Method, device, medium and product for energy efficiency optimization of network interface devices

Abstract

The application provides an energy efficiency optimization method, device, medium and product of a network interface device, and relates to the technical field of network interface devices. The method comprises the following steps: determining a predicted energy efficiency ratio of the network interface device according to real-time performance parameters of the network interface device; if it is detected that the network interface device is in an energy efficiency valley state according to the real-time performance parameters and the predicted energy efficiency ratio, determining a target voltage and a target frequency according to the real-time performance parameters and the predicted energy efficiency ratio, and adjusting the voltage and the frequency of the network interface device by using a dynamic working point jumping mechanism; and if it is detected that the network interface device is not in the energy efficiency valley state, and it is determined that the voltage and frequency adjustment mode is voltage and frequency increase or voltage and frequency decrease according to the real-time performance parameters and the predicted energy efficiency ratio, feeding back the adjustment of the voltage and the frequency of the network interface device according to the voltage and frequency adjustment mode until it is determined that the voltage and frequency adjustment mode is stop adjustment. The method of the application realizes linear and stable output of the energy efficiency ratio of the network interface device.

Description

Technical Field

[0001] This application relates to the field of network interface device technology, and in particular to a method, device, medium and product for optimizing the energy efficiency of a network interface device. Background Technology

[0002] Network interface devices are infrastructure computing platforms designed for software-defined networking, storage acceleration, and network security tasks. They are suitable for various application scenarios, including accelerating artificial intelligence (AI), hybrid cloud, high-performance computing, and 5G wireless networks.

[0003] In related technologies, the connection state (such as active or idle state) of a network interface device is dynamically switched based on its bandwidth utilization to achieve power consumption mode switching. However, this method ignores the possibility that the network interface device may be in an energy efficiency low-efficiency state, and thus ignores the nonlinear energy efficiency characteristics in the medium bandwidth utilization range. Therefore, related technologies struggle to achieve a linear and stable output of the energy efficiency ratio of the network interface device. Summary of the Invention

[0004] This application provides a method, device, medium, and product for optimizing the energy efficiency of a network interface device, which achieves linearized and stable output of the energy efficiency ratio of the network interface device.

[0005] This application provides a method for optimizing the energy efficiency of a network interface device, including:

[0006] Obtain real-time performance parameters of network interface devices;

[0007] The predicted energy efficiency ratio of the network interface device is determined based on its real-time performance parameters.

[0008] Based on real-time performance parameters and predicted energy efficiency ratios, detect whether network interface devices are in an energy efficiency gap.

[0009] If the network interface device is detected to be in an energy efficiency depression state, the target voltage and target frequency are determined based on real-time performance parameters and predicted energy efficiency ratio; where the target voltage and target frequency are the voltage and frequency of the network interface device when it leaves the energy efficiency depression state.

[0010] A dynamic operating point skipping mechanism is adopted to adjust the voltage and frequency of the network interface device according to the target voltage and target frequency;

[0011] If the network interface device is detected to be not in an energy efficiency low-level state, the voltage frequency adjustment method is determined based on real-time performance parameters and predicted energy efficiency ratio;

[0012] If the voltage and frequency adjustment method is determined to be either boost or buck, then the voltage and frequency of the network interface device will be adjusted according to the voltage and frequency adjustment method until the voltage and frequency adjustment method is determined to be stop.

[0013] This application also provides an energy efficiency optimization device for a network interface device, comprising:

[0014] The first acquisition module is used to acquire the real-time performance parameters of the network interface device;

[0015] The energy efficiency ratio calculation module is used to determine the predicted energy efficiency ratio of the network interface device based on the real-time performance parameters of the network interface device.

[0016] The energy efficiency gap detection module is used to detect whether network interface devices are in an energy efficiency gap state based on real-time performance parameters and predicted energy efficiency ratio.

[0017] The target value calculation module is used to determine the target voltage and target frequency based on real-time performance parameters and predicted energy efficiency ratio if the network interface device is detected to be in an energy efficiency depression state; wherein, the target voltage and target frequency are the voltage and frequency of the network interface device when it leaves the energy efficiency depression state;

[0018] The jump adjustment module is used to adjust the voltage and frequency of the network interface device according to the target voltage and target frequency by adopting a dynamic operating point jump mechanism.

[0019] The adjustment determination module is used to determine the voltage and frequency adjustment method based on real-time performance parameters and predicted energy efficiency ratio if the network interface device is detected to be not in an energy efficiency low-level state.

[0020] The adaptive adjustment module is used to adjust the voltage and frequency of the network interface device according to the voltage and frequency adjustment method if it is determined that the voltage and frequency adjustment method is boost or buck, until the voltage and frequency adjustment method is determined to be stop adjustment.

[0021] This application also provides an electronic device, including: a memory for storing a computer program; and a processor for implementing the energy efficiency optimization method of any of the above-described network interface devices when executing the computer program.

[0022] This application also provides a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, it implements the steps of the energy efficiency optimization method for any of the above-described network interface devices.

[0023] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the energy efficiency optimization method for any of the above-described network interface devices.

[0024] This application provides a method, device, medium, and product for optimizing the energy efficiency of a network interface device. The method includes: determining the predicted energy efficiency ratio of the network interface device based on its real-time performance parameters; if the network interface device is detected to be in an energy efficiency depression state based on the real-time performance parameters and the predicted energy efficiency ratio, then determining the target voltage and target frequency based on the real-time performance parameters and the predicted energy efficiency ratio, and adjusting the voltage and frequency of the network interface device using a dynamic operating point skipping mechanism; if the network interface device is not detected to be in an energy efficiency depression state, and the voltage and frequency adjustment method is determined to be either boost or buck based on the real-time performance parameters and the predicted energy efficiency ratio, then adjusting the voltage and frequency of the network interface device according to the voltage and frequency adjustment method until the voltage and frequency adjustment method is determined to be stop adjustment. The following technical effects were achieved: Through continuous prediction, detection, and feedback adjustment, the energy efficiency fluctuations of network interface devices under different loads were controlled, and the linearized and stable output of the energy efficiency ratio of network interface devices was realized; based on real-time performance parameters and predicted energy efficiency ratio, it was detected whether the network interface device was in an energy efficiency depression state, and through a dynamic operating point jump mechanism, the voltage and frequency of the network interface device were adjusted, realizing the active detection and avoidance of the energy efficiency depression state, thereby optimizing the nonlinear energy efficiency in the medium load area. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 A flowchart illustrating the energy efficiency optimization method for network interface devices provided in this application embodiment. Figure 1 ;

[0027] Figure 2 A flowchart illustrating the energy efficiency optimization method for network interface devices provided in this application embodiment. Figure 2 ;

[0028] Figure 3 A flowchart illustrating the energy efficiency optimization method for network interface devices provided in this application embodiment. Figure 3 ;

[0029] Figure 4 A schematic diagram of the structure of the energy efficiency optimization device for the network interface device provided in the embodiments of this application;

[0030] Figure 5 A schematic diagram of the structure of the electronic device provided in this application.

[0031] Figure label:

[0032] 410 - First Acquisition Module; 420 - Energy Efficiency Ratio Calculation Module; 430 - Low-lying Area Detection Module; 440 - Target Value Calculation Module; 450 - Jump Adjustment Module; 460 - Adjustment Determination Module; 470 - Adaptive Adjustment Module;

[0033] 510 - Processor; 520 - Memory; 530 - Communication components. Detailed Implementation

[0034] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the protection scope of this application.

[0035] It should be noted that, in the description of this application, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. The terms "first," "second," etc., in this application are used to distinguish similar objects and are not used to describe a specific order or sequence. The user information (including but not limited to user device information and user personal information) and data (including but not limited to data used for analysis, stored data, and displayed data) involved in one or more embodiments of this application are all information and data authorized by the user or fully authorized by the parties. Furthermore, the collection, use, and processing of related data must comply with relevant laws, regulations, and standards, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0036] To clearly understand the technical solution of this application, the solutions of the prior art will be described in detail first.

