A method for intelligent detection and maintenance early warning of network operation status

By calculating the Shannon information entropy of time delay and adaptive weighted modulation filtering, and combining it with hardware cache capacity, a traffic shaping correction amount is generated, which solves the problem of delayed intervention instructions in traditional early warning schemes, realizes self-healing truncation of the network under extreme traffic, and avoids packet loss at the underlying level.

CN123123209APending Publication Date: 2026-11-17GUIZHOU SHENGHUACHENG TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611331332.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-31
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Traditional network operation status early warning schemes are detached from the objective constraints of the physical limits of the underlying hardware cache, resulting in delayed intervention instructions under extreme micro-burst traffic conditions, and the inability to achieve timely network self-healing and interception and underlying packet loss prevention.

Method used

By acquiring the queuing delay sequence and calculating the Shannon information entropy of the delay, the sampling interval is dynamically generated. Combined with the buffer queue depth and hardware buffer capacity, adaptive weight modulation filtering and feedback control are performed to generate the traffic shaping correction amount, thereby realizing traffic self-healing truncation.

Benefits of technology

It improves upon the control failure caused by the algorithm's time consumption far exceeding the hardware cache breakdown time in traditional early warning schemes, and achieves timely traffic self-healing and truncation under extreme traffic conditions, avoiding underlying packet loss.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123123209A_ABST
    Figure CN123123209A_ABST
Patent Text Reader

Abstract

This invention relates to the fields of network communication and intelligent operation and maintenance technology, and in particular to a method for intelligent detection and maintenance early warning of network operation status, comprising: acquiring the queuing delay sequence of the target network node to calculate the delay Shannon information entropy, and dynamically driving the generation of dynamic sampling intervals; acquiring the current and previous cache queue depths based on the dynamic sampling intervals, and calculating the first derivative of the cache depth; performing adaptive weighted modulation filtering on the derivative using the delay Shannon information entropy, and outputting the filtered smooth erosion rate; calculating the remaining absolute physical space of the cache in combination with the upper limit of the hardware cache capacity, and using the smooth erosion rate to deduce the hardware cache overflow countdown; synchronously inputting the information entropy and the countdown into the feedback controller for nested constraint calculation, generating the underlying traffic shaping correction amount, and sending it to the queue scheduling register to perform traffic self-healing truncation, effectively improving the network's underlying congestion prevention capability.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the fields of network communication and intelligent operation and maintenance technology, and in particular to a method for intelligent detection and maintenance early warning of network operation status. Background Technology

[0002] With the development of software-defined networking and core network data forwarding technologies, underlying switching equipment faces extreme micro-burst traffic congestion challenges. Traditional network operation status early warning schemes mostly employ purely data-driven predictive models or static empirical thresholds for status assessment and intervention. These traditional schemes disregard the objective constraints of the physical limits of the underlying hardware cache. Because the complex matrix operations of the algorithm model take far longer than the physical time it takes for the physical network card cache queue to be instantly overwhelmed, the generated intervention commands lag significantly behind the actual moment of congestion at the physical layer. This results in the inability to achieve timely network self-healing truncation and underlying packet loss prevention under extreme micro-burst traffic conditions. Summary of the Invention

[0003] To overcome the above shortcomings, this invention provides an intelligent detection and maintenance early warning method for network operation status, aiming to improve the problem that traditional network early warning mostly adopts data prediction models without physical constraints. Since the algorithm time is much longer than the hardware cache breakdown time, it causes the intervention command to lag and the control to fail.

[0004] This invention provides the following technical solution: a method for intelligent detection and maintenance early warning of network operation status, comprising the following steps:

[0005] S1. Obtain the queuing delay sequence of the target network node, calculate the delay Shannon information entropy based on the queuing delay sequence, and use the delay Shannon information entropy to dynamically drive the generation of the dynamic sampling interval of the target network node.

[0006] S2. Based on the dynamic sampling interval, synchronously obtain the current cache queue depth and the previous cache queue depth of the target network node, and calculate the first derivative of the cache depth based on the current cache queue depth, the previous cache queue depth and the dynamic sampling interval;

[0007] S3. Based on the time-delay Shannon information entropy, perform adaptive weight modulation filtering on the first derivative of the cache depth to remove high-frequency observation differential noise caused by extremely small time steps, and output the filtered smooth erosion rate.

[0008] S4. Calculate the remaining absolute physical space of the cache based on the current cache queue depth and the upper limit of the hardware cache capacity preset by the target network node, and use the filtered smooth erosion rate to perform physical constraint deduction on the remaining absolute physical space of the cache, and calculate the hardware cache overflow countdown of the target network node.

[0009] S5. The time-delay Shannon information entropy and the hardware cache overflow countdown are synchronously input into a preset feedback controller for nested constraint calculation to generate the underlying traffic shaping correction amount for the target network node, and the underlying traffic shaping correction amount is sent to the queue scheduling register of the target network node to perform traffic self-healing truncation.

[0010] By adopting the above technical solution, and by using information entropy-driven sampling and deducing overflow countdown to execute nested feedback control, deterministic traffic self-healing truncation is achieved. This improves the problem that traditional network early warning mostly uses data prediction models without physical constraints, and the algorithm time is much longer than the hardware cache breakdown time, resulting in delayed intervention instructions and control failure.

[0011] The present invention has the following beneficial effects:

[0012] 1. In this invention, by using information entropy to drive the generation of dynamic sampling intervals and deducing the buffer overflow countdown to execute feedback control, traffic self-healing truncation is achieved. This improves the problem that traditional network early warning mostly adopts fixed-period detection, which cannot adapt to transient micro-burst traffic due to static mechanisms, resulting in the lag in the issuance of underlying intervention commands and control failure.

[0013] 2. In this invention, by using information entropy to perform adaptive weight modulation filtering on the first derivative of the cache depth, a smooth erosion rate is output. This improves the problem that traditional differential observations mostly use direct discrete difference operations, which cause high-frequency observation noise explosion and system control oscillations because the extremely small time step will infinitely amplify the underlying hardware clock error.

