Method, device and equipment for failure prediction of PCIE link and storage medium

Abstract

The application provides a PCIE link fault prediction method, device, equipment and storage medium, wherein the method comprises the following steps: obtaining a target monitoring parameter, a historical performance characteristic parameter, a historical average fault probability and a link fault evaluation mean value corresponding to the historical average fault probability of a target PCIE link; obtaining a first evaluation value based on the target monitoring parameter; obtaining a second evaluation value based on the historical performance characteristic parameter; obtaining a target link evaluation value based on the first evaluation value and the second evaluation value; establishing a probability distribution model based on the target link evaluation value and a historical link evaluation value of the target PCIE link; adjusting an edge probability associated with the link fault evaluation mean value in the probability distribution model based on the historical average fault probability, and performing fault prediction based on the adjusted probability distribution model. Through the technical scheme of the application, the prediction accuracy of the PCIE link fault probability can be improved.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a method, apparatus, device and storage medium for predicting PCIe link failures. Background Technology

[0002] As the interconnection hub for core hardware such as servers and data centers, the stability of PCIe links directly determines the reliability of system operation. With the surge in computing power demand and the increase in PCIe link transmission rates to the 5.0 standard, issues such as signal attenuation and crosstalk have intensified, and sudden failures can easily lead to data loss or service interruption. Existing fault prediction technologies mainly rely on traditional methods of threshold monitoring, which can only identify anomalies that have already occurred and cannot provide early warnings. Furthermore, fault prediction relies on single characteristic parameters, focusing primarily on explicit parameters such as signal amplitude, resulting in low prediction accuracy and difficulty in coping with complex and ever-changing link operating environments. Summary of the Invention

[0003] This application aims to at least partially address one of the technical problems in the related art.

[0004] In a first aspect, this application proposes a method for predicting the failure of a PCIe link. The method includes: obtaining target monitoring parameters, historical performance characteristic parameters, historical average failure probability, and the link failure evaluation mean corresponding to the historical average failure probability of a target PCIe link; obtaining a first evaluation value based on the target monitoring parameters; obtaining a second evaluation value based on the historical performance characteristic parameters; obtaining a target link evaluation value based on the first evaluation value and the second evaluation value; establishing a probability distribution model based on the target link evaluation value and the historical link evaluation value of the target PCIe link; adjusting the marginal probability associated with the link failure evaluation mean in the probability distribution model based on the historical average failure probability; and performing failure prediction based on the adjusted probability distribution model.

[0005] In one implementation, the target monitoring parameters include M types of parameters. Obtaining a first evaluation value based on the target monitoring parameters includes: determining the load state of the target device, wherein the target device is the device to which the target PCIe link belongs; classifying the target monitoring parameters into at least one first type of monitoring parameter and at least one second type of monitoring parameter based on the load state, and determining a first total weight and a second total weight of the at least one first type of monitoring parameter; wherein the at least one first type of monitoring parameter is the top N target monitoring parameters with a higher probability of degradation under the load state; M and N are positive integers, and M is greater than N; standardizing the target monitoring parameters to obtain corresponding standard values; determining a first weight for each type of first type of monitoring parameter based on the first total weight of the at least one first type of monitoring parameter, and determining a second weight for each type of second type of monitoring parameter based on the standard value and the second total weight of the at least one second type of monitoring parameter; and performing a weighted summation of the standard values ​​based on the first weight and the second weight to obtain the first evaluation value.

[0006] In one optional implementation, the standardization of the target monitoring parameters to obtain corresponding standard values ​​includes: dividing the target monitoring parameters into at least one positive monitoring parameter and at least one negative monitoring parameter based on the parameter characteristics of the target monitoring parameters; standardizing each of the negative monitoring parameters according to a preset first processing method to obtain a first standard value corresponding to each of the positive monitoring parameters; and standardizing each of the positive monitoring parameters according to a preset second processing method to obtain a second standard value corresponding to each of the negative monitoring parameters.

[0007] Optionally, the step of standardizing each of the positive monitoring parameters according to a preset first processing method to obtain a first standard value corresponding to each of the positive monitoring parameters includes: for each of the positive monitoring parameters, obtaining a first monitoring parameter of at least one first PCIe link, wherein the first PCIe link and the target PCIe link belong to the same target device, and the first monitoring parameter and the positive monitoring parameter are monitoring parameters of the same type; for each of the positive monitoring parameters, obtaining the maximum value between the target monitoring parameter and the first monitoring parameter as a reference value corresponding to the positive monitoring parameter; and for each of the positive monitoring parameters, standardizing the positive monitoring parameter according to a preset standardization processing method, combined with the corresponding reference value, to obtain the corresponding first standard value.

[0008] In one optional implementation, determining the first weight of each of the first type of monitoring parameters based on the first total weight of the at least one type of monitoring parameter includes: obtaining the first weight of each of the first type of monitoring parameters based on the differences in the degree of influence of different first type of monitoring parameters on link performance and the first total weight.

[0009] Optionally, obtaining the first weight for each of the first type of monitoring parameters based on the differences in the degree of influence of different first type of monitoring parameters on link performance and the first total weight includes: for each of the first type of monitoring parameters, obtaining at least one second monitoring parameter for a second PCIE link; wherein the second PCIE link and the target PCIE link belong to the same target device, and the second monitoring parameter and the first type of monitoring parameter are of the same type; for each of the first type of monitoring parameters, obtaining the variance value corresponding to the first type of monitoring parameter based on the second monitoring parameter and the first type of monitoring parameter; for each of the first type of monitoring parameters, performing parameter injection on the target device according to the variance value to obtain the corresponding link performance change value; and obtaining the first weight for each of the first type of monitoring parameters based on the link performance change value corresponding to different first type of monitoring parameters and the first total weight.

[0010] In one implementation, determining the second weight corresponding to each of the second type of monitoring parameters based on the standard value and the second total weight of the at least one second type of monitoring parameter includes: obtaining the sum of the standard values ​​corresponding to each of the second type of monitoring parameters; for each of the second type of monitoring parameters, obtaining the ratio of the standard value corresponding to the second type of monitoring parameter to the sum of the standard values, as the second weight ratio of the second type of monitoring parameter; and for each of the second type of monitoring parameters, obtaining the corresponding second weight of the second type of monitoring parameter based on the second weight ratio of the second type of monitoring parameter and the second total weight.

[0011] In one implementation, the historical performance characteristic parameters include multiple parameters. Obtaining a second evaluation value based on the historical performance characteristic parameters includes: determining a third weight corresponding to each historical performance characteristic parameter based on the degree of influence of different historical performance characteristic parameters on the transmission performance of the target PCIE link; normalizing the historical performance characteristic parameters to obtain corresponding third standard values; and performing weighted summation on the third standard values ​​based on the third weights to obtain the second evaluation value.

[0012] In one implementation, adjusting the marginal probability associated with the mean link failure evaluation in the probability distribution model based on the historical average failure probability includes: obtaining the value among multiple predicted link evaluation values ​​output by the probability distribution model that is closest to the mean link failure evaluation, and using it as the target predicted link evaluation value; adjusting the probability corresponding to the target predicted link evaluation value in the probability distribution model based on the historical average failure probability.

[0013] In one implementation, the fault prediction based on the adjusted probability distribution model includes: obtaining multiple predicted link evaluation values ​​and a first predicted probability for each predicted link evaluation value based on the adjusted probability distribution model; obtaining a fault prediction threshold; wherein the fault prediction threshold includes at least one of the following: a baseline threshold, a trend threshold, and a burst threshold; and performing fault probability prediction based on the first predicted probability for each predicted link evaluation value and the fault prediction threshold.

[0014] In one optional implementation, the fault prediction threshold includes the baseline threshold, and obtaining the fault prediction threshold includes: obtaining a first target historical link evaluation value in which the target PCIE link is in normal operating condition from the historical link evaluation values; and obtaining the baseline threshold based on the first target historical link evaluation value and a preset confidence interval.

[0015] In one implementation, the fault prediction threshold includes the burst threshold, and obtaining the fault prediction threshold includes: obtaining the standard deviation between the historical link evaluation value and the baseline threshold; adjusting the baseline threshold based on the standard deviation and a preset adjustment coefficient to obtain the burst threshold.

[0016] In one implementation, the fault prediction threshold includes the trend threshold, and the fault probability prediction based on the first prediction probability of each predicted link evaluation value and the fault prediction threshold includes: obtaining the rate of change of multiple link evaluation values ​​of the target PCIE link within a target time period and the prediction probability of each link evaluation value based on the multiple predicted link evaluation values, the prediction probability of each predicted link evaluation value, and at least a portion of the target historical link evaluation values; and performing fault probability prediction based on the rate of change of multiple link evaluation values, the prediction probability of each link evaluation value, and the trend threshold.