[0037] Network interface devices are infrastructure computing platforms designed for software-defined networking, storage acceleration, and network security tasks. They are suitable for various application scenarios, including accelerating artificial intelligence (AI), hybrid cloud, high-performance computing, and 5G wireless networks.

[0038] In related technologies, the connection state (such as active or idle state) of a network interface device is dynamically switched based on its bandwidth utilization to achieve power consumption mode switching. However, this method ignores the possibility that the network interface device may be in an energy-efficient low-efficiency state, and thus ignores the nonlinear energy efficiency characteristics in the medium bandwidth utilization range.

[0039] Furthermore, the related technologies rely on software interrupts to switch power consumption modes, resulting in high response latency and an inability to cope with microsecond-level traffic bursts.

[0040] Furthermore, the related technology employs an autoregressive integral moving average (ARIMA) model algorithm combined with dynamic voltage-frequency regulation (DVFS) to optimize energy efficiency. Specifically, this algorithm achieves a fixed proportional adjustment of voltage and frequency through a fixed multi-level voltage switching mechanism, such as 0.8V, 0.9V, 1.0V, 1.1V, and 1.2V. However, this algorithm suffers from slow prediction speed and poor flexibility in fixed voltage levels.

[0041] In summary, the relevant technologies cannot achieve a stable linear output of the energy efficiency ratio of network interface devices.

[0042] Therefore, in response to the above technical problems, the research found that, in order to solve this problem, when the network interface device is detected to be in an energy efficiency depression state, the inefficient operating point is intelligently skipped; otherwise, the voltage and frequency of the network interface device are independently adjusted through multi-voltage domain fast switching and dual-clock architecture to break through the linear constraint of DVFS; at the same time, a high-precision energy efficiency model is established through nonlinear multi-order polynomials to achieve dynamic prediction of the predicted energy efficiency ratio.

[0043] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0044] The specific application environment architecture or specific hardware architecture on which the energy efficiency optimization method for network interface devices depends is described here.

[0045] The core of the energy efficiency optimization method for network interface devices provided in this application embodiment lies in the server equipped with the network interface device. It is widely used in cutting-edge application scenarios with high network energy efficiency requirements. The network interface device can be a smart network card supporting adaptive voltage-frequency regulation (AVFS), a field-programmable gate array (FPGA), or an application-specific integrated circuit (ASIC), etc.; the server can be a data center server, an edge computing node, a high-performance computing cluster, or a virtualization platform, etc.

[0046] This method is applicable to accelerating artificial intelligence scenarios. Specifically, in AI training and inference clusters, servers frequently exchange large amounts of model parameters and data, which are characterized by high bursts and high bandwidth requirements. Servers equipped with network interface devices can predict bursts of traffic in real time and, combined with current temperature and packet characteristics, precisely adjust the voltage and frequency of the network interface devices to ensure sufficient performance during peak AI task periods and quickly enter an energy-efficient state during off-peak periods. This avoids energy inefficiency during medium loads, thereby significantly reducing the total cost of ownership for the entire AI cluster.

[0047] This method is also applicable to hybrid cloud scenarios. Specifically, hybrid cloud environments involve complex data migration, synchronization, and security policy enforcement between public and private clouds. Network interface devices play a crucial role in software-defined network storage and security acceleration. Servers equipped with network interface devices, through nonlinear multi-order polynomial modeling, can accurately adapt to the variable and unpredictable network load patterns in hybrid cloud scenarios, achieving a linear and stable output of energy efficiency ratios. This ensures the stability and high efficiency of cross-cloud services while optimizing cloud resource energy consumption.

[0048] This method is also applicable to high-performance computing (HPC) scenarios. Specifically, HPC applications typically require massively parallel computing, where inter-node communication becomes a performance bottleneck, making the line-speed processing capability of network interface devices crucial. Servers equipped with network interface devices leverage their low latency and high precision to dynamically adjust the power consumption of these devices, ensuring that they operate at their optimal energy efficiency point during long-duration, high-intensity computing tasks, effectively controlling data center heat dissipation pressure and operating costs.

[0049] This method is also applicable to 5G wireless network scenarios. Specifically, 5G core networks and edge computing nodes need to handle massive amounts of connection requests and data streams from mobile devices, with highly dynamic traffic patterns. Servers equipped with network interface devices can quickly respond to microsecond-level traffic bursts and effectively suppress temperature drift, ensuring high reliability and low latency services for 5G networks while improving the energy efficiency of network infrastructure.

[0050] Figure 1 A flowchart illustrating the energy efficiency optimization method for network interface devices provided in this application embodiment. Figure 1 .like Figure 1 As shown in the embodiments of this application, the executing entity can be an energy efficiency optimization device for a network interface device. This device can be located in an electronic device, and the device can be a server equipped with a network interface device. Therefore, the energy efficiency optimization method for a network interface device provided in this application embodiment includes the following steps:

[0051] S101. Obtain the real-time performance parameters of the network interface device.

[0052] Specifically, real-time performance parameters refer to multiple dimensions of parameters that can instantly reflect the current operating status of network interface devices. These parameters may include key indicators such as normalized bandwidth utilization, operating temperature, average packet length, queue depth, and normalized processor utilization.

[0053] S102. Determine the predicted energy efficiency ratio of the network interface device based on its real-time performance parameters.

[0054] Specifically, the predicted energy efficiency ratio (RER) refers to the energy efficiency level that a network interface device is expected to achieve under current real-time performance parameters, reflecting its processing power per unit of power consumption. Servers can determine the RER of network interface devices using machine learning models such as lightweight neural networks or random forests (which can be pre-trained using a large amount of historical running data), or through pre-configured nonlinear multi-order polynomials.

[0055] S105. Based on real-time performance parameters and predicted energy efficiency ratio, detect whether the network interface device is in an energy efficiency low-efficiency state.

[0056] Specifically, an "energy efficiency depression" state refers to a situation where a network interface device, when operating within a specific performance parameter threshold range, exhibits an actual energy efficiency ratio significantly lower than its theoretical optimal value, showing a marked decrease in energy efficiency. Servers operating in this state will experience increased energy consumption and decreased performance. The server determines whether a network interface device is in an energy efficiency depression state by considering real-time performance parameters, whether these parameters fall within the performance parameter threshold range, and whether the predicted energy efficiency ratio is significantly lower than the expected value under those real-time performance parameters.

[0057] If the network interface device is detected to be in an energy efficiency low state, then proceed to S106; if the network interface device is detected not to be in an energy efficiency low state, then proceed to S108.

[0058] S106. Determine the target voltage and target frequency based on real-time performance parameters and predicted energy efficiency ratio.

[0059] Specifically, the target voltage and target frequency are calculated to enable network interface devices to quickly escape energy efficiency deficits; that is, the voltage and frequency at which the network interface device escapes energy efficiency deficits. Once the server detects that the network interface device is in an energy efficiency deficit state, it will accurately calculate the target voltage and target frequency required to escape the energy efficiency deficit state based on the current real-time performance parameters and predicted energy efficiency ratio, using an energy efficiency gradient optimization algorithm.

[0060] S107. A dynamic operating point jumping mechanism is adopted to adjust the voltage and frequency of the network interface device according to the target voltage and target frequency.

[0061] Specifically, the dynamic operating point skipping mechanism is a voltage and frequency regulation strategy. When a network interface device is detected to be in an energy-inefficient state, the server quickly adjusts the voltage and frequency to pre-calculated target values, thereby effectively avoiding inefficient areas. During the voltage and frequency adjustment process, it can either directly adjust to the target value in one step, or adopt a multi-stage gradual skipping approach, rapidly reaching the target value in multiple steps to further reduce the risk of instantaneous current surges or clock instability caused by voltage or frequency mutations.

[0062] S108. Determine the voltage frequency adjustment method based on real-time performance parameters and predicted energy efficiency ratio.

