External centralized power supply device for immersed liquid cooling computing power module and control method of external centralized power supply device

Abstract

The embodiment of the invention provides an external centralized power supply device for an immersed liquid cooling computing power module and a control method of the external centralized power supply device, and relates to the technical field of an immersed liquid cooling computing power module power supply technology. The device comprises a liquid cooling box body; the external centralized power supply cabinet is arranged outside the liquid cooling box body, and the external centralized power supply cabinet internally comprises a central control unit and a plurality of hot plug type power supply modules which are connected in parallel; the diversion cable is arranged on the side wall of the liquid cooling box body in a penetrating manner and is used for introducing the direct-current electric energy output by the external centralized power supply cabinet into the liquid cooling box body; and the busbar is electrically connected with the diversion cable and is used for transmitting electric energy in an immersion liquid environment. According to the invention, the problem of low space utilization rate of the box body is solved, so that the effect of improving the space utilization rate of the box body is achieved.

Description

Technical Field

[0001] The present invention relates to the field of power supply technology for immersion liquid-cooled computing modules, and more specifically, to an external centralized power supply device for immersion liquid-cooled computing modules and its control method. Background Technology

[0002] With the explosive growth in demand for high-performance computing and the continuous increase in computing power density in data centers, traditional air cooling methods are no longer sufficient to meet the heat dissipation needs of high power density chips. Immersion liquid cooling technology has gradually become the industry mainstream due to its superior heat dissipation efficiency.

[0003] However, in existing immersion liquid cooling system architectures, the power supply scheme typically follows the traditional "rack-mounted power distribution unit (PDU) + distributed power modules (CRPS)" model. That is, each computing node (computing blade) immersed in the coolant is equipped with a set of CRPS power modules (usually with an "N+N" redundant configuration). This architecture suffers from high hardware costs and low space utilization in immersion scenarios.

[0004] There is currently no good solution to the above problems. Summary of the Invention

[0005] This invention provides an external centralized power supply device and control method for an immersion liquid-cooled computing module, which at least solves the problem of low hardware space utilization in related technologies.

[0006] According to one embodiment of the present invention, an external centralized power supply device for an immersion liquid-cooled computing module is provided, comprising:

[0007] Liquid-cooled enclosure;

[0008] An external centralized power supply cabinet is located outside the liquid-cooled box. The external centralized power supply cabinet contains a central control unit and several hot-swappable power modules connected in parallel.

[0009] A current-carrying cable is installed through the side wall of the liquid-cooled box to introduce the DC power output from the external centralized power supply cabinet into the interior of the liquid-cooled box.

[0010] The busbar, electrically connected to the current-carrying cable, is used to transmit electrical energy in an immersion liquid environment.

[0011] In one exemplary embodiment, the busbar includes a third portion extending along the length of the liquid-cooled housing, and the third portion is provided with a plurality of conductive contact areas at intervals.

[0012] According to another embodiment of the present invention, a control method for an external centralized power supply device for an immersion liquid-cooled computing module is provided, applied to the aforementioned external centralized power supply device for an immersion liquid-cooled computing module, comprising:

[0013] Acquire the total load current data on the busbar, as well as the real-time operating status of each power module in the external centralized power supply cabinet;

[0014] Based on the total load current data and the preset power module efficiency model, the target number of power modules to be turned on is determined. The target number of power modules to be turned on indicates the number of power modules that, under the condition of meeting the preset redundancy level, make the average load rate of the operating power modules in the preset high efficiency range.

[0015] Compare the number of currently active power modules with the target number of active power modules, and if the two are inconsistent, determine the specific power module identifier that needs to be switched based on the historical cumulative running time of each power module.

[0016] Send a state switching command to the power module corresponding to the specific power module identifier to adjust the output power configuration of the external centralized power supply cabinet; and / or monitor the voltage change rate of the bus, and when the voltage change rate exceeds a preset disturbance threshold, block the state switching command and forcibly maintain or increase the number of power modules that are turned on.

[0017] In an exemplary embodiment, determining the target number of power modules to be turned on based on the total load current data and a preset power module efficiency model includes:

[0018] Based on the total load current data and the rated voltage of the bus, the real-time total power of the computing module is obtained;

[0019] Based on the real-time total power, the rated power of a single power module, and the preset redundancy level, determine the minimum number of modules required to meet the power supply safety baseline.