[0017] In one implementation, acquiring at least one target monitoring parameter of the target PCIe link includes: acquiring the transmission rate of the target PCIe link; determining the sampling rate of the target monitoring parameter based on the transmission rate; and sampling to acquire the target monitoring parameter according to the sampling rate.

[0018] Secondly, this application proposes a PCIe link fault prediction device, the device comprising: an acquisition module, configured to acquire target monitoring parameters, historical performance characteristic parameters, historical average fault probability, and the link fault evaluation mean corresponding to the historical average fault probability of a target PCIe link; a first processing module, configured to obtain a first evaluation value based on the target monitoring parameters; a second processing module, configured to obtain a second evaluation value based on the historical performance characteristic parameters; a third processing module, configured to obtain a target link evaluation value based on the first evaluation value and the second evaluation value; a fourth processing module, configured to establish a probability distribution model based on the target link evaluation value and the historical link evaluation values ​​of the target PCIe link; and a fifth processing module, configured to adjust the marginal probability associated with the link fault evaluation mean in the probability distribution model based on the historical average fault probability, and perform fault prediction based on the adjusted probability distribution model.

[0019] In one implementation, the target monitoring parameters include M types of parameters. The first processing module can be used to: determine the load state of the target device, wherein the target device is the device to which the target PCIe link belongs; divide the target monitoring parameters into at least one first type of monitoring parameter and at least one second type of monitoring parameter based on the load state, and determine a first total weight of the at least one first type of monitoring parameter and a second total weight of the at least one second type of monitoring parameter; wherein the at least one first type of monitoring parameter is the top N target monitoring parameters with a higher probability of degradation under the load state; M and N are positive integers, and M is greater than N; standardize the target monitoring parameters to obtain corresponding standard values; determine a first weight for each type of first type of monitoring parameter based on the first total weight of the at least one first type of monitoring parameter, and determine a second weight for each type of second type of monitoring parameter based on the standard value and the second total weight of the at least one second type of monitoring parameter; and perform a weighted summation of the standard values ​​based on the first weight and the second weight to obtain the first evaluation value.

[0020] In one optional implementation, the first processing module may be used to: divide the target monitoring parameters into at least one positive monitoring parameter and at least one negative monitoring parameter based on the parameter characteristics of the target monitoring parameters; standardize each of the negative monitoring parameters according to a preset first processing method to obtain a first standard value corresponding to each of the positive monitoring parameters; and standardize each of the positive monitoring parameters according to a preset second processing method to obtain a second standard value corresponding to each of the negative monitoring parameters.

[0021] Optionally, the first processing module may be configured to: for each of the forward monitoring parameters, obtain at least one first monitoring parameter of a first PCIe link, wherein the first PCIe link and the target PCIe link belong to the same target device, and the first monitoring parameter and the forward monitoring parameter are monitoring parameters of the same type; for each of the forward monitoring parameters, obtain the maximum value between the target monitoring parameter and the first monitoring parameter as a reference value corresponding to the forward monitoring parameter; for each of the forward monitoring parameters, perform standardization processing on the forward monitoring parameter according to a preset standardization processing method, combined with the corresponding reference value, to obtain the corresponding first standard value.

[0022] In one alternative implementation, the first processing module may be used to: obtain a first weight for each of the first type of monitoring parameters based on the differences in the degree of influence of different first type of monitoring parameters on link performance and the first total weight.

[0023] Optionally, the first processing module may be configured to: for each of the first type of monitoring parameters, obtain at least one second monitoring parameter of a second PCIE link; wherein the second PCIE link and the target PCIE link belong to the same target device, and the second monitoring parameter and the first type of monitoring parameters are of the same type; for each of the first type of monitoring parameters, obtain the variance value corresponding to the first type of monitoring parameter based on the second monitoring parameter and the first type of monitoring parameter; for each of the first type of monitoring parameters, perform parameter injection on the target device according to the variance value to obtain the corresponding link performance change value; and obtain a first weight for each of the first type of monitoring parameters based on the link performance change value corresponding to different first type of monitoring parameters and the first total weight.

[0024] In one implementation, the first processing module can be used to: obtain the sum of standard values ​​corresponding to each of the second type of monitoring parameters; for each of the second type of monitoring parameters, obtain the ratio of the standard value corresponding to the second type of monitoring parameter to the sum of the standard values, as a second weight ratio of the second type of monitoring parameter; for each of the second type of monitoring parameters, obtain the corresponding second weight of the second type of monitoring parameter based on the second weight ratio of the second type of monitoring parameter and the second total weight.

[0025] In one implementation, the historical performance characteristic parameters include multiple parameters, and the second processing module can be used to: determine a third weight corresponding to each of the historical performance characteristic parameters based on the degree of influence of different historical performance characteristic parameters on the transmission performance of the target PCIE link; normalize the historical performance characteristic parameters to obtain corresponding third standard values; and perform weighted summation on the third standard values ​​based on the third weights to obtain the second evaluation value.

[0026] In one implementation, the fifth processing module can be used to: obtain the value among the multiple predicted link evaluation values ​​output by the probability distribution model that is closest to the mean of the link failure evaluation, and use it as the target predicted link evaluation value; adjust the probability corresponding to the target predicted link evaluation value in the probability distribution model based on the historical average failure probability.

[0027] In one implementation, the fifth processing module can be used to: obtain multiple predicted link evaluation values ​​and a first predicted probability for each predicted link evaluation value based on the adjusted probability distribution model; obtain a fault prediction threshold; wherein the fault prediction threshold includes at least one of the following: a baseline threshold, a trend threshold, and a burst threshold; and perform fault probability prediction based on the first predicted probability for each predicted link evaluation value and the fault prediction threshold.

[0028] In one optional implementation, the fault prediction threshold includes the baseline threshold, and the fifth processing module can be used to: obtain a first target historical link evaluation value in which the target PCIE link is in normal operating condition from the historical link evaluation values; and obtain the baseline threshold based on the first target historical link evaluation value and a preset confidence interval.

[0029] In one implementation, the fault prediction threshold includes the burst threshold, and the fifth processing module can be used to: obtain the standard deviation between the historical link evaluation value and the baseline threshold; adjust the baseline threshold based on the standard deviation and a preset adjustment coefficient to obtain the burst threshold.

[0030] In one implementation, the fault prediction threshold includes the trend threshold, and the fifth processing module can be used to: obtain the rate of change of multiple link evaluation values ​​and the predicted probability of each link evaluation value of the target PCIE link within a target time period based on the multiple predicted link evaluation values, the predicted probability of each predicted link evaluation value, and at least a portion of the target historical link evaluation values; and perform fault probability prediction based on the multiple link evaluation value change rates, the predicted probability of each link evaluation value change rate, and the trend threshold.

[0031] In one implementation, the acquisition module can be used to: acquire the transmission rate of the target PCIe link; determine the sampling rate of the target monitoring parameters based on the transmission rate; and sample and acquire the target monitoring parameters according to the sampling rate.

[0032] Thirdly, this application proposes an electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the PCIE link fault prediction method as described in the first aspect.

[0033] Fourthly, this application proposes a storage medium storing instructions that, when executed on an electronic device, cause the electronic device to perform the method described in the first aspect.

[0034] Fifthly, this application proposes a program product comprising at least one of a program and instructions, wherein when the program and instructions are executed by an electronic device, they implement the steps of the method described in the first aspect.

[0035] The PCIe link fault prediction method, apparatus, device, and storage medium provided in this application can obtain a first evaluation value based on monitoring parameters of the target PCIe link and a second evaluation value based on historical performance characteristic parameters of the target PCIe link. A target link evaluation value is then obtained based on the first and second evaluation values. A probability distribution model of the link evaluation value is established based on the target link evaluation value and historical link evaluation values. The marginal probabilities associated with the mean link fault evaluation value in the probability distribution model are adjusted based on the historical average fault probability. Fault prediction is then performed based on the adjusted probability distribution model. This improves the accuracy of predicting PCIe link fault probabilities.

