Server performance prediction method, electronic device, storage medium and program product

Abstract

The invention discloses a server performance prediction method, electronic equipment, a storage medium and a program product, and relates to the technical field of server performance prediction.The server performance prediction method comprises the steps that during data collection, time series data of a server in a preset monitoring period is collected; a data variable quantity sequence determined by the server in a preset monitoring period according to a preset time interval is also collected, the sampling frequency is automatically adjusted according to real-time data changes, self-adaptive dynamic sampling is achieved, feature information under different frequencies can be effectively extracted through multi-scale decomposition, a deep data structure of data can be mined, and the mining efficiency is improved. High-quality input is provided for follow-up cross-modal information fusion, through cross-modal information fusion, key information can be extracted, the influence of irrelevant noise is reduced, the quality of feature representation is improved, accurate prediction of server performance is achieved, the technical problem that a server performance prediction result is inaccurate is solved, and the service life of the server is prolonged. The technical effect of improving the accuracy of the performance prediction result is achieved.

Description

Technical Field

[0001] The present application relates to the technical field of server performance prediction, and in particular, to a server performance prediction method, an electronic device, a storage medium, and a program product. Background Art

[0002] Server performance prediction technology aims to predict various performance indicators of a server in a future period of time through the analysis of historical data and model training, such as the utilization rate of a central processing unit (CPU), the memory usage rate, the input / output (I / O) of a disk, etc. This technology is of great significance for server resource management, fault prevention, load balancing, etc. Through accurate performance prediction, potential problems can be identified in advance, resource allocation can be optimized, and the stability and efficiency of the system can be improved.

[0003] Therefore, it is very necessary to implement a solution that can accurately predict server performance. Summary of the Invention

[0004] The present application provides a server performance prediction method, an electronic device, a storage medium, and a program product to at least solve the problem that the server performance prediction result in the related technology is inaccurate.

[0005] The present application provides a server performance prediction method, including: Obtaining time series data of at least one server in a server cluster in a preset monitoring period and a data change amount sequence determined at a preset time interval, where the end time of the preset monitoring period is the current time; Based on the time series data and the data change amount sequence, determining a dual-coded feature vector corresponding to the at least one server, where the dual-coded feature vector includes a time domain feature obtained through time domain coding and a time domain feature obtained through frequency domain coding and conversion; Performing multi-scale decomposition on the dual-coded feature vector to obtain a time-frequency domain joint feature; Performing cross-modal information fusion on the time-frequency domain joint feature to obtain a spatio-temporal feature matrix; Based on the spatio-temporal feature matrix, performing server topology modeling to obtain a node-level state representation of the at least one server; Based on the node-level state representation, performing time series prediction to generate a performance prediction result corresponding to the at least one server.

[0006] The present application also provides a server performance prediction device, including: A data acquisition module, configured to acquire time series data of at least one server in a server cluster during a preset monitoring period and a data change amount sequence determined at a preset time interval, where the end time of the preset monitoring period is the current time; An encoding module, configured to determine a dual encoding feature vector corresponding to the at least one server based on the time series data and the data change amount sequence, where the dual encoding feature vector includes a time domain feature obtained through time domain encoding and a time domain feature obtained through frequency domain encoding and conversion; A decomposition module, configured to perform multi-scale decomposition on the dual encoding feature vector to obtain a time-frequency domain joint feature; A feature extraction module, configured to perform cross-modal information fusion on the time-frequency domain joint feature to obtain a spatio-temporal feature matrix; A modeling module, configured to perform server topology modeling based on the spatio-temporal feature matrix to obtain a node-level state representation of the at least one server; A prediction module, configured to perform time series prediction based on the node-level state representation to generate a performance prediction result corresponding to the at least one server.

[0007] This application also provides an electronic device, including: a memory, configured to store a computer program; a processor, configured to implement the steps of any of the above server performance prediction methods when executing the computer program.

[0008] This application also provides a computer-readable storage medium, in which a computer program is stored, and when the computer program is executed by a processor, the steps of any of the above server performance prediction methods are implemented.

[0009] This application also provides a computer program product, including a computer program, and when the computer program is executed by a processor, the steps of any of the above server performance prediction methods are implemented.

[0010] Through this application, when collecting data, not only the time series data of the server in the preset monitoring period is collected, but also the sequence of data change amounts determined by the server at preset time intervals in the preset monitoring period is collected. The sequence of data change amounts reflects the data fluctuation situation of the server in the preset monitoring period, realizing automatic adjustment of the sampling frequency according to real-time data changes and realizing adaptive dynamic sampling; by performing time-domain and frequency-domain encoding on the collected time series data and sequence of data change amounts to obtain a dual-encoded feature vector, combining the information of the data in the time domain and the frequency domain, enhancing the comprehensiveness and accuracy of feature representation, effectively extracting feature information at different frequencies through multi-scale decomposition, which helps to mine the deep data structure of the data and provides high-quality input for subsequent cross-modal information fusion. Through cross-modal information fusion, key information can be extracted, the influence of irrelevant noise can be reduced, and the quality of feature representation can be improved. By performing server topology modeling based on the high-quality spatio-temporal feature matrix to obtain the node-level state representation of the server, and then predicting the performance of the server based on the node-level state representation, accurate prediction of the server performance is realized. Therefore, the technical problem of inaccurate server performance prediction results can be solved, and the technical effect of improving the accuracy of performance prediction results can be achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] To more clearly illustrate the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0012] Figure 1 It is a schematic flowchart of a server performance prediction method provided in Embodiment 1 of the present application; Figure 2 It is a schematic flowchart of a server performance prediction method provided in Embodiment 2 of the present application; Figure 3 It is a schematic flowchart of a server performance prediction method provided in Embodiment 3 of the present application; Figure 4 It is a schematic flowchart of a server performance prediction method provided in Embodiment 4 of the present application; Figure 5 It is a schematic flowchart of a server performance prediction method provided in Embodiment 5 of the present application; Figure 6 It is a schematic flowchart of a server performance prediction method provided in Embodiment 6 of the present application; Figure 7 It is a schematic structural diagram of a server performance prediction device provided in the embodiments of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0013] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0014] It should be noted that in the description of the present application, the terms "including", "comprising" or any other variant thereof are intended to cover a non-exclusive inclusion, such that a process, method, article or device including a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device. The terms "first", "second", etc. in the present application are used to distinguish similar objects and are not used to describe a specific order or sequence.