[0020] The target number of modules to be enabled is determined based on the minimum number of modules.

[0021] In an exemplary embodiment, determining the identifier of the specific power module requiring a state switch based on the historical cumulative operating time of each of the power modules includes:

[0022] When the number of currently active power modules is greater than the target number of active power modules, all power modules currently active are sorted from longest to shortest according to their historical cumulative running time, and the power module at the top of the sort is selected as the module to be put into hibernation.

[0023] When the number of currently active power modules is less than the target number of active power modules, all power modules currently in hibernation are sorted from shortest to longest based on their historical cumulative running time, and the power module at the top of the sort is selected as the module to be woken up.

[0024] In an exemplary embodiment, monitoring the voltage change rate of the bus and, when the voltage change rate exceeds a preset disturbance threshold, disabling the state switching command and forcibly maintaining or increasing the number of power modules on includes:

[0025] Collect the instantaneous voltage value of the bus;

[0026] Based on a series of consecutive instantaneous voltage values, the time derivative of the voltage is calculated as the rate of change of voltage;

[0027] When the absolute value of the voltage change rate is greater than the disturbance threshold, the system is determined to be in a hot-plug disturbance state.

[0028] Under the hot-swap disturbance state, a forced power-on command is immediately sent to all power modules in sleep mode, and a delay timer is started.

[0029] Only after the delay timer expires and the voltage change rate returns to the normal range, the forced start command is released and the control logic based on the total load current data is restored.

[0030] In one exemplary embodiment, the method further includes:

[0031] Receive load prediction data from the upper-layer task scheduling system, wherein the load prediction data includes the expected increase in power consumption and the expected arrival time;

[0032] Calculate the remaining power margin of the currently active power modules;

[0033] If the expected increase in power consumption is greater than the remaining power margin, a power-on command will be sent to the power module in sleep mode at a preset start time point before the expected arrival time.

[0034] According to yet another embodiment of the present invention, a computer-readable storage medium is also provided, wherein a computer program is stored therein, wherein the computer program is configured to perform the steps in any of the above method embodiments when executed.

[0035] According to yet another embodiment of the present invention, an electronic device is also provided, including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to perform the steps in any of the above method embodiments.

[0036] This invention places the power supply cabinet outside the liquid-cooled enclosure, thereby creating redundancy in the space inside the enclosure and effectively increasing the usable space. Therefore, it can solve the problem of low space utilization inside the enclosure and achieve the effect of improving the space utilization rate of the enclosure. Attached Figure Description

[0037] Figure 1 This is a schematic diagram of an external centralized power supply device for an immersion liquid-cooled computing module according to an embodiment of the present invention;

[0038] Figure 2 This is a partial structural schematic diagram of an external centralized power supply device for an immersion liquid-cooled computing module according to an embodiment of the present invention;

[0039] Figure 3 This is a partial top view of an external centralized power supply device for an immersion liquid-cooled computing module according to an embodiment of the present invention;

[0040] Figure 4 This is a schematic diagram of the computing power module structure;

[0041] Figure 5 This is a flowchart of an external centralized power supply device and its control method for an immersion liquid-cooled computing module according to an embodiment of the present invention.

[0042] In the diagram, 1 is the liquid-cooled enclosure; 2 is the external centralized power supply cabinet; 3 is the power delivery cable; 41 is the first part; 42 is the second part; 43 is the third part; 5 is the computing module; and 51 is the power transmission interface. Detailed Implementation

[0043] The technical solutions of the embodiments of this application will be described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0044] In the following description, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0045] Furthermore, in this application, directional terms such as "upper," "lower," "left," and "right" may be defined relative to the orientation of the components shown in the accompanying drawings. It should be understood that these directional terms can be relative concepts, used for relative description and clarification, and may change accordingly depending on the orientation of the components in the accompanying drawings.

[0046] In this application, unless otherwise expressly specified and limited, the term "connection" should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral part; it can be a direct connection or an indirect connection through an intermediate medium. Furthermore, the term "coupled" can refer to an electrical connection that enables signal transmission.

[0047] As used herein, “about,” “approximately,” or “approximately” includes the stated value and the average value within an acceptable range of deviation from the given value, wherein the acceptable range of deviation is determined by a person skilled in the art taking into account the measurement under discussion and the error associated with the measurement of the given quantity (i.e., the limitations of the measurement system).