[0036] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0037] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0038] Figure 1 This is a flowchart illustrating a PCIe link fault prediction method provided in an embodiment of this application;

[0039] Figure 2 This is a flowchart illustrating another PCIe link fault prediction method provided in an embodiment of this application;

[0040] Figure 3 This is a schematic diagram of a benchmark value acquisition process provided in an embodiment of this application;

[0041] Figure 4 This is a schematic diagram of a weight value acquisition process provided in an embodiment of this application;

[0042] Figure 5 This is a flowchart illustrating another PCIe link fault prediction method provided in the embodiments of this application;

[0043] Figure 6 This is a flowchart illustrating another PCIe link fault prediction method provided in the embodiments of this application;

[0044] Figure 7 This is a schematic diagram of a fault prediction threshold acquisition process provided in an embodiment of this application;

[0045] Figure 8 This is a flowchart illustrating another PCIe link fault prediction method provided in the embodiments of this application;

[0046] Figure 9 This is a schematic diagram of a PCIE link fault prediction process provided in an embodiment of this application;

[0047] Figure 10 This is a schematic diagram of the structure of a PCIE link fault prediction device provided in an embodiment of this application;

[0048] Figure 11 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0049] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0050] The following describes a PCIe link fault prediction method and apparatus according to embodiments of this application with reference to the accompanying drawings.

[0051] Figure 1 This is a flowchart illustrating a PCIe link fault prediction method provided in an embodiment of this application. Figure 1 As shown, the method may include, but is not limited to, the following steps:

[0052] S101: Obtain the target monitoring parameters, historical performance characteristic parameters, historical average failure probability, and the average link failure evaluation value corresponding to the historical average failure probability of the target PCIE link.

[0053] In the embodiments of this application, the aforementioned PCIe link is the PCIe link of a storage device.

[0054] In the embodiments of this application, the target monitoring parameters include at least one of the following: number of bit errors, signal eye diagram opening, link transmission rate fluctuation value, differential signal voltage amplitude, channel crosstalk value, link chip temperature, number of data frame retransmissions, and link negotiation status code.

[0055] In the embodiments of this application, the aforementioned historical performance characteristic parameters include at least one of the following: attenuation characteristic parameters and signal optimization characteristic parameters.

[0056] Among them, the aforementioned attenuation-related characteristic parameters include at least one of the following: signal quality attenuation rate and bit error rate.

[0057] Among them, the above-mentioned signal optimization parameters include at least one of the following: eye opening increase and compensation voltage.

[0058] S102: Obtain the first evaluation value based on the target monitoring parameters.

[0059] For example, the target monitoring parameters are normalized and weighted based on preset weights to obtain the first evaluation value.

[0060] S103: Obtain the second evaluation value based on historical performance characteristic parameters.

[0061] For example, the historical performance characteristic parameters are normalized and weighted based on preset weights to obtain a second evaluation value.

[0062] In one implementation, the aforementioned historical performance characteristic parameters include multiple parameters. Based on the historical performance characteristic parameters, a second evaluation value is obtained. Based on the degree of influence of different historical performance characteristic parameters on the transmission performance of the target PCIE link, a third weight corresponding to each historical performance characteristic parameter is determined. The historical performance characteristic parameters are normalized to obtain the corresponding third standard value. Based on the third weight, the third standard value is weighted and summed to obtain the second evaluation value.

[0063] For example, for various historical performance characteristic parameters of the target PCIe link, the impact of their changes on the link transmission performance is analyzed, and a third weight is configured for each historical performance characteristic parameter according to the degree of impact, that is, the greater the degree of impact, the greater the corresponding weight; then, the parameters are normalized according to their units, ranges, and directions of advantage or disadvantage to obtain a third standard value; finally, the third standard value of each historical performance characteristic parameter is multiplied by its corresponding third weight, and all products are summed to obtain the second evaluation value.

[0064] S104: Obtain the target link evaluation value based on the first evaluation value and the second evaluation value.

[0065] For example, the target link evaluation value is obtained by adding the first evaluation value and the second evaluation value.

[0066] S105: Establish a probability distribution model based on the target link evaluation value and the historical link evaluation value of the target PCIE link.

[0067] For example, multiple historical link evaluation values ​​of the target PCIE link within a preset historical period are obtained, and a probability distribution model of the link evaluation value is established based on the target link evaluation value and the historical link evaluation values ​​of the target PCIE link.

[0068] S106: Adjust the marginal probability associated with the link failure evaluation mean in the probability distribution model based on the historical average failure probability, and perform failure prediction based on the adjusted probability distribution model.

[0069] For example, the link prediction evaluation value that is closest to the link failure evaluation mean in the probability distribution model is determined, and the historical average failure probability is added to the original prediction probability corresponding to the link prediction evaluation to obtain the new prediction probability of the link prediction evaluation, thereby completing the adjustment of the probability distribution model. Then, based on the adjusted probability distribution model and the preset link evaluation threshold, the prediction probability of the link prediction evaluation value being less than the threshold is obtained as the failure prediction result.

[0070] By implementing the embodiments of this application, a first evaluation value can be obtained based on the monitoring parameters of the target PCIe link, and a second evaluation value can be obtained based on the historical performance characteristic parameters of the target PCIe link. A target link evaluation value is then obtained based on the first and second evaluation values. A probability distribution model of the link evaluation value is established based on the target link evaluation value and the historical link evaluation values. The marginal probabilities associated with the mean link failure evaluation value in the probability distribution model are adjusted based on the historical average failure probability. Fault prediction is then performed based on the adjusted probability distribution model. This improves the accuracy of predicting PCIe link failure probabilities.

[0071] In some embodiments, the target monitoring parameters can be processed accordingly based on the operating scenario of the device to which the PCIe link belongs to obtain a first evaluation value. For example, please refer to [link to example]. Figure 2 , Figure 2 This is a flowchart illustrating another PCIe link fault prediction method provided in an embodiment of this application. For example... Figure 2 As shown, the method may include, but is not limited to, the following steps:

[0072] S201: Obtain the target monitoring parameters, historical performance characteristic parameters, historical average failure probability, and the average link failure evaluation value corresponding to the historical average failure probability of the target PCIE link.

[0073] In the embodiments of this application, step S201 can be implemented in any of the ways described in the embodiments of this application. The embodiments of this application do not limit this, nor will they be described in detail.

[0074] S202: Determine the load status of the target device, which is the device to which the target PCIe link belongs.

[0075] For example, the load status of the target device to which the target PCIe link belongs is determined.

[0076] In this embodiment of the application, the load state is any one of the following: high load, medium load, or low load.

[0077] In some embodiments, the historical performance characteristic parameters, historical average failure probability, and historical average failure probability obtained in the foregoing steps can be the historical performance characteristic parameters, historical average failure probability, and historical average failure probability obtained based on the historical data corresponding to the above-mentioned load states.

[0078] In other embodiments, the historical performance characteristic parameters, historical average failure probability, and historical average failure probability obtained in the aforementioned steps can be the historical performance characteristic parameters, historical average failure probability, and historical average failure probability obtained based on all historical data of the target device.

[0079] S203: Based on the load status, the target monitoring parameters are divided into at least one type of monitoring parameter and at least one type of monitoring parameter, and a first total weight of at least one type of monitoring parameter and a second total weight of at least one type of monitoring parameter are determined.

[0080] Among them, at least one type of first-class monitoring parameter is one of the top N target monitoring parameters with a high probability of degradation under load. M and N are positive integers, and M is greater than N.

[0081] For example, based on the load status, the target monitoring parameters are divided into at least one first type of monitoring parameters and at least one second type of monitoring parameters. The first type of monitoring parameters are the top N target monitoring parameters with a higher probability of degradation under the above load status among the M target monitoring parameters of the target PCIE link. The other monitoring parameters in the target monitoring parameters besides the first type of monitoring parameters are used as the second type of monitoring parameters. The first total weight of all first type of monitoring parameters and the second total weight of all second type of monitoring parameters are obtained. The sum of the first total weight and the second total weight is 1.

[0082] As an example, the first total weight mentioned above can be 0.5, and the second total weight mentioned above can be 0.5.

[0083] And determine the first total weight of at least one type of monitoring parameter and the second total weight of at least one type of monitoring parameter.

[0084] As an example, taking the target device's load state as high load, and the target monitoring parameters including bit error rate, signal eye diagram opening, link transmission rate fluctuation value, differential signal voltage amplitude, channel crosstalk value, link chip temperature, data frame retransmission count, and link negotiation status code as examples, the parameters most prone to deterioration under high load, such as rate fluctuation, eye diagram quality, temperature, and voltage weight, can be used as the first type of monitoring parameters, and the other target monitoring parameters besides the first type of monitoring parameters can be used as the second type of monitoring parameters.

[0085] S204: Standardize the target monitoring parameters to obtain the corresponding standard values.

[0086] For example, for each target monitoring parameter, a baseline value corresponding to the target monitoring parameter is obtained, and the target monitoring parameter is standardized based on the baseline value to obtain the corresponding standard value.

[0087] In one implementation, the above-mentioned standardization of the target monitoring parameters to obtain the corresponding standard values ​​may include the following steps S2041-S2043:

[0088] S2041: Based on the parameter characteristics of the target monitoring parameters, the target monitoring parameters are divided into at least one positive monitoring parameter and at least one negative monitoring parameter.

