CDN heterogeneous cache node resource estimation method and device

Abstract

The invention relates to the technical field of communication, and provides a CDN heterogeneous cache node resource estimation method and device, and the method comprises the steps: obtaining a data volume time sequence of a requested cache content in each CDN heterogeneous cache node and an access record text sequence of each user terminal in a recent preset time period; constructing a CDN network topological graph taking the CDN content center node, the CDN area nodes, the CDN edge service nodes and the user terminals as nodes; and inputting the data volume time sequence and the access record text sequence into a preset resource estimation model to obtain a resource demand estimated value of each CDN heterogeneous cache node at a future preset moment, wherein the resource demand estimated value is output by the resource estimation model. According to the method and the system, the resources required by each CDN heterogeneous cache node can be estimated more accurately and efficiently according to the use behavior preference of the user.

Description

Technical Field

[0001] The present invention relates to the field of communication technology, and in particular to a CDN heterogeneous cache node resource estimation method and device. Background Art

[0002] At present, in the Content Delivery Network (CDN), the estimation of resources (IT resources, such as CPU, memory, and bandwidth, etc.) required by each CDN heterogeneous cache node is mainly achieved through the expert experience of the maintenance team. However, CDN services are widely used by customers, across operators and regions, and have the characteristics of diverse and sudden demands. It is difficult to capture the demand patterns through manual experience, and there are problems such as low accuracy of resource estimation of CDN heterogeneous cache nodes and time-consuming and labor-intensive manual work. Summary of the invention

[0003] The present invention provides a CDN heterogeneous cache node resource estimation method and device, which are used to solve the problems in the prior art that it is difficult to capture demand rules through manual experience, the resource estimation accuracy of CDN heterogeneous cache nodes is low, and manual work is time-consuming and labor-intensive.

[0004] The present invention provides a CDN heterogeneous cache node resource estimation method, comprising the following steps.

[0005] Obtain the time series of the data volume requested for cached content in each CDN heterogeneous cache node and the access record text sequence of each user terminal within the most recent preset time period.

[0006] A CDN network topology diagram is constructed with CDN content center nodes, CDN regional nodes, CDN edge service nodes and user terminals as nodes. The data transmission relationship between the four types of heterogeneous nodes is used as the edge of the CDN network topology diagram. The data volume time series is used as the node feature of the corresponding CDN heterogeneous cache node, and the access record text sequence is used as the node feature of the corresponding user terminal.

[0007] The data volume time series and the access record text sequence are input into a preset resource estimation model to obtain the resource demand estimate of each CDN heterogeneous cache node at a preset time in the future output by the resource estimation model.

[0008] The resource estimation model is obtained by training based on data volume time series samples, access record text sequence samples, and resource demand labels at future preset moments corresponding to the data volume time series samples.

[0009] According to a CDN heterogeneous cache node resource estimation method provided by the present invention, the resource estimation model includes: a graph attention network and a long short-term memory neural network.

[0010] The graph attention network is used to fuse the target data volume time series and the neighbor reference sequence to obtain a target feature vector of the target node that is fused with the feature information of the neighbor nodes. The target node is any node among the CDN content center node, CDN regional node and CDN edge service node in the CDN network topology diagram; the target data volume time series is the data volume time series of the target node, and the neighbor reference sequence is the data volume time series or the access record text sequence of the neighbor nodes directly connected to the target node.

[0011] The long short-term memory neural network is used to output the estimated resource demand of the target node at a preset time in the future based on the target feature vector of the target node.

[0012] According to a CDN heterogeneous cache node resource estimation method provided by the present invention, the graph attention network includes: a word embedding layer, which is used to convert the target data volume time series into a fixed shape vector of a preset dimension to obtain a target data volume time series vector.

[0013] The first mapping layer is used to convert the dimension of the first-type neighbor reference sequence into a first-type neighbor reference sequence vector that is consistent with the dimension of the target data volume time series vector.

[0014] The second mapping layer is used to convert the dimension of the second-type neighbor reference sequence into a second-type neighbor reference sequence vector that is consistent with the dimension of the target data volume time series vector.

[0015] The first neighbor node level graph attention layer is used to fuse the target data volume time series vector and the first type of neighbor reference sequence vector and output a first type of intermediate feature vector of the target node that is fused with the feature information of the first type of neighbor nodes.

[0016] The second neighbor node level graph attention layer is used to fuse the target data volume time series vector and the second type of neighbor reference sequence vector and output a second type of intermediate feature vector of the target node that is fused with the feature information of the second type of neighbor nodes.

[0017] A target node-level graph attention layer is used to fuse the first-category intermediate feature vector and the second-category intermediate feature vector to obtain the target feature vector.

[0018] According to a CDN heterogeneous cache node resource estimation method provided by the present invention, the long short-term memory neural network includes: a long short-term memory layer, which is used to extract the time features in the target feature vector, and input the target feature vector with the extracted time features into a fully connected layer.