[0048] Example 1

[0049] like Figure 1-4 As shown, this application provides an immersion liquid-cooled external centralized power supply device, which has undergone system-level engineering optimization for immersion liquid-cooled environments (high insulation requirements, high fluid compatibility, and high maintenance difficulty).

[0050] The device includes a liquid-cooled housing 1 (Tank), an external centralized power supply cabinet 2 (Powershelf), a current-conducting cable 3 that penetrates the wall of the liquid-cooled housing, and a busbar. The external centralized power supply cabinet 2 can support output of two voltages, P12V and P54V. The low-voltage DC power supply is introduced into the liquid-cooled housing 1 through the current-conducting cable 3 and the busbar. This eliminates the need for installation space for the external centralized power supply cabinet, thereby effectively reducing the volume of the liquid-cooled housing 1.

[0051] Specifically, the external centralized power supply cabinet 2 is designed as a standard 19-inch or 21-inch rack-mount unit with a physical height of 4U (i.e., approximately 19-inch or 21-inch). ), depth set to It is equipped with several power module arrays, preferably 12 power module slots connected in parallel, so that 12 hot-swappable high-efficiency rectifier modules can be inserted when fully configured; each power module supports a wide range of AC input, with a voltage range of [missing information]. to The frequency range is to Each power module outputs a nominal voltage of DC power, the output voltage can be to Fine-tuning is performed within the range using software commands, with adjustment steps of [number missing]. The rated output power of a single power module is set to Therefore, the maximum total output power of the entire external centralized power supply cabinet is... Reachable The power module adopts a totem-pole bridgeless topology and an LLC resonant converter, with a peak conversion efficiency of [missing information]. exist The measured value under load rate is not lower than (Titanium-grade standard); In addition, the external centralized power supply cabinet integrates an independent central control unit board, whose processor core CCU adopts a... The core is a high-performance microcontroller, with a main frequency running at... For example, the chip model can be selected. Series; CCU is equipped with two independent The bus is used for internal communication. The first bus (Bus 0) is dedicated to connecting 12 power modules, and its operating frequency is set to [frequency value missing]. One bus is used to read voltage and current and send sleep commands. The second bus (Bus 1) connects to the Ethernet controller on the rear panel for data interaction with the upper-level Data Center Infrastructure Management System (DCIM). The CCU also has... Memory, used to store historical running data (such as...) (and a preset efficiency curve lookup table.)

[0052] The current-carrying cable 3 is a critical conduit for safely transferring large currents from the air environment (outside the enclosure) to the liquid environment (inside the enclosure), due to the electronic fluorinated liquid (e.g., fluorinated liquid) filled inside the immersion liquid-cooled enclosure (Tank). (or similar mineral oil), and the operating temperature is usually within to Therefore, the current-carrying cable 3 must possess extremely high sealing reliability and conductivity. Specifically, the current-carrying cable 3 can be a DC power supply cable. The current-carrying cable 3 is externally wrapped with an insulating sleeve made of polyetheretherketone (PEEK) material to reduce the chemical corrosion effect of the fluorinated liquid. Simultaneously, double O-ring seals are provided between the insulating sleeve and the wall of the liquid-cooled housing 1, and between the insulating sleeve and the busbar. The sealing ring material is fluororubber (FKM) with a Shore A hardness. Spend.

[0053] The busbar includes a conductive copper pillar, which comprises a vertically arranged first part 41, a horizontally arranged second part 42 located on top of the first part 41, and a third part 43 laid at the bottom of the liquid cooling box 1. The third part 43 has multiple conductive contact areas and extends along the length of the liquid cooling box 1. The power cable 3 of the external centralized power supply cabinet 2 passes through the liquid box 1 and is electrically connected to the second part 42 (usually by plugging). The computing module 5 (or server module) is plugged into the conductive contact area of ​​the third part 43 through the power transmission interface 51 (Clip). Inside the computing module 5, the power transmitted by the power transmission interface 51 is transferred to the various functional modules (motherboard / hard drive backplane / GPU, etc.) within the node through the PDB board, thereby realizing power supply. The third part 43 and the first part 41 can be integrally formed or separately plugged and fixed to adapt to different needs.

