5G collaborative SDN network load balancing scheduling method and storage medium

By using delay and hierarchical information interaction algorithms to divide the receiving clusters in the 5G collaborative SDN network, combining matrix decomposition and Istio virtual services, the problem of network traffic-guided load concurrency is solved, and load balancing is optimized and cost reduction is achieved.

CN118921701APending Publication Date: 2024-11-08CHINA TELECOM DIGITAL INTELLIGENCE TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410840769.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-06-27
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

There is a problem of traffic-guided load concurrency in the 5G collaborative SDN network. Traditional network functions based on hardware devices are complex and costly, making it difficult to achieve effective load balancing.

Method used

In the 5G collaborative SDN network, the receiving nodes are arranged using delay units, and the correlation analysis is performed using a layered information interaction algorithm, and the receiving cluster and the suboptimal receiving cluster are divided. The initiating node forwards traffic to each cluster according to the preset proportion. Combined with matrix decomposition, optimize traffic allocation and deploy Istio virtual services to achieve traffic shunt.

Benefits of technology

It effectively alleviates network congestion, reduces the cost of load balancing hardware expansion, improves the rationality and efficiency of traffic shunts, and solves the problem of traffic-guided load concurrency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118921701A_ABST
    Figure CN118921701A_ABST
Patent Text Reader

Abstract

The invention discloses a 5G collaborative SDN (Software Defined Network) load balancing scheduling method, which comprises the following steps of: obtaining delay units from an initiating node to connectable receiving nodes in a network through command execution, and arranging the connectable receiving nodes from small to large according to the delay units; performing association degree analysis on all connectable receiving nodes by adopting a hierarchical information interaction algorithm, and according to the magnitude of the association degrees, forming a receiving cluster with the association degree higher than a weight preset threshold value in all connectable receiving nodes, and forming a suboptimal receiving cluster with the association degree lower than the weight preset threshold value in all connectable receiving nodes; and the initiating node forwards traffic to the receiving cluster and the suboptimal receiving cluster according to a preset proportion. Different from a traditional network function based on hardware equipment, the method provided by the invention can more reasonably perform flow distribution, thereby relieving network congestion caused by concurrent pressure, and solving the problem of flow guide load concurrency of the network function.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of 5G collaborative SDN networks, and in particular to a method and storage medium for load balancing scheduling of a 5G collaborative SDN network. Background Art

[0002] With the rapid development of computer technology, information networks have become an important guarantee for social development. In the era of cloud-network integration, flexible and agile network service capabilities directly affect the competitiveness of operators. The closed and non-programmable nature of traditional hardware-based network functions makes management very complicated. Deploying network functions in network terminals based on SDN switching devices, the control and data separation architecture makes the management of network terminal functions more convenient, and the efficient control of network functions can be achieved by expanding the OpenFlow protocol. Among them, SDN is an emerging network architecture that separates network control functions from forwarding functions and realizes programmable control. This architecture will transfer the control layer from network devices to external computing devices, making the underlying infrastructure transparent and abstract to applications and network services, and the network can be regarded as a logical or virtual entity. However, the problem of traffic-guided load concurrency in network functions has always existed.

[0003] Therefore, a 5G collaborative SDN network load balancing scheduling method is urgently needed to solve the traffic guidance load concurrency problem of network functions. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention provides a method and storage medium for 5G collaborative SDN network load balancing scheduling to solve the problem of traffic guidance load concurrency of network functions.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] A method for 5G collaborative SDN network load balancing scheduling, characterized by comprising the following steps:

[0007] The delay unit from the initiating node to the connected receiving node in the network is obtained through command execution, and the connected receiving nodes are arranged from small to large according to the delay unit; the hierarchical information interaction algorithm is used to analyze the correlation of all connected receiving nodes, and according to the size of the correlation, all the connected receiving nodes with a correlation higher than the preset weight threshold are grouped into a receiving cluster, and those with a correlation lower than the preset weight threshold are grouped into a suboptimal receiving cluster; the initiating node forwards traffic to the receiving cluster and the suboptimal receiving cluster according to the preset ratio.

