Packet loss prediction method, device, equipment, medium and product
By constructing a network topology graph and using graph convolutional networks to predict future packet loss behavior, the problem of the inability to predict network packet loss in existing technologies is solved, thereby improving the computing power utilization and energy efficiency of data centers.
Patent Information
- Application Number
- CN202410998562.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-24
- Publication Date
- 2025-12-05
- Estimated Expiration
- 2044-07-24
AI Technical Summary
Existing technologies cannot predict network packet loss behavior in the future, and automatic dial-up testing or manual ping testing methods consume network bandwidth and resources.
By acquiring the network topology graph, a graph convolutional network is used to transform it into an adjacency matrix and a feature matrix. These are then input into a network quality prediction model to predict network performance data at future times to determine packet loss behavior.
Network packet loss behavior can be predicted in advance without test data, improving the energy efficiency of data center computing power and saving energy resources.
Smart Images
Figure CN118802588B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of computer technology, and more specifically, to a method, apparatus, device, medium, and product for predicting packet loss. Background Technology
[0002] Currently, network packet loss is detected through automatic dial-up testing or manual ping testing. This not only fails to predict whether packet loss will occur in the future, but also consumes a large amount of network bandwidth and forwarding resources for the test data.
[0003] Therefore, how to predict packet loss behavior in future networks has become a technical problem that the industry urgently needs to solve. Summary of the Invention
[0004] This application provides a packet loss prediction method, apparatus, device, medium, and product to solve the technical problem of how to predict packet loss behavior of a network in the future.
[0005] Firstly, this application provides a packet loss prediction method, including:
[0006] Obtain the network topology of the network to be predicted at the current time.
[0007] The network topology graph is transformed into the adjacency matrix and feature matrix of the network to be predicted based on the graph convolutional network.
[0008] The adjacency matrix and the feature matrix are input into the network quality prediction model to obtain the network performance data of the network to be predicted at future times, output by the network quality prediction model.
[0009] Based on the network performance data, predict the packet loss behavior of the network to be predicted at the future time.
[0010] In some embodiments, obtaining the network topology map of the network to be predicted at the current moment includes:
[0011] Real-time acquisition of channel network data for each virtual channel in the network to be predicted;
[0012] The network topology diagram of the network to be predicted is generated based on the channel network data at each time point.
[0013] Obtain the network topology of the network to be predicted from the current time.
[0014] The channel network data includes the priority of each virtual channel, network latency, buffer utilization, bandwidth utilization, deadlock detection status of priority-based flow control, link layer discovery protocol and configuration information of the network nodes to be predicted.
[0015] In some embodiments, generating the network topology map of the network to be predicted at each time step based on the channel network data at each time step includes:
[0016] The channel network data is cleaned and aggregated to obtain the first data; the data cleaning includes removing duplicate data, supplementing missing data values, and processing abnormal data.
[0017] The topology of the network to be predicted is generated based on the channel network data;
[0018] The network topology graph is obtained by mapping the first data of each virtual channel to the topology using a network mapping algorithm.
[0019] In some embodiments, the graph convolutional network-based method transforms the network topology graph into an adjacency matrix and a feature matrix corresponding to the network to be predicted, including:
[0020] Define the adjacency matrix and feature matrix of the network to be predicted; the adjacency matrix represents the connection relationship of virtual channels between the topological nodes of the network to be predicted; the feature matrix is used to represent the attributes of the virtual channels between the topological nodes.
[0021] The network topology graph is transformed into the adjacency matrix and the feature matrix based on the graph convolutional network.
[0022] In some embodiments, obtaining the network performance data of the network to be predicted at future times, output by the network quality prediction model, includes:
[0023] The time series corresponding to the spatial features of the network topology graph are determined based on the adjacency matrix and the feature matrix;
[0024] Based on the time series, information on the change of network latency over time is captured;
[0025] Based on the change information, predict the network performance data of the network to be predicted at future times.
[0026] In some embodiments, the network performance data includes the network latency occurrence time and network latency magnitude of the virtual channel of the network to be predicted; predicting the packet loss behavior of the network to be predicted at the future time based on the network performance data includes:
[0027] If the network performance data is greater than a preset threshold, it is determined that the network to be predicted will exhibit packet loss behavior at the future time.