[0054] The conductive copper pillars are preferably made of oxygen-free copper (TU1) with a purity of [missing information]. To ensure that the resistivity is lower than The cross-sectional areas of the first and second parts According to current density Calculations show that a single copper pillar can safely support approximately The current is sufficient to meet the power supply requirements of the Tank; in addition, the surface of the conductive copper pillar is plated with a thickness of A silver layer is applied to reduce contact resistance and prevent oxidation in a high-temperature oil-cooled environment.

[0055] It should be noted that the third part 43 consists of two parallel copper busbars: a positive busbar (Bus+) and a negative busbar (Bus-). The cross-sectional dimension of each copper busbar is [width]. thick The length is the same as the Tank length; the center distance between the positive and negative busbars is set to... It is fixed above the bottom plate of the liquid-cooled box 1 by an insulating bracket. This allows the coolant to flow freely below section 43, carrying away Joule heat.

[0056] The above-mentioned centralized power supply method has the following beneficial effects:

[0057] 1. It can effectively solve the cable management problem caused by the large number of power cables in the "PDU with CRPS" power supply method. At the same time, the structure of the bus and the power transmission interface 51 can greatly improve the maintainability of the computing module 5.

[0058] 2. After using the external centralized power supply cabinet 2 to supply power to the computing module 5, the power control strategy of the external centralized power supply cabinet 2 can achieve "N+N" power redundancy for the entire liquid cooling box 1 (because the number of CRPS power supplies is reduced relatively), thereby reducing the basic power consumption of the power module and improving energy efficiency.

[0059] 3. Using an external centralized power supply cabinet can effectively reuse existing external centralized power supply cabinet products. The space saved in the liquid cooling cabinet can be used to deploy more computing modules, reserving space for further improvement of computing power density in the future.

[0060] Example 2

[0061] like Figure 5 As shown, this application also provides a control method for an external centralized power supply device for an immersion liquid-cooled computing module, used in the aforementioned external centralized power supply device for an immersion liquid-cooled computing module, comprising:

[0062] Step S11: Obtain the total load current data on the busbar and the real-time operating status of each power module in the external centralized power supply cabinet 2.

[0063] In this embodiment, a Hall effect current sensor (e.g.) is used. The system collects current data and uses a resistor divider network to collect voltage data. A Hall effect current sensor is installed at the input of the busbar. The sensor's range is... Linearity is better than Response time is less than Its output is to The analog voltage signal is linearly proportional to the current value; the voltage division ratio of the precision resistor voltage divider network is set to... It can The bus voltage dropped to To adapt to the input range of the ADC; the ADC sampling frequency within the CCU Set as That is, each A single point is sampled to effectively capture the transient waveform at the moment of hot-plugging; simultaneously, to improve the signal-to-noise ratio (SNR), the system employs... Double oversampling, that is, after collecting 16 raw data points, the average value is taken as the effective sample, and the effective data update rate is... The corresponding time resolution is Since the raw acquired data usually contains high-frequency noise (ripple from the switching power supply) and random interference, filtering is necessary to obtain a stable DC component. A moving average filter can be used here. Specifically, for the total load current used in energy efficiency calculations... Using window length The moving average filter is calculated using the following formula: In the formula, k is the current sampling point number, I raw (k−i) represents the original sampled values ​​i times backward from the current time k.

[0064] For example, suppose recently The current value at each sampling point arrive Fluctuations between these values, after filtering, result in the output... Will stabilize at Nearby, ripple interference was eliminated, and so on.

[0065] Alternatively, a first-order low-pass filter (LPF) can be used, with the cutoff frequency set to... This will not be elaborated upon here.

[0066] In addition to bus-level analog signals, the CCU also needs to know the health status of each PSU. Specifically, based on the PMBus communication protocol stack, the CCU acts as the bus master, and every... Initiate a full polling round for each address. (Corresponding to 12 PSUs), send the command words READ_STATUS_WORD (0x79) and READ_IOUT (0x8C); then parse the returned STATUS_WORD; if Bit 1 (OverTemperature) or Bit 4 (Output Over Current) is set in the status word returned by a module, the CCU immediately marks the module as "Faulty", and so on.