[0008] To optimize the above technical solutions, the specific measures taken also include:

[0009] Furthermore, the hierarchical information interaction algorithm is used to perform correlation analysis on all connected receiving nodes, specifically including:

[0010] The log service is used to obtain the association relationship between the virtual machines of the initiating node and the connectable receiving nodes, and between the physical machines and virtual machines of the initiating node and the connectable receiving nodes, as the first layer;

[0011] Based on the first-layer association relationship, the transmission data between the initiating node and the connected receiving node is tracked and obtained as the second layer, thereby forming a receiving node-classification keyword interaction matrix R, and forming a corresponding weight matrix for the corresponding keyword tags. The receiving node-classification keyword interaction matrix R is multiplied by the weight matrix to obtain the association matrix between the connected receiving node and the initiating node.

[0012] Furthermore, according to the magnitude of the correlation, among all the connectable receiving nodes, those with correlation higher than a preset threshold form a receiving cluster, and those with correlation lower than the preset threshold form a suboptimal receiving cluster, specifically including:

[0013] According to the obtained association matrix, the numerical value corresponding to each classification keyword in the association matrix is ​​compared with the corresponding weight preset threshold value respectively. The association matrix in which the numerical value corresponding to each classification keyword in the association matrix is ​​higher than the weight preset threshold value, and the corresponding connected receiving nodes are divided into receiving clusters. The association matrix in which the numerical value corresponding to each classification keyword in the association matrix is ​​lower than the weight preset threshold value, and the corresponding connected receiving nodes are divided into suboptimal receiving clusters.

[0014] Furthermore, based on the preset ratios of traffic forwarding by the initiating node to the receiving cluster and the suboptimal receiving cluster respectively, the percentage of being assigned to the receiving cluster or the optimal receiving cluster is determined, and the preset weight threshold of each classification keyword is determined through the percentage of being assigned to the receiving cluster or the optimal receiving cluster.

[0015] Furthermore, the relative percentage ratio between the receiving cluster and the optimal receiving cluster can be dynamically adjusted according to the delay, so that the cluster where the corresponding receiving node is located is changed in sequence according to the degree of compliance between the correlation matrix corresponding to the receiving node and the preset threshold of the weight.

[0016] Furthermore, after obtaining the correlation matrix between the connected receiving nodes and the initiating nodes, the matrix similarity integration optimization is performed through the matrix decomposition method to confirm the optimal allocation value of the classification keyword in the receiving node-classification keyword interaction matrix R.

[0017] Furthermore, the similarity integration optimization of the matrix is ​​performed by matrix decomposition method to confirm the optimal allocation value of the classification keyword in the receiving node-classification keyword interaction matrix R, specifically including:

[0018] Assume that there are m receiving nodes and n classification keywords, that is, the receiving node-classification keyword interaction matrix R is mxn. Using the idea of ​​matrix decomposition, R is decomposed into two low-rank matrices U and V, and the product of the two is used to approximate R. The formula is as follows:

[0019] R≈U T V

[0020] in, and k<<min(m,n); where the column vector U of the matrix U and V i and V j , using the idea of ​​linear regression, the objective function is as follows:

[0021]

[0022] where K is the (i,j) pair with known scores, and R ij is the column vector of matrix R, λ is the regularization factor, represents the matrix norm;

[0023] According to the obtained minimum U and V in L(U,V), R' is obtained again, and the optimal distribution value of the classification keyword is obtained according to R'.

[0024] Furthermore, when the initiating node forwards traffic to the receiving cluster and the suboptimal receiving cluster according to a preset ratio, the initiating node traffic is preferentially forwarded to the adjacent receiving node with small delay in the receiving cluster or the suboptimal receiving cluster.

[0025] Furthermore, Istio "Virtual Service" receivers are deployed on connected receiving nodes.

[0026] Furthermore, a computer-readable storage medium stores a computer program, characterized in that: the computer program enables a computer to execute a method for 5G collaborative SDN network load balancing scheduling as described above.

[0027] The beneficial effects of the present invention are:

[0028] The present invention is different from the traditional network function based on hardware devices, whose closedness and non-programmability lead to very complex management and high cost of load balancing hardware expansion. Innovatively, the receiving nodes are arranged from small to large through the delay unit from the initiating node to each connected node in the network. Secondly, a hierarchical information interaction algorithm is used to analyze the correlation of all connected receiving nodes. According to the size of the correlation, all connected receiving nodes are divided into receiving clusters and suboptimal receiving clusters, and the clusters can be dynamically adjusted according to the delay between clusters; then, when the initiating node traffic is forwarded to the receiving node, each forwarding node can be regarded as a receiver and a "virtual service". In this way, the traffic can be diverted more reasonably, thereby alleviating the network congestion caused by concurrent pressure and solving the traffic guidance load concurrency problem of network functions. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 A flow chart of a method for 5G collaborative SDN network load balancing scheduling proposed in the present invention;