[0089] In the embodiments of this application, the above-mentioned parameter characteristics may include at least one of the following: unit, range, and superiority / inferiority direction.

[0090] For example, the number of bit errors, crosstalk value, and number of retransmissions can be classified as negative monitoring parameters, while eye diagram opening, transmission rate stability, and differential voltage amplitude can be used as positive monitoring parameters.

[0091] S2042: Standardize each negative monitoring parameter according to the preset first processing method to obtain the first standard value corresponding to each positive monitoring parameter.

[0092] For example, the first standard value corresponding to each negative monitoring parameter is obtained by standardizing each negative monitoring parameter based on the following formula:

[0093] First standard value = 100 - (Actual value of negative monitoring parameter / Maximum threshold) × 100

[0094] In one optional implementation, standardizing each negative monitoring parameter according to a preset first processing method to obtain a first standard value corresponding to each negative monitoring parameter may include the following steps: for each negative monitoring parameter, obtaining at least one first monitoring parameter of a first PCIe link, wherein the first PCIe link and the target PCIe link belong to the same target device, and the first monitoring parameter and the negative monitoring parameter are monitoring parameters of the same type; for each negative monitoring parameter, obtaining the maximum value between the target monitoring parameter and the first monitoring parameter as the reference value corresponding to the negative monitoring parameter; for each negative monitoring parameter, standardizing the negative monitoring parameter according to a preset standardization processing method and in combination with the corresponding reference value to obtain the corresponding first standard value.

[0095] For example, taking the negative monitoring parameter as the number of bit errors, the number of bit errors of the first PCIE link belonging to the same target device as the target PCIE link is obtained. The maximum number of bit errors between the number of bit errors of the target PCIE link and the number of bit errors of the first PCIE link is obtained as a benchmark value. This benchmark value is used as the maximum threshold in the aforementioned calculation formula to standardize the number of bit errors and obtain the corresponding first standard value.

[0096] S2043: Standardize each positive monitoring parameter according to the preset second processing method to obtain the second standard value corresponding to each negative monitoring parameter.

[0097] For example, the negative monitoring parameters are standardized based on the following formula to obtain the first standard value corresponding to each positive monitoring parameter:

[0098] Second standard value = (Actual value of positive monitoring parameter / Ideal value) × 100

[0099] The aforementioned ideal value can be a pre-set value.

[0100] In one alternative implementation, for each positive monitoring parameter, at least one second monitoring parameter of the first PCIE link is obtained, wherein the first PCIE link and the target PCIE link belong to the same target device, and the second monitoring parameter and the positive monitoring parameter are the same type of monitoring parameter; for each positive monitoring parameter, the maximum value between the target monitoring parameter and the first monitoring parameter is obtained as the reference value corresponding to the positive monitoring parameter (i.e., the ideal value in the aforementioned calculation formula).

[0101] As an example, please see Figure 3 , Figure 3 This is a schematic diagram of a benchmark value acquisition process provided in an embodiment of this application. For example... Figure 3 As shown, firstly, eight core parameters, such as the number of bit errors at the CPU receiver and the signal eye diagram opening, are collected. Then, the eight parameters are normalized according to differences in unit, range, and direction of superiority or inferiority. Next, initial weights are assigned to the eight normalized parameters based on the operating status of the devices. Finally, the core parameters of each device are shared in the cluster, and the highest value is selected as the standardized score to calculate the maximum threshold or ideal value, thereby calculating the standardized score as the baseline value.

[0102] S205: Determine the first weight of each type of monitoring parameter based on the first total weight of at least one type of monitoring parameter, and determine the second weight of each type of monitoring parameter based on the second total weight of at least one type of monitoring parameter.

[0103] For example, a corresponding weight ratio can be set for each type of first-class monitoring parameter and each type of second-class monitoring parameter, thereby determining the first weight of each type of first-class monitoring parameter based on the first total weight of at least one type of first-class monitoring parameter, and determining the second weight of each type of second-class monitoring parameter based on the second total weight of at least one type of second-class monitoring parameter.

[0104] As an example, taking the first type of monitoring parameters, including parameter rate fluctuation, eye diagram quality, temperature, and voltage weights, and sorting their influence from largest to smallest as parameter rate fluctuation, eye diagram quality, temperature, and voltage weights, we can set the weight ratio of parameter rate fluctuation to 0.4, the weight ratio of eye diagram quality to 0.3, the weight ratio of temperature to 0.2, and the weight ratio of voltage to 0.1. If the first total weight is 0.5, then the weight of parameter rate fluctuation is 0.2, the weight of eye diagram quality is 0.15, the weight of temperature is 0.1, and the weight of voltage is 0.05.

[0105] In one implementation, determining the first weight of each first type of monitoring parameter based on the first total weight of at least one first type of monitoring parameter includes: obtaining the first weight of each first type of monitoring parameter based on the differences in the degree of influence of different first type of monitoring parameters on link performance and the first total weight.

[0106] For example, the impact of each type of first-class monitoring parameter on the target PCIE link is sorted from largest to smallest. The greater the impact, the greater the corresponding weight ratio, thereby obtaining the weight ratio of each type of first-class monitoring parameter. Based on the weight ratio of each type of first-class monitoring parameter and the first total weight, the first weight of each type of first-class monitoring parameter is obtained.

[0107] In one alternative implementation, obtaining the first weight for each type of monitoring parameter based on the differences in the degree of influence of different first-type monitoring parameters on link performance and the first total weight may include the following steps A1-A4:

[0108] A1: For each type of first-class monitoring parameter, obtain at least one second monitoring parameter of the second PCIE link; wherein the second PCIE link and the target PCIE link belong to the same target device, and the second monitoring parameter and the first-class monitoring parameter are of the same type.

[0109] For example, taking the link transmission rate fluctuation value as the first type of monitoring parameter, the link transmission rate fluctuation value of at least one second PCIE link belonging to the target device and the target PCIE link is obtained.

[0110] A2: For each type of first-class monitoring parameter, obtain the variance value corresponding to the first-class monitoring parameter based on the second-class monitoring parameter and the first-class monitoring parameter.

[0111] For example, taking the link transmission rate fluctuation value as the first type of monitoring parameter, the average link transmission rate fluctuation value is obtained based on the link transmission rate fluctuation value of the target PCIE link and the link transmission rate fluctuation value of at least one second PCIE link. Then, according to the variance calculation formula, the variance value corresponding to the link transmission rate fluctuation value of the target PCIE link is obtained.

[0112] A3: For each type of first-class monitoring parameter, the corresponding link performance change value is obtained by injecting parameters into the target device according to the variance value.

[0113] For example, for each type of first-class monitoring parameter, + / -5% of the corresponding variance value is used as the parameter injection value. The parameter injection value is injected into the target device. After the injection is completed, the data IOPS (Input / Output Operations Per Second) is monitored. The change in data IOPS before and after the injection is used as the link performance change value.

[0114] A4: Based on the link performance change value and the first total weight corresponding to different first-class monitoring parameters, obtain the first weight of each first-class monitoring parameter.

[0115] For example, the link performance change values ​​of each type of first-class monitoring parameter are sorted from largest to smallest. The larger the link performance change value, the larger the corresponding weight ratio, thereby obtaining the weight ratio of each type of first-class monitoring parameter. Based on the weight ratio of each type of first-class monitoring parameter and the first total weight, the first weight of each type of first-class monitoring parameter is obtained.

[0116] As an example, the second type of monitoring parameters can be standardized to obtain corresponding standard values. Then, based on the proportional relationship between the standard values ​​of different second type of monitoring parameters and combined with the second total weight, the second weight corresponding to each type of second type of monitoring parameter can be obtained.

[0117] In one implementation, determining the second weight for each type of monitoring parameter based on the weight value of at least one type of second-class monitoring parameter and the second total weight includes: obtaining the sum of the standard values ​​corresponding to each type of second-class monitoring parameter; for each type of second-class monitoring parameter, obtaining the ratio of the standard value corresponding to the second-class monitoring parameter to the sum of the standard values, as the second weight ratio of the second-class monitoring parameter; for each type of second-class monitoring parameter, obtaining the corresponding second weight based on the second weight ratio and the second total weight of the second-class monitoring parameter. Normalization is then performed based on the standard values ​​corresponding to each type of second-class monitoring parameter and the second total weight corresponding to the second-class monitoring parameter to obtain the second weight corresponding to each type of second-class monitoring parameter.