[0015] In order to enable those skilled in the art of the present technology to better understand the solution of the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0016] The server performance prediction technology aims to predict various performance indicators of the server in a future period through the analysis of historical data and model training. At present, in the field of server performance prediction, in the related technologies, when collecting data, a fixed sampling frequency is usually used for data collection, which easily leads to insufficient sampling frequency when the data changes violently, and over-sampling when the data is relatively stable. Moreover, traditional performance prediction methods usually only focus on surface features and ignore the deep information structure inside the data, resulting in inaccurate results for server performance prediction.

[0017] In view of the above problems, the present disclosure provides a server performance prediction solution. By collecting the sequence of data change amounts determined at preset time intervals within a preset monitoring period, adaptive dynamic data collection is achieved, and the sampling frequency can be automatically adjusted according to real-time data changes. This not only improves the flexibility and efficiency of data collection, but also reduces unnecessary resource consumption while ensuring data accuracy, and is particularly suitable for scenarios where the data change rates in a server cluster are different, ensuring more refined data is obtained when the data fluctuates violently and reducing the sampling frequency during stable data periods to save resources. Through dual-stream time encoding, spatio-temporal alignment processing is performed on the collected time-aligned mixed-frequency metric data, realizing the ability to extract subtle change features, long-term trends, and periodic components from the original data. The enhanced time-domain encoding function captures short-term changes, while the frequency-domain encoding reveals long-term patterns in the data. Through the use of a trainable wavelet neural network to perform multi-scale decomposition on the feature vectors with a unified time reference, fine analysis of different frequency components is achieved. The application of the Morlet wavelet not only enhances the ability to capture signal details, but also further improves flexibility and adaptability by introducing a trainable mechanism to optimize parameter settings. By adopting a gated cross-attention mechanism to perform cross-modal fusion on the frequency-domain - time-domain joint features, effective integration of different modal information is realized. This mechanism assigns attention weights by calculating the similarity between queries, keys, and values, and combines a gated unit to regulate the information flow, enhancing the attention to key information and reducing the influence of irrelevant noise, thereby significantly improving the quality of feature representation. Using the finally obtained high-quality features for modeling and predicting server performance ensures that the prediction results are both highly accurate and can handle complex actual application scenarios.

[0018] An embodiment of the present application provides a server performance prediction method. In combination with the execution process of the server performance prediction method, the method is described in detail.

[0019] Figure 1 FIG. is a schematic flowchart of a server performance prediction method provided in Embodiment 1 of the present application. This method can be executed by the server performance prediction device provided in the embodiments of the present disclosure. The device can be implemented in software and / or hardware and can be integrated in an electronic device.

[0020] As Figure 1 shown, the server performance prediction method includes the following steps: Step 101, obtain the time series data of at least one server in the server cluster within a preset monitoring period and the sequence of data change amounts determined at preset time intervals. The end moment of the preset monitoring period is the current moment.

[0021] Among them, the preset monitoring period is greater than the preset time interval. The preset monitoring period can be set according to actual needs. For example, the preset monitoring period can be set to one week, one day, 3 hours, etc.; the preset time interval can also be set according to actual needs. It can be combined with the data collection frequency of the server and set the preset time interval according to actual needs and resource conditions. For example, if the data collection frequency of the server is once per second, the preset time interval can be set to 3 seconds, 5 seconds, etc. The present disclosure places no restrictions on the values of the preset monitoring period and the preset time interval.

[0022] In this embodiment, the end moment of the preset monitoring period is the current moment. That is to say, the time series data of at least one server from a certain historical moment to the current moment is obtained, as well as the data change amount sequence determined at the preset time interval within the preset monitoring period. The time interval between a certain historical moment and the current moment is the preset monitoring period. For example, assuming that the preset monitoring period is 1 hour, the time series data from the previous hour to the current moment and the data change amount sequence determined at each preset time interval within this hour are obtained.

[0023] Among them, each server in the server cluster collects data according to its own data collection frequency, and obtains the status data at each collection moment (different time points), such as CPU utilization rate, memory usage rate, etc. In this embodiment, the status data of at least one server in the server cluster at different time points within the preset monitoring period is monitored to obtain the time series data, and at the preset time interval, the data change amount at each time point within the preset monitoring period is determined to form the data change amount sequence. For the server i , for each time point (denoted as t ) within the preset monitoring period, its corresponding data change amount can be expressed by the following formula (1): (1) Among them, represents the data change amount of the i th server at the time point t , represents the status data of the i th server at the time point t , represents the status data of the i th server at the time point ( ), represents the preset time interval.

[0024] Through the above data acquisition method of determining the data change amount sequence at preset time intervals, it is possible to collect multi-source heterogeneous data of the server cluster, obtain time-aligned hybrid frequency index data, and achieve adaptive dynamic sampling. This sampling method can automatically adjust the sampling frequency according to real-time data changes, which not only improves the flexibility and efficiency of data acquisition, but also can reduce unnecessary resource consumption while ensuring data accuracy. It is especially suitable for the situation where the data change rates in the server cluster are different, ensuring more refined data is obtained when the data fluctuates violently and reducing the sampling frequency during data stability periods to save resources.