[0067] It should be noted that modules marked as "unavailable" will be removed from the available resource pool. Removed from the list and not included in subsequent load balancing calculations; for example, existing systems... One PSU was found during polling. The report indicated an undervoltage fault; therefore, in subsequent calculations, the total number of available modules in the system was determined. Revised to This ensures that the scheduling algorithm does not attempt to wake up a faulty module, and so on.

[0068] To prevent erroneous decisions caused by sensor drift, it is necessary to calculate the drift variation:

[0069] .

[0070] Assuming the verification threshold at this time (i.e., 5%)

[0071] like They deemed the data reliable and proceeded with the next steps.

[0072] like This indicates a serious deviation in the metering circuit of the shunt or PSU. At this time, the sensor fault safety mode is triggered: the system is forced to operate in the maximum redundancy configuration (wake up all available PSUs), report alarm information, and stop all energy-saving sleep operations.

[0073] For example, The reading is Currently open Each PSU reported separately. The sum is ,deviation The verification passed; if the total PSU is only ,deviation If this happens, an alarm will be triggered immediately, and so on.

[0074] Step S12: Based on the total load current data and the preset power module efficiency model, determine the target number of power modules to be turned on. The target number of power modules to be turned on indicates the number of power modules that, under the condition of satisfying the preset redundancy level, ensure that the average load rate of the operating power modules is in the preset high efficiency range.

[0075] In this embodiment, to achieve adaptive centralized power supply, it is necessary to maximize the overall energy efficiency of the system (objective function) while satisfying reliability constraints (hard constraints), and then control the corresponding power modules according to the calculation results.

[0076] Specifically:

[0077] Step S121: Discretize the continuous physical efficiency curve into a high-precision look-up table (LUT).

[0078] In the power module efficiency model used in this embodiment, the conversion efficiency of the power module is... With load rate ( The relationship between the two curves exhibits typical asymmetric parabolic characteristics, particularly in the lightly loaded region (…). Due to the large proportion of fixed losses, efficiency drops sharply (e.g.) hour ); in the half-load area ( Efficiency reaches its peak. In the fully loaded area ( ),because Increased losses and a slight decrease in efficiency (e.g.) (This is not limited to)

[0079] At the same time, the CCU memory is pre-loaded with... The efficiency lookup table for each point is Eff_Table

[101] , with indexes... Corresponding load rate to .

[0080] For example:

[0081] Eff_Table

[10] (10% load) = ;

[0082] Eff_Table

[20] (20% load) = ;

[0083] Eff_Table

[50] (50% load) = ;

[0084] Eff_Table

[100] (100% load) = ;

[0085] This model is applicable to any measured load rate. Efficiency is calculated by linear interpolation of two adjacent LUT points. :

[0086]

[0087] in For adjacent integer load rates, This represents the corresponding efficiency value.

[0088] Step S122: Determine the boundaries of the constraints.

[0089] Before performing optimization calculations, it is necessary to determine the system's safety boundaries, for which the system is configured with a redundancy mode. (i.e., redundancy level), this mode includes different redundancy level conditions:

[0090] Pattern A (N+1): It is suitable for general business operations and allows for the failure of any module without downtime.

[0091] Pattern B (N+N): Suitable for financial-grade mission-critical operations, providing 100% double computing power redundancy.

[0092] Meanwhile, to prevent overload of a single module, a maximum permissible continuous load rate for a single module is set. (Reserve 10% as dynamic adjustment space), and determine the basic requirement quantity and minimum number of safety openings based on actual needs.

[0093] Step S123: Traverse the execution flow of the optimization algorithm.

[0094] The CCU executes the following logic to find the optimal solution. :

[0095] Search scope from arrive (Total number of physical installations in the system, e.g., 12), for each possible number of startups m( ):

[0096] a. Calculate the average load rate of a single module: In the formula, This represents the current total power (i.e., the real-time total power). This refers to the rated power of a single module.

[0097] b. Efficiency of single module for query / interpolation calculation: .

[0098] c. Calculate the total system loss: .

[0099] Comparison and Decision Making: Choice Minimum (or The maximum value of m is used as .

[0100] For example:

[0101] Input conditions:

[0102]

[0103]

[0104] Redundancy mode:

[0105]

[0106] Step 1: Determine the lower limit

[0107] .

[0108] .

[0109] Step 2: Traverse and calculate

[0110] Option A ( ):

[0111] Single module load .

[0112] load rate .