[0019] The fully connected layer is used to output the estimated resource demand of the target node at a preset time in the future based on the target feature vector of the extracted time feature.

[0020] According to a CDN heterogeneous cache node resource estimation method provided by the present invention, the loss function of the resource estimation model during training is: .

[0021] in, It represents the resource demand forecast value of the i-th CDN heterogeneous cache node at a preset time in the future output by the resource estimation model during training. It represents the resource demand label of the i-th CDN heterogeneous cache node at a preset time in the future, and n represents the total number of CDN heterogeneous cache nodes.

[0022] According to a CDN heterogeneous cache node resource estimation method provided by the present invention, before the data volume time series and the access record text sequence are input into a preset resource estimation model, it also includes: standardizing the data volume time series and encoding the access record text sequence of each user terminal into sequences of the same length.

[0023] The present invention also provides a CDN heterogeneous cache node resource estimation device, comprising the following modules.

[0024] The data acquisition module is used to obtain the time series of the data volume requested for cached content in each CDN heterogeneous cache node and the access record text sequence of each user terminal within the recent preset time period.

[0025] A topology construction module is used to construct a CDN network topology diagram with CDN content center nodes, CDN regional nodes, CDN edge service nodes and user terminals as nodes. The data transmission relationship between the four types of heterogeneous nodes is used as the edge of the CDN network topology diagram, the data volume time series is used as the node feature of the corresponding CDN heterogeneous cache node, and the access record text sequence is used as the node feature of the corresponding user terminal.

[0026] The model execution module is used to input the data volume time series and the access record text sequence into a preset resource estimation model to obtain the resource demand estimate of each CDN heterogeneous cache node at a preset time in the future output by the resource estimation model.

[0027] The resource estimation model is obtained by training based on data volume time series samples, access record text sequence samples, and resource demand labels at future preset moments corresponding to the data volume time series samples.

[0028] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the CDN heterogeneous cache node resource estimation method as described in any one of the above is implemented.

[0029] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the CDN heterogeneous cache node resource estimation method as described in any one of the above is implemented.

[0030] The present invention also provides a computer program product, including a computer program, wherein when the computer program is executed by a processor, the method for estimating CDN heterogeneous cache node resources as described in any one of the above is implemented.

[0031] The CDN heterogeneous cache node resource estimation method and device provided by the present invention obtain the data volume time series of cache content requested in each CDN heterogeneous cache node and the access record text sequence of each user terminal in the most recent preset time period; construct a CDN network topology diagram with CDN content center node, CDN regional node, CDN edge service node and user terminal as nodes, and the data transmission relationship between the four types of heterogeneous nodes as the edge of the CDN network topology diagram, and use the data volume time series as the node feature of the corresponding CDN heterogeneous cache node, and use the access record text sequence as the node feature of the corresponding user terminal; input the data volume time series and the access record text sequence into a preset resource estimation model to obtain the resource demand estimation value of each CDN heterogeneous cache node at a preset time in the future output by the resource estimation model; wherein the resource estimation model is trained based on the data volume time series samples, the access record text sequence samples and the resource demand volume labels at the preset time in the future corresponding to the data volume time series samples. Since the resource estimation model is trained based on data volume time series samples, access record text sequence samples and resource demand labels at future preset moments corresponding to data volume time series samples, the model can reflect the relationship between the demand of each CDN heterogeneous cache node and the access behavior of the user terminal, so as to more accurately and efficiently estimate the resources required for each CDN heterogeneous cache node according to the user's usage behavior preferences, thereby minimizing the impact on CDN services caused by resource shortages. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0033] Figure 1 It is a flow chart of the CDN heterogeneous cache node resource estimation method provided by the present invention.

[0034] Figure 2It is a schematic diagram of the CDN network topology constructed in the CDN heterogeneous cache node resource estimation method provided by the present invention.

[0035] Figure 3 It is a schematic diagram of the resource estimation model structure in the CDN heterogeneous cache node resource estimation method provided by the present invention.

[0036] Figure 4 It is a structural schematic diagram of a CDN heterogeneous cache node resource estimation device provided by the present invention.

[0037] Figure 5 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0039] In order to better understand the content of the embodiments of the present invention, the network structure of the CDN network and the functions of the main service nodes are introduced as follows: The main service nodes of the CDN network include: CDN content center nodes, CDN regional nodes and CDN edge service nodes. CDN edge service nodes are connected to user terminals. CDN content center nodes, CDN regional nodes and CDN edge service nodes form a three-level network structure from top to bottom, which together constitute the CDN business system and realize data service functions. The data service functions in the CDN network include: access, storage / caching, processing (data slicing, transcoding, etc.) and distribution of user-accessed content.

[0040] CDN content center node: mainly responsible for connecting with the business system to achieve CDN content access, management, storage and active distribution to CDN regional nodes, including the following functions.