[0118] For example, the standard values ​​corresponding to all the second type of monitoring parameters are summed to obtain the sum of the standard values. Then, for each second type of monitoring parameter, the standard value of the second type of parameter is divided by the sum of the standard values ​​to obtain the corresponding proportion value. The proportion value is then multiplied by the second total weight to obtain the second weight of the second type of monitoring parameter.

[0119] As an example, please see Figure 4 , Figure 4 This is a schematic diagram illustrating a weight value acquisition process provided in an embodiment of this application. For example... Figure 4 As shown, monitoring parameters that are prone to deterioration under different scenarios can be extracted, sorted, normalized, and scored to obtain corresponding built-in weights. The scores can be adjusted according to the operating scenario of the target device to ensure that the data is closer to the actual state of the target device.

[0120] S206: The standard values ​​are weighted and summed based on the first weight and the second weight to obtain the first evaluation value.

[0121] In the embodiments of this application, step S206 can be implemented in any of the ways described in the embodiments of this application. The embodiments of this application do not limit this, nor will they be described in detail.

[0122] S207: Obtain the second evaluation value based on historical performance characteristic parameters.

[0123] In the embodiments of this application, step S207 can be implemented in any of the embodiments of this application. The embodiments of this application do not limit this and will not be described in detail.

[0124] S208: Obtain the target link evaluation value based on the first evaluation value and the second evaluation value.

[0125] In the embodiments of this application, step S208 can be implemented in any of the ways described in the embodiments of this application. The embodiments of this application do not limit this, nor will they be described in detail.

[0126] S209: Establish a probability distribution model based on the target link evaluation value and the historical link evaluation value of the target PCIE link.

[0127] In the embodiments of this application, step S209 can be implemented in any of the ways described in the embodiments of this application. The embodiments of this application do not limit this, nor will they be described in detail.

[0128] S210: Adjust the marginal probability associated with the mean of link failure evaluation in the probability distribution model based on the historical average failure probability, and perform failure prediction based on the adjusted probability distribution model.

[0129] In the embodiments of this application, step S210 can be implemented in any of the embodiments of this application. The embodiments of this application do not limit this and will not be described in detail.

[0130] By implementing the embodiments of this application, target monitoring parameters can be divided into a first type of monitoring parameters and a second type of monitoring parameters based on the load status of the target device to which the target PCIe link belongs. The first and second types of monitoring parameters are then processed accordingly to obtain a weight for each type of target monitoring parameter. Based on this weight, a weighted sum of the standard values ​​of the target monitoring parameters is then performed to obtain a first evaluation value. This can further improve the accuracy of predicting the failure probability of PCIe links.

[0131] In one implementation, the marginal probabilities associated with the link failure evaluation mean in the probability distribution model can be adjusted based on the historical average failure probability, so that the probability distribution model better reflects the actual state of the target device. As an example, please refer to... Figure 5 , Figure 5 This is a flowchart illustrating another PCIe link fault prediction method provided in an embodiment of this application. For example... Figure 5 As shown, the method may include, but is not limited to, the following steps:

[0132] S501: Obtain the target monitoring parameters, historical performance characteristic parameters, historical average failure probability, and the average link failure evaluation value corresponding to the historical average failure probability of the target PCIe link.

[0133] In the embodiments of this application, step S501 can be implemented in any of the embodiments of this application. The embodiments of this application do not limit this and will not be described in detail.

[0134] S502: Obtain the first evaluation value based on the target monitoring parameters.

[0135] In the embodiments of this application, step S502 can be implemented in any of the embodiments of this application. The embodiments of this application do not limit this and will not be described in detail.

[0136] S503: Obtain the second evaluation value based on historical performance characteristic parameters.

[0137] In the embodiments of this application, step S503 can be implemented in any of the embodiments of this application. The embodiments of this application do not limit this and will not be described in detail.

[0138] S504: Obtain the target link evaluation value based on the first evaluation value and the second evaluation value.

[0139] In the embodiments of this application, step S504 can be implemented in any of the embodiments of this application. The embodiments of this application do not limit this and will not be described in detail.

[0140] S505: Establish a probability distribution model based on the target link evaluation value and the historical link evaluation value of the target PCIE link.

[0141] In the embodiments of this application, step S505 can be implemented in any of the embodiments of this application. The embodiments of this application do not limit this and will not be described in detail.

[0142] S506: Obtain the value that is closest to the average value of the link failure evaluation among the multiple predicted link evaluation values ​​output by the probability prediction model, and use it as the target predicted link evaluation value.

[0143] It is understandable that the probability distribution model includes multiple prediction link evaluation values ​​and the prediction probability corresponding to each prediction link evaluation value.

[0144] S507: Adjust the probability corresponding to the target predicted link evaluation value in the probability distribution model based on the historical average failure probability.

[0145] For example, the original probability of the target predicted link evaluation value obtained based on the probability prediction model is added to the historical average failure probability to obtain the new predicted probability of the target predicted link evaluation value.

[0146] S508: Fault prediction based on the adjusted probability distribution model.

[0147] In the embodiments of this application, step S508 can be implemented in any of the ways described in the embodiments of this application. The embodiments of this application do not limit this, nor will they be described in detail.

[0148] By implementing the embodiments of this application, the marginal probabilities associated with the mean of link fault evaluation in the probability distribution model can be adjusted based on the historical average fault probability, so that the predicted probability output by the probability distribution model is more consistent with the historical operating status of the target equipment, thereby further improving the accuracy of fault prediction.

[0149] In some embodiments, multiple fault probability thresholds can be obtained to perform fault prediction based on a probability distribution model combined with these thresholds. For an example, please refer to [link to example]. Figure 6 , Figure 6 This is a flowchart illustrating another PCIe link fault prediction method provided in an embodiment of this application. For example... Figure 6 As shown, the method may include, but is not limited to, the following steps:

[0150] S601: Obtain the target monitoring parameters, historical performance characteristic parameters, historical average failure probability, and the average link failure evaluation value corresponding to the historical average failure probability of the target PCIe link.

[0151] In the embodiments of this application, step S601 can be implemented in any of the ways described in the embodiments of this application. The embodiments of this application do not limit this, nor will they be described in detail.

[0152] S602: Obtain the first evaluation value based on the target monitoring parameters.

[0153] In the embodiments of this application, step S602 can be implemented in any of the ways described in the embodiments of this application. The embodiments of this application do not limit this, nor will they be described in detail.

[0154] S603: Obtain a second evaluation value based on historical performance characteristic parameters.

[0155] In the embodiments of this application, step S603 can be implemented in any of the embodiments of this application. The embodiments of this application do not limit this and will not be described in detail.

[0156] S604: Obtain the target link evaluation value based on the first evaluation value and the second evaluation value.

[0157] In the embodiments of this application, step S604 can be implemented in any of the ways described in the embodiments of this application. The embodiments of this application do not limit this, nor will they be described in detail.

[0158] S605: Establish a probability distribution model based on the target link evaluation value and the historical link evaluation value of the target PCIE link.

[0159] In the embodiments of this application, step S605 can be implemented in any of the embodiments of this application. The embodiments of this application do not limit this and will not be described in detail.

[0160] S606: Based on the adjusted probability distribution model, obtain multiple predicted link evaluation values ​​and the predicted probability of each predicted link evaluation value.

[0161] S607: Obtain the fault prediction threshold.

[0162] In the embodiments of this application, the fault prediction threshold includes at least one of the following: baseline threshold, trend threshold, and burst threshold.

[0163] It should be noted that the baseline threshold can be set based on a preset confidence interval (e.g., 95%) of historical link evaluation values ​​to reflect the threshold of normal operation of the PCIE link; the trend threshold is set by calculating the slope change of historical link evaluation values ​​through a time series model; and the burst threshold is a threshold used to capture abnormal spikes in the PCIE link.

[0164] In one implementation, the aforementioned fault prediction threshold includes a baseline threshold. Obtaining the fault prediction threshold includes: obtaining a first target historical link evaluation value in which the target PCIE link is in normal operating condition from the historical link evaluation values; and obtaining the baseline threshold based on the first target historical link evaluation value and a preset confidence interval.

[0165] In some embodiments, for each load state of the target device, a first target historical link evaluation value can be obtained when the target PCIe link is in normal operation under that load state; based on the first target historical link evaluation value and a preset confidence interval, a baseline threshold corresponding to that load state can be obtained.

[0166] S608: Fault probability prediction is performed based on the predicted probability of each predicted link evaluation value and the fault prediction threshold.

[0167] For example, taking the fault prediction threshold including the baseline threshold as an example, the first predicted link evaluation value that is greater than the baseline threshold is obtained from the predicted link evaluation values, and the prediction probabilities corresponding to all the first predicted link evaluation values ​​are added together as the probability that the target PCIe link is operating normally.