[0028] Secondly, this application provides a packet loss prediction device.
[0029] Thirdly, embodiments of this application provide an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to implement the above-described method when executing the program through the computer program.
[0030] Fourthly, embodiments of this application provide a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described method.
[0031] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the above-described method.
[0032] The packet loss prediction method, apparatus, device, medium, and product provided in this application can construct the adjacency matrix and feature matrix corresponding to the network to be predicted through graph convolutional networks and network topology graphs. By inputting the adjacency matrix and feature matrix into the network quality prediction model, it can be determined whether the network to be predicted will have packet loss behavior in the future. The packet loss behavior of the network to be predicted can be predicted in advance without the need for test data, thereby improving the energy efficiency ratio of data center computing power, improving the utilization rate of data center computing power, and saving data center energy resources. Attached Figure Description
[0033] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0034] Figure 1 This is one of the flowcharts illustrating the packet loss prediction method provided in the embodiments of this application;
[0035] Figure 2 A schematic diagram of the data acquisition process provided for embodiments of this application;
[0036] Figure 3 This is a schematic diagram illustrating the process of constructing a network topology diagram provided in an embodiment of this application;
[0037] Figure 4 This is a second schematic flowchart of the packet loss prediction method provided in the embodiments of this application;
[0038] Figure 5 This is a schematic diagram of the structure of a neuron provided in an embodiment of this application;
[0039] Figure 6 This is a schematic diagram of the packet loss prediction device provided in the embodiments of this application;
[0040] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation
[0041] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0042] It should be noted that the terms "first," "second," etc., used in this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or modules is not necessarily limited to those steps or modules explicitly listed, but may include other steps or modules not explicitly listed or inherent to such processes, methods, products, or devices.
[0043] The packet loss prediction method provided in this application is applicable to terminals. Terminals can be various electronic devices with displays and web browsing capabilities, including but not limited to servers, smartphones, tablets, laptops, and desktop computers.
[0044] Figure 1 This is one of the flowcharts illustrating the packet loss prediction method provided in the embodiments of this application, such as... Figure 1 As shown, the method includes steps 110, 120, 130, and 140. These method steps are merely one possible implementation of this application.
[0045] Step 110: Obtain the network topology of the network to be predicted at the current moment.
[0046] Specifically, the packet loss prediction method provided in this application is executed by a packet loss prediction device, which can be a hardware device independently installed in the terminal or a software program running in the terminal. For example, when the terminal is a mobile phone, the packet loss prediction device can be manifested as software or other applications in the mobile phone.
[0047] The network to be predicted in this embodiment is an intelligent lossless network. The intelligent lossless network can be implemented based on RoCEv2, where both switches and server network interface cards support the RoCEv2 protocol. Since the Remote Direct Memory Access (RDMA) protocol is highly sensitive to network packet loss—a packet loss rate of 0.01% can cause RDMA throughput to drop to 0—it is necessary to predict network packet loss in the intelligent lossless network to minimize the risk of RDMA throughput dropping to 0, thereby improving the data center's computing power efficiency. RoCEv2 is the second version of RDMA with Ethernet convergence.
[0048] The data collectors of the operation and maintenance management platform in the intelligent lossless network can collect channel network data of different virtual channels in the intelligent lossless network in real time, including the priority of different virtual channels, the network latency of different virtual channels, the buffer utilization of different virtual channels, the bandwidth utilization of different virtual channels, the deadlock detection status of priority-based flow control (PFC), and the node link layer discovery protocol (LLDP) and configuration information of the intelligent lossless network.
[0049] PFC, also known as Per Priority Pause or Class Based Flow Control (CBFC), allows the creation of multiple virtual channels on an Ethernet link and assigns a priority level to each virtual channel, allowing individual suspension and restart of any one of the virtual channels.
[0050] The analyzer of the operation and maintenance management platform in the intelligent lossless network analyzes the real-time channel network data collected by the collector to generate and save the intelligent lossless logical network topology diagram.
[0051] For example, the analyzer of the intelligent lossless network operation and maintenance management platform cleans and aggregates the collected channel network data, automatically generates an intelligent lossless logical network topology based on the node's LLDP information, PFC deadlock detection status information and configuration information, and maps the indicator information of each virtual channel to the intelligent lossless logical network topology through a network mapping algorithm to obtain the network topology diagram of the network to be predicted.