[0014] 3. In this invention, by dividing the constant by the overflow countdown to generate a crisis multiplier with the same dimension to participate in the feedback control operation, and then outputting the flow shaping correction amount, the problem of feature evaluation weight distortion and underlying flow control deviation caused by the inconsistent physical dimensions of network state parameters, which is mostly adopted by traditional data feature fusion, is improved.

[0015] 4. In this invention, by comparing the overflow countdown with the sampling interval to trigger the escape mechanism and issue a clear command, the internal integral accumulation link is blocked. This improves the problem that most traditional flow controllers use conventional proportional integral mechanisms, which cause integral saturation due to the continuous accumulation of historical errors during extreme congestion, resulting in the locking and delay of the underlying rate limiting strategy. Attached Figure Description

[0016] Figure 1 This is a flowchart of a method for intelligent detection and maintenance early warning of network operation status proposed in this invention;

[0017] Figure 2This is a flowchart illustrating the parallel computation process for queuing delay entropy deduction and sampling interval generation in the intelligent detection and maintenance early warning method for network operation status proposed in this invention.

[0018] Figure 3 This is a flowchart of the data cleaning process involving calculus calculation and cross-modulation filtering for a network operation status intelligent detection and maintenance early warning method proposed in this invention.

[0019] Figure 4 This is a flowchart of the extreme value extraction and physical boundary countdown deduction of the intelligent detection and maintenance early warning method for network operation status proposed in this invention;

[0020] Figure 5 This is a flowchart illustrating the state machine decomposition and physical action implementation of a network operation status intelligent detection and maintenance early warning method proposed in this invention. Detailed Implementation

[0021] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Example 1:

[0023] The target network node is physically connected to the underlying network interface card (NIC) via a peripheral component interconnect bus. The central processing unit integrates a memory-mapped input / output module. The underlying NIC internally integrates a physical layer chip, hardware timers, a hardware interrupt controller, and a queue scheduling register. The system constructs an out-of-band data path directly to the underlying NIC based on the peripheral component interconnect bus. Each logic module uses memory-mapped input / output technology to bypass the operating system kernel and directly perform read / write operations on the underlying logic hardware.

[0024] In the first embodiment of the present invention, the present invention provides a method for intelligent detection and maintenance early warning of network operating status, such as... Figures 1-5 As shown, it includes the following steps:

[0025] S1. Obtain the queuing delay sequence of the target network node, calculate the delay Shannon information entropy based on the queuing delay sequence, and use the delay Shannon information entropy to dynamically drive the generation of the dynamic sampling interval of the target network node.

[0026] Furthermore, in S1, the Shannon information entropy of the delay is calculated based on the queuing delay sequence, including:

[0027] Extract the enqueue and dequeue timestamps generated when the physical layer chip parses the packet frame header;

[0028] Calculate the physical time difference obtained by subtracting the dequeue timestamp from the enqueue timestamp;

[0029] Multiple physical time differences are aggregated in chronological order to form a queuing delay sequence;

[0030] Construct a preset sliding time window and perform histogram binning on the queuing delay sequence within the preset sliding time window;

[0031] Statistically analyze the frequency of queuing delay sequences falling into each time bin;

[0032] Divide the frequency of occurrence by the total number of samples in the queuing delay sequence to obtain the probability distribution characteristics;

[0033] Based on the probability distribution characteristics, a logarithmic summation operation is performed to obtain the time-delayed Shannon information entropy.

[0034] In S1, the dynamic sampling interval of the target network nodes is dynamically generated using the time-delayed Shannon information entropy, including:

[0035] Extract the peak value of the underlying hardware interrupt response capability of the target network node;

[0036] The boundary constraint threshold is obtained by dividing the peak value of the underlying hardware interrupt response capability by a preset constant.

[0037] During the factory initialization phase of the equipment, positive real-value parameters are written into the static storage area as the preset baseline frequency;

[0038] The absolute value of the preset baseline frequency is strictly less than the boundary constraint threshold.

[0039] Transform the time-delayed Shannon information entropy into a non-linear scaling factor;

[0040] The product term is obtained by multiplying the preset baseline frequency by the nonlinear scaling factor;

[0041] Calculate the ratio obtained by dividing the constant unit by the product term, and use the ratio as the dynamic sampling interval.

[0042] Specifically, step S1 obtains the queuing delay sequence of the target network nodes. The underlying logic unit calculates the delay Shannon information entropy based on the queuing delay sequence. The system uses the delay Shannon information entropy to dynamically drive the generation of the dynamic sampling interval of the target network nodes.

[0043] Regarding the execution logic for calculating the Shannon information entropy of delay based on the queuing delay sequence, the target network node receives the underlying network data packets. The system directly schedules the physical layer chip to parse the packet frame header. The physical layer chip triggers its hardware clock during processing. The system extracts the enqueue timestamp of the packet arriving at the physical layer chip and the dequeue timestamp of the packet leaving the physical layer chip. The system performs a subtraction operation to calculate the physical time difference obtained by subtracting the dequeue timestamp from the enqueue timestamp. The physical time difference represents the dwell time of a single data packet at the hardware level. The system aggregates multiple consecutively generated physical time differences in chronological order to form a queuing delay sequence.

[0044] The queuing delay sequence is input into the delay disorder evaluation module. To avoid interference from drastic fluctuations in the total number of samples caused by the system's dynamic sampling frequency, the system constructs a sliding feature window based on a fixed sample size on the buffer sequence. The system extracts data segments within the current sliding feature window and performs histogram binning on the constant number of queuing delay sequences within the sliding feature window. The system counts the frequency of occurrence of the queuing delay sequence falling into each time bin. The system extracts the total number of samples of the queuing delay sequence within the sliding feature window. Due to the fixed sample size, this total number of samples is a constant. The system divides the frequency of occurrence by the total number of samples of the queuing delay sequence to obtain the probability distribution characteristics corresponding to each time bin.