[0030] Figure 2 A flow chart of a layered information interaction algorithm for a 5G collaborative SDN network load balancing scheduling method proposed in the present invention. DETAILED DESCRIPTION

[0031] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0032] A method for 5G collaborative SDN network load balancing scheduling according to an embodiment of the present invention includes the following steps:

[0033] Firstly, the delay unit from the initiating node to the connected receiving node in the network is obtained through command execution: milliseconds, and the connected receiving nodes are arranged from small to large according to the delay unit, i.e. milliseconds; secondly, a hierarchical information interaction algorithm is used to analyze the correlation of all connected receiving nodes, and according to the size of the correlation, all the connected receiving nodes with a correlation higher than the preset weight threshold are grouped into a receiving cluster, and those with a correlation lower than the preset weight threshold are grouped into a suboptimal receiving cluster; the initiating node forwards traffic to the receiving cluster and the suboptimal receiving cluster according to the preset ratio.

[0034] In another embodiment, the initiating node preferentially forwards traffic to each receiving node in the receiving cluster. If the load balance of each receiving node in the receiving cluster is higher than 80% on average, data is distributed to each receiving node in the suboptimal receiving cluster to further alleviate network congestion caused by concurrent pressure.

[0035] The above-mentioned hierarchical information interaction algorithm is used to perform correlation analysis on all connected receiving nodes, specifically including:

[0036] In order to prevent the large and small model operations in some large systems that provide global services, the amount of daily log data may reach tens of terabytes (TBs). Therefore, the received data is divided into two layers for analysis.

[0037] The log service is used to obtain the association relationship between the virtual machines of the initiating node and the connected receiving nodes, as well as between the physical machines and virtual machines of the initiating node and the connected receiving nodes (such as databases, middleware, server basic monitoring indicators, etc.), as the first layer;

[0038] On the basis of the first-layer association relationship, in order to discover the receiving nodes of other potentially hidden association relationships, the transmission data between the initiating node and the connected receiving node is tracked and obtained. In addition to the connectivity relationships between virtual machines and between physical machines and virtual machines involved in the first layer, the device weights are divided through different devices through business weight data as the second layer, thereby forming a receiving node-classification keyword interaction matrix R, and forming a corresponding weight matrix for the corresponding keyword tags. The receiving node-classification keyword interaction matrix R is multiplied by the weight matrix to obtain the association matrix between the connected receiving node and the initiating node.

[0039] Among them, according to the magnitude of the correlation, all the connected receiving nodes with correlation higher than a preset threshold form a receiving cluster, and those with correlation lower than the preset threshold form a suboptimal receiving cluster, specifically including:

[0040] According to the obtained association matrix, the numerical value corresponding to each classification keyword in the association matrix is ​​compared with the corresponding weight preset threshold value respectively. The association matrix in which the numerical value corresponding to each classification keyword in the association matrix is ​​higher than the weight preset threshold value, and the corresponding connected receiving nodes are divided into receiving clusters. The association matrix in which the numerical value corresponding to each classification keyword in the association matrix is ​​lower than the weight preset threshold value, and the corresponding connected receiving nodes are divided into suboptimal receiving clusters.

[0041] Among them, according to the preset ratio of traffic forwarding by the initiating node to the receiving cluster and the suboptimal receiving cluster respectively, the percentage of being assigned to the receiving cluster or the optimal receiving cluster is determined, and the preset threshold of the weight of each classification keyword is determined through the percentage of being assigned to the receiving cluster or the optimal receiving cluster.

[0042] Among them, the relative percentage ratio between the receiving cluster and the optimal receiving cluster can be dynamically adjusted according to the delay, so that the cluster where the corresponding receiving node is located is changed in sequence according to the degree of compliance between the correlation matrix corresponding to the receiving node and the preset threshold of the weight.

[0043] Among them, after obtaining the correlation matrix between the connected receiving nodes and the initiating nodes, the matrix similarity integration optimization is performed through the matrix decomposition method to confirm the optimal distribution value of the classification keyword in the receiving node-classification keyword interaction matrix R. The matrix decomposition model can reduce the matrix error and thus improve the accuracy of the model.