[0041] Content access: Obtain the specified content according to the content injection instructions of the Content Management System (CMS) and inject it into the content storage, and register it in the content management system; or do not inject content through the CMS, but use the back-to-source method to access the content source.

[0042] Content distribution: Distribute and deliver content according to the scheduling strategy of the content management system.

[0043] Local load balancing: Receives content location and requests from lower-level nodes (CDN regional nodes), performs content addressing, and allocates appropriate devices to provide services according to load balancing strategies, and performs load balancing on devices within the node.

[0044] Content storage: Content is stored according to the policies in the content management system, and content can be stored and updated according to the cache policy.

[0045] Content processing: pre-process the injected content, such as slicing, transcoding, and transpackaging.

[0046] CDN regional node: mainly responsible for aggregating the traffic of lower-level nodes (CDN edge service nodes) and reducing the return traffic. Specifically, it includes the following functions.

[0047] Content distribution: Distribute and deliver content according to the scheduling strategy of the content management system.

[0048] Local load balancing: Receive content location and requests from lower-level nodes, perform content addressing, and allocate appropriate devices to provide services according to load balancing strategies, and perform load balancing on devices within the node.

[0049] Content caching: Store and update content according to caching strategies.

[0050] CDN edge service node: As the main entity of CDN service, it is mainly responsible for receiving user terminal requests, verifying and providing local cached content services to users. If the content is not hit, it obtains and caches the content from the superior node, or redirects to the superior to provide services. Specifically, it includes the following functions.

[0051] Content service control: Search for content service nodes based on content ID.

[0052] Global load balancing: Receives terminal service requests and controls edge service nodes to provide edge services to terminals based on content routing results, node status, load balancing policies, etc., performs load balancing on devices within the node, and provides node fault switching.

[0053] Content services: Provide various content services for various business terminals, such as application services, streaming media services and download services.

[0054] Content processing: slicing, transcoding, and repackaging of content.

[0055] Content caching: Store and update content according to caching strategies.

[0056] Based on the above CDN network, the CDN heterogeneous cache node resource estimation method of the embodiment of the present invention is as follows: Figure 1 As shown, the process includes the following steps S110 to S130.

[0057] Step S110: Obtain the time series of the data volume requested for cached content in each CDN heterogeneous cache node and the access record text sequence of each user terminal in the recent preset time period. In the CDN network, CDN content center nodes, CDN regional nodes, CDN edge service nodes and user terminals are four types of nodes that are heterogeneous (node ​​types are heterogeneous).

[0058] For user terminals, when requesting resources, they will access the CDN edge service node. The access record text sequence is the access record text recorded by time. For example, within the most recent preset T period, each access is recorded in the order of the access time. The recorded content includes: access content and access duration, among which the access content reflects the user's behavioral preferences.

[0059] For the three types of CDN heterogeneous cache nodes, namely CDN content center nodes, CDN regional nodes and CDN edge service nodes, CDN edge service nodes will receive content requests from user terminals. If the content is not hit, the CDN edge service node will request the content from the CDN regional node. If the content is not hit in the CDN regional node, the CDN regional node will request the content from the CDN content center node. Therefore, for these three types of nodes, the time series of the amount of data requested for cached content in each CDN node in the most recent preset time period can be obtained. The elements in the series are the amount of data requested for cached content in the CDN node recorded in chronological order in the preset time period.

[0060] Step S120: Construct a CDN network topology graph with CDN content center nodes, CDN regional nodes, CDN edge service nodes and user terminals as nodes, and use the data transmission relationship between the four types of heterogeneous nodes as the edge of the CDN network topology graph, use the data volume time series as the node feature of the corresponding CDN heterogeneous cache node, and use the access record text sequence as the node feature of the corresponding user terminal. Figure 2 As shown, this is the CDN network topology diagram constructed in this step, and the CDN network topology diagram is constructed according to the actual CDN network connection structure.

[0061] Specifically, for Figure 2 The CDN network topology diagram, the characteristic of the CDN content center node is the amount of data requested for the cached content of the CDN content center node within a preset time period, expressed as A={a 1 、a 2 、a 3 , …, a u}, where a u Represents the time series of the amount of data requested for cached content in the u-th CDN content center node.

[0062] The characteristic of a CDN regional node is the amount of data requested for the cached content of the CDN regional node within a preset time period, expressed as B={b 1 、b 2 、b 3 , …, b v}, where b v Represents the time series of the amount of data requested for cached content in the vth CDN region node.

[0063] The characteristic of a CDN edge service node is the amount of data requested for the cached content of the CDN edge service node within a preset time period, expressed as C = {c 1 、c 2 、c 3 ,…,c s}, where c s Represents the time series of the amount of data requested for cached content in the s-th CDN edge service node.