[0052] The network topology graph at the current moment can include the intelligent lossless logical network topology graph of the most recent N time steps.
[0053] Step 120: Based on the graph convolutional network, transform the network topology graph into the adjacency matrix and feature matrix corresponding to the network to be predicted.
[0054] Specifically, the adjacency matrix represents the connection relationship of virtual channels between nodes in the intelligent lossless logic network topology, the feature matrix represents the attributes of virtual channels between nodes in the intelligent lossless logic network topology at a certain time step t, and the attributes of virtual channels are the channel network data corresponding to the channel indicators of virtual channels.
[0055] Data from the intelligent lossless logical network topology graph of the most recent N time steps can be obtained from the operation and maintenance management platform of the intelligent lossless network. This data can be preprocessed, and the intelligent lossless logical network topology graph can be transformed into an adjacency matrix and a feature matrix through a graph convolutional network. The virtual channel attribute values can then be normalized.
[0056] Step 130: Input the adjacency matrix and feature matrix into the network quality prediction model to obtain the network performance data of the network to be predicted at future times, as output by the network quality prediction model.
[0057] Step 140: Predict the packet loss behavior of the network to be predicted at future moments based on network performance data.
[0058] Specifically, the network quality prediction model is a model used to predict the network performance at future times. Network performance data includes the occurrence time and magnitude of network latency in the virtual channels of the network to be predicted.
[0059] For example, by inputting the adjacency matrix and feature matrix into the pre-trained network quality prediction model, the network quality prediction model will output the i-th virtual channel of the intelligent lossless logic network at the N-th time step in the future, the network latency occurrence time of the i-th virtual channel, and the network latency magnitude of the i-th virtual channel.
[0060] Based on the model output of the i-th virtual channel of the intelligent lossless logic network at the N-th future time step, the network latency occurrence time of the i-th virtual channel, and the network latency magnitude of the i-th virtual channel, it is determined whether there is packet loss in the intelligent lossless network at the N-th future time step.
[0061] Predicting packet loss behavior of a network under test in the future based on network performance data includes: determining that the network under test will experience packet loss behavior in the future when the network performance data is greater than a preset threshold.
[0062] A threshold can be set to predict whether the network to be predicted will experience packet loss in the future when the network performance data exceeds the preset threshold.
[0063] The packet loss prediction method provided in this application can construct the adjacency matrix and feature matrix of the network to be predicted through graph convolutional networks and network topology graphs. By inputting the adjacency matrix and feature matrix into the network quality prediction model, it can determine whether the network to be predicted will have packet loss behavior in the future. The packet loss behavior of the network to be predicted can be predicted in advance without the need for test data, thereby improving the energy efficiency ratio of data center computing power, improving the utilization rate of data center computing power, and saving data center energy resources.
[0064] It should be noted that each implementation method of this application can be freely combined, rearranged, or executed individually, and does not need to rely on or depend on a fixed execution order.
[0065] In some embodiments, step 110 includes:
[0066] Real-time acquisition of channel network data for each virtual channel in the network to be predicted;
[0067] Generate a network topology diagram of the network to be predicted at each time step based on the channel network data at each time step;
[0068] Obtain the network topology of the network to be predicted at the current time.
[0069] The channel network data includes the priority of each virtual channel, network latency, buffer utilization, bandwidth utilization, deadlock detection status of priority-based flow control, link layer discovery protocol and configuration information of the network nodes to be predicted.
[0070] Based on the channel network data at each time step, generate the network topology diagram of the network to be predicted at each time step, including:
[0071] The channel network data is cleaned and aggregated to obtain the first data; data cleaning includes removing duplicate data, supplementing missing data values, and handling abnormal data.
[0072] Generate the topology of the network to be predicted based on channel network data;
[0073] The network topology graph is obtained by mapping the first data of each virtual channel to the topology using a network mapping algorithm.
[0074] Specifically, Figure 2 A schematic diagram of the data acquisition process provided in the embodiments of this application; such as Figure 2As shown, the intelligent lossless network operation and maintenance management platform can use Telemetry technology to push network data of different virtual channels to the physical devices in the intelligent lossless network in a push mode. This includes pushing the priority of different virtual channels, the network latency of different virtual channels, the buffer utilization of different virtual channels, the bandwidth utilization of different virtual channels, the deadlock detection status of PFC, and the node LLDP and configuration information of the intelligent lossless network to the collector.