[0063] Specifically, voltage and frequency adjustment methods refer to the direction of voltage and frequency adjustment determined based on real-time performance parameters and predicted energy efficiency ratios. These methods include three modes: boost and frequency increase, buck and frequency decrease, and no adjustment. Boost and frequency increase refers to enhancing performance by increasing voltage and frequency when the load increases; buck and frequency decrease refers to saving power by reducing voltage and frequency when the load decreases; and no adjustment indicates that the current voltage and frequency have reached or are close to optimal, requiring no further adjustment.

[0064] If the voltage frequency adjustment method is determined to be either boost or buck, then continue executing S109.

[0065] S109. Based on the voltage and frequency adjustment method, adjust the voltage and frequency of the network interface device according to feedback until the voltage and frequency adjustment method is determined to be "stop adjustment".

[0066] Specifically, if it is determined that a voltage boost / frequency increase or voltage buck / frequency decrease operation is needed, the server will gradually adjust the voltage and frequency accordingly and continuously monitor the adjustment effect. This process is a feedback mechanism, until the server determines that the current voltage and frequency have reached or are close to the optimal state, at which point it confirms that the voltage and frequency adjustment will stop.

[0067] The voltage and frequency can be adjusted by adaptive voltage and frequency regulation (AVFS), which dynamically adjusts the voltage and frequency based on real-time performance parameters, aiming to meet performance requirements while minimizing power consumption.

[0068] Voltage and frequency can also be adjusted using proportional-integral-derivative (PID) control. By analyzing energy efficiency deviations, their cumulative values, and rates of change, the voltage / frequency adjustment is calculated using a linear combination of proportional, integral, and derivative functions to achieve stable and precise feedback control.

[0069] Voltage and frequency can also be adjusted using fuzzy logic control, which uses fuzzy states of inputs such as energy efficiency deviation and temperature, along with preset expert rules, to reason and defuzzify, resulting in nonlinear voltage / frequency adjustment commands.

[0070] Voltage and frequency can also be adjusted using adaptive step size adjustment based on lookup tables (LUTs). This means that the corresponding voltage / frequency adaptive step size is retrieved from a preset multidimensional lookup table based on real-time performance parameters to achieve fast, offline optimized feedback adjustment.

[0071] Voltage and frequency can also be adjusted online using reinforcement learning (RL). This treats voltage / frequency adjustment as a decision-making action, and learns the optimal strategy online by interacting with network interface devices to obtain reward signals such as energy efficiency improvement, thereby maximizing long-term energy efficiency.

[0072] After executing S107 or S109, or after executing S108, if the voltage and frequency adjustment method is determined to be "stop adjustment," the server will return to the beginning of the loop, i.e., re-execute S101 to optimize the energy efficiency of the network interface device for the next time period, thereby achieving continuous closed-loop control. Simultaneously, the voltage and frequency parameters of the network interface device after this adjustment will directly affect the operating mode of the next loop, determining whether to execute dynamic jump decisions again or perform adaptive energy efficiency fine-tuning. This mechanism ensures that the network interface device always operates stably in an optimal or near-optimal energy efficiency state, achieving a linear and stable output of its energy efficiency ratio.

[0073] This application provides an energy efficiency optimization method for a network interface device, comprising: determining the predicted energy efficiency ratio of the network interface device based on its real-time performance parameters; if the network interface device is detected to be in an energy efficiency depression state based on the real-time performance parameters and the predicted energy efficiency ratio, then determining the target voltage and target frequency based on the real-time performance parameters and the predicted energy efficiency ratio, and adjusting the voltage and frequency of the network interface device using a dynamic operating point skipping mechanism; if the network interface device is not detected to be in an energy efficiency depression state, and the voltage and frequency adjustment method is determined to be either boost or buck based on the real-time performance parameters and the predicted energy efficiency ratio, then adjusting the voltage and frequency of the network interface device according to the voltage and frequency adjustment method until the voltage and frequency adjustment method is determined to be stop adjustment. The following technical effects were achieved: Through continuous prediction, detection, and feedback adjustment, the energy efficiency fluctuations of network interface devices under different loads were controlled, and the linearized and stable output of the energy efficiency ratio of network interface devices was realized; based on real-time performance parameters and predicted energy efficiency ratio, it was detected whether the network interface device was in an energy efficiency depression state, and through a dynamic operating point jump mechanism, the voltage and frequency of the network interface device were adjusted, realizing the active detection and avoidance of the energy efficiency depression state, thereby optimizing the nonlinear energy efficiency in the medium load area.

[0074] Figure 2 A flowchart illustrating the energy efficiency optimization method for network interface devices provided in this application embodiment. Figure 2 .like Figure 2 As shown, in one possible design, S109 includes adjusting the voltage of the network interface device based on the voltage frequency adjustment method, which includes:

[0075] S201. Based on the voltage frequency adjustment method, determine the adjusted target voltage domain from multiple preset voltage domains.

[0076] Each voltage domain corresponds to an independent power supply network.

[0077] If the current original voltage domain is found to be inconsistent with the target voltage domain, then continue to execute S202; otherwise, end the adjustment.

[0078] S202. Switch to the power supply network corresponding to the target voltage domain for power supply.

[0079] Specifically, a power supply network refers to a physically isolated circuit path that provides power to a specific voltage domain. It includes independent power lines, voltage regulators, and control logic to ensure electrical isolation between different voltage domains and prevent mutual interference. A voltage domain, on the other hand, divides the power supply system of network interface devices into multiple independent power supply areas with different voltage levels. Each voltage domain can be independently turned on, off, or its output voltage can be adjusted.

[0080] Based on a predetermined voltage frequency adjustment method, the server selects a voltage domain from a set of preset voltage domains that matches the required voltage level as the target voltage domain. For example, if a voltage boost is needed, the next voltage domain with a higher voltage level than the current domain is selected; conversely, if a voltage buck is needed, the next voltage domain with a lower voltage level than the current domain is selected. After determining the target voltage domain, the server checks whether the current voltage domain and the target voltage domain are consistent. If they are inconsistent, the server initiates a power switching process, switching the power supply source for the network interface devices from the power supply network corresponding to the current voltage domain to the power supply network corresponding to the target voltage domain, where the latter will provide power.

[0081] The technical effect of this application embodiment is that by dividing the power supply system into multiple independent voltage domains, independent and dynamic voltage adjustment requirements are achieved.

[0082] In one possible design, S202 includes switching to the power supply network corresponding to the target voltage domain for power supply, including:

[0083] S2021. Precharge the power supply network corresponding to the target voltage domain until the preset first duration is reached.

[0084] S2022. Switch to the power supply network corresponding to the target voltage domain for power supply, and disconnect the power supply connection of the power supply network corresponding to the original voltage domain after a preset second time period after the switch.

[0085] Specifically, by pre-charging, formal switching, and ending pre-charging, the impact of power switching on the stability of network interface devices is reduced, preventing logic errors or system resets caused by voltage surges. The first duration can be 100ns; after the first duration of pre-charging, the charge pre-balance of the target voltage domain is determined. The second duration can be 10ns; after the second duration, the power supply network corresponding to the original voltage domain is disconnected, thus ensuring that the power switch of the target voltage domain is completely stable.

[0086] This operation can be implemented by the voltage domain switching controller using the hardware description language Verilog, and the relevant code can be represented as follows:

[0087] verilog

[0088] / / Voltage Domain Switching Controller

[0089] always @(posedge clk) begin / / Executes on the rising edge of clock clk

[0090] if (target_domain != current_domain) begin

[0091] enable_precharge <= 1; / / Initiate the precharge phase

[0092] #100ns; / / Charge pre-balance

[0093] switch_control <= target_domain; / / Initiate the actual switchover

[0094] #10ns;

[0095] enable_precharge <= 0; / / End the precharge phase

[0096] end

[0097] end

[0098] Here, target_domain refers to the target voltage domain, and current_domain refers to the original voltage domain.

[0099] In one possible design, there are multiple power supply networks, including: a first power supply network, a second power supply network, and a third power supply network;

[0100] The power supply types of the first power supply network include: low dropout linear regulators and switched capacitors;

[0101] The power supply types of the second power supply network include: multiphase buck converters;

[0102] The power supply types of the third power supply network include: direct-connected power management integrated circuits;

[0103] The minimum voltage of the voltage domain corresponding to the second power supply network is greater than the maximum voltage of the voltage domain corresponding to the first power supply network.