[0025] Step 102: Based on the time series data and the data change amount sequence, determine the dual-encoding feature vectors corresponding to at least one server. The dual-encoding feature vectors include time domain features obtained through time domain encoding and time domain features obtained through frequency domain encoding and conversion.

[0026] In this embodiment, after obtaining the time series data and the data change amount sequence of at least one server, for each server, the time series data and the data change amount sequence corresponding to the server can be encoded in two different ways, namely time domain and frequency domain, to obtain dual-encoding feature vectors.

[0027] Exemplarily, a two-stream time encoding method can be used to perform spatio-temporal alignment processing on the time-aligned hybrid frequency index data obtained in the previous step to obtain dual-encoding feature vectors with a unified time reference. The two-stream time encoding includes time domain encoding and frequency domain encoding, so the obtained dual-encoding feature vectors include time domain features obtained through time domain encoding and time domain features obtained through frequency domain encoding and conversion.

[0028] Step 103: Perform multi-scale decomposition on the dual-encoding feature vectors to obtain time-frequency domain joint features.

[0029] In this embodiment, after obtaining the dual-encoding feature vectors corresponding to each server, the dual-encoding feature vectors of each server can be further decomposed at multiple scales to obtain time-frequency domain joint features. Since the dual-encoding feature vectors are obtained through two encoding methods, namely time domain encoding and frequency domain encoding, and include features in two dimensions, namely time domain and frequency domain, the features obtained after decomposition still contain time domain and frequency domain features, which are called time-frequency domain joint features.

[0030] Among them, a commonly used multi-scale decomposition algorithm can be adopted to decompose the dual-encoded feature vectors of each server. The present disclosure places no limitation on the specific multi-scale decomposition algorithm used. For example, wavelet transform can be used to perform multi-scale decomposition on the dual-encoded feature vectors of each server to separate the short-term fluctuations and long-term trends therein. Alternatively, a pre-trained time series decomposition model can also be used to perform multi-scale decomposition on the dual-encoded feature vectors of each server.

[0031] Step 104: Perform cross-modal information fusion on the time-frequency domain joint features to obtain a spatio-temporal feature matrix.

[0032] In this embodiment, for the time-frequency domain joint features of each server obtained by decomposition, cross-modal information fusion can be performed to obtain an enhanced spatio-temporal feature matrix.

[0033] Exemplarily, based on the attention mechanism, cross-modal information fusion can be performed on the time-frequency domain joint features of each server to obtain a spatio-temporal feature matrix corresponding to each server.

[0034] Step 105: Perform server topology modeling based on the spatio-temporal feature matrix to obtain the node-level state representations of at least one server.

[0035] In this embodiment, after obtaining the spatio-temporal feature matrices of each server, server topology modeling can be performed based on these spatio-temporal feature matrices to obtain the node-level state representations of these servers. Among them, the node-level state representations reflect the topological relationships between server nodes. By aggregating the feature information of neighbor nodes through the node-level state representations and updating the features of the current node, the feature representations of each server node are made more accurate, providing strong support for subsequent performance prediction and ensuring the reliability and accuracy of server performance prediction.

[0036] Exemplarily, the spatio-temporal feature matrices of each server can be input into a pre-trained graph convolutional network for modeling, and the graph convolutional network outputs the corresponding node-level state representations.

[0037] Step 106: Perform time series prediction based on the node-level state representations to generate performance prediction results corresponding to at least one server.

[0038] In this embodiment, after obtaining the node-level state representations of the servers, time series prediction can be performed based on the node-level state representations to obtain performance prediction results corresponding to each server.

[0039] Exemplarily, a multi-scale time series Transformer model can be used to perform time series prediction on the node-level state representation. After the encoder of the multi-scale time series Transformer model extracts the input node-level state representation, the decoder of the multi-scale time series Transformer model makes predictions based on the input features to generate server performance indicators of each server in a certain time period in the future as performance prediction results.

[0040] The server performance prediction method of the embodiment of the present disclosure not only collects the time series data of the server in the preset monitoring period, but also collects the data change sequence determined by the server in the preset monitoring period according to the preset time interval. The data change sequence reflects the data fluctuation of the server in the preset monitoring period, realizes automatic adjustment of the sampling frequency according to the real-time data change, and realizes adaptive dynamic sampling; by encoding the collected time series data and the data change sequence in the time domain and frequency domain to obtain a double-encoded feature vector, the information of the data in the time domain and frequency domain is combined, and the comprehensiveness and accuracy of the feature representation are enhanced; through multi-scale decomposition, feature information at different frequencies can be effectively extracted, which is helpful to mine the deep data structure of the data and provide high-quality input for subsequent cross-modal information fusion; through cross-modal information fusion, key information can be extracted, the influence of irrelevant noise can be reduced, and the quality of feature representation can be improved; by performing server topology modeling based on the high-quality spatiotemporal feature matrix, the node-level state representation of the server is obtained, and then the performance of the server is predicted based on the node-level state representation, and the server performance is accurately predicted. Therefore, the technical problem of inaccurate server performance prediction results can be solved, and the technical effect of improving the accuracy of performance prediction results can be achieved.

[0041] In an optional embodiment of the present disclosure, Figure 2 As shown, based on the above embodiment, step 102 may include the following sub-steps: Step 201, perform time domain coding processing on the time series data to obtain a first time domain feature.

[0042] In this embodiment, for each server, for the time series data of the server, an enhanced time domain coding function can be used to capture the subtle change characteristics in the original time series data to obtain a time domain feature (for ease of description and distinction, referred to as a first time domain feature).