[0075] The collected channel network data undergoes data cleaning and aggregation, with the cleaned and aggregated data serving as the primary data set. Data cleaning can be performed by removing duplicate, missing, and outlier data.
[0076] Remove duplicate data: Remove duplicates from each indicator data point based on its unique identifier, such as the time data in each indicator data point.
[0077] Handling missing data: Missing data is filled using mean interpolation;
[0078] Handling outlier data: Use standard deviation to determine whether the data is outlier based on a threshold, and remove the detected outlier data.
[0079] Figure 3 This is a schematic diagram illustrating the process of constructing a network topology diagram provided in an embodiment of this application. In the diagram, "Device" refers to a device, such as... Figure 3 As shown, the collected channel network data is aggregated according to virtual channels. Priority, network latency, buffer utilization, and bandwidth utilization metrics are grouped according to virtual channels, and the average, maximum, and minimum values of each metric for each virtual channel are calculated. The analyzer of the intelligent lossless network operation and maintenance management platform automatically generates an intelligent lossless logical network topology based on the node's LLDP information, configuration information, and PFC deadlock detection status information. It then uses a network mapping algorithm to map the metric information of each virtual channel into the intelligent lossless network topology, resulting in a network topology diagram.
[0080] The packet loss prediction method provided in this application improves the accuracy of packet loss prediction by constructing a network topology graph, which can better extract the adjacency matrix and feature moments.
[0081] In some embodiments, step 120 includes:
[0082] Define the adjacency matrix and feature matrix of the network to be predicted; the adjacency matrix represents the connection relationship of virtual channels between the topological nodes of the network to be predicted; the feature matrix is used to represent the attributes of the virtual channels between the topological nodes.
[0083] Graph convolutional networks transform the network topology graph into an adjacency matrix and a feature matrix.
[0084] Specifically, the topology information of the intelligent lossless logical network at the current Nth time step can be collected from the intelligent lossless network operation and maintenance management platform. The intelligent lossless logical network topology can be represented as G = (V, E), where V is the set of nodes in the intelligent lossless logical network. , , ,....., }, where N is the number of nodes in the intelligent lossless logic network, and E is the set of edges, i.e., the virtual channels between nodes in the intelligent lossless logic network topology. If the first... Network nodes and the Network nodes If there is a virtual channel connection between them, then =1, otherwise =0. The intelligent lossless logic network topology is transformed into an adjacency matrix X and a feature matrix Y, as shown below:
[0085] X= ;Y= ;
[0086] Wherein, the N×N adjacency matrix X represents the connection relationship of virtual channels between nodes of the intelligent lossless logic network, and the N×M feature matrix Y represents the attributes (indicator information of virtual channels) of virtual channels between nodes of the intelligent lossless logic network at a certain time step t, where M represents the number of virtual channel attributes between nodes of the intelligent lossless logic network. Since the number of features of each virtual channel attribute of the intelligent lossless logic network is the same, and the feature dimension remains unchanged at each time step, the time series length of the virtual channel attributes of the intelligent lossless logic network is also set to M.
[0087] Since the priority, network latency, buffer utilization, and bandwidth utilization metrics of the intelligent lossless logic network virtual channel attributes have different units of measurement, Min-Max normalization is used to eliminate the influence of these units and improve the convergence speed and accuracy of the model. This normalization maps the results to the range of 0-1. The normalization formula is shown below:
[0088] ;
[0089] in, For the normalized data, The maximum value of the indicator data. This represents the minimum value of the indicator data.
[0090] The packet loss prediction method provided in this application can effectively extract the data features of the channel network data of the virtual channel by constructing an adjacency matrix and a feature matrix, thereby improving the accuracy of packet loss prediction.
[0091] In some embodiments, step 130 includes:
[0092] Determine the time series corresponding to the spatial features of the network topology graph based on the adjacency matrix and feature matrix;
[0093] Capture information on network latency changes over time based on time series analysis;
[0094] Predict network performance data of the network to be predicted at future moments based on changing information.