[0113] Table lookup interpolation efficiency .

[0114] Total loss .

[0115] Option B ( ):

[0116] Single module load .

[0117] load rate .

[0118] Table lookup interpolation efficiency (Near the peak).

[0119] Total loss .

[0120] Option C ( ):

[0121] Single module load .

[0122] load rate .

[0123] Table lookup interpolation efficiency .

[0124] Total loss .

[0125] Step 3: Decision Making

[0126] Comparison of losses: .

[0127] Conclusion: Although It satisfies N+1 redundancy, but Higher energy efficiency; at this time, the algorithm intelligently selects... As a target number of units to be activated, this not only satisfies redundancy requirements but also saves additional costs by allowing the power supply to operate at a more efficient load point. The power consumption is reduced (equivalent to a reduction of approximately 10% in heat loss), and so on.

[0128] Step S13: Compare the number of currently active power modules with the target number of active power modules. If the two are inconsistent, determine the specific power module identifier that needs to be switched based on the historical cumulative running time of each power module.

[0129] In this embodiment, the required number of modules is determined. After that, the next step is to determine which specific modules are involved.

[0130] Specifically, it includes the following steps:

[0131] S131, State machine determination and difference calculation.

[0132] The CCU maintains a state machine for each PSU, and the state machine contains three main states:

[0133] ACTIVE: The main power path is on, the output voltage is normal, and the PWM waveform is normal.

[0134] SLEEP: Main power stage shuts down (PWM stops), only auxiliary power supply operates to maintain communication, power consumption... .

[0135] FAULT: The module has experienced a hardware failure and has been isolated.

[0136] Calculate the difference: ,in, This represents the number of power modules currently in operation.

[0137] S132, fatigue-prioritized sleep strategy.

[0138] when When a hibernation module is needed, the system executes a "let the most tired ones rest" strategy, which specifically involves reading all active systems. Cumulative runtime of each module (Unit: hours), then on Sort the list in descending order (bubble sort or quick sort), then select the first element from the head of the list. Each ID is used as a target to be put into hibernation.

[0139] For example, ,at this time Then select and As a target to be put into hibernation, its corresponding ID is the identifier of a specific power module, and so on; in particular, to avoid sudden current changes, the CCU does not immediately shut down two modules at the same time, but sends a command to shut them down first. ,wait (Let the voltage stabilize and rebalance), then send a command to shut down. .

[0140] S133, Freshness-first wake-up strategy.

[0141] when When a module needs to be woken up, the system executes either the "make the module that has been resting the longest work" or "use the newest one" strategy. To maintain consistent aging across all modules, the system prioritizes the module that has been resting the longest. The smallest sleep module allows for faster uptime of the "new" module, ultimately ensuring consistency across the entire array's lifespan.

[0142] For example, the current hibernation pool At this point, the system needs to wake up one module; a comparison reveals... Therefore, wake up And so on.

[0143] It should be noted that, in order to prevent the inrush current generated by the simultaneous startup of multiple power supplies from triggering the tripping of the upstream circuit breaker, the wake-up operation must be set with a random delay or a fixed interval, such as an interval time. The wake-up sequence at this time is wake-up. ->Wait -> (If necessary) Wake up the next one.

[0144] S134, Oscillation Suppression Logic.

[0145] To prevent frequent PSU starts and stops due to load fluctuations near the critical point (e.g., when the load is high...). and The jump between them causes the optimal solution to be in and (repeatedly jumping between different points), a hysteresis loop mechanism can be set:

[0146] Upgrade threshold: only when the calculated Persistently greater than Exceeding the upgrade stabilization time The wake-up will only be executed at that time.

[0147] Reduction threshold: only when the calculated consistently less than Exceeding the downgraded shading time Sleep mode is only executed when the downgrade delay is much longer than the upgrade delay to ensure safety.

[0148] For example, if in The load drops briefly within seconds, and calculations show that it can reduce [the impact of load reduction]. There are 1 module, but since it did not reach the target... During the stabilization period, the CCU will ignore the downgrade recommendation and maintain the existing redundancy, thereby avoiding unnecessary mechanical wear and electrical fluctuations, and so on.

[0149] Example 3

[0150] Unlike Example 2, the actual operating environment of an immersion liquid-cooled data center is full of dynamic disturbances (such as hot-swapping of computing modules). Therefore, in order to improve control accuracy, it is also necessary to eliminate the effects of these disturbances.