[0044] The above-mentioned matrix similarity integration optimization by matrix decomposition method is used to determine the optimal allocation value of the classification keyword in the receiving node-classification keyword interaction matrix R, which specifically includes:

[0045] Assume that there are m receiving nodes for Istio data transmission and n classification keywords, that is, the receiving node-classification keyword interaction matrix R is mxn. The idea of ​​matrix decomposition is used to decompose R into two low-rank matrices U and V. The product of the two is used to approximate R. The formula is as follows:

[0046] R≈U T V

[0047] in, and k<<min(m,n); where the column vector U of the matrix U and V i and V j Represent the slice and classification threshold latent vectors respectively. Using the idea of ​​linear regression to minimize the square error, the objective function is as follows:

[0048]

[0049] where K is the (i,j) pair with known scores, and R ij is the column vector of matrix R. To prevent overfitting, a regularization term is added. λ is the regularization factor. represents the matrix norm;

[0050] According to the obtained minimum U and V in L(U,V), R' is obtained again, and the optimal distribution value of the classification keyword is obtained according to R'.

[0051] In a specific embodiment, when the initiating node forwards traffic to the receiving cluster and the suboptimal receiving cluster according to a preset ratio, the initiating node traffic is preferentially forwarded to the adjacent receiving node with small delay in the receiving cluster or the suboptimal receiving cluster.

[0052] In a specific embodiment, an Istio "Virtual Service" receiver is deployed on a connectable receiving node.

[0053] Istio: An open platform for connecting, managing, and protecting microservices. Istio provides an easy way to create a network of deployed services with load balancing, service-to-service authentication, monitoring, and more, without making any changes to the service code. You can add Istio support services by deploying a special sidecar proxy (sidecarproxy) throughout the environment, which intercepts all network communications between microservices and is configured and managed using Istio's control plane capabilities.

[0054] In this solution, the SIMPLE architecture control plane mainly consists of three parts: resource manager, dynamic processor and rule generator. The "Virtual Service" control end of Istio is deployed on the resource manager, so that Istio is used to replace the problem of hard load expansion limitations and high costs. Among them, the resource manager is responsible for managing, updating and issuing network topology and user policies, and load balancing according to the performance constraints of network functions and switches. The dynamic processor automatically infers the connection mapping in and out of the network function and modifies the packet header. To achieve this function, the dynamic processor receives data packets from each switch directly connected to the network function, and uses a light load similarity detection algorithm to determine the connection mapping in and out of the network function, and finally passes it to the rule generator. The rule generator configures the data plane according to the network topology and user policies issued by the resource manager and the connection mapping relationship issued by the dynamic processor, and guides the network traffic to pass through the required network functions in the correct order. The resource manager and dynamic processor can be regarded as controller applications, and the rule generator can be regarded as an extension of the network operating system. SIMPLE can be regarded as an active controller for ordinary network functions. Since the packet header content is not modified, the establishment time of each flow is reduced.

[0055] In order to adapt to the services represented by different platforms, Istio abstracts the concept of service into a call set VirtualService. The biggest advantage of this approach is that it integrates the service concept with the business operation. For example, a service that provides user portraits has three containers named UserPortrait-vl. Now it needs to be tested online. In order not to affect the normal online access, only one of them can be selected for testing. In the traditional microservice platform, one of them can be taken down and published separately as another service: but in Kubernetes, in this case, only one Service needs to be established and the Pod is filtered through the sclector to form two groups of applications on the virtual machine. Only simple configuration is required, and the business side does not need to make any changes to implement service changes.

[0056] Istio's "Virtual Service" draws on its ideas. It does not rely heavily on the deployment structure, and does not need to care about the specific isolation strategy of the deployment structure. Instead, it directly divides the terminal in the order. It also allows overlap and can be configured arbitrarily, making it very flexible. The service corresponding to the virtual service is specified by the Host; in the Kubernetes environment, the Host is an A record in CoreDNS.

[0057] Due to the design of virtual services, Istio can adapt to different platforms, that is, whether it is Kubernetes or a virtual machine plane based on Consul, Istio only needs to map the Host in VirtualService to the corresponding resources in the platform. In the Consul environment, how are the above two concepts mapped? Then, the platform service lifecycle management is adapted, that is, Istio needs to receive notifications to correctly maintain the status of the entire service grid. For example, after a service is newly established, Istio must listen to this event, inject Sidear into it, and perform some initialization configuration. In order to solve this problem, Istio engineers implement different listeners for different platforms.