[0168] For example, taking the fault prediction threshold including the burst threshold as an example, the first predicted link evaluation value that is greater than the burst threshold is obtained from the predicted link evaluation values, and the prediction probabilities corresponding to all the second predicted link evaluation values ​​are added together as the probability that the target PCIE link will not experience a burst failure.

[0169] In one implementation, the fault prediction threshold includes a baseline threshold. Obtaining the fault prediction threshold may include the following steps: obtaining the evaluation value of a first target historical link in which the target PCIE link is in normal operating condition from the historical link evaluation values; and obtaining the baseline threshold based on the first target historical link evaluation value and a preset confidence interval.

[0170] For example, firstly, the evaluation values ​​of the target PCIe link under normal operation are selected from the historical link evaluation values, and the result is the first target historical link evaluation value under normal operation; then, based on the set of the first target historical link evaluation values, combined with a preset confidence interval (e.g., 95%), the boundary value corresponding to the confidence interval is determined by statistical calculation, and the baseline threshold reflecting the normal operation boundary of the link is obtained.

[0171] In one optional implementation, the fault prediction threshold includes a burst threshold. Obtaining the fault prediction threshold may include the following steps: obtaining the standard deviation between historical link evaluation values ​​and a baseline threshold; adjusting the baseline threshold based on the standard deviation and a preset adjustment coefficient to obtain the burst threshold.

[0172] For example, the deviation between each historical link evaluation value and a predetermined baseline threshold is obtained. Then, the standard deviation is calculated for all deviations to obtain the standard deviation between the historical link evaluation value and the baseline threshold. Next, the obtained standard deviation is multiplied by a preset adjustment factor (e.g., 3), and the calculation result is added to the baseline threshold. The final value is the burst threshold used to monitor abnormal spikes in real-time data.

[0173] In one implementation, the fault prediction threshold includes a trend threshold. Fault probability prediction is performed based on a first prediction probability of each predicted link evaluation value and the fault prediction threshold, including: obtaining the rate of change of multiple link evaluation values ​​and the prediction probability of the rate of change of each link evaluation value of the target PCIE link within a target time period based on multiple predicted link evaluation values, the prediction probability of each predicted link evaluation value, and at least a portion of the target historical link evaluation values; and performing fault probability prediction based on the rate of change of multiple link evaluation values, the prediction probability of the rate of change of each link evaluation value, and the trend threshold.

[0174] As an example, please see Figure 7 , Figure 7This is a schematic diagram of a fault prediction threshold acquisition process provided in an embodiment of this application. Figure 7 As shown, a probability distribution can be generated based on historical link evaluation values, with each day as the time unit, and the probability distribution generated the previous day is used as the initial value for the next day's probability distribution. Then, the link evaluation value of each PCIE link is recorded every hour, and the current record is used to overwrite the data of the corresponding PCIE link at the corresponding time of the previous day, so as to realize the real-time refresh of the probability distribution. Then, the features extracted from the historical link feature data are divided into attenuation category (signal quality attenuation rate, bit error rate), historical fault category (fault record label), and signal optimization category (signal quality optimization parameters). The attenuation category and signal optimization category parameters are standardized and weighted and superimposed on the link evaluation values ​​of the previous and next 5 minutes. The historical fault category parameters are mapped to the edge of the probability distribution and associated with similar signal quality parameters to improve the probability of the corresponding edge. Finally, a three-level dynamic threshold system including baseline threshold, trend threshold and burst threshold is constructed.

[0175] In some embodiments, the sampling frequency of the monitoring parameters can be adjusted according to the link status of the target PCIe link. For example, please refer to [link to example]. Figure 8 , Figure 8 This is a flowchart illustrating another PCIe link fault prediction method provided in an embodiment of this application. For example... Figure 8 As shown, the method may include, but is not limited to, the following steps:

[0176] S801: Obtain the transmission rate of the target PCIe link.

[0177] For example, the actual transmission rate of the target PCIe link can be collected in real time by using a bandwidth probe deployed on the link gateway of the target PCIe link.

[0178] S802: Determine the sampling rate of target monitoring parameters based on the transmission rate.

[0179] For example, multiple transmission rate ranges and a preset sampling frequency corresponding to each transmission rate range are preset, so that the preset sampling frequency corresponding to the transmission rate range to which the transmission rate of the target PCIE link belongs is used as the sampling frequency for collecting the target monitoring parameters of the target PCIE link.

[0180] As an example, when the transmission rate is detected to be stable at 10Gbps or above, the sampling frequency is automatically increased to once every 100 milliseconds, which is equivalent to generating 10 sets of sampling data per second, ensuring "no blind spots" in capturing instantaneous signal distortions; while at lower transmission rates (e.g., below 1Gbps), the sampling frequency can be reduced to once every 500 milliseconds or once per second, thereby significantly reducing the computational load of the nodes while ensuring basic monitoring needs.

[0181] S803: Obtain target monitoring parameters by sampling according to the sampling rate.

[0182] S804: Obtain the historical performance characteristic parameters, historical average failure probability, and the average link failure evaluation value corresponding to the historical average failure probability of the target PCIe link.

[0183] In the embodiments of this application, step S804 can be implemented in any of the embodiments of this application. The embodiments of this application do not limit this and will not be described in detail.

[0184] S805: Obtain the first evaluation value based on the target monitoring parameters.

[0185] In the embodiments of this application, step S805 can be implemented in any of the embodiments of this application. The embodiments of this application do not limit this and will not be described in detail.

[0186] S806: Obtain the second evaluation value based on historical performance characteristic parameters.

[0187] In the embodiments of this application, step S806 can be implemented in any of the embodiments of this application. The embodiments of this application do not limit this and will not be described in detail.

[0188] S807: Obtain the target link evaluation value based on the first evaluation value and the second evaluation value.

[0189] In the embodiments of this application, step S807 can be implemented in any of the embodiments of this application. The embodiments of this application do not limit this and will not be described in detail.

[0190] S808: Establish a probability distribution model based on the target link evaluation value and the historical link evaluation value of the target PCIE link.

[0191] In the embodiments of this application, step S808 can be implemented in any of the ways described in the embodiments of this application. The embodiments of this application do not limit this, nor will they be described in detail.

[0192] S809: Adjust the marginal probability associated with the mean of link failure evaluation in the probability distribution model based on the historical average failure probability, and perform failure prediction based on the adjusted probability distribution model.

[0193] In the embodiments of this application, step S809 can be implemented in any of the embodiments of this application. The embodiments of this application do not limit this and will not be described in detail.

[0194] By implementing the embodiments of this application, an adaptive adjustment mechanism for the sampling frequency of target monitoring parameters can be implemented based on the transmission rate of the target PCIE link, avoiding the sampling frequency from lagging behind changes in link status, thereby improving the real-time performance and accuracy of fault prediction.

[0195] As an example, please see Figure 9 , Figure 9 This is a schematic diagram of a PCIe link fault prediction process provided in an embodiment of this application. Figure 9 As shown, real-time data sampling is first performed on the PCIe link, and dynamic scenario adjustments are made in conjunction with the device's operating scenario. At the same time, link bandwidth sampling is carried out through the bandwidth probe of the link gateway. After the real-time sampled data is acquired, it is coupled with the previously collected and processed historical data to form a dataset that is closer to the current link status. Finally, a probability distribution is generated on a daily basis based on this dataset, with the probability distribution of the previous day as the initial value for the next day. The link evaluation value record of the PCIe link is updated every hour to build a probability distribution model, which provides a basis for subsequent fault probability judgment and early warning.

[0196] To implement the above embodiments, this application also proposes a PCIE link fault prediction system. The system includes a data acquisition device and a processing device. The data acquisition device is used to acquire target monitoring parameters of the target PCIE link and send the target monitoring parameters to the processing device. The processing device is used to predict the fault probability based on the PCIE link fault prediction method provided in any embodiment of this application.

[0197] During signal modulation and transmission, the acquisition device monitors the amplitude, phase, and other characteristics of the transmitted signal containing target monitoring parameters in real time. It captures abnormal fluctuations through a sliding window, suppressing potential signal distortion at the transmitting end in advance. The processing device employs a sliding mean filtering algorithm to dynamically adjust the time window based on link noise characteristics, continuously calculating the mean of the signal within the window. This mean replaces the original value at the center of the window, smoothing high-frequency jitter, effectively eliminating isolated pulses shorter than the window length, preserving the true signal trend, and pushing the preprocessed target monitoring parameters to a local cache queue in real time via a lightweight protocol. It retains the most recent preset duration (e.g., 1 hour) of original data, providing low-latency data input for real-time monitoring.