[0151] Step S14, hot-plug disturbance suppression based on voltage differential

[0152] In this embodiment, the extreme concurrency conflict problem of "a high-power node being inserted just as the energy-saving algorithm attempts to put the power module into hibernation" is solved by the following steps:

[0153] S141, voltage differential sensing mechanism.

[0154] Because immersion computing nodes typically contain large-capacity input capacitors, they draw a huge charging current the moment they come into contact with the bus, causing the bus voltage to drop instantaneously. To simulate this situation, the CCU's internal hardware timer... An extremely short-cycle interrupt is triggered to read the bus voltage. Simultaneously calculate the first-order difference In order to filter out quantization noise, a consistency check is performed at three consecutive points, at which point the instantaneous voltage drop rate of the bus is measured. It is directly proportional to the capacitive characteristics of the inserted load.

[0155] S142, Disturbance threshold determination and mode switching.

[0156] Set two thresholds:

[0157] Drop threshold: If a value below this is detected, it means there is a heavy load of insertions.

[0158] Rebound threshold: If a value higher than this is detected, it means that a heavy load has been unplugged (inductive backlash).

[0159] At this moment once The state machine immediately and forcibly switches from "energy-saving mode" to "defense mode".

[0160] S143, Dynamic Locking and Emergency Support.

[0161] In defensive mode, the CCU performs the following sequence of operations:

[0162] Command Freeze: Immediately discard all pending hibernation commands, even if the previous steps determined that 3 modules should hibernate, they must be forcibly retained at this time.

[0163] Full wake-up: (optional strategy) To cope with the possible successive insertion operations, a wake-up command is immediately sent to all modules in a dormant state; although this sacrifices energy efficiency in the short term, it ensures voltage stability.

[0164] Lock timer: Start a timer with a duration of The countdown begins, during which no more sleep calculations are performed.

[0165] Recovery mechanism: When Return to zero and Once the system returns to the preset steady-state range, it automatically switches back to energy-saving mode, and so on.

[0166] Furthermore, to eliminate the inherent hysteresis of feedback control (power module startup takes time), feedforward processing can be performed based on task scheduling. Specifically, the scheduler pre-processes the system training tasks before distributing them to the nodes. The CCU is notified every second, and its communication protocol and data packet definition are as follows:

[0167] Interface: The CCU's Ethernet interface listens on a specific port (such as Port 5000).

[0168] Data packet format (JSON):

[0169] { "event": "PREDICT_LOAD", "timestamp": 1670000000, "power_delta":4500, / / Expected increase of 4500W "time_to_active": 2000 / / Expected load to arrive in 2000ms}

[0170] Detailed implementation process:

[0171] Analysis: The CCU receives the data packet and parses it. and .

[0172] Prediction: Calculate the current remaining margin .

[0173] Assuming four 3000W modules are currently running, the total load is 6000W, and the single module is limited to 90% load.

[0174] .

[0175] decision making:

[0176] if ( If so, no action is needed, as the existing modules are sufficient.

[0177] if Then calculate the gap. And calculate the number of additional wake-ups required. .

[0178] Execution timing: To complete power-on just before the load arrives, the CCU sets up a delayed task:

[0179] In the formula Cold start time of power module (typical value) ).

[0180] exist After completion, immediately send a wake-up command, and so on.

[0181] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0182] It should be noted that the above modules can be implemented by software or hardware. For the latter, they can be implemented in the following ways, but are not limited to: all the above modules are located in the same processor; or, the above modules are located in different processors in any combination.

[0183] Embodiments of the present invention also provide a computer-readable storage medium storing a computer program, wherein the computer program is configured to perform the steps in any of the above method embodiments when executed.

[0184] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard disk, magnetic disk, or optical disk.

[0185] Embodiments of the present invention also provide an electronic device including a memory and a processor, the memory storing a computer program and the processor being configured to run the computer program to perform the steps in any of the above method embodiments.

[0186] In one exemplary embodiment, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor and the input / output device is connected to the processor.