[0104] The maximum voltage of the voltage domain corresponding to the second power supply network is less than the minimum voltage of the voltage domain corresponding to the third power supply network.

[0105] Specifically, a low-dropout linear regulator (LDO) is a linear regulated power supply that uses a variable resistor to dissipate excess energy between the input and output voltages, thus providing a stable, low-noise output voltage. A switched-capacitor is an inductorless DC-DC converter that uses the charging and discharging of a capacitor to transfer energy and achieves boost, buck, or inverting voltage conversion by changing the capacitor's connection method. A multiphase buck converter is a high-performance switching-mode power supply consisting of multiple parallel phases operating in an interleaved manner. A direct-connect power management integrated circuit (PMIC) refers to a power management integrated circuit that is mounted next to the network interface device and directly powered through short, wide printed circuit board (PCB) traces or interconnects within the package.

[0106] Table 1 is a parameter table of multiple power supply networks provided in the embodiments of this application.

[0107] Table 1:

[0108]

[0109] The technical effect of this application embodiment is that the power supply requirements under different voltage ranges are realized through the first power supply network, the second power supply network and the third power supply network.

[0110] Figure 3 A flowchart illustrating the energy efficiency optimization method for network interface devices provided in this application embodiment. Figure 3 .like Figure 3 As shown, in one possible design, S109 includes adjusting the frequency of the network interface device based on the voltage frequency adjustment method, including:

[0111] S301. Obtain frequency stepping information used to indicate the frequency adjustment step size.

[0112] If the frequency adjustment step size is greater than the preset step size threshold, it means that a larger frequency adjustment is needed, so continue to execute S302; if the frequency adjustment step size is less than or equal to the step size threshold, it means that fine frequency adjustment is needed, so continue to execute S303.

[0113] S302. Adjust the frequency division coefficient of the pre-configured phase-locked loop module according to the frequency adjustment step size.

[0114] S303. Adjust the step size according to the frequency to adjust the delay time of the pre-configured delay lock-in loop module.

[0115] After S302 or S303 is executed, continue to execute S304.

[0116] S304. Adjust the frequency of the network interface device through a phase-locked loop module or a time-locked loop module.

[0117] Specifically, the frequency adjustment step size refers to the specific amount of change or step size (ΔF) required for this frequency adjustment calculated by the server. This value is derived by the energy efficiency optimization algorithm based on factors such as the current load, temperature, and predicted energy efficiency ratio.

[0118] The step size threshold is a pre-set threshold for frequency change, used to distinguish between large and small step size adjustments. When the frequency adjustment step size is greater than the step size threshold, a method with lower precision but a larger range is used to adjust the frequency; otherwise, a method with higher precision but a smaller range is used to adjust the frequency. For example, the step size threshold can be 2%.

[0119] A phase-locked loop (PLL) module is a circuit used to generate and regulate clock signals. By adjusting its internal divider, the output clock frequency can be changed. PLLs typically achieve a wide range of frequency regulation, but the adjustment precision is relatively low. For example, a PLL can achieve a frequency adjustment of ±15% with a response time of <100ns.

[0120] The division factor is a configurable parameter in a PLL that determines the multiplication or division factor of the input reference clock, directly controlling the output frequency. The division factor is negatively correlated with the frequency adjustment step size; to achieve a larger frequency increase, the division factor usually needs to be reduced.

[0121] A delay-locked loop (DLL) is a circuit used to precisely control the phase and delay of a clock signal. By adjusting its internal delay time, the clock period can be fine-tuned. DLLs can achieve high-precision frequency fine-tuning within a small range, but the adjustment range is limited; the frequency adjustment accuracy of a DLL is lower than that of a PLL. For example, a DLL can compensate for ±2% jitter, achieving an accuracy of 0.1%, with a delay of <20ns.

[0122] The frequency divider is a configurable parameter in a DLL (Dynamic Logic Controller). By increasing or decreasing the delay time, the clock cycle can be fine-tuned. The delay time is positively correlated with the frequency adjustment step size; to increase the frequency, the delay time usually needs to be decreased.

[0123] Adjusting the frequency of a network interface device via a PLL or DLL can be achieved using a clock adaptive regulator through the hardware description language SystemVerilog. The relevant code can be represented as follows:

[0124] systemverilog

[0125] module clock_adapter (

[0126] input [3:0] freq_step, / / 0.5% per step, achieving frequency adjustment step size from 0% to 7.5%.

[0127] output reg clk_out

[0128] / / There are 4 inputs freq_step and one output clk_out

[0129] always @(*) begin

[0130] if (|freq_step[3:2]) begin / / Step size > 2%

[0131] pll_mode <= 1; / / Enable PLL mode

[0132] pll_div <= 64 - freq_step;

[0133] end else begin

[0134] dll_mode <= 1; / / Enable DLL mode

[0135] dll_delay <= freq_step[1:0] * 2;

[0136] end

[0137] end

[0138] endmodule

[0139] Here, freq_step refers to the frequency adjustment step size, pll_div refers to the frequency division coefficient, and dll_delay refers to the frequency division coefficient.

[0140] The technical effect of this application embodiment is that by using a hierarchical adjustment strategy, the wide-range adjustment capability of the phase-locked loop module and the high-precision fine-tuning capability of the delay-locked loop module are combined, enabling the server to cope with drastic load changes and achieve fine optimization of energy efficiency, taking into account both dynamic response and steady-state accuracy.

[0141] It should be noted that since the voltage adjustment and frequency adjustment of the network interface device are decoupled, the execution order of S201 to S202 and S301 to S304 is not limited in this embodiment. S201 to S202 can be executed first, followed by S301 to S304; or S301 to S304 can be executed first, followed by S201 to S202; or S201 to S202 and S301 to S304 can be executed simultaneously. Furthermore, when executing S301 to S304, only PLL mode can be enabled, only DLL mode can be enabled, or PLL mode can be enabled first, followed by DLL mode.

[0142] In one possible design, S102 includes:

[0143] The real-time performance parameters are input into a pre-configured nonlinear multi-order polynomial energy efficiency model, which outputs the predicted energy efficiency ratio of the network interface device. The nonlinear multi-order polynomial energy efficiency model is used to describe the nonlinear relationship between the predicted energy efficiency ratio and the real-time performance parameters.

[0144] Furthermore, real-time performance parameters include: normalized bandwidth utilization U, operating temperature T, and average packet length S.

[0145] Normalized bandwidth utilization U refers to the bandwidth utilization rate normalized to the interval [0,1]. Bandwidth utilization rate is the ratio of the currently used bandwidth to its theoretical maximum bandwidth. When U=0, it indicates that the network interface device is completely idle; when U=1, it indicates that the network interface device is in a fully loaded saturated state.

[0146] Operating temperature T refers to the core temperature of a network interface device during operation, measured in degrees Celsius (°C), and it directly affects the physical characteristics of semiconductor devices.

[0147] Average packet length S refers to the average length of all data packets transmitted through a network interface device over a period of time, measured in bytes. Average packet length is used to enable nonlinear multi-order polynomial energy efficiency models to perceive and adapt to different network traffic patterns.

[0148] The nonlinear multi-order polynomial energy efficiency model is expressed as:

[0149]

[0150] in, The predicted energy efficiency ratio of the network interface device; k is an integer between 0 and 3, and k can take the values ​​0, 1, 2 and 3 in sequence; These are the preset coefficients; This is a preset function related to the operating temperature; It is the load change rate obtained based on the normalized bandwidth utilization.

[0151] It is a third-order polynomial concerning normalized bandwidth utilization. It can be a linear function or a quadratic function, etc., used to dynamically reflect the influence of operating temperature on the nonlinear characteristics of energy efficiency.

[0152] This term describes the impact of average packet length on energy efficiency and is a packet length penalty term designed to match the direct memory access (DMA) engine characteristics of network interface devices. It reflects how the energy efficiency gain gradually saturates as the average packet length increases, until it approaches its maximum value. .