[0095] Specifically, Figure 4 This is a second flowchart illustrating the packet loss prediction method provided in the embodiments of this application, as shown below. Figure 4 As shown, a network quality prediction model is constructed, consisting of two graph convolutional layers (GCN), two long short-term memory layers (LSTM), and one fully connected layer (Dense). The number of layers in the network quality prediction model can be set according to actual conditions; the number shown here is only an example.
[0096] To obtain the historical label matrix, we need to collect historical intelligent lossless logical network topology maps from the intelligent lossless network operation and maintenance management platform as the total dataset. The intelligent lossless logical network topology maps are then transformed into historical adjacency matrices X and historical feature matrices Y, as well as the real network latency attribute values of the intelligent lossless logical network virtual channels predicted for the Nth time step in the future, forming the label matrix Z. 80% of the total dataset can be allocated as the training set, and 20% as the test set. The training set is used to train the model, and the test set is used to test the performance of the model.
[0097] The historical adjacency matrix X and feature matrix Y are input into a graph convolutional layer. Two graph convolutional layers are used to obtain the time series corresponding to the spatial features of the intelligent lossless logic network topology graph. The time series corresponding to the spatial features of the intelligent lossless logic network topology graph is then input into a subsequent long short-term memory layer. Two long short-term memory layers are used to obtain the feature vectors of the spatial and temporal features of the intelligent lossless logic network topology graph. Finally, the feature vectors of the spatial and temporal features of the intelligent lossless logic network topology graph are input into a fully connected layer to output the i-th virtual channel of the intelligent lossless logic network at the N-th time step, the network delay occurrence time t of the i-th virtual channel, and the network delay magnitude d of the i-th virtual channel.
[0098] The two graph convolutional layers have 64 kernels (i.e., the output dimension), and the activation function is set to "ReLU". Each neural network layer can be written as the following non-linear function:
[0099] ;
[0100] in, =K That is, the input data. =Z That is, the output data. Choose different numbers of layers for the neural network. The parameters also determine the different models.
[0101] ;
[0102] in, It is the parameter matrix of the l-th neural network layer. () is the non-linear activation function ReLU. It is a symmetric normalization of the adjacency matrix X. D It is the diagonal matrix of the node degree of X.
[0103] The number of neurons in the two long short-term memory layers is set to 128, and the activation function is set to "relu". Long short-term memory is a special type of recurrent neural network that can remember long-term information by controlling the retention time of values in the cache, making it suitable for time series prediction. Each neuron has four inputs and one output, and each neuron has a cell that stores the remembered values. Figure 5 This is a schematic diagram of the neuron structure provided in the embodiments of this application, such as... Figure 5 As shown, the neurons in the Long Short-Term Memory (LSTM) layer involve the following formula:
[0104] (1)
[0105] (2)
[0106] (3)
[0107] + (4)
[0108] (5)
[0109] (6)
[0110] (7)
[0111] Each LSTM neuron contains three gates: a forget gate, an input gate, and an output gate. Equation (1) represents the forget gate, equations (2) and (3) represent the addition of new information, equation (4) combines new and old information, and equations (5) and (6) output the information learned by the LSTM unit about the next time step. Each connection in the LSTM unit contains corresponding weights. Represents the input vector. Represents a hidden state. Represents the neuron state at time t. denoted by , where W is the trainable weight matrix and b is the bias vector.
[0112] The fully connected layer has 3 neurons, which correspond to the i-th virtual channel of the output intelligent lossless logic network, the network delay time t of the i-th virtual channel, and the network delay size d of the i-th virtual channel. The activation function is set to "relu".
[0113] The network quality prediction model can be trained for 2000 epochs (epochs=2000), with a batch size of 64 (batch_size=64). Mean Squared Error (MSE) can be chosen as the loss function and objective function (loss='mean_squared_error'). The formula is shown below:
[0114] ;
[0115] The gradient descent optimization algorithm can choose the Adam optimizer to improve the learning speed of traditional gradient descent (optimizer = 'adam'). Neural networks, through gradient descent, can find the optimal weight parameters that minimize the objective function; the neural network learns these weight parameters autonomously through training. Training is performed using a training set to minimize the objective function as much as possible, and a test set is used after each training round to evaluate and validate the network's predictive model quality.