[0187] Through the above description of the embodiments, those skilled in the art can clearly understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0188] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another apparatus, or some features may be ignored or not executed. Furthermore, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0189] The units described as separate components may or may not be physically separate. A component shown as a unit can be one or more physical units; that is, it can be located in one place or distributed in multiple different locations. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0190] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0191] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, in essence, or the parts that contribute to the prior art, or all or part of the technical solutions, can be embodied in the form of a software product. This software product is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0192] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An external centralized power supply device for an immersion liquid-cooled computing module, characterized in that, include: Liquid-cooled enclosure; An external centralized power supply cabinet is located outside the liquid-cooled box. The external centralized power supply cabinet contains a central control unit and several hot-swappable power modules connected in parallel. A current-carrying cable is installed through the side wall of the liquid-cooled box to introduce the DC power output from the external centralized power supply cabinet into the interior of the liquid-cooled box. The busbar, electrically connected to the current-carrying cable, is used to transmit electrical energy in an immersion liquid environment.

2. The apparatus according to claim 1, characterized in that, The busbar includes a third portion extending along the length of the liquid-cooled housing, and the third portion is provided with a plurality of conductive contact areas at intervals.

3. A control method for an external centralized power supply device for an immersion liquid-cooled computing module, characterized in that, The external centralized power supply device for the immersion liquid-cooled computing module as described in any one of claims 1-2 includes: Acquire the total load current data on the busbar, as well as the real-time operating status of each power module in the external centralized power supply cabinet; Based on the total load current data and the preset power module efficiency model, the target number of power modules to be turned on is determined. The target number of power modules to be turned on indicates the number of power modules that, under the condition of meeting the preset redundancy level, make the average load rate of the operating power modules in the preset high efficiency range. Compare the number of currently active power modules with the target number of active power modules, and if the two are inconsistent, determine the specific power module identifier that needs to be switched based on the historical cumulative running time of each power module. Send a state switching command to the power module corresponding to the specific power module identifier to adjust the output power configuration of the external centralized power supply cabinet; and / or monitor the voltage change rate of the bus, and when the voltage change rate exceeds a preset disturbance threshold, block the state switching command and forcibly maintain or increase the number of power modules that are turned on.

4. The method according to claim 3, characterized in that, The determination of the target number of power modules to be turned on based on the total load current data and the preset power module efficiency model includes: Based on the total load current data and the rated voltage of the bus, the real-time total power of the computing module is obtained; Based on the real-time total power, the rated power of a single power module, and the preset redundancy level, determine the minimum number of modules required to meet the power supply safety baseline. The target number of modules to be enabled is determined based on the minimum number of modules.

5. The method according to claim 3, characterized in that, The determination of the specific power module identifier requiring state switching based on the historical cumulative operating time of each power module includes: When the number of currently active power modules is greater than the target number of active power modules, all power modules currently active are sorted from longest to shortest according to their historical cumulative running time, and the power module at the top of the sort is selected as the module to be put into hibernation. When the number of currently active power modules is less than the target number of active power modules, all power modules currently in hibernation are sorted from shortest to longest based on their historical cumulative running time, and the power module at the top of the sort is selected as the module to be woken up.

6. The method according to claim 3, characterized in that, The step of monitoring the voltage change rate of the bus and, when the voltage change rate exceeds a preset disturbance threshold, disabling the state switching command and forcibly maintaining or increasing the number of power modules on includes: Collect the instantaneous voltage value of the bus; Based on a series of consecutive instantaneous voltage values, the time derivative of the voltage is calculated as the rate of change of voltage; When the absolute value of the voltage change rate is greater than the disturbance threshold, the system is determined to be in a hot-plug disturbance state. Under the hot-swap disturbance state, a forced power-on command is immediately sent to all power modules in sleep mode, and a delay timer is started. Only after the delay timer expires and the voltage change rate returns to the normal range, the forced start command is released and the control logic based on the total load current data is restored.

7. The method according to claim 3, characterized in that, The method further includes: Receive load prediction data from the upper-layer task scheduling system, wherein the load prediction data includes the expected increase in power consumption and the expected arrival time; Calculate the remaining power margin of the currently active power modules; If the expected increase in power consumption is greater than the remaining power margin, a power-on command will be sent to the power module in sleep mode at a preset start time point before the expected arrival time.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, wherein the computer program is configured to perform the method described in any one of claims 3 to 7 when executed.

9. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to run the computer program to perform the method as described in any one of claims 3 to 7.

Images