[0153] It is the absolute value of the gradient, used to suppress false triggering of negative gradients.

[0154] Furthermore, the nonlinear multi-order polynomial energy efficiency model can also be expressed as:

[0155] math

[0156]

[0157] In one possible design, the coefficients are calibrated, and the calibration results are shown in Table 2. Table 2 is a parameter table for the calibration of each coefficient provided in the embodiments of this application.

[0158] Table 2:

[0159]

[0160] The technical effect of this application embodiment is that, based on the normalized bandwidth utilization, operating temperature and average packet length, the predicted energy efficiency ratio of the network interface device is determined through a nonlinear multi-order polynomial energy efficiency model, thereby improving the accuracy of the predicted energy efficiency ratio.

[0161] In one possible design, after executing S102, the method further includes:

[0162] S103. Obtain the actual energy efficiency ratio.

[0163] If the error between the predicted energy efficiency ratio and the actual energy efficiency ratio is greater than the preset error threshold, it is determined that the nonlinear multi-order polynomial energy efficiency model needs to be calibrated, and then S104 is executed; otherwise, it is determined that the nonlinear multi-order polynomial energy efficiency model is accurate enough, and then S105 is executed.

[0164] S104. Iteratively update based on the error and the preset learning rate. Continue until the error is less than or equal to the error threshold.

[0165] in, .

[0166] Specifically, the error threshold is a pre-set, acceptable maximum error value, which can be 3%. The learning rate is used to control the step size or speed of model parameter updates. The larger the learning rate, the faster the model responds to new data, but it may lead to insufficient stability; conversely, the smaller the learning rate, the smoother the parameter updates and the more stable the convergence process, but the slower the speed.

[0167] The server then uses hardware sensors or performance counters to measure the actual power consumption and performance of the network interface device, and calculates the actual energy efficiency ratio (EER). It then calculates the difference between the predicted EER and the actual EER, i.e., the error between the two. If it is determined that the nonlinear multi-order polynomial energy efficiency model needs calibration, the parameters in the model are adjusted based on the current error magnitude and the preset learning rate.

[0168] The update rule can be expressed as:

[0169]

[0170] in, The adjustment amount, the adjusted sum. This represents the actual energy efficiency ratio. For learning rate, The value can be between 0.00005 and 0.0002, for example, it can be 0.0001; The previous minus sign was used to implement negative feedback regulation; specifically, if... , If it is a negative number, the adjusted value is... The value will decrease, causing the model's next prediction to decrease and move closer to the actual value; conversely, the model's next prediction will increase.

[0171] Furthermore, The update rule can also be expressed as:

[0172] math

[0173] \Delta a_1 = -0.0001 \times (\eta_{pred} - \eta_{meas})

[0174] The technical effect of this application embodiment is that by introducing negative feedback adjustment based on the actual energy efficiency ratio, the long-term effectiveness of the nonlinear multi-order polynomial energy efficiency model after actual deployment is improved.

[0175] In one possible design, the learning rate is iteratively updated based on the error and a preset learning rate. and This continues until the error is less than or equal to the respective error threshold.

[0176] The update methods for the above parameters are the same as those for... The update method is similar, and will not be described again in the embodiments of this application.

[0177] In one possible design, if the error between the predicted energy efficiency ratio and the actual energy efficiency ratio is greater than a preset error threshold, and it is determined that the voltage and frequency adjustment of the network interface device does not meet expectations, then a rollback mechanism is executed.

[0178] The rollback mechanism is used to quickly restore the system to a known, stable, and efficient operating state to mitigate risks and ensure service quality. Specifically, before each voltage / frequency adjustment, the server automatically records the current critical operating state as a baseline. If a rollback is initiated, the voltage and frequency of the network interface devices are immediately restored to the baseline state recorded before the adjustment. Furthermore, after the rollback is executed, the server remeasures the actual energy efficiency ratio and performance to confirm that a stable state has been achieved.

[0179] The technical effect of this application embodiment is that if the error between the predicted energy efficiency ratio and the actual energy efficiency ratio is greater than the preset error threshold, the stability and reliability of the network interface device are improved through the rollback mechanism.

[0180] In one possible design, when performing AVFS adjustment, S108 includes:

[0181] S1081. Based on the predicted energy efficiency ratio, the energy efficiency gradient is obtained.

[0182] If the energy efficiency gradient is less than the preset first gradient threshold and the load change rate is greater than the preset change rate threshold, then execute S1082; if the energy efficiency gradient is greater than the preset second gradient threshold and the operating temperature is greater than the preset temperature threshold, then execute S1083.

[0183] S1082. Determine the voltage frequency adjustment method as boost and frequency boost.

[0184] The first gradient threshold is negative.

[0185] S1083. Determine the voltage and frequency adjustment method as voltage reduction and frequency reduction.

[0186] The second gradient threshold is a positive number and is less than the absolute value of the first gradient threshold.

[0187] Specifically, the energy efficiency gradient refers to the rate of change of the predicted energy efficiency ratio relative to the normalized bandwidth utilization, i.e., the partial derivative. It means that when U changes slightly, The changing trend and rate. Among them, the positive gradient This means that increasing U will improve Network interface devices are in a region where energy efficiency increases with load; negative gradient. This means that an increase in U will decrease Network interface devices are in a region where energy efficiency decreases as load increases.

[0188] The first gradient threshold is a pre-set negative energy efficiency gradient threshold, which can take the value of -0.01, used to determine whether the network interface device is in the region where energy efficiency decreases as load increases; the second gradient threshold is a pre-set positive energy efficiency gradient threshold, which can take the value of 0.005, used to determine whether the network interface device is in the region where energy efficiency increases as load increases; the rate of change threshold is a pre-set load change rate threshold, which can take the value of 0. When the load change rate is greater than the rate of change threshold, it is considered that the load is increasing rapidly; the temperature threshold is a pre-set higher operating temperature threshold, which can take the value of 85°C. When the operating temperature is greater than the temperature threshold, it indicates that the heat dissipation pressure is high, there is thermal throttling or reliability risk, and cooling should be given priority.

[0189] The server checks whether the following two conditions are met simultaneously: Condition 1 is that the energy efficiency gradient is less than the first gradient threshold, and condition 2 is that the load change rate is greater than the change rate threshold. If both conditions are met, the voltage and frequency adjustment method is determined to be boost and frequency boost. The purpose is to significantly improve performance to break out of the inefficiency zone and meet sudden performance demands.

[0190] The server checks whether the following two additional conditions are met simultaneously: Condition 1 is that the energy efficiency gradient is greater than the second gradient threshold, and Condition 2 is that the operating temperature is greater than the temperature threshold. If both conditions are met, the voltage and frequency adjustment method is determined to be "voltage and frequency reduction". The purpose is to actively cool down by reducing power consumption to prevent overheating. At the same time, since energy efficiency is still increasing, a small frequency reduction will not immediately lead to a significant drop in energy efficiency.

[0191] The above conditional judgment can also be expressed as:

[0192] math

[0193] Decision function = case studies

[0194] \text{ } & \text{if} \frac{\partial \eta}{\partial U} < -0.01 \text{ AND} \frac{dU}{dt} > 0 \\

[0195] \text{Buck voltage reduction and frequency reduction} & \text{if} \frac{\partial \eta}{\partial U} > 0.005\text{ OR} T > 85℃

[0196] \end{cases}

[0197] The technical effect of this application embodiment is that, by using the energy efficiency gradient, combined with the load change rate and operating temperature, the voltage frequency adjustment method is determined to be either boosting and increasing frequency or bucking and decreasing frequency.

[0198] In one possible design, the formula for calculating the load change rate is expressed as:

[0199]

[0200] in, The sampling period; Let be the normalized bandwidth utilization at time t; Let be the normalized bandwidth utilization rate three sampling periods prior to time t.

[0201] In addition, to achieve noise filtering, if If the result is positive, output the calculated value; otherwise, output 0.