[0043] The time domain coding function can be expressed as shown in the following formula (2): (2) in, x t Represents a time point in time series data t The corresponding data points; Denotes the eigenvalue obtained by performing time-domain encoding on x t ; the obtained eigenvalue after time-domain encoding. α is a preset amplitude parameter used to control the amplitude of the sine wave; is a preset frequency parameter, and is a preset phase offset.

[0044] Step 202: After performing frequency-domain encoding on the data change amount sequence, convert it to the time domain to obtain the second time-domain feature.

[0045] In this embodiment, for each server, for the data change amount sequence of the server, a frequency-domain encoding function can be introduced to extract the long-term trend and periodic components of the data in the data change amount sequence.

[0046] Assume y ( ω ) is the frequency-domain representation obtained from through the Fast Fourier Transform (FFT), then the frequency-domain encoding can be expressed by the following formula (3): (3) where y ( ω ) represents the frequency-domain representation form of the data, j represents the imaginary unit, ω and represents the angular frequency variable.

[0047] In this embodiment, by performing frequency-domain encoding on the data change amount sequence, different frequency components in the data can be better separated and processed.

[0048] Next, for the obtained frequency-domain encoding feature, by calculating the inverse transform from the frequency domain to the time domain, it can be converted to the time domain to obtain the time-domain feature (for ease of description and distinction, it is called the second time-domain feature). Among them, the conversion formula from the frequency domain to the time domain can be expressed by the following formula (4): (4) where F -1 represents the inverse Fourier transform operation, represents the second time-domain feature value of the i th server at the time point t .

[0049] Step 203: Based on the first time-domain feature and the second time-domain feature, determine the dual-encoding feature vector.

[0050] In this embodiment, after determining the first time-domain feature and the second time-domain feature of each server, the dual-coding feature vector of the server can be determined based on the first time-domain feature and the second time-domain feature of the same server.

[0051] Suppose the i -th server's dual-coding feature vector value at time point t is denoted as V i ( t ), then: .

[0052] Wherein, represents the enhanced data obtained after time-domain encoding of the data of the i -th server at time point t , represents the feature representation obtained after frequency-domain encoding and conversion to the time domain of the data change amount of the i -th server at time point t . It can be seen that the dual-coding feature vector is the concatenation of the first time-domain feature and the second time-domain feature at the same time point.

[0053] In an alternative embodiment of the present disclosure, for the obtained dual-coding feature vector, all the feature vectors in the dual-coding feature vector can also be adjusted to the same time basis through a commonly used interpolation method at present, so as to obtain a dual-coding feature vector with a unified time basis.

[0054] The server performance prediction method of the embodiments of the present disclosure captures the subtle change features in the original time series data by adopting an enhanced time-domain coding function, and introduces a frequency-domain coding function to extract the long-term trend and periodic components of the data in the data change amount sequence, constructs the dual-coding feature vector of each server. The dual-stream time coding enhances the comprehensiveness and accuracy of the feature representation by combining time-domain and frequency-domain information. The enhanced time-domain coding captures the subtle changes in the data, while the frequency-domain coding reveals the long-term trend and periodic components. The dual-coding strategy can not only improve the model's ability to understand complex patterns, but also enable subsequent analysis to be based on richer feature information, thereby improving the robustness and reliability of server performance prediction.

[0055] In an alternative embodiment of the present disclosure, as Figure 3 shown, on the basis of the foregoing embodiment, step 103 may include the following sub-steps: Step 301, based on the data change amount sequence, determine the decision variable corresponding to the current moment.

[0056] In this embodiment, for the data change amount sequence of each server, the decision variable corresponding to each server at the current moment can be determined.

[0057] In an alternative embodiment of the present disclosure, for each server, when calculating the decision variable, the data change amounts in the data change amount sequence can be traversed, and the current traversed data change amount can be converted to obtain the corresponding conversion parameter. For example, each data change amount in the data change amount sequence of each server can be input into the adaptive adjustment basis function shown in the following formula (5) to determine the conversion parameter: (5) Where, x That is, it represents a data change amount in the data change amount sequence of the foregoing embodiment, k and x 0 represents a preset parameter. It can be understood that for each data change amount in the data change amount sequence, the corresponding conversion parameter is obtained through the above formula (5), forming a conversion parameter sequence.

[0058] Then, the decision variable can be determined based on the conversion parameter. For example, an integral operation can be performed on the obtained conversion parameter to obtain the decision variable, and the calculation formula of the decision variable can be expressed as shown in the following formula (6): (6) Where, D i ( t ) represents the decision variable of the i th server at time t , represents the preset time interval, τ represents the time within a preset time interval, F () function is the above-mentioned adaptive adjustment basis function, represents the i th server at time τ data change amount.

[0059] As an example, for each server, within the nearest preset time interval (corresponding to a time window) from the current moment, using the above formula (6), based on the data change amounts at all times within this time window, the decision variable corresponding to the current moment can be determined, and t in the above formula (6) represents the current moment.

[0060] As another example, for each server, using the above formula (6), the decision variables corresponding to each preset time interval can be calculated within the entire preset monitoring period, and the decision variable at the current moment can be determined based on the calculated decision variables. For example, the mean value of the decision variables can be calculated as the decision variable at the current moment.

[0061] Step 302: Determine the transformation scale parameter based on the decision variable.

[0062] In this embodiment, after determining the decision variable of each server at the current moment, the transformation scale parameter can be determined based on the decision variables of each server.

[0063] As an example, the transformation scale parameter can be determined by the following formula (7): (7) Where, represents the decision variable of the server at the current moment, represents the transformation scale parameter, represents the initial value of the preset scale parameter, γ represents the preset adjustment coefficient, and it is set that γ the value of is less than 0.

