A method for detecting handover latency in 5GC industrial private networks

Abstract

This invention discloses a 5GC industrial private network handover latency detection method. It utilizes a dual-probe measurement mechanism by deploying a first and second probe on the N2 and N3 interfaces respectively, allowing for more precise timestamp settings on the control plane and data plane. The method also incorporates a 5G synchronization clock to calibrate the clock deviation between the two probes. Furthermore, it calculates multi-level latency by acquiring the differences in start and end timestamps of wireless transmission latency, 5GC forwarding latency, and the processing start and end timestamps of each stage of the protocol stack composed of IP, UDP, and GTP-U. This allows for accurate calculation of the network quality of the operating device, improving detection accuracy. By compiling keepalived to support bidirectional forwarding detection and coordinating with VRRP primary / backup node handover, it achieves millimeter-level fault detection, improving fault detection efficiency, shortening the network fault detection cycle, increasing response speed, and enabling millisecond-level handover, thus enhancing handover efficiency.

Description

Technical Field

[0001] This invention relates to the field of network monitoring technology, and in particular to a method for detecting handover latency in a 5GC industrial private network. Background Technology

[0002] With the continuous advancement of the Industrial Internet of Things (IIoT), 5G technology is being applied more and more widely in the industrial field. As a new network architecture, 5GC industrial private networks can provide low-latency and highly reliable communication services for many industrial scenarios. Deployed independently in various industrial scenarios, the network and accessible users are confined to the industrial park, aiming to provide users with better services through high system reliability, security, and low latency.

[0003] However, in industrial environments, network failures and subsequent handovers are unavoidable due to environmental interference, equipment aging, and human factors. Traditional network failure detection methods, however, have long cycles, resulting in low handover efficiency. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method for detecting handover delay in 5GC industrial private networks, so as to shorten the network fault detection cycle and improve handover efficiency.

[0005] In a first aspect, the present invention provides a 5GC industrial private network handover delay detection method, comprising the following steps: calibrating the clock deviation between a first probe deployed on the N2 interface and a second probe deployed on the N3 interface using a 5G synchronous clock, and calculating the signaling processing delay and end-to-end path delay; estimating the wireless transmission delay using a 5G base station based on the reference signal received power and reference signal received quality in the air interface measurement report combined with a path loss model; calculating the 5GC forwarding delay using a UPF based on the enqueue and dequeue times in the GTP-U tunnel forwarding logs stored in the UPF's primary and backup Redis; and obtaining the difference in start and end timestamps of each stage of the protocol stack composed of IP, UDP, and GTP-U using an eBPF probe deployed in the SMF or UPF kernel. The three data points are summed to obtain the end-to-end latency and stored. A Gaussian model is constructed based on the latency data stored within 48 hours. Anomaly detection is performed using the 3σ principle based on the constructed Gaussian model. When an anomaly occurs, the switching between the primary UPF and the backup UPF is triggered. Keepalived, which supports BFD, is compiled and configured with BFD instances, binding the VRRP instance to BFD. After the BFD session is established, millisecond-level probe packets are continuously sent at preset time intervals. When a link fault or node fault is detected, a fault event is sent to Keepalived, triggering VRRP state switching. After receiving the fault event sent by BFD, Keepalived notifies the 5GC's control plane network element CPF and data plane network element UPF of the switching instruction.

[0006] The beneficial technical effects of this invention are as follows: The 5GC industrial private network handover latency detection method of this invention utilizes a dual-probe measurement mechanism by deploying a first probe and a second probe on the N2 and N3 interfaces respectively, enabling more precise timestamp settings for the control plane and data plane. Combined with 5G synchronization clock calibration of the clock deviation between the dual probes, and by calculating multi-level latency using the differences in start and end timestamps of the processing at each stage of the protocol stack composed of IP, UDP, and GTP-U, the network quality of the operating device can be accurately calculated, improving detection accuracy. Furthermore, by compiling keepalived to support bidirectional forwarding detection and coordinating with VRRP primary / backup node handover, millimeter-level fault detection can be achieved, improving fault detection efficiency, shortening the network fault detection cycle, increasing response speed, and achieving millisecond-level handover, thus improving handover efficiency. Attached Figure Description

[0007] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0008] Figure 1 A flowchart illustrating the 5GC industrial private network handover latency detection method provided in this embodiment of the invention;

[0009] Figure 2 This is a schematic diagram of a sub-process of the 5GC industrial private network handover delay detection method provided in an embodiment of the present invention. Detailed Implementation

[0010] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0011] Please see Figure 1 , Figure 1 This is a flowchart illustrating the 5GC industrial private network handover latency detection method provided in this embodiment of the invention. The 5GC industrial private network handover latency detection method includes the following steps S10-S60:

[0012] S10. The clock deviation between the first probe deployed on the N2 interface and the second probe deployed on the N3 interface is calibrated using a 5G synchronous clock.