[0202] The formula for calculating the load change rate can also be expressed as:

[0203] def efficiency_gradient(U_history):

[0204] grad=(U_history[-1]-U_history[-4]) / (3*Δt)#3 point difference

[0205] return grad if abs(grad) > 0.001 else 0

[0206] In one possible design, when performing dynamic operating point jumping, the method further includes:

[0207] If the normalized bandwidth utilization rate is within the preset utilization threshold range, and the predicted energy efficiency ratio is less than the preset energy efficiency ratio threshold, then it is determined that the network interface device is in an energy efficiency low-end state.

[0208] Accordingly, S106 determines the target voltage based on real-time performance parameters and predicted energy efficiency ratio, including:

[0209] If the load change rate is positive, execute S1061; if the load change rate is negative, execute S1062.

[0210] S1061. Continuously increase the normalized bandwidth utilization rate by the first utilization rate threshold until the modified normalized bandwidth utilization rate is outside the utilization rate threshold range.

[0211] S1062. Continuously subtract the second utilization threshold from the normalized bandwidth utilization until the modified normalized bandwidth utilization is outside the utilization threshold range.

[0212] After executing S1061 or S1062, continue executing S1063.

[0213] S1063. Based on the modified normalized bandwidth utilization and predicted energy efficiency ratio, the target voltage is obtained.

[0214] Specifically, the utilization threshold range is a pre-defined range of utilization values, such as [20%, 70%]. This range is identified as an area where network interface devices are prone to falling into an energy efficiency trough. The first utilization threshold and the second utilization threshold represent two pre-defined utilization increment values, where the first utilization threshold can be set to 15% and the second utilization threshold can be set to 10%. Both thresholds serve as step adjustments to move out of the energy efficiency trough. Specifically, the first utilization threshold is used to increase the load, while the second utilization threshold is used to decrease the load.

[0215] If the load change rate is positive, the server believes the load is likely to continue increasing. To escape the energy efficiency low point, it adopts an upward escape strategy. This involves continuously increasing the current U value by the first utilization threshold until the modified U exceeds the maximum value of the utilization threshold range.

[0216] If the load change rate is negative, the server believes that the load is likely to continue to decrease. In this case, it will continuously subtract the second utilization threshold from the current U value until the modified U is less than the minimum value of the utilization threshold range.

[0217] Determining the state of energy efficiency deficiencies can be implemented using the C language, and the relevant code can be represented as follows:

[0218] c

[0219] if (U_current ∈ [20%,70%] && η_current < η_threshold) {

[0220] target_U = (dU_dt > 0) ? U_current + 15% : U_current - 10%;

[0221] enforce_voltage_freq(target_U);

[0222] }

[0223] After obtaining the modified normalized bandwidth utilization, the target voltage required to support this modified normalized bandwidth utilization can be derived in reverse through model calculation or table lookup.

[0224] Furthermore, in the voltage-related timing control of the network interface device adjusted according to the target voltage, the voltage domain switching delay is 18µs, the frequency locking delay is 2µs, and the stability detection delay is 5µs.

[0225] The technical effect of this application embodiment is that by intelligently selecting the escape direction and calculating the target voltage, the network interface device is freed from the energy efficiency depression state, thereby improving the overall energy efficiency performance of the network interface device.

[0226] In one possible design, after executing S106, which involves determining the target voltage based on real-time performance parameters and the predicted energy efficiency ratio, the method further includes:

[0227] S1064. Obtain the compensation voltage based on the operating temperature and the preset nominal voltage.

[0228] S1065. Correct the target voltage based on the compensation voltage.

[0229] Specifically, the nominal voltage is the rated operating voltage at the reference temperature; for example, if the reference temperature is 25°C, the corresponding nominal voltage is 0.9V. The compensation voltage is positively correlated with the difference between the operating temperature and the reference temperature; if the operating temperature is higher than the reference temperature, the compensation voltage is usually positive, indicating that the voltage needs to be increased to compensate for the degradation of transistor performance at high temperatures and maintain timing correctness; if the operating temperature is lower than the reference temperature, the compensation voltage is usually negative or zero, indicating that the voltage can be reduced because transistor performance is better at low temperatures.

[0230] Compensation voltage The formula for calculating can be expressed as:

[0231]

[0232] in, Nominal voltage, Reference temperature; This is a preset compensation coefficient, representing the amount of voltage adjustment required for every 1°C change in temperature. The value can be 0.8mV / ℃.

[0233] Compensation voltage The calculation formula can also be expressed as:

[0234] math

[0235] V_{adj} = V_{nom} + (T - 25℃) \times 0.0008 \text{ V / ℃}

[0236] The technical effect of this application embodiment is that by introducing a temperature compensation voltage to correct the initial target voltage, the practicality and robustness of the network interface device are improved.

[0237] In one possible design, when performing dynamic operating point jumping, the method further includes:

[0238] The frequency threshold is obtained based on the voltage of the network interface device.

[0239] Among them, the frequency threshold is positively correlated with the voltage of the network interface device.

[0240] If the frequency of the network interface device exceeds the frequency threshold, a preset security protection mechanism will be triggered.

[0241] Specifically, the frequency threshold is the upper limit of the highest clock frequency that the network interface device can operate safely and stably, which is dynamically calculated based on the current power supply voltage.

[0242] The server uses a hardware watchdog to check for illegal combinations of voltage and frequency every 1µs. Based on the voltage of the network interface device, it dynamically calculates the safe frequency threshold corresponding to the current voltage using a preset voltage-frequency mapping relationship, lookup table, or mathematical function. This mapping relationship is usually derived from the process, voltage, and temperature characteristics of the network interface device to ensure that all timing paths can meet the setup / hold time requirements under a given voltage.

[0243] If the frequency of the network interface device exceeds a frequency threshold, a preset security protection mechanism is triggered. This security protection mechanism is a pre-defined set of emergency measures to prevent damage to the network interface device due to overload or instability. The security protection mechanism may include:

[0244] Forced frequency reduction means immediately reducing the operating frequency to a safe level.

[0245] Voltage boosting refers to temporarily increasing the supply voltage to match the current high-frequency demand, if the power supply allows.

[0246] Alarms and logging: This involves sending alarms to the system management software and recording event logs.

[0247] Traffic throttling refers to temporarily reducing the data processing rate to alleviate the load on the equipment.

[0248] System reset is a last resort to restart the device in extremely unstable situations.

[0249] The technical effect of this application embodiment is that by establishing a positive correlation between voltage and frequency threshold, the frequency can be monitored in real time to prevent system instability caused by insufficient voltage.

[0250] In one possible design, the frequency threshold The formula for calculating can be expressed as:

[0251]

[0252] Where V refers to the voltage of the network interface device.

[0253] The formula for calculating the frequency threshold can also be expressed as:

[0254] math

[0255] f_{max}(V) = \begin{cases}

[0256] 0.8 \text{GHz} & \text{if} V < 0.7v \\

[0257] 1.5 \text{GHz} & \text{if} V \geq 1.0v

[0258] \end{cases}

[0259] Table 3 shows the parameters of the energy efficiency optimization method and the autoregressive integral moving average model algorithm for the network interface device provided in the embodiments of this application. As shown in Table 3, the parameters of the two methods in terms of dealing with burst traffic, energy efficiency ratio stability, control temperature drift, and control hardware resource usage are presented. It can be seen that the energy efficiency optimization method for the network interface device has a significant improvement over the autoregressive integral moving average model algorithm.

[0260] Table 3:

[0261]

[0262] Figure 4 This is a schematic diagram of the energy efficiency optimization device for a network interface device provided in an embodiment of this application. Figure 4 As shown in this embodiment, the energy efficiency optimization device for the network interface device can be located in the electronic device, and the device includes:

[0263] The first acquisition module 410 is used to acquire real-time performance parameters of the network interface device;

[0264] The energy efficiency ratio calculation module 420 is used to determine the predicted energy efficiency ratio of the network interface device based on the real-time performance parameters of the network interface device.