[0064] The characteristic of a user terminal is the text sequence of the access record of the user terminal accessing the CDN network within a preset time period, expressed as D={d 1 ,d 2 ,d 3 ,…,d z}, where d z Represents the access record text sequence of the zth user terminal.

[0065] Step S130: Input the data volume time series and access record text sequence into a preset resource estimation model to obtain the resource demand estimate of each CDN heterogeneous cache node at a preset future moment output by the resource estimation model. The resource estimation model is trained based on data volume time series samples, access record text sequence samples, and resource demand labels at preset future moments corresponding to the data volume time series samples. Specifically, the data volume time series samples and access record text sequence samples are historical data, i.e., data within the historical T' time interval. The resource demand labels can be the label matrix Y of the resource demand of each CDN heterogeneous cache node at the required predicted future (relative to the future of the historical T' time interval) L moments. The predicted resource demand of the i-th CDN heterogeneous cache node can be expressed as {y i t+1 , y i t+2 ,y i t+3 ,…, y i t+L-1 , y i t+L}, the shape (i.e., matrix dimension) is N*L, t represents the current moment, and N represents the number of resource types (including: CPU, storage, memory, and bandwidth, etc.).

[0066] In the CDN heterogeneous cache node resource estimation method provided in this embodiment, since a resource estimation model is adopted based on data volume time series samples, access record text sequence samples and resource demand labels at future preset moments corresponding to the data volume time series samples, the model can reflect the relationship between the demand of each CDN heterogeneous cache node and the access behavior of the user terminal, so that the resources required for each CDN heterogeneous cache node can be estimated more accurately and efficiently according to the user's usage behavior preferences, thereby minimizing the impact on CDN services caused by resource shortages.

[0067] In some embodiments, the resource estimation model includes: Graph Attention Networks (GAT) and long short-term memory neural networks (LSTM).

[0068] The graph attention network is used to fuse the target data volume time series and the neighbor reference sequence to obtain the target feature vector of the target node that is fused with the feature information of the neighbor nodes. The target node is any node in the CDN content center node, CDN regional node, and CDN edge service node in the CDN network topology diagram. The target data volume time series is the data volume time series of the target node, and the neighbor reference sequence is the data volume time series or access record text sequence of the neighbor nodes directly connected to the target node. That is, the spatial dependency between the three types of CDN heterogeneous cache nodes and user terminals is captured through the graph attention network.

[0069] The LSTM neural network is used to output the estimated resource demand of the target node at a preset time in the future based on the target feature vector of the target node. Specifically, the LSTM neural network is used to learn the time series of the data volume requested for cached content in three types of CDN heterogeneous cache nodes and the dynamic changes in user terminal behavior to capture time dependencies.

[0070] In this embodiment, the spatial dependencies between the three types of CDN heterogeneous cache nodes and user terminals are captured through the graph attention network, and the temporal dependencies are captured by learning the time series of the requested data volume of cached content in the three types of CDN heterogeneous environment nodes and the dynamic changes of user terminal access behavior through the long short-term memory neural network, so as to more accurately and efficiently estimate the resources required for CDN content cache nodes according to user usage behavior preferences.

[0071] In some embodiments, Figure 3 As shown, the graph attention network includes the following hierarchical structure.

[0072] The word embedding layer is used to convert the target data volume time series into a fixed shape vector of a preset dimension to obtain a target data volume time series vector.

[0073] The first mapping layer is used to convert the dimension of the first-type neighbor reference sequence into a first-type neighbor reference sequence vector that is consistent with the dimension of the target data volume time series vector.

[0074] The second mapping layer is used to convert the dimension of the second-type neighbor reference sequence into a second-type neighbor reference sequence vector that is consistent with the dimension of the target data volume time series vector.

[0075] The first neighbor node level graph attention layer is used to fuse the target data volume time series vector and the first type of neighbor reference sequence vector and output a first type of intermediate feature vector of the target node that is fused with the feature information of the first type of neighbor nodes.

[0076] The second neighbor node level graph attention layer is used to fuse the target data volume time series vector and the second type of neighbor reference sequence vector and output a second type of intermediate feature vector of the target node that is fused with the feature information of the second type of neighbor nodes.

[0077] A target node-level graph attention layer is used to fuse the first-category intermediate feature vector and the second-category intermediate feature vector to obtain the target feature vector.

[0078] Depend on Figure 2 It can be seen that there are other heterogeneous nodes in the upper and lower layers of CDN regional nodes and CDN edge service nodes. Therefore, there are two types of neighbor nodes for CDN regional nodes and CDN edge service nodes. The upper layer nodes can be regarded as the first type of neighbor nodes, and the lower layer nodes can be regarded as the second type of neighbor nodes.

[0079] Like all attention mechanisms, the calculation of GAT is also divided into two steps: calculating the attention coefficient and weighted summation. The GAT model is implemented by stacking graph attention layers, and the input of each graph attention layer is the text and indicator feature set generated by the node.