[0064] Through the above formula (7), the dynamic adjustment of the transformation scale parameter according to the value is realized. When is larger, a smaller scale parameter is used to capture more details; when is smaller, a larger scale parameter is used to capture the long-term trend.

[0065] Step 303: Perform a continuous wavelet transform on the double-coded feature vector based on the transformation scale parameter to obtain wavelet coefficients at different scales.

[0066] In this embodiment, for each server, the continuous wavelet transform can be performed on the double-coded feature vector of the server using the corresponding transformation scale parameter to obtain wavelet coefficients at different scales.

[0067] As an example, a trainable wavelet neural network can be used to perform multi-scale decomposition on the feature vector with a unified time reference to obtain wavelet coefficients at different scales. Among them, the Morlet wavelet can be selected as the basis function. For the double-coded feature vector of each server, the continuous wavelet transform (Continuous Wavelet Transform, CWT) is used to perform multi-scale decomposition on it, and the expression of CWT is shown in formula (8): (8) Where, It represents the determined transformation scale parameter, which determines the degree of wavelet scaling; b represents the translation parameter; Indicated in scale and location b The wavelet coefficients on ; represents the complex conjugate of the wavelet basis function; Indicates i The double-encoded feature vector of the servers.

[0068] Step 304: Generate time-frequency domain joint features based on wavelet coefficients at different scales.

[0069] In this embodiment, a trainable mechanism is introduced to optimize the parameters in the wavelet basis function and the transformation scale parameters. After multi-scale decomposition through the above-mentioned continuous wavelet transform, wavelet coefficients at different scales are obtained, and then the wavelet coefficients at different scales are reorganized into a new feature matrix to obtain frequency domain-time domain joint features, that is, time-frequency domain joint features.

[0070] The server performance prediction method of the embodiment of the present disclosure determines the decision variable corresponding to the current moment based on the data change sequence, and then determines the transformation scale parameter based on the decision variable, so as to realize the dynamic adjustment of the scale parameter of the wavelet transform according to the decision variable. By selecting Morlet wavelet as the basic function, a trainable wavelet neural network is used to perform continuous wavelet transform to realize multi-scale decomposition. Based on the parameters and scale parameters in the dynamically optimized wavelet basis function, the wavelet coefficients at different scales are reorganized to form a new feature matrix, and the joint features in the time-frequency domain are obtained. The feature information at different frequencies can be effectively extracted, and the expressiveness of the features can be further improved by optimizing the parameter setting. The application of Morlet wavelet can not only capture the detailed features of the signal, but also adapt to different frequency requirements by adjusting the scale parameters, which is helpful to explore the deep data structure, and provide high-quality input for subsequent cross-modal fusion, which significantly improves the accuracy and meticulousness of performance prediction.

[0071] In an optional embodiment of the present disclosure, Figure 4 As shown, based on the above embodiment, step 104 may include the following sub-steps: Step 401, input the time-frequency domain joint features into the attention module of the gated cross attention network, and obtain the attention feature matrix output by each attention branch of the attention module.

[0072] In this embodiment, for each server, the obtained joint time-frequency domain features are input into the attention module of the gated cross-attention network. The concept of query (denoted as Q), key (denoted as K), and value (denoted as V) is introduced into the attention module to construct an attention mechanism. For the input joint time-frequency domain features, through the attention mechanism of the attention module, the input joint time-frequency domain features are mapped to three vectors of query Q, key K, and value V. Based on Q and K, the attention score matrix (denoted as A) is calculated. For example, the dot product of Q and K can be used as the similarity measure between the two to obtain the attention score matrix, that is: 。

[0073] Among them, d k represents the dimensionality of the key vector K, which is used to scale the dot product result to stabilize the gradient, and K T represents the transpose matrix of the key vector K.

[0074] Next, the obtained attention score matrix is normalized by the Softmax function to convert the attention scores into a probability distribution, obtaining an attention weight matrix. Among them, the attention weight matrix can be determined by the following formula (9): (9) Among them, represents the attention weight at the i row j and column of the determined attention weight matrix, i row j and

[0075] column of the attention score matrix.

[0076] Next, the obtained attention weight matrix is multiplied by the value vector V to achieve the fusion of cross-modal information, obtaining an attention feature matrix.

[0077] Through the above method, each attention branch of the attention module can determine the corresponding attention feature matrix for each branch and output it.

[0077] Step 402: Concatenate the attention feature matrices output by each attention branch along the feature dimension to obtain a feature concatenation matrix.

[0078] In this embodiment, for the attention feature matrices output by each attention branch, these attention feature matrices can be concatenated along the feature dimension to obtain a feature concatenation matrix concatenated along the feature dimension.

[0079] Step 403: Input the feature concatenation matrix into the gated unit of the gated cross-attention network to obtain the spatio-temporal feature matrix output by the gated unit.

[0080] In this embodiment, the gating unit of the gated cross-attention network includes a fully connected layer and a Sigmoid activation function. The feature concatenation matrix is input into the gating unit, and the trainable weight parameters in the fully connected layer and the Sigmoid activation function constitute a dynamic weight matrix. Then, this weight matrix is used to perform weighted summation on the input feature concatenation matrix to obtain a spatio-temporal feature matrix.

[0081] It can be understood that when the server load fluctuates periodically, the weight parameters of the gating unit automatically increase the weight of the frequency domain branch, while in the case of burst traffic scenarios, the weight of the frequency domain branch is correspondingly reduced to achieve dynamic feature selection. The whole process is continuous. As more and more data is input, the weight parameters of the gating unit become more optimal.