[0265] The low-efficiency detection module 430 is used to detect whether the network interface device is in an energy efficiency low-efficiency state based on real-time performance parameters and predicted energy efficiency ratio;

[0266] The target value calculation module 440 is used to determine the target voltage and target frequency based on real-time performance parameters and predicted energy efficiency ratio if the network interface device is detected to be in an energy efficiency depression state; wherein, the target voltage and target frequency are the voltage and frequency of the network interface device when it leaves the energy efficiency depression state;

[0267] The jump adjustment module 450 is used to adjust the voltage and frequency of the network interface device according to the target voltage and target frequency using a dynamic operating point jump mechanism.

[0268] The adjustment determination module 460 is used to determine the voltage frequency adjustment method based on real-time performance parameters and predicted energy efficiency ratio if the network interface device is detected to be not in an energy efficiency depression state.

[0269] The adaptive adjustment module 470 is used to adjust the voltage and frequency of the network interface device according to the voltage and frequency adjustment method if the voltage and frequency adjustment method is determined to be either boost or buck, until the voltage and frequency adjustment method is determined to be stop adjustment.

[0270] The energy efficiency optimization device for network interface equipment provided in this application embodiment can perform... Figure 1 The technical solution of the method embodiment shown has the same implementation principle and technical effect as... Figure 1 The methods shown in the embodiments are similar, and will not be described again in the embodiments of this application.

[0271] Meanwhile, the energy efficiency optimization device for network interface equipment provided in this application embodiment is further refined based on the energy efficiency optimization device for network interface equipment provided in the previous application embodiment.

[0272] In one possible design, the adaptive adjustment module 470 includes:

[0273] The voltage domain selection module is used to determine the target voltage domain after adjustment from multiple preset voltage domains according to the voltage frequency adjustment method; wherein each voltage domain corresponds to an independent power supply network.

[0274] The voltage domain switching module is used to switch to the power supply network corresponding to the target voltage domain if the current original voltage domain is detected to be inconsistent with the target voltage domain.

[0275] In one possible design, the voltage domain switching module includes:

[0276] The pre-charge module is used to pre-charge the power supply network corresponding to the target voltage domain until the preset first duration is reached.

[0277] The network switching module is used to switch to the power supply network corresponding to the target voltage domain for power supply, and disconnect the power supply connection of the power supply network corresponding to the original voltage domain after a preset second time period after switching.

[0278] In one possible design, there are multiple power supply networks, including: a first power supply network, a second power supply network, and a third power supply network;

[0279] The power supply types of the first power supply network include: low dropout linear regulators and switched capacitors;

[0280] The power supply types of the second power supply network include: multiphase buck converters;

[0281] The power supply types of the third power supply network include: direct-connected power management integrated circuits;

[0282] The minimum voltage of the voltage domain corresponding to the second power supply network is greater than the maximum voltage of the voltage domain corresponding to the first power supply network.

[0283] The maximum voltage of the voltage domain corresponding to the second power supply network is less than the minimum voltage of the voltage domain corresponding to the third power supply network.

[0284] In one possible design, the adaptive adjustment module 470 includes:

[0285] The step determination module is used to acquire frequency step information that indicates the frequency adjustment step size;

[0286] The phase-locked loop adjustment module is used to adjust the frequency division coefficient of the pre-configured phase-locked loop module according to the frequency adjustment step size if the frequency adjustment step size is greater than the preset step size threshold; wherein, the frequency division coefficient is negatively correlated with the frequency adjustment step size;

[0287] The delay-locked loop adjustment module is used to adjust the delay time of the pre-configured delay-locked loop module according to the frequency adjustment step size if the frequency adjustment step size is less than or equal to the step size threshold; wherein, the delay time is positively correlated with the frequency adjustment step size; the frequency adjustment accuracy of the phase-locked loop module is lower than that of the delay-locked loop module.

[0288] The frequency adjustment module is used to adjust the frequency of network interface devices via a phase-locked loop module or a time-locked loop module.

[0289] In one possible design, the energy efficiency ratio calculation module 420 is used to input real-time performance parameters into a pre-configured nonlinear multi-order polynomial energy efficiency model and output the predicted energy efficiency ratio of the network interface device; wherein, the nonlinear multi-order polynomial energy efficiency model is used to describe the nonlinear relationship between the predicted energy efficiency ratio and the real-time performance parameters.

[0290] In one possible design, real-time performance parameters include: normalized bandwidth utilization U, operating temperature T, and average packet length S;

[0291] The nonlinear multi-order polynomial energy efficiency model is expressed as:

[0292]

[0293] in, Predicted energy efficiency ratio for network interface devices; These are the preset coefficients; This is a preset function related to the operating temperature; It is the load change rate obtained based on the normalized bandwidth utilization.

[0294] One possible design also includes:

[0295] The second acquisition module is used to acquire the actual energy efficiency ratio;

[0296] The parameter update module is used to iteratively update the parameters based on the error and a preset learning rate if the error between the predicted energy efficiency ratio and the actual energy efficiency ratio exceeds a preset error threshold. Continue until the error is less than or equal to the error threshold.

[0297] In one possible design, the adjustment determining module 460 includes:

[0298] The gradient calculation module is used to obtain the energy efficiency gradient based on the predicted energy efficiency ratio.

[0299] The first determining module is used to determine the voltage frequency adjustment method as boost and frequency increase if the energy efficiency gradient is less than a preset first gradient threshold and the load change rate is greater than a preset change rate threshold; wherein, the first gradient threshold is a negative number.

[0300] The second determining module is used to determine the voltage frequency adjustment method as voltage reduction and frequency reduction if the energy efficiency gradient is greater than the preset second gradient threshold and the operating temperature is greater than the preset temperature threshold; wherein the second gradient threshold is a positive number and the absolute value of the second gradient threshold is less than the first gradient threshold.

[0301] One possible design also includes:

[0302] The third determining module is used to determine that the network interface device is in an energy efficiency depression state if the normalized bandwidth utilization rate is within a preset utilization rate threshold range and the predicted energy efficiency ratio is less than a preset energy efficiency ratio threshold.

[0303] Accordingly, the target value calculation module 440 includes:

[0304] The first modification module is used to continuously increase the normalized bandwidth utilization by a first utilization threshold if the load change rate is positive, until the modified normalized bandwidth utilization is outside the utilization threshold range.

[0305] The second modification module is used to continuously subtract the second utilization threshold from the normalized bandwidth utilization if the load change rate is negative, until the modified normalized bandwidth utilization is outside the utilization threshold range.

[0306] The voltage calculation module is used to obtain the target voltage based on the modified normalized bandwidth utilization and the predicted energy efficiency ratio.

[0307] One possible design also includes:

[0308] The compensation calculation module is used to obtain the compensation voltage based on the operating temperature and the preset nominal voltage; where the nominal voltage is the rated operating voltage at the reference temperature; and the compensation voltage is positively correlated with the difference between the operating temperature and the reference temperature.

[0309] The voltage repair module is used to correct the target voltage based on the compensation voltage.

[0310] One possible design also includes:

[0311] The threshold calculation module is used to obtain the frequency threshold based on the voltage of the network interface device; wherein, the frequency threshold is positively correlated with the voltage of the network interface device.

[0312] The security protection module is used to trigger a preset security protection mechanism if the frequency of the network interface device exceeds the frequency threshold.

[0313] The energy efficiency optimization device for network interface equipment provided in this application embodiment can perform... Figure 2 and Figure 3 The technical solution of the method embodiment shown has the same implementation principle and technical effect as... Figure 2 and Figure 3 The methods shown in the embodiments are similar, and will not be described again in the embodiments of this application.

[0314] Figure 5 A schematic diagram of the structure of the electronic device provided in this application. Figure 5 As shown, the electronic device provided in this embodiment includes at least one processor 510 and a memory 520. Optionally, the electronic device further includes a communication component 530. The processor 510, memory 520, and communication component 530 are connected via a bus.

[0315] In the specific implementation process, at least one processor 510 executes computer execution instructions stored in memory 520, causing at least one processor 510 to execute the above-described embodiment of the energy efficiency optimization method for network interface devices.