[0080] .

[0081] The output is a new set of text and indicator features.

[0082] .

[0083] In order to calculate the weight of each neighbor node, a shared weight matrix W of F×F' is applied to each node, and then the attention coefficient can be calculated, which can represent the node j Relative to the node i Importance e ij .

[0084] .

[0085] In order to make the attention coefficient easier to calculate and compare, a softmax is introduced for all i Neighboring nodes j Perform regularization.

[0086] .

[0087] Therefore, we get the nodes after the graph attention layer. i The feature representation is: that is, GAT outputs for each vertex i New features (integrating neighborhood information), σ is the activation function.

[0088] .

[0089] in N i Representation Node i The neighbor node set of α That is, the coefficient used for weighted summation during each convolution.

[0090] In some embodiments, the long short-term memory neural network includes the following hierarchical structure.

[0091] The long short-term memory layer is used to extract the time feature in the target feature vector, and input the target feature vector with the extracted time feature into the fully connected layer.

[0092] It should be noted that the target feature vector after the graph attention network is a target feature vector with spatial features. After passing through the long short-term memory layer, the target feature vector contains the spatial features and temporal features of the CDN network topology.

[0093] The fully connected layer is used to output the estimated resource demand of the target node at a preset time in the future based on the target feature vector of the extracted time feature.

[0094] like Figure 3 As shown, taking the prediction of CDN edge service node resource demand as an example, the resource estimation model of the above embodiment is described in detail as follows.

[0095] The graph attention network has two branches, branch 1 and branch 2. The structure of branch 1 includes the first word embedding layer, the first mapping layer and the first neighbor node level graph attention layer mentioned above. The structure of branch 2 includes: the second word embedding layer, the second mapping layer and the second neighbor node level graph attention layer.

[0096] Branch 1: Learn the weight of the first-class neighbor nodes (i.e., CDN regional nodes) of CDN edge service node i in evaluating its importance. Input the time series h of the requested data volume of CDN edge service node i ci and the time series of the requested data volume of CDN regional nodes h bi , which are converted into vectors through the word embedding layer and the first mapping layer and then input into the neighbor node level graph attention layer, and output the new CDN edge service node feature h after integrating the CDN regional node feature information. ci 1 .

[0097] Branch 2: Learn the weight of the second type of neighbor nodes (i.e., user terminals) of CDN edge service node i in evaluating its importance. Input the time series h of the requested data volume of CDN edge service node i ci And the terminal user's APP access record text sequence h di , which are converted into vectors through the word embedding layer and the second mapping layer and then input into the neighbor node level graph attention layer, and output the new feature h of the CDN edge service node after integrating the feature information of the end user. ci 2 .

[0098] The word embedding layer (embedding): uses word embedding to convert each word in the target data time series into a vector. The input data dimension is z, and the output is set to the 64-dimensional space vector that needs to convert the word. The input sequence length is F, so the shape of the output data of this layer is (None, F, 64). The function of this layer is to perform vector mapping on the input words and convert the index of each word into a 64-dimensional fixed-shape vector.

[0099] The mapping layer uses the conversion matrix of the mapping layer to convert the dimension of the operating indicator characteristics of the neighbor node (the first-type neighbor reference sequence or the second-type neighbor reference sequence) to be consistent with the dimension of the target data volume time series vector.

[0100] Among them, the neighbor node level graph attention layer: the number of convolution kernels is 256, and the activation function is set to "relu".

[0101] Then the new feature h of the CDN edge service node after integrating the feature information of the CDN regional node output from the above two branches is ci 1and the new feature h of CDN edge service node after integrating the terminal user feature information ci 2 The input is sent to the central node-level graph attention layer for central node-level attention aggregation, and the new feature h of the CDN edge service node is outputted by integrating the features of two types of neighboring nodes. ci '.

[0102] Among them, the central node level graph attention layer: the number of convolution kernels is 128, and the activation function is set to "relu".

[0103] In some embodiments, the loss function of the resource estimation model during training is: .

[0104] in, It represents the resource demand forecast value of the i-th CDN heterogeneous cache node at a preset time in the future output by the resource estimation model during training. It represents the resource demand label of the i-th CDN heterogeneous cache node at a preset time in the future, and n represents the total number of CDN heterogeneous cache nodes.

[0105] Specifically, the model can be trained for 1000 iterations (epochs=1000), the batch size is set to 32 (batch_size=32), and the mean squared error (MSE) is selected as the loss function. The total sample data set can be divided into training data and test data, 80% of the entire sample data set is taken as training data, and the remaining 20% ​​is taken as test data. The training set is used for training, and the test set is used to evaluate and verify the model. During training, only a set of data volume time series samples, access record text sequence samples, and resource demand labels at future preset moments corresponding to the data volume time series samples need to be input into the resource estimation model, and the model's predicted value is substituted into the above loss function. The model weight value is updated by back propagation according to the value of the loss function, thereby completing an iteration, and then the next set of data volume time series samples, access record text sequence samples, and resource demand labels at future preset moments corresponding to the data volume time series samples are input into the resource estimation model until 1000 iterations are made and the loss function converges, and the model training is completed.