[0198] For example, the processing device can deploy a local cache queue at the edge node through the system, using a "first-in, first-out" mechanism to retain the most recent raw data for a preset time period. This provides low-latency data input for real-time fault prediction and provides original evidence for anomaly tracing. That is, when a link anomaly is detected, the raw data at the corresponding time point in the cache can be quickly retrieved to help engineers locate the root cause of the anomaly.

[0199] Please see Figure 10 , Figure 10 This is a schematic diagram of the structure of a PCIe link fault prediction device provided in an embodiment of this application. Figure 10 As shown, the device 1000 includes: an acquisition module 1001, used to acquire target monitoring parameters, historical performance characteristic parameters, historical average failure probability, and the average link failure evaluation value corresponding to the historical average failure probability of the target PCIE link; a first processing module 1002, used to obtain a first evaluation value based on the target monitoring parameters; a second processing module 1003, used to obtain a second evaluation value based on the historical performance characteristic parameters; a third processing module 1004, used to obtain a target link evaluation value based on the first and second evaluation values; a fourth processing module 1005, used to establish a probability distribution model based on the target link evaluation value and the historical link evaluation values ​​of the target PCIE link; and a fifth processing module 1006, used to adjust the marginal probability associated with the average link failure evaluation value in the probability distribution model based on the historical average failure probability, and to perform fault prediction based on the adjusted probability distribution model.

[0200] In one implementation, the target monitoring parameters include M types of parameters. The first processing module 1002 can be used to: determine the load status of the target device, where the target device is the device to which the target PCIE link belongs; divide the target monitoring parameters into at least one first type of monitoring parameter and at least one second type of monitoring parameter based on the load status, and determine the first total weight of the at least one first type of monitoring parameter and the second total weight of the at least one second type of monitoring parameter; wherein, the at least one first type of monitoring parameter is the top N target monitoring parameters with a high probability of degradation under load status; M and N are positive integers, and M is greater than N; standardize the target monitoring parameters to obtain corresponding standard values; determine the first weight of each first type of monitoring parameter based on the first total weight of the at least one first type of monitoring parameter, and determine the second weight of each second type of monitoring parameter based on the standard value and the second total weight of the at least one second type of monitoring parameter; and perform a weighted summation of the standard values ​​based on the first weight and the second weight to obtain a first evaluation value.

[0201] In one optional implementation, the first processing module 1002 can be used to: divide the target monitoring parameters into at least one positive monitoring parameter and at least one negative monitoring parameter based on the parameter characteristics of the target monitoring parameters; standardize each negative monitoring parameter according to a preset first processing method to obtain a first standard value corresponding to each positive monitoring parameter; and standardize each positive monitoring parameter according to a preset second processing method to obtain a second standard value corresponding to each negative monitoring parameter.

[0202] Optionally, the first processing module 1002 can be used to: for each positive monitoring parameter, obtain at least one first monitoring parameter of a first PCIE link, wherein the first PCIE link and the target PCIE link belong to the same target device, and the first monitoring parameter and the positive monitoring parameter are monitoring parameters of the same type; for each positive monitoring parameter, obtain the maximum value between the target monitoring parameter and the first monitoring parameter as the reference value corresponding to the positive monitoring parameter; for each positive monitoring parameter, perform standardization processing on the positive monitoring parameter according to a preset standardization processing method, combined with the corresponding reference value, to obtain the corresponding first standard value.

[0203] In one alternative implementation, the first processing module 1002 may be used to: obtain the first weight of each type of monitoring parameter based on the differences in the degree of influence of different first-type monitoring parameters on link performance and the first total weight.

[0204] Optionally, the first processing module 1002 may be used to: for each type of first-class monitoring parameter, obtain at least one second monitoring parameter of a second PCIE link; wherein the second PCIE link and the target PCIE link belong to the same target device, and the second monitoring parameter and the first-class monitoring parameter are of the same type; for each type of first-class monitoring parameter, obtain the variance value corresponding to the first-class monitoring parameter based on the second monitoring parameter and the first-class monitoring parameter; for each type of first-class monitoring parameter, perform parameter injection on the target device according to the variance value to obtain the corresponding link performance change value; and obtain the first weight of each type of first-class monitoring parameter based on the link performance change value corresponding to different first-class monitoring parameters and the first total weight.

[0205] In one implementation, the first processing module 1002 can be used to: obtain the sum of the standard values ​​corresponding to each second type of monitoring parameter; for each second type of monitoring parameter, obtain the ratio of the standard value corresponding to the second type of monitoring parameter to the sum of the standard values, as the second weight ratio of the second type of monitoring parameter; for each second type of monitoring parameter, obtain the corresponding second weight of the second type of monitoring parameter based on the second weight ratio and the second total weight of the second type of monitoring parameter.

[0206] In one implementation, the historical performance characteristic parameters include multiple parameters. The second processing module 1003 can be used to: determine the third weight corresponding to each historical performance characteristic parameter based on the degree of influence of different historical performance characteristic parameters on the transmission performance of the target PCIE link; normalize the historical performance characteristic parameters to obtain the corresponding third standard value; and perform weighted summation on the third standard value based on the third weight to obtain the second evaluation value.

[0207] In one implementation, the fifth processing module 1006 can be used to: obtain the value among the multiple predicted link evaluation values ​​output by the probability distribution model that is closest to the average link failure evaluation value, and use it as the target predicted link evaluation value; adjust the probability corresponding to the target predicted link evaluation value in the probability distribution model based on the historical average failure probability.

[0208] In one implementation, the fifth processing module 1006 can be used to: obtain multiple predicted link evaluation values ​​and a first predicted probability of each predicted link evaluation value based on the adjusted probability distribution model; obtain a fault prediction threshold; wherein the fault prediction threshold includes at least one of the following: a baseline threshold, a trend threshold, and a burst threshold; and perform fault probability prediction based on the first predicted probability of each predicted link evaluation value and the fault prediction threshold.

[0209] In one optional implementation, the fault prediction threshold includes a baseline threshold, and the fifth processing module 1006 can be used to: obtain the first target historical link evaluation value in which the target PCIE link is in normal operating condition from the historical link evaluation values; and obtain the baseline threshold based on the first target historical link evaluation value and a preset confidence interval.

[0210] In one implementation, the fault prediction threshold includes a burst threshold. The fifth processing module 1006 can be used to: obtain the standard deviation between the historical link evaluation value and the baseline threshold; adjust the baseline threshold based on the standard deviation and a preset adjustment coefficient to obtain the burst threshold.

[0211] In one implementation, the fault prediction threshold includes a trend threshold, and the fifth processing module 1006 can be used to: obtain the rate of change of multiple link evaluation values ​​and the predicted probability of the rate of change of each link evaluation value of the target PCIE link within the target time period based on multiple predicted link evaluation values, the predicted probability of each predicted link evaluation value, and at least some of the target historical link evaluation values ​​from the historical link evaluation values; and perform fault probability prediction based on the rate of change of multiple link evaluation values, the predicted probability of the rate of change of each link evaluation value, and the trend threshold.

[0212] In one implementation, the acquisition module 1001 can be used to: acquire the transmission rate of the target PCIe link; determine the sampling rate of the target monitoring parameters based on the transmission rate; and acquire the target monitoring parameters by sampling according to the sampling rate.

[0213] The apparatus of this application embodiment can obtain a first evaluation value based on the monitoring parameters of the target PCIe link, and a second evaluation value based on the historical performance characteristic parameters of the target PCIe link. A target link evaluation value is then obtained based on the first and second evaluation values. A probability distribution model of the link evaluation value is established based on the target link evaluation value and the historical link evaluation values. The marginal probabilities associated with the mean link failure evaluation value in the probability distribution model are adjusted based on the historical average failure probability. Fault prediction is then performed based on the adjusted probability distribution model. This improves the accuracy of predicting PCIe link failure probabilities.

[0214] It should be noted that the foregoing explanation of the PCIe link fault prediction method embodiment also applies to the PCIe link fault prediction device of this embodiment, and will not be repeated here.

[0215] To implement the above embodiments, this application also proposes an electronic device. Please see [link to relevant documentation]. Figure 11 , Figure 11 This is a schematic diagram of the structure of the electronic device provided in an embodiment of this application. For example... Figure 11 As shown, the electronic device 1100 includes: a processor 1101 and a memory 1102 communicatively connected to the processor 1101; the memory 1102 stores computer-executable instructions; the processor 1101 executes the computer-executable instructions stored in the memory to implement the method provided in the foregoing embodiments.

[0216] To implement the above embodiments, this application also proposes a storage medium storing instructions that, when executed on an electronic device, cause the electronic device to perform the methods provided in the foregoing embodiments.