[0116] After the network quality prediction model is trained, the adjacency matrix and feature matrix of the network to be predicted can be input into the graph convolutional layer. The graph convolutional layer obtains the time series corresponding to the spatial features of the network topology graph, and inputs the time series corresponding to the spatial features of the network topology graph into the subsequent long short-term memory layer. The long short-term memory layer captures the information on the change of network latency over time based on the time series. The fully connected layer predicts the network performance data of the network to be predicted at future times based on the change information.
[0117] The packet loss prediction method provided in this application can predict the network performance data of the network to be predicted at future times through a network quality prediction model, thereby improving the efficiency of packet loss prediction.
[0118] In some embodiments, the packet loss prediction method includes the following steps:
[0119] 1) The collector of the operation and maintenance management platform in the intelligent lossless network collects channel network data in real time, such as the priority of different virtual channels, the network latency of different virtual channels, the buffer utilization of different virtual channels, the bandwidth utilization of different virtual channels, the deadlock detection status of PFC, and the node LLDP and configuration information of the intelligent lossless network.
[0120] 2) The analyzer of the intelligent lossless network operation and maintenance management platform cleans and aggregates the collected channel network data.
[0121] 3) The analyzer of the intelligent lossless network operation and maintenance management platform automatically generates an intelligent lossless logical network topology based on the node's LLDP information, PFC deadlock detection status information and configuration information, and maps the indicator information of each virtual channel to the intelligent lossless logical network topology through a network mapping algorithm to obtain a network topology diagram.
[0122] 4) Preprocess the intelligent lossless logic network topology graph data, transforming it into an adjacency matrix X and a feature matrix Y. The adjacency matrix X represents the connection relationship of virtual channels between nodes in the intelligent lossless logic network topology, and the feature matrix Y represents the attributes of the virtual channels between nodes in the intelligent lossless logic network topology at a certain time step t. The attribute values of the virtual channels are then normalized.
[0123] 5) Input the adjacency matrix and feature matrix into the pre-trained network quality prediction model, and output the i-th virtual channel of the intelligent lossless logic network at the N-th time step in the fully connected layer, the network delay time t of the i-th virtual channel, and the network delay d of the i-th virtual channel.
[0124] 6) Based on the output of the model output layer, the i-th virtual channel of the intelligent lossless logic network at the N-th time step, the network delay time t of the i-th virtual channel, and the network delay magnitude d of the i-th virtual channel, determine whether there is packet loss in the intelligent lossless network at the N-th time step.
[0125] The packet loss prediction method provided in this application can effectively improve the energy efficiency ratio of data center computing power, increase the utilization rate of data center computing power, and save data center energy resources.
[0126] The packet loss prediction device provided in the embodiments of this application is described below. The packet loss prediction device described below can be referred to in correspondence with the packet loss prediction method described above.
[0127] Figure 6 This is a schematic diagram of the packet loss prediction device provided in the embodiments of this application, as shown below. Figure 6 As shown, the device includes an acquisition module 610, a conversion module 620, an input module 630, and a prediction module 640.
[0128] The acquisition module is used to obtain the network topology of the network to be predicted at the current moment.
[0129] The transformation module is used to transform the network topology graph into the adjacency matrix and feature matrix of the network to be predicted based on the graph convolutional network.
[0130] The input module is used to input the adjacency matrix and feature matrix into the network quality prediction model to obtain the network performance data of the network to be predicted at future time points output by the network quality prediction model.
[0131] The prediction module is used to predict the packet loss behavior of the network to be predicted at future times based on network performance data.
[0132] Specifically, according to the embodiments of this application, any and multiple modules among the acquisition module, conversion module, input module, and prediction module can be merged into one module, or any one of the modules can be split into multiple modules.
[0133] Alternatively, at least some of the functionality of one or more of these modules can be combined with at least some of the functionality of other modules and implemented in a single module.
[0134] According to embodiments of this application, at least one of the acquisition module, conversion module, input module, and prediction module can be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or any other reasonable means of integrating or packaging the circuitry, or implemented in any one of the three methods of software, hardware, and firmware, or in a suitable combination of any of these.
[0135] Alternatively, at least one of the acquisition module, transformation module, input module, and prediction module can be at least partially implemented as a computer program module, which can perform corresponding functions when the computer program module is run.