[0082] The server performance prediction method of the present disclosure embodiment calculates the attention feature matrix of the time-frequency domain joint features through the attention mechanism, and combines the gating mechanism to adjust the information flow to obtain an enhanced spatio-temporal feature matrix. The gated cross-attention mechanism effectively integrates information of different modalities, enhances the network's attention to key information, and at the same time reduces the influence of irrelevant noise, enabling the network to flexibly select important information for processing, not only improving the quality of feature representation, but also enhancing the network's ability to cope with complex environments, and finally achieving high robustness and stability of the prediction results.

[0083] In an alternative embodiment of the present disclosure, as Figure 5 shown, on the basis of the foregoing embodiment, step 105 may include the following sub-steps: Step 501, obtain the adjacency matrix corresponding to the at least one server.

[0084] Among them, the adjacency matrix describes the connection relationship between at least one server. The initial adjacency matrix can be determined according to the actual topological structure of the at least one server, and the initial adjacency matrix is updated by adding the self-loop matrix (identity matrix) to obtain the adjacency matrix.

[0085] Step 502, determine the degree matrix based on the adjacency matrix.

[0086] In this embodiment, based on the adjacency matrix of the at least one server, the corresponding degree matrix can be determined. The degree matrix is a diagonal matrix, and the elements on its diagonal are the degrees of the corresponding nodes (i.e., servers), that is, the number of connections to other nodes.

[0087] Among them, the calculation method of the degree matrix is as follows: .

[0088] Among them, Denote the element at the i th row j and th column in the adjacency matrix, i Denote the element at the i th row j and th column in the degree matrix, i Denote the indices of all other nodes connected to node in the adjacency matrix i . By summing all elements in the i th row of the adjacency matrix, the degree of node

[0089] is obtained.

[0090] In this embodiment, the graph convolutional network may include at least one layer of graph convolution. When constructing a multi-layer graph convolutional network, the output of each layer is used as the input of the next layer. Node information is propagated and node-level state representations (node feature matrices) are learned through graph convolution operations, gradually refining more abstract and high-level node representations. Before each graph convolution, the adjacency matrix and the corresponding degree matrix are re-evaluated and updated according to the server topology at the current moment. Define the graph convolution layer as: .

[0091] Wherein, H (m+1) represents the node feature matrix of the ( m + 1)-th layer, H (m) represents the node feature matrix of the m th layer, and the node feature matrix of the input layer is the spatio-temporal feature matrix of at least one server node; represents the adjacency matrix with self-loops added, represents the degree matrix, W (m) represents the weight matrix of the m th layer, σ represents the activation function (e.g., ReLU function).

[0092] The output obtained after the last layer of graph convolution L is the desired node-level state representation, where

[0093] The server performance prediction method according to the embodiments of the present disclosure determines an adjacency matrix and a degree matrix, and applies graph convolution operations to propagate node information and learn node-level state representations, extracts high-level node representations, and obtains node-level state representations, enabling the network to adapt to the dynamic changes in the relationships between servers, thereby more accurately reflecting the actual situation. Thus, not only the accuracy of node representations is improved, but also strong support is provided for subsequent time series prediction, ensuring the reliability and accuracy of future server performance prediction.

[0094] In an alternative embodiment of the present disclosure, as Figure 6 shown, based on the foregoing embodiment, step 106 may include the following sub-steps: Step 601, perform multi-scale feature extraction on the node-level state representation to obtain a multi-scale fusion feature.

[0095] In this embodiment, the obtained node-level state representation is a feature matrix, and multi-scale feature extraction can be performed on the node-level state representation to obtain a multi-scale fusion feature.

[0096] As an example, a multi-scale temporal Transformer network can be used to perform multi-scale feature extraction on the node-level state representation to obtain a multi-scale fusion feature. The multi-scale temporal Transformer network includes a multi-scale feature extraction module and a Transformer architecture. The multi-scale feature extraction module is used to capture the pattern changes at different time scales in the input node-level state representation, and the Transformer architecture is used to encode the features at each time scale extracted by the multi-scale feature extraction module after multi-scale feature extraction to obtain a multi-scale fusion feature. Among them, the multi-scale feature extraction module contains multiple parallel branches, and each branch is responsible for processing the data of the node-level state representation at a specific time scale to obtain features at different time scales. For the features output by each branch of the multi-scale feature extraction module, the Transformer architecture first performs positional encoding to add positional information to each element in the time series; then, the multi-head self-attention mechanism is used to capture the dependencies within the time series. The input of the multi-head self-attention mechanism is the output of the positional encoding. The output of the multi-head self-attention mechanism passes through a feed-forward neural network to further refine the feature representation. Finally, the feature representations refined for the features output by each branch are fused by weighted summation to form a unified feature representation, obtaining a multi-scale fusion feature.

[0097] By designing a multi-scale feature extraction module to capture pattern changes at different time scales, and using the Transformer architecture to encode the features at each time scale, and then performing feature fusion through weighted summation. The multi-scale temporal Transformer network captures pattern changes at different time scales by designing a multi-scale feature extraction module, enabling the network to comprehensively understand the temporal characteristics of the data. Using the Transformer architecture to encode the features at each time scale further enhances the network's ability to capture long-term and short-term dependencies.

[0098] Step 602, based on the multi-scale fusion features, use a regression model to predict the performance metric data of at least one server in the next monitoring period.

[0099] In this embodiment, after obtaining the multi-scale fusion features, a regression model can be used to predict the performance metric data of at least one server within a certain future time period (such as the next monitoring period).

[0100] It should be noted that commonly used regression models at present can be adopted for performance prediction, and the present disclosure does not limit the specific form of the regression model.

[0101] Step 603, based on the node-level state representation, use a traffic pulse generative adversarial network to generate the performance pulse data of at least one server.