[0058] (1) SIMPLE data plane design

[0059] SIMPLE data plane design has two key features: first, the switch does not rely on the traditional five-tuple for matching and forwarding; second, it is necessary to ensure that the rules can be compatible with the limited TCAM because the distribution of network functions runs through the entire large network. To solve these problems, Zaffa et al. proposed a data plane design that combines tunnels and labels. Although this combination of tunnels and labels is common in networks, especially SDN, it is very novel in the scenario of executing network function policies.

[0060] (2) Resource Constraint Modeling The biggest challenge in resource management is dealing with network functions and SDN switch flow table capacity constraints, which is an NP-hard resource valuation problem. Figure 2 As shown in the figure, SIMPLE divides the optimization problem into two parts: the first step is to process the switch constraints offline, which is only required when the network topology, switch and network configuration, and network policy change; the second step is to process the constraints at several locations.

[0061] (3) Flow matching based on similarity test

[0062] In SDN-based network function management, when issuing forwarding rules, the network function may dynamically modify the incoming traffic, especially when the network function modifies the flow packet header, the forwarding rules of the downstream switch must recognize the new packet header. Therefore, it is necessary to be able to identify whether the data packets initially flowing into the network function and the data packets flowing out of the network function belong to the same flow. Here, the SIMPLE dynamic processor uses a similarity algorithm to solve this problem, which is divided into 3 steps.

[0063] 1 Data packet collection: When a new flow enters the network function, the switch sends the first P packets of the new flow to the controller. At the same time, within a time window W, the first P packets are collected from all flows of the network function.

[0064] 2 Calculating Load Similarity: Network functions may change the load, so the load cannot be directly compared. A similarity score is proposed here to calculate the overlap between each flow.

[0065] 3 Identify similar flows: By calculating the similarity score between the data packets leaving the network function and the newly incoming data packets, the data packets with the highest scores are selected as the same flow. If there are multiple flows with the highest scores, these flows can be considered to be related to the incoming flow.

[0066] In a specific embodiment, this solution belongs to the field of 5G collaborative SDN network. The closedness and non-programmability of traditional network functions based on hardware devices lead to very complex management and high cost of load balancing hardware expansion. Innovatively, first, the delay unit from the initiating node to the connected receiving node in the network is obtained through command execution: milliseconds, and the connected receiving nodes are arranged from small to large according to milliseconds, and the "Virtual Service" client of Istio is deployed on the connected receiving nodes. Secondly, the maximum information interaction algorithm is used to analyze the correlation of all connected receiving nodes with business data as the correlation indicator, and the nodes with a correlation higher than 50% are formed into a receiving cluster, and the nodes with a correlation lower than 50% are formed into a suboptimal receiving cluster. The clusters can dynamically adjust the clusters where the nodes are located according to the delay. Then, the initiating node traffic is preferentially forwarded to the adjacent receiving nodes with small delay in the receiving cluster, and each forwarded receiving node can be regarded as a receiver and a "virtual service". If the average load balancing of each node in the receiving cluster is higher than 80%, data distribution is performed to the suboptimal receiving cluster to alleviate the network congestion caused by concurrent pressure. Finally, when the average load balancing of the receiving cluster or the suboptimal receiving cluster is higher than 80%, in order to maximize the efficiency of load balancing and reduce the cost of load balancing resources, the matrix decomposition method is innovatively used to distribute the traffic after similarity integration of the traffic within the cluster according to the traffic protocol (http, tcp, udp, https, dns, etc.), thus providing a distribution processing method for fast data transmission when the load is high.

[0067] In another embodiment, the present invention proposes a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute a method for 5G collaborative SDN network load balancing scheduling as described above.

[0068] In another embodiment, the present invention proposes 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, a method for 5G collaborative SDN network load balancing scheduling as described above is implemented.

[0069] In the embodiments disclosed in the present application, the computer storage medium can be a tangible medium that can contain or store a program for use by or in conjunction with an instruction execution system, device or equipment. The computer storage medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the above. More specific examples of computer storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above.

[0070] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0071] The above are only preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should be regarded as the protection scope of the present invention.

Claims

1. A method for 5G collaborative SDN network load balancing scheduling, characterized in that: The steps include: Obtain the delay unit from the initiating node to the connected receiving node in the network through command execution, and arrange the connected receiving nodes from small to large according to the delay unit; use the hierarchical information interaction algorithm to analyze the correlation of all connected receiving nodes, and according to the size of the correlation, form a receiving cluster with a correlation higher than the preset weight threshold among all the connected receiving nodes, and form a suboptimal receiving cluster with a correlation lower than the preset weight threshold; The initiating node forwards traffic to the receiving cluster and the suboptimal receiving cluster according to the preset ratio.