[0045] The system performs logarithmic summation based on probability distribution characteristics to obtain the time-delay Shannon information entropy. The specific mathematical model for discretization and information entropy calculation is as follows:

[0046] ;

[0047] ;

[0048] in The probability distribution characteristic of the queuing delay sequence falling into the i-th time bin is represented by a value that is greater than or equal to 0 and less than or equal to 1. This represents the queuing delay sequence falling into the first position. Frequency of occurrence within each time box. This represents the total number of samples in the queue delay sequence within the preset sliding time window. This represents the calculated Shannon information entropy for the time delay. This represents the total number of time bins divided within a preset sliding time window by histogram binning. This represents a logarithmic function operation. The calculated Shannon information entropy of the time delay is output as a state control variable to the hardware sampling driver unit.

[0049] Regarding the execution logic for dynamically driving the generation of the dynamic sampling interval of the target network node using the time-delay Shannon information entropy, the system extracts the peak value of the underlying hardware interrupt response capability of the target network node. The peak value of the underlying hardware interrupt response capability defines the physical limit of the network card chip's ability to process interrupt signals per second. The system divides the peak value of the underlying hardware interrupt response capability by a preset constant to obtain the boundary constraint threshold.

[0050] During the device's factory initialization phase, the system writes positive real-valued parameters to the static storage area of ​​the physical motherboard. These parameters serve as the preset baseline frequency for system operation. The system executes low-level security logic to enforce that the absolute value of the preset baseline frequency is strictly less than the boundary constraint threshold. The system then cuts off illegal frequency configurations that exceed the hardware's processing limits.

[0051] The hardware sampling driver unit receives the time-delay Shannon information entropy. The system converts the time-delay Shannon information entropy into a nonlinear scaling factor. The system performs a multiplication operation between the preset baseline frequency and the nonlinear scaling factor to obtain a product term. The system calculates the ratio obtained by dividing a constant unit by the product term, assigns the ratio as a value, and outputs it as the dynamic sampling interval.

[0052] The mathematical model for calculating the frequency modulation and time interval output is as follows:

[0053] ;

[0054] ;

[0055] in This represents the nonlinear scaling factor derived from the time-delayed Shannon information entropy. This represents an exponential function with the natural constant as its base. This represents the dimensionless sensitivity coefficient pre-written into the system register. The numerical range of this dimensionless sensitivity coefficient is set between 0.1 and 10.0, with the system preferably configured to 1.0 to match the logarithmic convergence boundary. This represents the Shannon information entropy of the time delay calculated from the output of the preceding steps. This represents the preset baseline frequency read from the static storage area. The preset baseline frequency is set to a range of 10 Hz to 1000 Hz, with 100 Hz being the preferred configuration for the system. The peak value of the underlying hardware interrupt response capability, which is defined in the preceding steps, is obtained through actual testing by injecting an extreme packet sending stress test program into the physical ports during the system initialization phase. This represents the dynamically generated sampling interval.

[0056] The dynamic sampling interval constitutes the final output variable of the local data flow closed loop. The system sends the dynamic sampling interval to the underlying hardware timer, directly triggering the target network node to perform the next cycle of synchronous sampling of the buffer queue depth.

[0057] S2. Based on the dynamic sampling interval, synchronously obtain the current cache queue depth and the previous cache queue depth of the target network node, and calculate the first derivative of the cache depth based on the current cache queue depth, the previous cache queue depth and the dynamic sampling interval.

[0058] Furthermore, in S2, the first derivative of the cache depth is calculated based on the current cache queue depth, the previous cache queue depth, and the dynamic sampling interval, including:

[0059] Start the underlying hardware timer that is bound to the dynamic sampling interval;

[0060] Based on the physical time period triggered by the underlying hardware timer, the current cache queue depth corresponding to the current sampling moment and the previous cache queue depth corresponding to the immediately preceding sampling moment are synchronously collected.

[0061] Calculate the physical byte difference between the current cache queue depth and the previous cache queue depth;

[0062] Divide the physical byte difference by the dynamic sampling interval to obtain the first derivative of the cache depth.

[0063] Specifically, the system receives the dynamic sampling interval calculated in the preceding steps. This dynamic sampling interval is directly input to the underlying hardware control plane of the target network node as a timing control parameter. The underlying hardware timer is bound to the dynamic sampling interval and initiates the countdown logic.

[0064] The underlying hardware timer generates a hardware trigger pulse based on the value of the dynamic sampling interval. This pulse defines a strict physical time period. The system performs synchronous data acquisition operations at the level edge triggered by the physical time period. The system bypasses the host operating system's upper-layer software protocol stack through memory-mapped input / output technology, directly issuing read commands to the underlying chip registers of the target network node. The system obtains the current buffer queue depth corresponding to the current sampling moment. The system accesses the local temporary status register to retrieve the saved previous buffer queue depth corresponding to the immediately preceding sampling moment.

[0065] The current cache queue depth and the previous cache queue depth, measured in bytes, constitute a discrete-state physical observation data pair.

[0066] Physical observation data and the dynamic sampling interval are input together to the differential calculation unit. The differential calculation unit performs hardware-level discrete difference operations. The system calculates the physical byte difference obtained by subtracting the previous buffer queue depth from the current buffer queue depth. The physical byte difference reflects the accumulation or dissipation of data packet queuing within a specific time window. The system directly divides the obtained physical byte difference by the dynamic sampling interval. After performing the division operation, the differential calculation unit outputs the first derivative of the buffer depth.

[0067] The specific mathematical model for discrete differential calculation is as follows:

[0068] ;

[0069] ;

[0070] in This represents the difference in physical bytes generated by the system calculations within the current physical time period. This represents the current cache queue depth read from the underlying chip register at the current sampling moment. This represents the depth of the preceding cache queue, which was acquired and stored at the immediately preceding sampling time. This represents the dynamic sampling interval that drives the hardware timer. The first derivative of the cache depth represents the output of the differential computation unit. The first derivative of the cache depth characterizes the current instantaneous physical erosion rate of the target network node's underlying cache.