[0106] During the training process, the gradient descent optimization algorithm selects the adam optimizer to improve the learning speed of traditional gradient descent (optimizer='adam'). Through gradient descent, the neural network can find the optimal weight value that minimizes the objective function. The neural network will learn the weight value autonomously through training, and the weight of the model will be derived after the model converges.

[0107] In some embodiments, before the data volume time series and the access record text sequence are input into a preset resource estimation model, it also includes: normalizing the data volume time series and encoding the access record text sequence of each user terminal into sequences of the same length.

[0108] Specifically, the data volume time series of cache content requests in the three types of CDN heterogeneous cache nodes are standardized: (X-mean) / std. The calculation is performed separately for each dimension, and the data is subtracted from its mean by attribute (by column) and divided by its variance std. For data volume time series samples, the standardized data will improve the convergence speed and accuracy of the model during model training. When applying the model, the data volume time series also needs to be standardized.

[0109] In this embodiment, the access record text sequences of each user terminal are encoded as sequence representations, and the encoding sequence length of each node feature is defined as F. The longest length F in the data set is taken as its encoding sequence length, and the length of each data is filled with F. The dictionary size is taken as z. The same length of encoding sequence length is helpful for model training and improves the convergence speed and accuracy of the model.

[0110] The CDN heterogeneous cache node resource estimation device provided by the present invention is described below. The CDN heterogeneous cache node resource estimation device described below and the CDN heterogeneous cache node resource estimation method described above can be referenced to each other.

[0111] like Figure 4 As shown, the CDN heterogeneous cache node resource estimation device of the present invention includes the following modules.

[0112] The data acquisition module 410 is used to acquire the time series of the data volume requested for cached content in each CDN heterogeneous cache node and the access record text sequence of each user terminal within the latest preset time period.

[0113] The topology construction module 420 is used to construct a CDN network topology diagram with CDN content center nodes, CDN regional nodes, CDN edge service nodes and user terminals as nodes, the data transmission relationship between the four types of heterogeneous nodes as the edge of the CDN network topology diagram, the data volume time series as the node feature of the corresponding CDN heterogeneous cache node, and the access record text sequence as the node feature of the corresponding user terminal.

[0114] The model execution module 430 is used to input the data volume time series and the access record text sequence into a preset resource estimation model to obtain the resource demand estimate of each CDN heterogeneous cache node at a preset time in the future output by the resource estimation model.

[0115] The resource estimation model is obtained by training based on data volume time series samples, access record text sequence samples, and resource demand labels at future preset moments corresponding to the data volume time series samples.

[0116] In the CDN heterogeneous cache node resource estimation device of the present invention, since a resource estimation model is adopted based on data volume time series samples, access record text sequence samples and resource demand labels at future preset moments corresponding to the data volume time series samples, the model can reflect the relationship between the demand of each CDN heterogeneous cache node and the access behavior of the user terminal, so that the resources required for each CDN heterogeneous cache node can be estimated more accurately and efficiently according to the user's usage behavior preferences, thereby avoiding the impact on CDN services caused by resource shortages to the greatest extent.

[0117] In some embodiments, the resource estimation model includes: a graph attention network and a long short-term memory neural network.

[0118] The graph attention network is used to fuse the target data volume time series and the neighbor reference sequence to obtain a target feature vector of the target node that is fused with the feature information of the neighbor nodes. The target node is any node among the CDN content center node, CDN regional node and CDN edge service node in the CDN network topology diagram; the target data volume time series is the data volume time series of the target node, and the neighbor reference sequence is the data volume time series or the access record text sequence of the neighbor nodes directly connected to the target node.

[0119] The long short-term memory neural network is used to output the estimated resource demand of the target node at a preset time in the future based on the target feature vector of the target node.

[0120] In some embodiments, the graph attention network includes the following hierarchical structure.

[0121] The word embedding layer is used to convert the target data volume time series into a fixed shape vector of a preset dimension to obtain a target data volume time series vector.

[0122] The first mapping layer is used to convert the dimension of the first-type neighbor reference sequence into a first-type neighbor reference sequence vector that is consistent with the dimension of the target data volume time series vector.

[0123] The second mapping layer is used to convert the dimension of the second-type neighbor reference sequence into a second-type neighbor reference sequence vector that is consistent with the dimension of the target data volume time series vector.

[0124] The first neighbor node level graph attention layer is used to fuse the target data volume time series vector and the first type of neighbor reference sequence vector and output a first type of intermediate feature vector of the target node that is fused with the feature information of the first type of neighbor nodes.