2. A method for 5G collaborative SDN network load balancing scheduling according to claim 1, characterized in that: The hierarchical information interaction algorithm is used to perform correlation analysis on all connected receiving nodes, specifically including: The log service is used to obtain the association relationship between the virtual machines of the initiating node and the connectable receiving nodes, and between the physical machines and virtual machines of the initiating node and the connectable receiving nodes, as the first layer; Based on the first-layer association relationship, the transmission data between the initiating node and the connected receiving node is tracked and obtained as the second layer, thereby forming a receiving node-classification keyword interaction matrix R, and forming a corresponding weight matrix for the corresponding keyword tags. The receiving node-classification keyword interaction matrix R is multiplied by the weight matrix to obtain the association matrix between the connected receiving node and the initiating node.

3. A method for 5G collaborative SDN network load balancing scheduling according to claim 2, characterized in that: According to the magnitude of the correlation, among all the connectable receiving nodes, those with correlation higher than a preset threshold form a receiving cluster, and those with correlation lower than the preset threshold form a suboptimal receiving cluster, specifically including: According to the obtained association matrix, the numerical value corresponding to each classification keyword in the association matrix is ​​compared with the corresponding weight preset threshold value respectively. The association matrix in which the numerical value corresponding to each classification keyword in the association matrix is ​​higher than the weight preset threshold value, and the corresponding connected receiving nodes are divided into receiving clusters. The association matrix in which the numerical value corresponding to each classification keyword in the association matrix is ​​lower than the weight preset threshold value, and the corresponding connected receiving nodes are divided into suboptimal receiving clusters.

4. A method for 5G collaborative SDN network load balancing scheduling according to claim 3, characterized in that: According to the preset ratio of traffic forwarding by the initiating node to the receiving cluster and the suboptimal receiving cluster respectively, the percentage of being assigned to the receiving cluster or the optimal receiving cluster is determined, and the preset threshold of the weight of each classification keyword is determined through the percentage of being assigned to the receiving cluster or the optimal receiving cluster.

5. A method for 5G collaborative SDN network load balancing scheduling according to claim 4, characterized in that: The relative percentage ratio between the receiving cluster and the optimal receiving cluster can be dynamically adjusted according to the delay, so that the cluster where the corresponding receiving node is located is changed in sequence according to the degree of compliance between the correlation matrix corresponding to the receiving node and the preset threshold of the weight.

6. A method for 5G collaborative SDN network load balancing scheduling according to claim 2, characterized in that: After obtaining the association matrix between the connected receiving nodes and the initiating nodes, the matrix similarity integration optimization is performed through the matrix decomposition method to confirm the optimal allocation value of the classification keyword in the receiving node-classification keyword interaction matrix R.

7. A method for 5G collaborative SDN network load balancing scheduling according to claim 6, characterized in that: The similarity integration optimization of the matrix is ​​performed by the matrix decomposition method to confirm the optimal allocation value of the classification keyword in the receiving node-classification keyword interaction matrix R, specifically including: Assume that there are m receiving nodes and n classification keywords, that is, the receiving node-classification keyword interaction matrix R is mxn. Using the idea of ​​matrix decomposition, R is decomposed into two low-rank matrices U and V, and the product of the two is used to approximate R. The formula is as follows: R≈U T In in, and k < min(m, n); where the column vectors U of the matrices U and V i and V j , using the idea of ​​linear regression, the objective function is as follows: where K is the (i, j) pair with known scores, and R ij is the column vector of matrix R, λ is the regularization factor, represents the matrix norm; According to the obtained minimum U and V in L(UV), R' is obtained again, and the optimal distribution value of the classification keyword is obtained according to R'.

8. A method for 5G collaborative SDN network load balancing scheduling according to claim 1, characterized in that: When the initiating node forwards traffic to the receiving cluster and the suboptimal receiving cluster according to a preset ratio, the initiating node traffic is preferentially forwarded to the adjacent receiving node with small delay in the receiving cluster or the suboptimal receiving cluster.

9. A method for 5G collaborative SDN network load balancing scheduling according to claim 1, characterized in that: Deploy an Istio "VirtualService" receiver on a connected receiving node.

10. A computer-readable storage medium storing a computer program, characterized in that: The computer program enables the computer to execute a method for 5G collaborative SDN network load balancing scheduling as described in any one of claims 1-9.