[0071] The first derivative of the cache depth, as an independent data state characteristic with physical dimensions, is continuously pushed into the data flow bus by the system. The first derivative of the cache depth enters the adaptive filtering module, serving as the core pre-input variable for the next step of observation noise suppression and state estimation.

[0072] S3. Based on the time-delay Shannon information entropy, the first derivative of the cache depth is adaptively weighted and filtered to remove high-frequency observation differential noise caused by extremely small time steps, and the output is filtered and smoothed erosion rate.

[0073] Furthermore, in S3, adaptive weighted modulation filtering is performed on the first derivative of the cache depth based on the delay Shannon information entropy, including:

[0074] Construct a state-space transition matrix with the first derivative of cache depth as the observed variable and the actual physical erosion rate as the implicit state variable;

[0075] Extract the thermal noise baseline variance obtained through the bottom-level unloaded calibration;

[0076] Transform the time-delayed Shannon information entropy into an exponential amplification factor;

[0077] The thermal noise baseline variance is multiplied and modulated using an exponential amplification factor to generate the dynamic observation noise covariance.

[0078] In S3, the output filter smooths the erosion rate, including:

[0079] By combining the dynamic observation noise covariance and the prior error covariance of the state space transition matrix, the smoothing filter gain is calculated and generated.

[0080] Calculate the observational physical bias between the first derivative of the cache depth and the prior state estimate;

[0081] The smoothing filter gain is used to perform a multiplicative feedback correction operation on the observed physical bias, and the output is a smoothed erosion rate.

[0082] Synchronously monitor the absolute value of the smoothing filter gain;

[0083] When the absolute value of the smoothing filter gain is lower than the preset confidence threshold, an over-limit interrupt level is generated;

[0084] The target network node's underlying network card is triggered to perform a register latching action by utilizing an over-limit interrupt level.

[0085] Specifically, the module pre-constructs a state-space transition matrix. The system sets the first derivative of the cache depth, which contains high-frequency hardware jitter noise, as the observed variable in the state space. The system sets the actual physical erosion rate occurring at the bottom layer of the target network node as the implicit state variable in the state space. Since micro-burst traffic conforms to the physical characteristics of a random walk within an extremely short time window, the system configures the state-space transition matrix as an identity matrix. The system extracts the process noise covariance based on the statistical results of the variance of the erosion rate during historical stationary periods.

[0086] The adaptive filtering module triggers a read instruction from the underlying memory. The system extracts the thermal noise baseline variance, which is pre-obtained through the underlying no-load calibration program. The thermal noise baseline variance defines the static error baseline of the hardware measurement path. The system introduces the time-delay Shannon information entropy as an input parameter into the matrix modulation logic. The system performs an exponential operation on the time-delay Shannon information entropy with the natural constant as the base. This operation converts the time-delay Shannon information entropy into an exponential amplification factor. The system uses the exponential amplification factor to perform a scalar multiplication modulation operation on the thermal noise baseline variance. The modulation process generates the dynamic observation noise covariance.

[0087] The mathematical model for the above covariance modulation is as follows:

[0088] ;

[0089] ;

[0090] in This represents the exponential amplification factor generated by the system's calculation and transformation. This represents an exponential function with the natural constant as its base. The Shannon information entropy representing the time delay extracted from the input source. This represents the baseline variance of thermal noise extracted from the underlying calibration file. This represents the dynamic observation noise covariance of the output of the system performing multiplication modulation operations.

[0091] The dynamic observation noise covariance is input to the state estimation unit. The system operates on the state space transition matrix and extracts the prior error covariance left over from the previous update cycle. The system divides the prior error covariance by the sum of the prior error covariance and the dynamic observation noise covariance. The state estimation unit generates the smoothing filter gain based on the result of this division operation.

[0092] The system retrieves the prior state estimate from the register. It performs a subtraction operation, calculating the difference between the first derivative of the buffer depth and the prior state estimate. This difference is defined as the observation physical bias. The system uses the smoothing filter gain to perform a multiplicative feedback correction operation on the observation physical bias. The system then sums the output of the multiplicative feedback correction operation with the prior state estimate. The state estimation unit completes data cleaning for the current sampling time and outputs the filtered and smoothed erosion rate.

[0093] The mathematical model for the above state estimation and correction is as follows:

[0094] ;

[0095] ;

[0096] ;

[0097] ;

[0098] in The smoothing filter gain generated by the state estimation unit represents the smoothing filter gain. This represents the prior error covariance in the state-space transition matrix. The dynamic observation noise covariance represents the modulation of the preceding equation. This represents the observed physical bias generated by the division operation. This represents the first derivative of the cache depth as an input to the observed variable. This represents the estimated prior state that the system retains. This represents the filtered smoothing erosion rate of the final output after the feedback correction is performed. This represents the posterior error covariance generated after the measurement update. While the system outputs a smoothed erosion rate via filtering, it simultaneously performs a posterior update of the error covariance to ensure the recursive closed loop of the state machine. The system uses a constant 1 to subtract the smoothing filter gain and multiplies it by the prior error covariance to generate the posterior error covariance. Before entering the next sampling period, the system adds the posterior error covariance to the preset process noise covariance to generate the prior error covariance for the next period for subsequent iterations.

[0099] The filtered and smoothed erosion rate enters the data bus and flows to the subsequent physical constraint deduction module. Status monitoring logic is deployed at the data transmission outgoing node. The system extracts the absolute value of the smoothing filter gain. The system compares the absolute value of the smoothing filter gain with a preset trust threshold limit in memory.

[0100] A smoothing filter gain absolute value below a preset confidence threshold indicates severe physical jitter divergence in the underlying sampled data. The system hardware interrupt controller responds to the out-of-limit state. The hardware interrupt controller generates an out-of-limit interrupt level on the control bus. The out-of-limit interrupt level is sent along the physical pin to the underlying network interface card (NIC) of the target network node. The underlying NIC control unit captures the out-of-limit interrupt level. The underlying NIC control unit directly revokes the execution privileges of the regular update procedure. The underlying NIC forcibly executes a register latch. The register latch action blocks external write operations to the queue scheduling status register. The target network node is physically locked and maintains execution of the control signaling from the previous cycle.