[0125] The second neighbor node level graph attention layer is used to fuse the target data volume time series vector and the second type of neighbor reference sequence vector and output a second type of intermediate feature vector of the target node that is fused with the feature information of the second type of neighbor nodes.

[0126] A target node-level graph attention layer is used to fuse the first-category intermediate feature vector and the second-category intermediate feature vector to obtain the target feature vector.

[0127] In some embodiments, the long short-term memory neural network includes the following hierarchical structure.

[0128] The long short-term memory layer is used to extract the time feature in the target feature vector, and input the target feature vector with the extracted time feature into the fully connected layer.

[0129] The fully connected layer is used to output the estimated resource demand of the target node at a preset time in the future based on the target feature vector of the extracted time feature.

[0130] In some embodiments, the loss function of the resource estimation model during training is: .

[0131] in, It represents the resource demand forecast value of the i-th CDN heterogeneous cache node at a preset time in the future output by the resource estimation model during training. It represents the resource demand label of the i-th CDN heterogeneous cache node at a preset time in the future, and n represents the total number of CDN heterogeneous cache nodes.

[0132] In some embodiments, the CDN heterogeneous cache node resource estimation device also includes: a standardization processing module, which is used to standardize the data volume time series and the access record text sequence before inputting the data volume time series and the access record text sequence into a preset resource estimation model, and encode the access record text sequence of each user terminal into a sequence of the same length.

[0133] Figure 5 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 5 As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530 and a communication bus 540, wherein the processor 510, the communication interface 520 and the memory 530 communicate with each other through the communication bus 540. The processor 510 may call the logic instructions in the memory 530 to execute the CDN heterogeneous cache node resource estimation method, which includes the following steps.

[0134] Obtain the time series of the data volume requested for cached content in each CDN heterogeneous cache node and the access record text sequence of each user terminal within the most recent preset time period.

[0135] A CDN network topology diagram is constructed with CDN content center nodes, CDN regional nodes, CDN edge service nodes and user terminals as nodes. The data transmission relationship between the four types of heterogeneous nodes is used as the edge of the CDN network topology diagram. The data volume time series is used as the node feature of the corresponding CDN heterogeneous cache node, and the access record text sequence is used as the node feature of the corresponding user terminal.

[0136] The data volume time series and the access record text sequence are input into a preset resource estimation model to obtain the resource demand estimate of each CDN heterogeneous cache node at a preset time in the future output by the resource estimation model.

[0137] The resource estimation model is obtained by training based on data volume time series samples, access record text sequence samples, and resource demand labels at future preset moments corresponding to the data volume time series samples.

[0138] In addition, the logic instructions in the above-mentioned memory 530 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0139] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the CDN heterogeneous cache node resource estimation method provided by the above methods, and the method includes the following steps.

[0140] Obtain the time series of the data volume requested for cached content in each CDN heterogeneous cache node and the access record text sequence of each user terminal within the most recent preset time period.

[0141] A CDN network topology diagram is constructed with CDN content center nodes, CDN regional nodes, CDN edge service nodes and user terminals as nodes. The data transmission relationship between the four types of heterogeneous nodes is used as the edge of the CDN network topology diagram. The data volume time series is used as the node feature of the corresponding CDN heterogeneous cache node, and the access record text sequence is used as the node feature of the corresponding user terminal.

[0142] The data volume time series and the access record text sequence are input into a preset resource estimation model to obtain the resource demand estimate of each CDN heterogeneous cache node at a preset time in the future output by the resource estimation model.

[0143] The resource estimation model is obtained by training based on data volume time series samples, access record text sequence samples, and resource demand labels at future preset moments corresponding to the data volume time series samples.

[0144] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for estimating CDN heterogeneous cache nodes resources provided by the above methods is implemented, and the method includes the following steps.

[0145] Obtain the time series of the data volume requested for cached content in each CDN heterogeneous cache node and the access record text sequence of each user terminal within the most recent preset time period.

[0146] A CDN network topology diagram is constructed with CDN content center nodes, CDN regional nodes, CDN edge service nodes and user terminals as nodes. The data transmission relationship between the four types of heterogeneous nodes is used as the edge of the CDN network topology diagram. The data volume time series is used as the node feature of the corresponding CDN heterogeneous cache node, and the access record text sequence is used as the node feature of the corresponding user terminal.

[0147] The data volume time series and the access record text sequence are input into a preset resource estimation model to obtain the resource demand estimate of each CDN heterogeneous cache node at a preset time in the future output by the resource estimation model.

[0148] The resource estimation model is obtained by training based on data volume time series samples, access record text sequence samples, and resource demand labels at future preset moments corresponding to the data volume time series samples.

[0149] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Those of ordinary skill in the art may understand and implement it without creative work.