[0102] In this embodiment, for the obtained node-level state representation, a traffic pulse generative adversarial network (TP-GAN) can be further used to generate the performance pulse data of at least one server.

[0103] Among them, the TP-GAN network consists of two parts: a generator and a discriminator. The task of the generator is to generate data simulating real server traffic pulses, while the discriminator is used to distinguish between actual data and generated data. During the optimization process of the TP-GAN network, the generator tries to deceive the discriminator into misclassifying the generated data as real data, while the discriminator tries to improve its recognition ability to accurately distinguish between real and forged data. By aiming to minimize the respective losses of the generator and the discriminator, the network parameters of the generator and the discriminator are updated.

[0104] In this embodiment, the node-level state representation is input into the generator of the TP-GAN network, and the generator synthesizes the performance pulse data of at least one server based on its own network parameters and the node-level state representation.

[0105] Step 604, based on the performance metric data and the performance pulse data, generate a performance prediction curve corresponding to at least one server.

[0106] In this embodiment, for the same server, the performance metric data and performance pulse data of the server can be combined to generate a performance prediction curve of the server. The performance metric data is adjusted by the performance pulse data to reflect the emerging burst traffic situation, and then the performance prediction curve of the server is drawn based on the adjusted data.

[0107] The server performance prediction method of the embodiments of the present disclosure captures the pattern changes at different time scales by performing multi-scale feature extraction on the node-level state representation to obtain multi-scale fusion features, so as to more comprehensively understand the time characteristics and long-term and short-term dependencies of the data. By combining the flow pulse adversarial generation network to generate a performance prediction curve, it can not only simulate the real burst traffic situation, but also adjust the prediction curve to better reflect the abnormal situations that may occur in the future, thereby improving the authenticity and practicality of the prediction curve, ensuring that the prediction results are both highly accurate and can cope with complex actual application scenarios.

[0108] Through the description of the above embodiments, those skilled in the art can clearly understand that the method according to the above embodiments can be implemented by means of software plus a necessary general hardware platform. Of course, it can also be implemented by hardware, but in many cases, the former is a better implementation method.

[0109] The embodiments of the present application also provide a server performance prediction device. Figure 7 As shown in the structural schematic diagram of a server performance prediction device provided by the embodiments of the present application, Figure 7 the server performance prediction device 70 includes: a data acquisition module 710, an encoding module 720, a decomposition module 730, a feature extraction module 740, a modeling module 750, and a prediction module 760.

[0110] Among them, the data acquisition module 710 is used to acquire the time series data of at least one server in the server cluster in a preset monitoring period and the data change amount sequence determined according to a preset time interval, and the end time of the preset monitoring period is the current time; The encoding module 720 is used to determine a dual-encoding feature vector corresponding to at least one server based on the time series data and the data change amount sequence. The dual-encoding feature vector includes the time domain features obtained through time domain encoding and the time domain features obtained through frequency domain encoding and conversion; The decomposition module 730 is used to perform multi-scale decomposition on the dual-encoding feature vector to obtain the time-frequency domain joint features; The feature extraction module 740 is used to perform cross-modal information fusion on the time-frequency domain joint features to obtain a spatio-temporal feature matrix; The modeling module 750 is used to perform server topology modeling based on the spatio-temporal feature matrix to obtain the node-level state representation of at least one server; A prediction module 760 for performing time series prediction based on the node-level state representation to generate a performance prediction result corresponding to at least one server.

[0111] Optionally, the encoding module 720 is further configured to: Perform time domain encoding processing on the time series data to obtain first time domain features; Perform frequency domain encoding on the data change amount sequence and then convert it to the time domain to obtain second time domain features; Determine a dual-encoding feature vector based on the first time domain features and the second time domain features.

[0112] Optionally, the decomposition module 730 includes: A first determination unit for determining a decision variable corresponding to the current moment based on the data change amount sequence; A second determination unit for determining a transformation scale parameter based on the decision variable; A transformation unit for performing continuous wavelet transform on the dual-encoding feature vector based on the transformation scale parameter to obtain wavelet coefficients at different scales; A generation unit for generating a time-frequency domain joint feature based on the wavelet coefficients at different scales.

[0113] Optionally, the first determination unit is further configured to: Traverse the data change amounts in the data change amount sequence, convert the currently traversed data change amount, and obtain a corresponding conversion parameter; Determine the decision variable based on the conversion parameter.

[0114] Optionally, the feature extraction module 740 is further configured to: Input the time-frequency domain joint feature into the attention module of the gated cross-attention network to obtain an attention feature matrix output by each attention branch of the attention module; Perform feature dimension splicing on the attention feature matrices output by each attention branch to obtain a feature splicing matrix; Input the feature splicing matrix into the gated unit of the gated cross-attention network to obtain a spatio-temporal feature matrix output by the gated unit.

[0115] Optionally, the modeling module 750 is further configured to: Obtain an adjacency matrix corresponding to at least one server; Determine a degree matrix based on the adjacency matrix; Input the spatio-temporal feature matrix, the degree matrix, and the adjacency matrix into a graph convolutional network for processing to obtain a node-level state representation output by the graph convolutional network.

[0116] Optionally, the prediction module 760 is further configured to: Perform multi-scale feature extraction on the node-level state representation to obtain multi-scale fusion features; Based on the multi-scale fusion features, use a regression model to predict the performance metric data of at least one server in the next monitoring period; Based on the node-level state representation, use a traffic pulse adversarial generation network to generate performance pulse data of at least one server; Based on the performance metric data and the performance pulse data, generate a performance prediction curve corresponding to at least one server.

[0117] For the description of the features in the corresponding embodiments of the server performance prediction device, reference can be made to the relevant descriptions in the corresponding embodiments of the server performance prediction method, which will not be elaborated here one by one.