[0101] S4. Calculate the remaining absolute physical space of the cache based on the current cache queue depth and the upper limit of the hardware cache capacity of the target network node, and use the filtering smooth erosion rate to perform physical constraint deduction on the remaining absolute physical space of the cache, and calculate the hardware cache overflow countdown of the target network node.

[0102] Furthermore, in S4, the hardware buffer overflow countdown of the target network node is calculated, including:

[0103] Extract the maximum hardware cache capacity recorded in the underlying data manual of the target network node;

[0104] Calculate the physical capacity difference between the upper limit of the hardware cache capacity and the current cache queue depth;

[0105] Use the difference in physical capacity as the remaining absolute physical space of the cache;

[0106] Compare the smoothed erosion rate with the preset avoidance of division by zero anomalies and the minimum positive real number;

[0107] Extract the largest value term from the terms "filtering smoothing erosion rate" and "avoiding division by zero abnormal minimum positive real number" as the denominator parameter;

[0108] Divide the remaining absolute physical space of the cache by the denominator parameter to output the hardware cache overflow countdown.

[0109] Specifically, the deduction module receives the current cache queue depth collected and retained in the preceding steps, as well as the filtering and smoothing erosion rate output by the adaptive filtering module. The system sends a configuration read signaling to the target network node through the device driver interface. The system extracts the hardware cache capacity limit recorded in the target network node's underlying datasheet. When the target network node does not provide underlying datasheet parameters, the system performs reverse calibration by continuously injecting line-rate traffic into the physical port until the first physical layer packet loss threshold occurs, thereby obtaining and solidifying the hardware cache capacity limit. The hardware cache capacity limit defines the absolute spatial boundary limit of the silicon chip storage crystal at the physical level.

[0110] The current cache queue depth and the maximum hardware cache capacity are fed into the arithmetic logic unit (ALU). The ALU performs a subtraction operation, calculating the physical capacity difference between the maximum hardware cache capacity and the current cache queue depth. The system defines this calculated physical capacity difference as the remaining absolute physical space of the cache. The remaining absolute physical space of the cache represents the maximum data carrying capacity that the target network node can currently accommodate.

[0111] The system's internal non-volatile register area is pre-written with a minimum positive real number to avoid division-by-zero exceptions. This minimum positive real number is used to prevent microprocessor-level division overflow crashes caused by net outflow decay of network traffic. The derivation module invokes an internal hardware numerical comparator. The numerical comparator extracts the filtering and smoothing erosion rate on the data bus and the minimum positive real number to avoid division-by-zero exceptions in the register. The numerical comparator compares the absolute magnitudes of the two parameters. The system directly extracts the largest value determined by the numerical comparator as the denominator parameter for the division operation.

[0112] The divider receives the remaining absolute physical space of the cache and the denominator parameter. The divider divides the remaining absolute physical space of the cache by the denominator parameter. The derivation module outputs a hardware cache overflow countdown based on the division instruction cycle.

[0113] The mathematical model for the above physical constraint derivation is as follows:

[0114] ;

[0115] ;

[0116] ;

[0117] in This represents the remaining absolute physical space of the cache generated by the arithmetic logic unit calculation. This represents the upper limit of the hardware cache capacity extracted from the underlying datasheet of the target network node. This represents the current cache queue depth obtained from sampling by the preceding system. This represents the denominator parameter of the output of the numerical comparator for comparison and filtering. This represents a mathematical comparison function that performs maximum value extraction. This represents the filtering smoothing erosion rate of the input from the preceding adaptive filtering module. This represents a system-preset mechanism to prevent division by zero, specifically avoiding extremely small positive real numbers. This represents the countdown to the hardware buffer overflow of the output of the final division operation performed by the deduction module.

[0118] The hardware cache overflow countdown carries a physical dimension including a time unit. The system writes the hardware cache overflow countdown to global shared memory. The hardware cache overflow countdown then leaves the inference module and flows to the subsequent feedback controller for nested constraint calculations.

[0119] S5. The time-delay Shannon information entropy and hardware buffer overflow countdown are synchronously input into the preset feedback controller for nested constraint calculation, generating the underlying traffic shaping correction amount for the target network node, and sending the underlying traffic shaping correction amount to the queue scheduling register of the target network node to perform traffic self-healing truncation.

[0120] Furthermore, in S5, nested constraint calculations are performed, including:

[0121] Extract the preset safety buffer countdown constant;

[0122] Divide the safety buffer countdown constant by the hardware buffer overflow countdown to obtain the crisis multiplier of the same dimension;

[0123] Calculate the square value of the crisis multiplier with the same dimension as the state penalty term;

[0124] Adding the time-delayed Shannon information entropy to the constant unit yields the proportional adjustment factor;

[0125] Extract the proportional control coefficient from the preset feedback controller;

[0126] The proportional control coefficient, proportional adjustment factor, and state penalty term are multiplied together to obtain the negative value of the proportional penalty.

[0127] In S5, the underlying traffic shaping correction for the target network node is generated, including:

[0128] Calculate the systematic error value obtained by subtracting the dimensionless crisis multiplier from the constant unit;

[0129] The integral cumulative value is obtained by performing an integral operation on the system error value as it slides over physical time;

[0130] Extract the integral control coefficients from the preset feedback controller;

[0131] Multiplying the integral control coefficient by the integral cumulative value yields the integral recovery positive value;

[0132] Add the negative value of the proportional penalty to the positive value of the integral restoration to obtain the result value;

[0133] The resulting value is used as the underlying flow shaping correction amount.

[0134] In S5, the underlying traffic shaping and correction amount is sent to the queue scheduling register of the target network node to perform traffic self-healing truncation, including:

[0135] Encapsulate the underlying traffic shaping and correction parameters into underlying hardware control instructions;

[0136] Within a single clock cycle, the underlying hardware control instructions are physically sent to the queue scheduling register of the target network node in an atomic write operation.