[0316] The specific implementation process of processor 510 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0317] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in the application can be directly manifested as being executed by a hardware processor, or executed by a combination of hardware and software modules within the processor.

[0318] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0319] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0320] Embodiments of this application also provide a computer-readable storage medium storing a computer program, wherein the computer program is configured to execute the steps in the embodiments of the energy efficiency optimization method for any of the network interface devices described above when running.

[0321] 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.

[0322] The embodiments of this application also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in the embodiments of the energy efficiency optimization method for any of the network interface devices described above.

[0323] Embodiments of this application also provide another computer program product, including a non-volatile computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps in the above-described embodiments of the energy efficiency optimization method for any of the network interface devices.

[0324] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0325] The above provides a detailed description of the energy efficiency optimization method, device, medium, and product for a network interface device provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are only intended to help understand the method and core ideas of this application. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of the claims of this application.

Claims

1. A method for optimizing the energy efficiency of a network interface device, characterized in that, include: Obtain real-time performance parameters of network interface devices; The predicted energy efficiency ratio of the network interface device is determined based on the real-time performance parameters of the network interface device. Based on the real-time performance parameters and the predicted energy efficiency ratio, it is detected whether the network interface device is in an energy efficiency low-efficiency state; If the network interface device is detected to be in the energy efficiency depression state, then the target voltage and target frequency are determined based on the real-time performance parameters and the predicted energy efficiency ratio; wherein, the target voltage and the target frequency are the voltage and frequency of the network interface device when it leaves the energy efficiency depression state; A dynamic operating point skipping mechanism is adopted to adjust the voltage and frequency of the network interface device according to the target voltage and the target frequency; If the network interface device is detected to be not in the energy efficiency low-efficiency state, the voltage frequency adjustment method is determined based on the real-time performance parameters and the predicted energy efficiency ratio. If the voltage frequency adjustment method is determined to be either boost or buck adjustment, then the voltage and frequency of the network interface device are adjusted according to the voltage frequency adjustment method until the voltage frequency adjustment method is determined to be stop adjustment. The frequency adjustment of the network interface device according to the voltage frequency adjustment method includes: acquiring frequency step information to indicate the frequency adjustment step size; if the frequency adjustment step size is greater than a preset step size threshold, adjusting the frequency division coefficient of a pre-configured phase-locked loop (PLL) module according to the frequency adjustment step size; wherein the frequency division coefficient is negatively correlated with the frequency adjustment step size; if the frequency adjustment step size is less than or equal to the step size threshold, adjusting the delay time of a pre-configured delay-locked loop (TLL) module according to the frequency adjustment step size; wherein the delay time is positively correlated with the frequency adjustment step size; the frequency adjustment accuracy of the PLL module is lower than the frequency adjustment accuracy of the TLL module; and adjusting the frequency of the network interface device through the PLL module or the TLL module.

2. The method according to claim 1, characterized in that, The voltage adjustment of the network interface device according to the voltage frequency adjustment method includes: According to the voltage frequency adjustment method, the adjusted target voltage domain is determined from a plurality of preset voltage domains; wherein each voltage domain corresponds to an independent power supply network; If the current original voltage domain is found to be inconsistent with the target voltage domain, the power supply network corresponding to the target voltage domain is switched to provide power.

3. The method according to claim 2, characterized in that, The step of switching to the power supply network corresponding to the target voltage domain for power supply includes: The power supply network corresponding to the target voltage domain is pre-charged until a preset first duration is reached. The power supply is switched to the power supply network corresponding to the target voltage domain, and after a preset second time period following the switch, the power supply connection of the power supply network corresponding to the original voltage domain is disconnected.

4. The method according to claim 3, characterized in that, The plurality of power supply networks include: a first power supply network, a second power supply network, and a third power supply network; The power supply types of the first power supply network include: low dropout linear regulators and switched capacitors; The power supply type of the second power supply network includes: a multiphase buck converter; The power supply type of the third power supply network includes: direct-connected power management integrated circuit; The minimum voltage of the voltage domain corresponding to the second power supply network is greater than the maximum voltage of the voltage domain corresponding to the first power supply network. The maximum voltage of the voltage domain corresponding to the second power supply network is less than the minimum voltage of the voltage domain corresponding to the third power supply network.

5. The method according to claim 1, characterized in that, The step of determining the predicted energy efficiency ratio of the network interface device based on its real-time performance parameters includes: The real-time performance parameters are input into a pre-configured nonlinear multi-order polynomial energy efficiency model, which outputs the predicted energy efficiency ratio of the network interface device; wherein, the nonlinear multi-order polynomial energy efficiency model is used to describe the nonlinear relationship between the predicted energy efficiency ratio and the real-time performance parameters.

6. The method according to claim 5, characterized in that, The real-time performance parameters include: normalized bandwidth utilization U, operating temperature T, and average packet length S; The nonlinear multi-order polynomial energy efficiency model is expressed as: in, The predicted energy efficiency ratio of the network interface device; and These are the preset coefficients; This is a preset function relating to the operating temperature; It is the load change rate obtained based on the normalized bandwidth utilization rate.

7. The method according to claim 6, characterized in that, After determining the predicted energy efficiency ratio of the network interface device based on its real-time performance parameters, the method further includes: Obtain the actual energy efficiency ratio; If the error between the predicted energy efficiency ratio and the actual energy efficiency ratio is greater than a preset error threshold, then the prediction is iteratively updated based on the error and a preset learning rate. until the error is less than or equal to the error threshold.

8. The method according to claim 6, characterized in that, The step of determining the voltage frequency adjustment method based on the real-time performance parameters and the predicted energy efficiency ratio includes: Based on the predicted energy efficiency ratio, the energy efficiency gradient is obtained; If the energy efficiency gradient is less than a preset first gradient threshold and the load change rate is greater than a preset change rate threshold, then the voltage frequency adjustment method is determined to be boost frequency adjustment; wherein, the first gradient threshold is a negative number. If the energy efficiency gradient is greater than a preset second gradient threshold and the operating temperature is greater than a preset temperature threshold, then the voltage frequency adjustment method is determined to be buck-frequency reduction; wherein, the second gradient threshold is a positive number and the absolute value of the second gradient threshold is less than that of the first gradient threshold.

9. The method according to claim 6, characterized in that, Also includes: If the normalized bandwidth utilization rate is within a preset utilization threshold range, and the predicted energy efficiency ratio is less than a preset energy efficiency ratio threshold, then it is determined that the network interface device is in the energy efficiency depression state. Accordingly, determining the target voltage based on the real-time performance parameters and the predicted energy efficiency ratio includes: If the load change rate is positive, the normalized bandwidth utilization rate is continuously increased by a first utilization threshold until the modified normalized bandwidth utilization rate is outside the utilization threshold range. If the load change rate is negative, the normalized bandwidth utilization rate is continuously subtracted from the second utilization threshold until the modified normalized bandwidth utilization rate is outside the utilization threshold range. The target voltage is obtained based on the modified normalized bandwidth utilization and the predicted energy efficiency ratio.

10. The method according to claim 9, characterized in that, After determining the target voltage based on the real-time performance parameters and the predicted energy efficiency ratio, the method further includes: The compensation voltage is obtained based on the operating temperature and the preset nominal voltage; wherein, the nominal voltage is the rated operating voltage at the reference temperature; and the compensation voltage is positively correlated with the difference between the operating temperature and the reference temperature. The target voltage is corrected based on the compensation voltage.

11. The method according to claim 9, characterized in that, Also includes: A frequency threshold is obtained based on the voltage of the network interface device; wherein the frequency threshold is positively correlated with the voltage of the network interface device. If the frequency of the network interface device exceeds the frequency threshold, a preset security protection mechanism is triggered.

12. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to implement the steps of the energy efficiency optimization method for the network interface device as described in any one of claims 1 to 11 when executing the computer program.

13. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, wherein when the computer program is executed by a processor, it implements the steps of the energy efficiency optimization method for the network interface device as described in any one of claims 1 to 11.

14. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the energy efficiency optimization method for the network interface device as described in any one of claims 1 to 11.

Images