[0136] The packet loss prediction device provided in this application embodiment can construct the adjacency matrix and feature matrix of the network to be predicted through graph convolutional networks and network topology graphs. By inputting the adjacency matrix and feature matrix into the network quality prediction model, it can determine whether the network to be predicted will have packet loss behavior in the future. It can predict the packet loss behavior of the network to be predicted in advance without the need for test data, thereby improving the energy efficiency ratio of data center computing power, improving the utilization rate of data center computing power, and saving data center energy resources.
[0137] In some embodiments, the acquisition module is specifically used for:
[0138] Real-time acquisition of channel network data for each virtual channel in the network to be predicted;
[0139] Generate a network topology diagram of the network to be predicted at each time step based on the channel network data at each time step;
[0140] Obtain the network topology of the network to be predicted at the current time.
[0141] The channel network data includes the priority of each virtual channel, network latency, buffer utilization, bandwidth utilization, deadlock detection status of priority-based flow control, link layer discovery protocol and configuration information of the network nodes to be predicted.
[0142] In some embodiments, generating a network topology map of the network to be predicted at each time step based on channel network data at each time step includes:
[0143] The channel network data is cleaned and aggregated to obtain the first data; data cleaning includes removing duplicate data, supplementing missing data values, and handling abnormal data.
[0144] Generate the topology of the network to be predicted based on channel network data;
[0145] The network topology graph is obtained by mapping the first data of each virtual channel to the topology using a network mapping algorithm.
[0146] In some embodiments, the conversion module is specifically used for:
[0147] Define the adjacency matrix and feature matrix of the network to be predicted; the adjacency matrix represents the connection relationship of virtual channels between the topological nodes of the network to be predicted; the feature matrix is used to represent the attributes of the virtual channels between the topological nodes.
[0148] Graph convolutional networks transform the network topology graph into an adjacency matrix and a feature matrix.
[0149] In some embodiments, the input module is specifically used for:
[0150] Determine the time series corresponding to the spatial features of the network topology graph based on the adjacency matrix and feature matrix;
[0151] Capture information on network latency changes over time based on time series analysis;
[0152] Predict network performance data of the network to be predicted at future moments based on changing information.
[0153] In some embodiments, network performance data includes the occurrence time and magnitude of network latency of the virtual channel in the network to be predicted; the prediction module is specifically used for:
[0154] If the network performance data exceeds a preset threshold, it is determined that the network to be predicted will experience packet loss in the future.
[0155] It should be noted that the packet loss prediction device provided in this application embodiment can implement all the method steps implemented in the above packet loss prediction method embodiment and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiment and the beneficial effects will not be described in detail.
[0156] Figure 7 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application, such as... Figure 7 As shown, the electronic device may include: a processor 710, a communication interface 720, a memory 730, and a communication bus 740, wherein the processor 710, the communication interface 720, and the memory 730 communicate with each other via the communication bus 740. The processor 710 can call a computer program stored in the memory 730 to execute the above-described methods, such as including:
[0157] Obtain the network topology of the network to be predicted at the current time.
[0158] Based on graph convolutional networks, the network topology graph is transformed into the adjacency matrix and feature matrix of the network to be predicted;
[0159] The adjacency matrix and feature matrix are input into the network quality prediction model to obtain the network performance data of the network to be predicted at future time points, which is output by the network quality prediction model.
[0160] Predict packet loss behavior of the network at future moments based on network performance data.
[0161] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional modules and sold or used as independent products, and can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0162] On the other hand, this application also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can perform the methods provided in the above embodiments.
[0163] On the other hand, embodiments of this application also provide a processor-readable storage medium storing a computer program for causing the processor to execute the methods provided in the above embodiments.
[0164] The processor-readable storage medium can be any available medium or data storage device that the processor can access, including but not limited to magnetic memory (e.g., floppy disk, hard disk, magnetic tape, magneto-optical disk (MO)), optical memory (e.g., CD, DVD, BD, HVD), and semiconductor memory (e.g., ROM, EPROM, EEPROM, non-volatile memory (NAND FLASH), solid-state drive (SSD)).
[0165] The device embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.
[0166] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.
[0167] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.