[0217] To implement the above embodiments, this application also proposes a program product, including at least one of a program and instructions, wherein when the program and instructions are executed by an electronic device, they implement the steps of the method provided in the foregoing embodiments.

[0218] It should be noted that the acquisition, transmission, storage, use, and processing of data in this application comply with the relevant provisions of national laws and regulations and do not violate public order and good morals.

[0219] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, data stored, data displayed, etc.) and signals involved in this application are all authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0220] It is worth noting that in the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, they do not mean that the applicant has used or necessarily used the solution.

[0221] In the description of this application, unless otherwise stated, " / " means "or", for example, A / B can mean A or B; "and / or" in this document is merely a description of the relationship between related objects, indicating that there can be three relationships, for example, A and / or B can mean: A exists alone, A and B exist simultaneously, and B exists alone.

[0222] In the foregoing descriptions of the embodiments, the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0223] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0224] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0225] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which programs can be printed, because programs can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0226] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0227] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by a program instructing related hardware, and the program can be stored in a computer-readable storage medium. When executed, the program includes one or a combination of the steps of the method embodiments.

[0228] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0229] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

Claims

1. A method for predicting faults in a PCIe link, characterized in that, include: Obtain the target monitoring parameters, historical performance characteristic parameters, historical average failure probability, and the average link failure evaluation value corresponding to the historical average failure probability of the target PCIE link; Based on the target monitoring parameters, a first evaluation value is obtained; Based on the aforementioned historical performance characteristic parameters, a second evaluation value is obtained; The target link evaluation value is obtained based on the first evaluation value and the second evaluation value; A probability distribution model is established based on the target link evaluation value and the historical link evaluation value of the target PCIE link; The marginal probabilities associated with the link failure evaluation mean in the probability distribution model are adjusted based on the historical average failure probability, and failure prediction is performed based on the adjusted probability distribution model. The adjustment of the marginal probability associated with the link failure evaluation mean in the probability distribution model based on the historical average failure probability includes: The value that is closest to the mean of the link failure evaluation among the multiple predicted link evaluation values ​​output by the probability distribution model is taken as the target predicted link evaluation value. The probability corresponding to the target predicted link evaluation value in the probability distribution model is adjusted based on the historical average failure probability.

2. The method according to claim 1, characterized in that, The target monitoring parameters include M types of parameters, and obtaining a first evaluation value based on the target monitoring parameters includes: Determine the load status of the target device, which is the device to which the target PCIe link belongs; Based on the load state, the target monitoring parameters are divided into at least one first type of monitoring parameters and at least one second type of monitoring parameters, and a first total weight and a second total weight of the at least one first type of monitoring parameters and the at least one second type of monitoring parameters are determined; wherein, the at least one first type of monitoring parameters are the top N target monitoring parameters with a higher probability of degradation under the load state; M and N are positive integers, and M is greater than N; The target monitoring parameters are standardized to obtain corresponding standard values; A first weight for each of the first type of monitoring parameters is determined based on a first total weight of the at least one type of monitoring parameter, and a second weight for each of the second type of monitoring parameters is determined based on the standard value and the second total weight of the at least one type of monitoring parameter. The first evaluation value is obtained by weighting and summing the standard values ​​based on the first weight and the second weight.

3. The method according to claim 2, characterized in that, The standardization process for the target monitoring parameters to obtain corresponding standard values ​​includes: Based on the parameter characteristics of the target monitoring parameters, the target monitoring parameters are divided into at least one positive monitoring parameter and at least one negative monitoring parameter; The negative monitoring parameters are standardized according to a preset first processing method to obtain the first standard value corresponding to each positive monitoring parameter; The positive monitoring parameters are standardized according to the preset second processing method to obtain the second standard value corresponding to each negative monitoring parameter.

4. The method according to claim 3, characterized in that, The step of standardizing each of the positive monitoring parameters according to a preset first processing method to obtain a first standard value corresponding to each of the positive monitoring parameters includes: For each of the positive monitoring parameters, at least one first monitoring parameter of a first PCIe link is obtained, wherein the first PCIe link and the target PCIe link belong to the same target device, and the first monitoring parameter and the positive monitoring parameter are the same type of monitoring parameter; For each of the positive monitoring parameters, the maximum value between the target monitoring parameter and the first monitoring parameter is obtained and used as the reference value corresponding to the positive monitoring parameter; For each of the positive monitoring parameters, the positive monitoring parameters are standardized according to a preset standardization method and in combination with the corresponding benchmark value to obtain the corresponding first standard value.

5. The method according to claim 2, characterized in that, Determining the first weight for each of the first type of monitoring parameters based on the first total weight of the at least one type of monitoring parameter includes: Based on the differences in the degree of impact of different first-type monitoring parameters on link performance and the first total weight, the first weight of each first-type monitoring parameter is obtained.

6. The method according to claim 5, characterized in that, The step of obtaining the first weight for each of the first type of monitoring parameters based on the differences in the degree of impact of different first-type monitoring parameters on link performance and the first total weight includes: For each of the first type of monitoring parameters, at least one second monitoring parameter of the second PCIe link is obtained; wherein the second PCIe link and the target PCIe link belong to the same target device, and the second monitoring parameter and the first type of monitoring parameter are of the same type. For each of the first type of monitoring parameters, the variance value corresponding to the first type of monitoring parameters is obtained based on the second monitoring parameters and the first type of monitoring parameters; For each of the first type of monitoring parameters, parameter injection is performed on the target device according to the variance value to obtain the corresponding link performance change value; Based on the link performance change value corresponding to different first-type monitoring parameters and the first total weight, obtain the first weight for each first-type monitoring parameter.

7. The method according to claim 2, characterized in that, The determination of the second weight corresponding to each of the second type of monitoring parameters based on the standard value and the second total weight of the at least one second type of monitoring parameter includes: Obtain the sum of the standard values ​​corresponding to each of the second type of monitoring parameters; For each of the second type of monitoring parameters, the ratio of the standard value corresponding to the second type of monitoring parameter to the sum of the standard values ​​is obtained as the second weight ratio of the second type of monitoring parameter; For each of the second type of monitoring parameters, the corresponding second weight is obtained based on the second weight ratio and the second total weight of the second type of monitoring parameters.

8. The method according to claim 1, characterized in that, The historical performance characteristic parameters include multiple parameters, and obtaining the second evaluation value based on the historical performance characteristic parameters includes: Based on the degree of influence of different historical performance characteristic parameters on the transmission performance of the target PCIE link, a third weight is determined for each of the historical performance characteristic parameters. The historical performance characteristic parameters are normalized to obtain the corresponding third standard values; The second evaluation value is obtained by weighting and summing the third standard value based on the third weight.

9. The method according to claim 1, characterized in that, The fault prediction based on the adjusted probability distribution model includes: Based on the adjusted probability distribution model, multiple predicted link evaluation values ​​and a first predicted probability for each predicted link evaluation value are obtained. Obtain fault prediction thresholds; wherein, the fault prediction thresholds include at least one of the following: baseline threshold, trend threshold, and burst threshold; Fault probability prediction is performed based on the first predicted probability of each predicted link evaluation value and the fault prediction threshold.

10. The method according to claim 9, characterized in that, The fault prediction threshold includes the baseline threshold, and obtaining the fault prediction threshold includes: Obtain the first target historical link evaluation value from the historical link evaluation values, indicating that the target PCIE link is in normal operating condition; The baseline threshold is obtained based on the historical link evaluation value of the first target and the preset confidence interval.

11. The method according to claim 10, characterized in that, The fault prediction threshold includes the suddenness threshold, and obtaining the fault prediction threshold includes: Obtain the standard deviation between the historical link evaluation value and the baseline threshold; The baseline threshold is adjusted based on the standard deviation and a preset adjustment coefficient to obtain the burst threshold.

12. The method according to claim 9, characterized in that, The fault prediction threshold includes the trend threshold, and the fault probability prediction based on the first prediction probability of each predicted link evaluation value and the fault prediction threshold includes: Based on the multiple predicted link evaluation values, the predicted probability of each predicted link evaluation value, and at least some of the target historical link evaluation values, the change rate of multiple link evaluation values ​​of the target PCIE link within the target time period and the predicted probability of each change rate of the link evaluation value are obtained. Fault probability prediction is performed based on the rate of change of the multiple link evaluation values, the predicted probability of the rate of change of each link evaluation value, and the trend threshold.

13. The method according to claim 1, characterized in that, The acquisition of at least one target monitoring parameter of the target PCIe link includes: Obtain the transmission rate of the target PCIe link; The sampling rate of the target monitoring parameters is determined based on the transmission rate; The target monitoring parameters are obtained by sampling at the stated sampling rate.

14. An electronic device, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement the method as described in any one of claims 1 to 13.

Images