[0150] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0151] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A CDN heterogeneous cache node resource estimation method, characterized in that: include: Obtaining the time series of the data volume requested for cached content in each CDN heterogeneous cache node and the access record text sequence of each user terminal within the most recent preset time period; Construct a CDN network topology graph with CDN content center nodes, CDN regional nodes, CDN edge service nodes and user terminals as nodes, use the data transmission relationship between the four types of heterogeneous nodes as the edge of the CDN network topology graph, use the data volume time series as the node feature of the corresponding CDN heterogeneous cache node, and use the access record text sequence as the node feature of the corresponding user terminal; Inputting the data volume time series and the access record text sequence into a preset resource estimation model to obtain an estimated resource demand value of each CDN heterogeneous cache node at a preset time in the future output by the resource estimation model; The resource estimation model is obtained by training based on data volume time series samples, access record text sequence samples, and resource demand labels at future preset moments corresponding to the data volume time series samples.

2. The CDN heterogeneous cache node resource estimation method according to claim 1, characterized in that: Resource estimation models include: graph attention network and long short-term memory neural network; The graph attention network is used to fuse the target data volume time series and the neighbor reference sequence to obtain a target feature vector of the target node that is fused with feature information of neighbor nodes, wherein the target node is any node among the CDN content center node, the CDN regional node and the CDN edge service node in the CDN network topology diagram; the target data volume time series is the data volume time series of the target node, and the neighbor reference sequence is the data volume time series or the access record text sequence of the neighbor node directly connected to the target node; The long short-term memory neural network is used to output the estimated resource demand of the target node at a preset time in the future based on the target feature vector of the target node.

3. The CDN heterogeneous cache node resource estimation method according to claim 2, characterized in that: The graph attention network includes: A word embedding layer, used to convert the target data volume time series into a fixed shape vector of a preset dimension to obtain a target data volume time series vector; A first mapping layer, used to convert the dimension of the first-type neighbor reference sequence into a first-type neighbor reference sequence vector consistent with the dimension of the target data volume time series vector; A second mapping layer, used for converting the dimension of the second type neighbor reference sequence into a second type neighbor reference sequence vector consistent with the dimension of the target data volume time series vector; A first neighbor node level graph attention layer, used for fusing the target data volume time series vector and the first type of neighbor reference sequence vector and outputting a first type of intermediate feature vector of the target node fused with feature information of the first type of neighbor nodes; A second neighbor node level graph attention layer is used to fuse the target data volume time series vector and the second type of neighbor reference sequence vector and output a second type of intermediate feature vector of the target node fused with feature information of the second type of neighbor nodes; A target node-level graph attention layer is used to fuse the first-category intermediate feature vector and the second-category intermediate feature vector to obtain the target feature vector.

4. The CDN heterogeneous cache node resource estimation method according to claim 2, characterized in that: The long short-term memory neural network comprises: A long short-term memory layer, used for extracting the time feature in the target feature vector, and inputting the target feature vector with the extracted time feature into the fully connected layer; The fully connected layer is used to output the estimated resource demand of the target node at a preset time in the future based on the target feature vector of the extracted time feature.

5. The CDN heterogeneous cache node resource estimation method according to claim 2, characterized in that: The loss function of the resource estimation model during training is: ； in, It represents the resource demand forecast value of the i-th CDN heterogeneous cache node at a preset time in the future output by the resource estimation model during training. It represents the resource demand label of the i-th CDN heterogeneous cache node at a preset time in the future, and n represents the total number of CDN heterogeneous cache nodes.

6. The CDN heterogeneous cache node resource estimation method according to any one of claims 1 to 5, characterized in that: Before inputting the data volume time series and the access record text sequence into a preset resource estimation model, the method further includes: The data volume time series is standardized, and the access record text sequence of each user terminal is encoded into a sequence with the same length.

7. A CDN heterogeneous cache node resource estimation device, characterized in that: include: A data acquisition module is used to obtain a time series of data volume requested for cached content in each CDN heterogeneous cache node and a text sequence of access records of each user terminal within a recent preset time period; A topology construction module is used to construct a CDN network topology diagram with CDN content center nodes, CDN regional nodes, CDN edge service nodes and user terminals as nodes, the data transmission relationship between the four types of heterogeneous nodes as the edge of the CDN network topology diagram, the data volume time series as the node feature of the corresponding CDN heterogeneous cache node, and the access record text sequence as the node feature of the corresponding user terminal; A model execution module, used to input the data volume time series and the access record text sequence into a preset resource estimation model, and obtain the resource demand estimate of each CDN heterogeneous cache node at a preset time in the future output by the resource estimation model; The resource estimation model is obtained by training based on data volume time series samples, access record text sequence samples, and resource demand labels at future preset moments corresponding to the data volume time series samples.

8. An electronic device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the CDN heterogeneous cache node resource estimation method as described in any one of claims 1 to 6 when executing the computer program.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for estimating CDN heterogeneous cache node resources is implemented as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for estimating CDN heterogeneous cache node resources is implemented as described in any one of claims 1 to 6.

Images