within the scope of implementation of the present invention.

[0121] 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 regulating steady-state performance test of a liquid-cooled storage system, characterized in that, Includes the following steps: S1. During the performance test of the liquid cooling storage system, the operating parameters of the liquid cooling system and the performance parameters of the storage system are collected simultaneously. Based on the operating parameters of the liquid cooling system and the performance parameters of the storage system, it is determined whether the operating state during the test meets the preset steady-state judgment condition. The time interval that continuously meets the preset steady-state judgment condition is locked as the test steady-state window. S2. For the test steady-state window, obtain the liquid cooling branch operating parameters, storage unit core temperature and thermal contribution parameters corresponding to the current test load of the storage node under test, and calculate the heat dissipation capacity index corresponding to the test steady-state window based on the liquid cooling branch operating parameters, the storage unit core temperature and the thermal contribution parameters. S3. Based on the heat dissipation load capacity index corresponding to the current test steady-state window, determine the target test load level corresponding to the next test steady-state window, and adjust the current test load level to the target test load level within the transition interval between the current test steady-state window and the next test steady-state window. S4. Based on the difference between the current test load level and the target test load level, generate a performance benchmark compensation coefficient, and perform compensation processing on the storage performance data collected in the next test steady-state window based on the performance benchmark compensation coefficient. S5. Label the storage performance data collected within each test steady-state window with the corresponding heat dissipation capacity index, test load level and performance benchmark compensation coefficient, and classify the steady-state effectiveness of the storage performance data according to the heat dissipation capacity index.

2. The steady-state performance test regulation method of a liquid-cooled storage system according to claim 1, characterized in that, In step S1, the operating parameters of the liquid cooling system include at least two of the following: liquid cooling inlet temperature, liquid cooling branch flow rate, and liquid cooling branch circulation pressure. The storage system performance parameters include at least two of the following: storage read / write load, IOPS, read / write latency, and storage cell core temperature. The synchronous acquisition refers to the corresponding acquisition of the operating parameters of the liquid cooling system and the performance parameters of the storage system according to the same sampling period or the sampling period aligned with the timestamp.

3. The steady-state performance test regulation method of a liquid-cooled storage system according to claim 1, characterized in that, In step S1, the preset steady-state determination conditions include: Within N consecutive sampling periods, the fluctuation range of the read / write load of the storage system does not exceed the preset load fluctuation threshold, and the fluctuation range of the inlet temperature of the liquid cooling system does not exceed the preset temperature fluctuation threshold. When the above conditions are met simultaneously, the time interval corresponding to the N consecutive sampling periods is locked as the test steady-state window; The sampling period is 1s-10s, N is 30-100, the preset load fluctuation threshold is no greater than 5%, and the preset temperature fluctuation threshold is no greater than 2℃.

4. The method for testing and controlling the steady-state performance of a liquid-cooled storage system according to claim 1, characterized in that, In step S2, the thermal contribution parameter corresponding to the current test load is a thermal contribution coefficient The thermal contribution coefficient is calculated according to the read proportion, the write proportion, the rated power consumption of the pure read load, the rated power consumption of the pure write load, and the maximum rated power consumption of the storage unit in the current test load. The thermal contribution coefficient The formula for calculation is: ； in, This represents the rated power consumption of the storage unit under a pure write load. This represents the write ratio in the current test load. This represents the rated power consumption of the storage unit under a pure read load. This represents the read ratio in the current test load. This represents the maximum rated power consumption of the storage unit.

5. The method for testing and controlling the steady-state performance of a liquid-cooled storage system according to claim 4, characterized in that, In step S2, the heat dissipation load capacity index Φ is calculated as follows: ； in, The core temperature limit is preset for the storage unit. To test the average core temperature of the memory cells within the steady-state window, This represents the real-time coolant flow rate of the liquid cooling branch corresponding to the tested storage node. The design rated flow rate of this liquid cooling branch is... This is the real-time circulating pressure of the liquid cooling branch. The design rated pressure of this liquid cooling branch is... The thermal contribution coefficient of the current test load is α, β, γ, and δ, which are preset weighting coefficients, and α+β+γ+δ=1.

6. The method for testing and controlling the steady-state performance of a liquid-cooled storage system according to claim 5, characterized in that, The rules for determining the preset weighting coefficients are as follows: The values ​​of α range from 0.4 to 0.6, β range from 0.15 to 0.25, γ range from 0.1 to 0.2, and δ range from 0.05 to 0.

15. Wherein, α is used to characterize the influence weight of the core temperature margin of the storage unit on the heat dissipation capacity index, β is used to characterize the influence weight of the liquid cooling branch flow rate on the heat dissipation capacity index, γ is used to characterize the influence weight of the liquid cooling branch circulation pressure on the heat dissipation capacity index, and δ is used to characterize the influence weight of the current test load thermal contribution on the heat dissipation capacity index.

7. The method for testing and controlling the steady-state performance of a liquid-cooled storage system according to claim 1, characterized in that, In step S3, the test load is pre-divided into multiple consecutive test load levels, and each test load level corresponds to a set of read and write strength parameters. The read / write intensity parameter combination includes read / write ratio, data block size, queue depth, and number of concurrent threads; A mapping relationship is established between the numerical range of the heat dissipation load capacity index and the test load level, and the target test load level corresponding to the next test steady state window is determined based on the numerical range of the heat dissipation load capacity index corresponding to the current test steady state window.

8. The method for testing and controlling the steady-state performance of a liquid-cooled storage system according to claim 7, characterized in that, In step S3, when the change in the heat dissipation load capacity index corresponding to two consecutive test steady-state windows exceeds the preset fluctuation threshold, the test load level adjustment is triggered. During the test load level adjustment process, the read and write strength parameter combination corresponding to the current test load level is smoothly adjusted to the read and write strength parameter combination corresponding to the target test load level; Within a single adjustment cycle, the parameter adjustment range shall not exceed 20% of the difference between the corresponding parameters between the current test load level and the target test load level.

9. The method for testing and controlling the steady-state performance of a liquid-cooled storage system according to claim 1, characterized in that, In step S4, the performance benchmark compensation coefficient Calculate as follows: ； in, Test load level for target Rated IOPS of the lower storage unit This represents the rated IOPS of the storage unit under the current test load level L0. The rated read / write latency of the storage unit is tested under the target test load level Lx. This represents the rated read / write latency of the storage unit under the current test load level L0. The compensation process includes: normalizing the IOPS and read / write latency collected in the next test steady-state window based on the performance benchmark compensation coefficient.

10. The method for testing and controlling the steady-state performance of a liquid-cooled storage system according to claim 1, characterized in that, In step S5, the steady-state effectiveness classification includes: When the heat dissipation capacity index Φ≥0.8, the storage performance data within the corresponding steady-state test window will be marked as the steady-state valid core data; When 0.6≤Φ<0.8, the storage performance data within the corresponding steady-state test window will be marked as the valid steady-state reference data; When Φ < 0.6, the storage performance data within the corresponding steady-state test window will be marked as unsteady-state heat-limited data. The final performance test conclusion is generated based on the steady-state effective core data and the steady-state effective reference data, and the non-steady-state heat dissipation-limited data is not included in the calculation of the final performance test conclusion.

Images