[0118] An embodiment of the present application also provides an electronic device, including a memory and a processor. A computer program is stored in the memory, and the processor is configured to run the computer program to execute the steps in any of the above embodiments of the server performance prediction method.

[0119] An embodiment of the present application also provides a computer-readable storage medium, in which a computer program is stored. The computer program is configured to execute the steps in any of the above embodiments of the server performance prediction method when running.

[0120] In an exemplary embodiment, the above computer-readable storage medium may include, but is not limited to: USB flash drive, read-only memory (ROM for short), random access memory (RAM for short), mobile hard disk, magnetic disk, or optical disc and other various media that can store computer programs.

[0121] An embodiment of the present application also provides a computer program product. The above computer program product includes a computer program, and when the computer program is executed by a processor, it implements the steps in any of the above embodiments of the server performance prediction method.

[0122] An embodiment of the present application also provides another computer program product, including a non-volatile computer-readable storage medium. The non-volatile computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements the steps in any of the above embodiments of the server performance prediction method.

[0123] Those skilled in the art may further realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered as exceeding the scope of this application.

[0124] The above has introduced in detail a server performance prediction method, an electronic device, a storage medium, and a program product provided by this application. Specific examples have been used herein to elaborate on the principles and implementation manners of this application. The description of the above embodiments is only used to help understand the method and its core idea of this application. It should be noted that for those of ordinary skill in the art in this technical field, without departing from the principle of this application, several improvements and modifications can still be made to this application, and these improvements and modifications also fall within the protection scope of the claims of this application.

Claims

1. A server performance prediction method, characterized in that Including: Obtain the time series data of at least one server in the server cluster during a preset monitoring period and the data change amount sequence determined at preset time intervals, where the end time of the preset monitoring period is the current time; Based on the time series data and the data change amount sequence, determine the dual-coded feature vector corresponding to the at least one server, where the dual-coded feature vector includes the time domain features obtained through time domain coding and the time domain features obtained through frequency domain coding and conversion; Perform multi-scale decomposition on the dual-coded feature vector to obtain the time-frequency domain joint features; Perform cross-modal information fusion on the time-frequency domain joint features to obtain the spatio-temporal feature matrix; Based on the spatio-temporal feature matrix, perform server topology modeling to obtain the node-level state representation of the at least one server; Based on the node-level state representation, perform time series prediction to generate the performance prediction result corresponding to the at least one server.

2. The server performance prediction method according to claim 1, wherein The determining the dual-coded feature vector corresponding to the at least one server based on the time series data and the data change amount sequence includes: Perform time domain coding processing on the time series data to obtain the first time domain features; Perform frequency domain coding on the data change amount sequence and then convert it to the time domain to obtain the second time domain features; Based on the first time domain features and the second time domain features, determine the dual-coded feature vector.

3. The server performance prediction method according to claim 1, wherein The performing multi-scale decomposition on the dual-coded feature vector to obtain the time-frequency domain joint features includes: Based on the data change amount sequence, determine the decision variable corresponding to the current time; Based on the decision variable, determine the transformation scale parameter; Perform continuous wavelet transform on the dual-coded feature vector based on the transformation scale parameter to obtain wavelet coefficients at different scales; Based on the wavelet coefficients at different scales, generate the time-frequency domain joint features.

4. The server performance prediction method according to claim 3, wherein The determining the decision variable corresponding to the current time based on the data change amount sequence includes: Traverse the data change amounts in the data change amount sequence, perform conversion on the currently traversed data change amount to obtain the corresponding conversion parameter; Based on the conversion parameter, determine the decision variable.

5. The server performance prediction method according to claim 1, characterized in that, The performing cross-modal information fusion on the time-frequency domain joint features to obtain the spatio-temporal feature matrix includes: Input the time-frequency domain joint features into the attention module of the gated cross-attention network to obtain the attention feature matrix output by each attention branch of the attention module; Perform feature dimension splicing on the attention feature matrix output by each attention branch to obtain the feature splicing matrix; Input the feature splicing matrix into the gated unit of the gated cross-attention network to obtain the spatio-temporal feature matrix output by the gated unit.

6. The server performance prediction method according to claim 1, wherein The performing server topology modeling based on the spatio-temporal feature matrix to obtain the node-level state representation of the at least one server includes: Obtain the adjacency matrix corresponding to the at least one server; Based on the adjacency matrix, determine the degree matrix; Input the spatio-temporal feature matrix, the degree matrix, and the adjacency matrix into the graph convolutional network for processing to obtain the node-level state representation output by the graph convolutional network.

7. The server performance prediction method according to claim 1, wherein Performing time series prediction based on the node-level state representation to generate a performance prediction result corresponding to the at least one server, including: Performing multi-scale feature extraction on the node-level state representation to obtain multi-scale fusion features; Based on the multi-scale fusion features, using a regression model to predict the performance metric data of the at least one server in the next monitoring period; Based on the node-level state representation, using a traffic pulse adversarial generation network to generate performance pulse data of the at least one server; Based on the performance metric data and the performance pulse data, generating a performance prediction curve corresponding to the at least one server.

8. An electronic device, characterized in that, Including: A memory for storing a computer program; A processor for implementing the steps of the server performance prediction method according to any one of claims 1 to 7 when executing the computer program.

9. A computer-readable storage medium, characterized in that, A computer program is stored in the computer-readable storage medium, wherein the computer program implements the steps of the server performance prediction method according to any one of claims 1 to 7 when executed by a processor.

10. A computer program product, comprising a computer program, characterized in that, The computer program implements the steps of the server performance prediction method according to any one of claims 1 to 7 when executed by a processor.

Images