[0137] Real-time acquisition and comparison of the countdown value of the hardware cache overflow with the value of the dynamic sampling interval;

[0138] When the countdown value of the hardware cache overflow is determined to be strictly less than the value of the dynamic sampling interval, the emergency escape mechanism is triggered.

[0139] The emergency escape mechanism is used to block the integral accumulation control loop inside the preset feedback controller and send a zero-rate limiting command to the underlying network card of the target network node.

[0140] Specifically, a pre-configured feedback controller serves as the core execution unit, receiving the Shannon information entropy of the delay and the hardware buffer overflow countdown input. The system's arithmetic logic unit extracts the pre-configured safety buffer countdown constant from memory. The divider divides the safety buffer countdown constant by the hardware buffer overflow countdown, generating a dimensionless crisis multiplier of the same magnitude.

[0141] The system separates proportional control and integral control branches based on the same-dimensional crisis multiplier. In the proportional control branch, the system calculates the square of the same-dimensional crisis multiplier as the state penalty term, and adds the time delay Shannon information entropy to the constant unit to obtain the proportional adjustment factor. The multiplier performs a multiplication operation on the extracted proportional control coefficient, proportional adjustment factor, and state penalty term, and outputs the negative value of the proportional penalty after the sign bit is inverted.

[0142] In the integral control branch, the subtractor calculates the constant unit 1 minus the crisis multiplier of the same dimension, outputting the system error value. Considering the discrete digital characteristics of the underlying network switching chip and the non-uniform length of the dynamic sampling interval generated in the preceding steps, the integrator abandons continuous integration in its internal accumulation register and instead performs discrete integral accumulation based on the dynamic physical time step. Specifically, the system extracts the system error value of the current period and the dynamic sampling interval of the current period, multiplies them, and adds them to the historical integral accumulation value of the previous period to generate the currently updated integral accumulation value. The multiplier multiplies the extracted integral control coefficient with the integral accumulation value, and outputs the integral value back to a positive value.

[0143] The end adder of the arithmetic logic unit adds the proportional penalty negative value to the integral recovery positive value. The result of the addition is the underlying flow shaping correction.

[0144] The overall computational mathematical model for generating the above nested constraints and correction values ​​is as follows:

[0145] ;

[0146] in It represents the underlying traffic shaping correction amount generated by the addition of arithmetic logic units, and its value directly maps to the specific size of the traffic limit of the underlying network card. This represents the proportional control coefficient extracted from the system's self-configuration space. The Shannon information entropy represents the delay of the input feedback controller. This represents the countdown constant for the safety buffer that the system extracts from memory. This represents the countdown to the overflow of the hardware buffer that is calculated and fed into the input during the inference and prediction phase. This represents the integral control coefficient extracted by the system. This represents the physical time point at which the system executes the current control input calculation. This represents the physical time sliding variable upon which the integral operation of the execution error relies.

[0147] The aforementioned continuous mathematical model is implemented in the underlying digital hardware through discrete state equations. The continuous integral terms in the model are mapped to discrete iterative operation logic in the accumulation register of the underlying chip. That is, the system uses the dynamic sampling interval of the current cycle as the discrete differential time step, multiplies the system error value of the current cycle by the dynamic sampling interval, and accumulates it into the value of the integral register of the previous cycle, thereby effectively eliminating the control integral distortion caused by the non-uniform dynamic sampling frequency.

[0148] The underlying traffic shaping and correction parameters detach from the feedback controller and flow into the hardware driver layer. The protocol encapsulation module encapsulates the underlying traffic shaping and correction parameters into underlying hardware control commands. The system triggers the internal high-speed bus controller. The bus controller initiates an atomic write operation within a single clock cycle. During execution, the atomic write operation excludes externally initiated business logic interruptions at the hardware bus level. The underlying hardware control commands are physically sent to the queue scheduling register of the target network node. The register value overwrite operation directly performs self-healing truncation of traffic on the network interface.

[0149] Outside of the main cycle of the feedback controller, a separate hardware monitoring comparator is deployed in the system bypass. The hardware monitoring comparator collects the value of the hardware buffer overflow countdown in real time, extracts the value of the dynamic sampling interval, and performs value comparison.

[0150] When the judgment logic unit detects that the hardware cache overflow countdown value is strictly less than the value of the dynamic sampling interval, it indicates that physical cache breakdown is highly likely to occur before the next sampling. The system triggers the emergency escape mechanism.

[0151] The emergency escape mechanism sends a hardware blocking level. This blocking level interrupts the integrator's internal integration accumulation control, preventing the integrator from continuing to accumulate erroneous states. The emergency escape mechanism triggers a non-maskable interrupt on the underlying network interface card (NIC), forcibly bypassing the scheduling queue of the regular software call stack and directly issuing a zero-rate-limit command to the underlying NIC of the target network node. Based on this command, the underlying NIC instantly clears the port's forwarding token, performing a fallback operation of physical link blocking in extreme conditions.

[0152] Example 2:

[0153] The application scenario is set as edge switches in a data center software-defined network (SDN) carrying massive concurrent services. The underlying network traffic exhibits extreme differences in disorder between stable and micro-burst periods. Traditional network operation and maintenance mechanisms use fixed-frequency probe sampling, which excessively consumes device computing resources during stable periods and is prone to missing traffic characteristics due to sampling frequency lag during micro-burst periods. When faced with microsecond-level micro-burst traffic, the extremely small sampling time step amplifies the underlying hardware clock jitter and queue depth measurement errors, causing a high-frequency observation differential noise explosion. This leads to the divergence and failure of prediction models based on instantaneous rate of change, resulting in oscillations in rate limiting strategies. Traditional data-driven prediction models lack extreme constraints on the underlying hardware cache capacity; matrix operations take far longer than the physical cache breakdown time, causing severe control lag. Existing early warning mechanisms mostly rely on static empirical threshold triggers. When facing extreme congestion, the feedback controller is prone to integral lock-up, and the system cannot generate physically deterministic rate-limiting signaling within nanosecond clock cycles, resulting in underlying intervention commands being issued and executed only after physical layer packet loss occurs. To address the above problems, this invention provides a method for intelligent detection and maintenance early warning of network operating status, such as... Figure 1 As shown. The specific implementation process of this system is as follows:

[0154] The system first executes step S1 to detect precursors of network micro-bursts. The target network node receives underlying data packets. The system intercepts the enqueue and dequeue timestamps generated when the physical layer chip parses the packet frame header. The processing unit calculates the physical time difference by subtracting the dequeue timestamp from the enqueue timestamp. The system aggregates multiple sets of physical time differences according to a time sequence to form a queuing delay sequence. The system extracts the queuing delay sequence within a preset sliding time window and performs histogram binning. The system statistically analyzes the probability distribution of sequence data falling into each time bin. Based on the probability distribution, the system performs a logarithmic summation operation to output the delay Shannon information entropy. The delay Shannon information entropy is pushed into the time mapping module. The time mapping module converts the delay Shannon information entropy into a nonlinear scaling factor. The system adjusts the preset baseline frequency according to the nonlinear scaling factor and calculates and generates a dynamic sampling interval. The larger the delay Shannon information entropy, the smaller the generated dynamic sampling interval value. The dynamic sampling interval is output to the underlying hardware timer.

[0155] The system executes step S2 to acquire physical observation data. The underlying hardware timer continuously outputs trigger pulses based on the received dynamic sampling interval. The system bypasses the upper-layer software protocol stack at the pulse trigger moment. The system directly reads the storage unit to obtain the current buffer queue depth. The system extracts the previous buffer queue depth stored in the local status register. The arithmetic logic unit calculates the physical byte difference obtained by subtracting the previous buffer queue depth from the current buffer queue depth. The divider divides the physical byte difference by the dynamic sampling interval. The system outputs the first derivative of the buffer depth.

[0156] To address the hardware clock jitter amplification problem caused by extremely small time steps, the system performs data cleaning in step S3. The adaptive filtering module synchronously receives the first derivative of the buffer depth and the Shannon information entropy of the time delay. The system constructs a state-space transition matrix with the actual physical erosion rate as the implicit state variable. The system converts the Shannon information entropy of the time delay into an exponential amplification factor. The system uses the exponential amplification factor to perform multiplicative modulation on the underlying thermal noise baseline variance. This operation generates the dynamic observation noise covariance. The system combines the dynamic observation noise covariance with the prior error covariance to calculate the smoothing filter gain. The system uses the smoothing filter gain to perform feedback correction on the observed physical bias. The adaptive filtering module removes high-frequency observation differential noise. The system outputs a filtered smoothed erosion rate that approximates the actual physical state.

[0157] The data inflow simulation module executes step S4. The system extracts the upper limit of the hardware cache capacity burned into the device firmware. The system calculates the upper limit of the hardware cache capacity minus the current cache queue depth. The calculated physical capacity difference is defined by the system as the remaining absolute physical space of the cache. The system extracts the larger of the filtering smoothing erosion rate and the preset avoidance of division by zero anomalies and extremely small positive real numbers as the denominator parameter. The divider divides the remaining absolute physical space of the cache by the denominator parameter. The simulation module outputs a hardware cache overflow countdown carrying a second-level physical time dimension.

[0158] The system finally executes step S5 to complete the control strategy distribution. The feedback controller synchronously receives the delay Shannon information entropy and the hardware buffer overflow countdown. The system extracts the preset safety buffer countdown constant. The system divides the safety buffer countdown constant by the hardware buffer overflow countdown to generate a dimensionless crisis multiplier of the same dimension. The feedback controller uses the delay Shannon information entropy to calculate the proportional adjustment factor. The system substitutes the proportional adjustment factor and the squared penalty term of the crisis multiplier of the same dimension into the control loop to perform nested constraint calculations. The feedback controller calculates and generates the underlying traffic shaping correction. The protocol encapsulation module converts the underlying traffic shaping correction into underlying hardware control instructions. The system triggers the internal high-speed bus controller to initiate an atomic write operation. The underlying hardware control instructions overwrite the queue scheduling register of the target network node within a single clock cycle. The underlying network interface card enforces physical-layer traffic self-healing truncation based on the updated register value.

[0159] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent detection and maintenance early warning of network operation status, characterized in that, Includes the following steps: S1. Obtain the queuing delay sequence of the target network node, calculate the delay Shannon information entropy based on the queuing delay sequence, and use the delay Shannon information entropy to dynamically drive the generation of the dynamic sampling interval of the target network node. S2. Based on the dynamic sampling interval, synchronously obtain the current cache queue depth and the previous cache queue depth of the target network node, and calculate the first derivative of the cache depth based on the current cache queue depth, the previous cache queue depth and the dynamic sampling interval; S3. Based on the time-delay Shannon information entropy, perform adaptive weight modulation filtering on the first derivative of the cache depth to remove high-frequency observation differential noise caused by extremely small time steps, and output the filtered smooth erosion rate. S4. Calculate the remaining absolute physical space of the cache based on the current cache queue depth and the upper limit of the hardware cache capacity preset by the target network node, and use the filtered smooth erosion rate to perform physical constraint deduction on the remaining absolute physical space of the cache, and calculate the hardware cache overflow countdown of the target network node. S5. The time-delay Shannon information entropy and the hardware cache overflow countdown are synchronously input into a preset feedback controller for nested constraint calculation to generate the underlying traffic shaping correction amount for the target network node, and the underlying traffic shaping correction amount is sent to the queue scheduling register of the target network node to perform traffic self-healing truncation.

2. The intelligent detection and maintenance early warning method for network operation status according to claim 1, characterized in that, In S1, calculating the delay Shannon information entropy based on the queuing delay sequence includes: Extract the enqueue and dequeue timestamps generated when the physical layer chip parses the packet frame header; Calculate the physical time difference obtained by subtracting the dequeue timestamp from the enqueue timestamp; The queuing delay sequence is formed by aggregating multiple sets of physical time differences in chronological order. A preset sliding time window is constructed, and the queuing delay sequence within the preset sliding time window is subjected to histogram binning. Statistically count the frequency of the queuing delay sequence falling into each time bin; Divide the occurrence frequency by the total number of samples in the queuing delay sequence to obtain the probability distribution characteristics; The time-delay Shannon information entropy is obtained by performing a logarithmic summation operation based on the probability distribution characteristics.