Claims
1. A packet loss prediction method, characterized by, The method comprises the following steps: obtaining a network topology graph of a network to be predicted at a current time; converting the network topology graph into an adjacency matrix and a feature matrix corresponding to the network to be predicted based on a graph convolution network; inputting the adjacency matrix and the feature matrix into a network quality prediction model to obtain network performance data of the network to be predicted at a future time output by the network quality prediction model; predicting a packet loss behavior of the network to be predicted at the future time based on the network performance data; wherein the obtaining of the network topology graph of the network to be predicted at the current time comprises: real-time collection of channel network data of each virtual channel in the network to be predicted; generation of a network topology graph of the network to be predicted at each time based on the channel network data at each time; and obtaining of the network topology graph of the network to be predicted at the current time; the channel network data comprising priority, network delay, buffer utilization, bandwidth utilization, deadlock detection state based on priority flow control, network node link layer discovery protocol and configuration information of each virtual channel; the conversion of the network topology graph into the adjacency matrix and the feature matrix corresponding to the network to be predicted based on the graph convolution network comprises: defining the adjacency matrix and the feature matrix of the network to be predicted; the adjacency matrix representing the connection relationship of the virtual channels between the topology nodes of the network to be predicted; and the feature matrix representing the attributes of the virtual channels between the topology nodes; and converting the network topology graph into the adjacency matrix and the feature matrix based on the graph convolution network.
2. The packet loss prediction method of claim 1, wherein, The generation of the network topology graph of the network to be predicted at each time based on the channel network data at each time comprises: data cleaning and data aggregation of the channel network data to obtain first data; the data cleaning comprising removal of duplicate data, supplement of missing data and processing of abnormal data; generating a topology structure of the network to be predicted based on the channel network data; projecting the first data of each virtual channel into the topology structure based on a network projection algorithm to obtain the network topology graph.
3. The packet loss prediction method of claim 1, wherein, The obtaining of the network performance data of the network to be predicted at the future time output by the network quality prediction model comprises: determining a time sequence corresponding to a spatial feature of the network topology graph based on the adjacency matrix and the feature matrix; capturing change information of network delay over time based on the time sequence; predicting network performance data of the network to be predicted at the future time based on the change information.
4. The packet loss prediction method of claim 1, wherein, The network performance data comprises network delay occurrence time and network delay size of the virtual channels of the network to be predicted; The prediction of the packet loss behavior of the network to be predicted at the future time based on the network performance data comprises: determining that the network to be predicted has the packet loss behavior at the future time in the case that the network performance data is greater than a preset threshold.
5. A packet loss prediction apparatus characterized by comprising: The method comprises the following steps: an obtaining module, configured to obtain a network topology graph of a network to be predicted at a current time; a conversion module, configured to convert the network topology graph into an adjacency matrix and a feature matrix corresponding to the network to be predicted based on a graph convolution network; The input module is configured to input the adjacency matrix and the feature matrix into a network quality prediction model to obtain network performance data of the to-be-predicted network at a future time output by the network quality prediction model. The prediction module is configured to predict packet loss behavior of the to-be-predicted network at the future time based on the network performance data. The acquisition module is specifically configured to collect channel network data of each virtual channel in the to-be-predicted network in real time, generate a network topology graph of the to-be-predicted network at each time based on the channel network data at each time, and acquire the network topology graph of the to-be-predicted network at the current time. The channel network data includes priority, network delay, buffer utilization, bandwidth utilization, deadlock detection state based on priority-based flow control, to-be-predicted network node link layer discovery protocol, and configuration information of each virtual channel. The conversion module is specifically configured to define an adjacency matrix and a feature matrix of the to-be-predicted network, wherein the adjacency matrix represents a connection relationship of a virtual channel between topology nodes of the to-be-predicted network, and the feature matrix is used to represent attributes of the virtual channel between the topology nodes. The network topology graph is converted into the adjacency matrix and the feature matrix based on the graph convolution network. 6.An electronic device comprising a memory and a processor, the electronic device characterized by, The memory stores a computer program, and the processor is configured to execute the computer program to perform the packet loss prediction method in any one of claims 1 to 4.
7. A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the packet loss prediction method in any one of claims 1 to 4.
8. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the packet loss prediction method in any one of claims 1 to 4.
Citation Information
Patent Citations
VNF demand prediction method and system based on data driving
CN113923129A