3. The intelligent detection and maintenance early warning method for network operation status according to claim 1, characterized in that, In S1, the dynamic sampling interval of the target network node is generated by dynamically driving the generation of the time-delay Shannon information entropy, including: Extract the peak value of the underlying hardware interrupt response capability of the target network node; The boundary constraint threshold is obtained by dividing the peak value of the underlying hardware interrupt response capability by a preset constant. During the factory initialization phase of the equipment, positive real-value parameters are written into the static storage area as the preset baseline frequency; The absolute value of the preset baseline frequency is verified to be strictly less than the boundary constraint threshold. The time-delayed Shannon information entropy is converted into a nonlinear scaling factor; The product term is obtained by multiplying the preset baseline frequency by the nonlinear scaling factor; The ratio obtained by dividing the constant unit by the product term is calculated, and the ratio is used as the dynamic sampling interval.

4. The intelligent detection and maintenance early warning method for network operation status according to claim 1, characterized in that, In S2, the calculation of the first derivative of the cache depth based on the current cache queue depth, the previous cache queue depth, and the dynamic sampling interval includes: Start the underlying hardware timer bound to the dynamic sampling interval; Based on the physical time period triggered by the underlying hardware timer, the current cache queue depth corresponding to the current sampling moment and the previous cache queue depth corresponding to the immediately preceding sampling moment are synchronously collected; Calculate the physical byte difference between the current cache queue depth and the previous cache queue depth; Divide the physical byte difference by the dynamic sampling interval to obtain the first derivative of the cache depth.

5. The intelligent detection and maintenance early warning method for network operation status according to claim 1, characterized in that, In S3, the adaptive weighted modulation filtering of the first derivative of the cache depth based on the time-delay Shannon information entropy includes: Construct a state-space transition matrix with the first derivative of the cache depth as the observation variable and the actual physical erosion rate as the implicit state variable; Extract the thermal noise baseline variance obtained through the bottom-level unloaded calibration; The time-delayed Shannon information entropy is converted into an exponential amplification factor; The thermal noise baseline variance is multiplied and modulated using the exponential amplification factor to generate the dynamic observation noise covariance.

6. The intelligent detection and maintenance early warning method for network operation status according to claim 5, characterized in that, In S3, the output filtering smoothing erosion rate includes: By combining the dynamic observation noise covariance with the prior error covariance of the state space transition matrix, the smoothing filter gain is calculated and generated. Calculate the observation physical deviation between the first derivative of the buffer depth and the prior state estimate; The smoothing filter gain is used to perform a multiplication feedback correction operation on the observed physical bias, and the smoothed erosion rate is output. Simultaneously monitor the absolute value of the smoothing filter gain; When the absolute value of the smoothing filter gain is determined to be lower than a preset trust threshold, an over-limit interrupt level is generated; The over-limit interrupt level is used to trigger the underlying network card of the target network node to perform a register latching action.

7. The intelligent detection and maintenance early warning method for network operation status according to claim 1, characterized in that, In S4, the calculation of the hardware cache overflow countdown of the target network node includes: Extract the maximum hardware cache capacity recorded in the underlying data manual of the target network node; Calculate the physical capacity difference obtained by subtracting the current cache queue depth from the upper limit of the hardware cache capacity; The difference in physical capacity is taken as the remaining absolute physical space of the cache. Compare the filtered smoothing erosion rate with the preset minimum positive real number to avoid division by zero anomalies; Extract the maximum value term from the two terms of the filtered smoothing erosion rate and the avoidance of the smallest positive real number in the division-by-zero anomaly as the denominator parameter; Divide the remaining absolute physical space of the cache by the denominator parameter to output the hardware cache overflow countdown.

8. The intelligent detection and maintenance early warning method for network operation status according to claim 1, characterized in that, In S5, the nested constraint calculation includes: Extract the preset safety buffer countdown constant; Divide the security buffer countdown constant by the hardware buffer overflow countdown to obtain a crisis multiplier of the same dimension. The square value of the dimensionless crisis multiplier is calculated as the state penalty term; Add the time-delay Shannon information entropy to the constant unit to obtain the proportional adjustment factor; Extract the proportional control coefficient from the preset feedback controller; The proportional control coefficient, the proportional adjustment factor, and the state penalty term are multiplied together to obtain a negative proportional penalty value.

9. The intelligent detection and maintenance early warning method for network operation status according to claim 8, characterized in that, In S5, generating the underlying traffic shaping correction for the target network node includes: Calculate the systematic error value obtained by subtracting the dimensionless crisis multiplier from the constant unit; The system error value is integrated over physical time to obtain the cumulative integral value; Extract the integral control coefficients from the preset feedback controller; Multiplying the integral control coefficient by the integral cumulative value yields the integral recovery positive value; The negative value of the proportional penalty is added to the positive value of the integral recovery to obtain the result value; The resulting value is used as the underlying flow shaping correction amount.

10. The intelligent detection and maintenance early warning method for network operation status according to claim 1, characterized in that, In S5, the step of sending the underlying traffic shaping correction amount to the queue scheduling register of the target network node to perform traffic self-healing truncation includes: The underlying flow shaping and correction amount is encapsulated into underlying hardware control instructions; Within a single clock cycle, the underlying hardware control instructions are physically sent to the queue scheduling register of the target network node via an atomic write operation. The value of the hardware cache overflow countdown is collected and compared with the value of the dynamic sampling interval in real time. When the value of the hardware cache overflow countdown is determined to be significantly less than the value of the dynamic sampling interval, an emergency escape mechanism is triggered. The emergency escape mechanism is used to block the integral accumulation control loop inside the preset feedback controller, and a zero-rate limiting command is sent to the underlying network card of the target network node.