[0013] The N2 interface is the control plane interface between the 5G RAN (Radio Access Network) and the 5GC, used to transmit control plane messages, including processes related to connection establishment, mobility management, and session management. The 5GC is the 5G core network. The N3 interface is the user plane interface between the 5G RAN and the UPF (User Plane Function), used to transmit user plane data. The UPF is the unit in the 5GC responsible for processing user data packets. By deploying first and second probes on the N2 and N3 interfaces respectively, a dual-probe measurement mechanism is used to refine the timestamp settings for the control plane and data plane. Combined with the 5G synchronization clock to calibrate the clock deviation between the two probes, the accuracy of network quality detection is improved.

[0014] Combination Figure 2 Preferably, the step S10 of calibrating the clock deviation between the first probe deployed on the N2 interface and the second probe deployed on the N3 interface using a 5G synchronization clock includes:

[0015] S11: Monitor the handover preparation time and resource allocation time by deploying the first probe on the N2 interface;

[0016] S12: Monitor the data forwarding time between the source UPF and the target UPF using a second probe deployed on the N3 interface;

[0017] S13: Calibrate the clock deviation between the first and second probes using a 5G synchronous clock.

[0018] Preferably, in some embodiments, step S11, namely, monitoring the handover preparation time and resource allocation time through the first probe deployed on the N2 interface, specifically includes:

[0019] By embedding a 10ns precision timestamp in the signaling through the first probe deployed on the N2 interface, the timestamps of the handover preparation signaling event and the resource allocation signaling event are collected respectively.

[0020] The signaling processing delay can be calculated based on the difference between the timestamp of the resource allocation signaling event and the timestamp of the handover preparation signaling event.

[0021] Preferably, in some embodiments, step S12, namely, monitoring the data forwarding time between the source UPF and the target UPF using the second probe deployed on the N3 interface, specifically includes:

[0022] By inserting a timestamped OAM detection packet into the data stream sent from the source UPF to the target UPF using a second probe deployed on the N3 interface, the timestamps of the source UPF completing data stream encapsulation and sending it to the target UPF, and the timestamp of the target UPF successfully receiving the data stream are collected.

[0023] OAM (Operation Administration and Maintenance) refers to the division of network management work into three main categories based on the actual needs of operator network operations: Operation, Administration, and Maintenance. Operation primarily involves daily network and service analysis, prediction, planning, and configuration. Maintenance mainly involves routine operational activities such as network and service testing and fault management. OAM supports discovering link monitoring, remote fault indication, and remote switchback testing for scalability. It can calculate end-to-end path delay based on the difference between the timestamp of successful data stream reception at the target UPF and the timestamp of data stream encapsulation and transmission from the source UPF to the target UPF.

[0024] S20: The wireless transmission delay is estimated by the 5G base station based on the reference signal received power and reference signal received quality in the air interface measurement report, combined with the path loss model. The 5GC forwarding delay is calculated by the UPF based on the enqueue time and dequeue time in the GTP-U tunnel forwarding log stored in the main Redis and backup Redis of the UPF. The difference between the start and end timestamps of each stage of the protocol stack consisting of IP, UDP and GTP-U is obtained by the eBPF probe deployed in the kernel of SMF or UPF. The end-to-end delay is obtained by adding the wireless transmission delay, the 5GC forwarding delay and the difference between the start and end timestamps of each stage of the protocol stack, and then saved.

[0025] In this context, a 5G base station (gNB) refers to a next-generation base station in a 5G network. It is a 5GC device responsible for signal coverage and user connectivity. Air interface measurement reports are obtained by the UE (User Equipment) periodically reporting measurement results, triggered by the measurement intervals configured in the 5G base station (gNB). The SMF (Session Management function) is responsible for tunnel maintenance, IP address allocation and management, UP function selection, policy enforcement, and QoS (Quality of Service) control, billing data collection, and roaming. Redis is an open-source (BSD licensed) in-memory data structure store used as a database, cache, and message broker. 5GC forwarding latency is the difference between dequeue time and enqueue time. The deployed eBPF probes can acquire and calculate the start and end timestamps of each stage of the protocol stack, which consists of IP (Internet Protocol), UDP (User Datagram Protocol), and GTP-U (GPRS Tunneling Protocol-User Plane). These timestamp differences include the difference between the timestamps of data packets entering and leaving the IP layer, the difference between the start and end timestamps of UDP packet processing in the UDP layer, and the difference between the time taken for GTP tunnel encapsulation and decapsulation. End-to-end latency can be stored in the UPF's service processing module. End-to-end latency is the same as end-to-end path latency.

[0026] S30: The UPF's service processing module constructs a Gaussian model based on latency data stored over 48 hours. This Gaussian model is then used to perform anomaly detection using the 3σ principle. When an anomaly occurs, a switchover between the primary and backup UPF is triggered. The UPF's service processing module also performs anomaly detection based on stored latency data to detect service congestion. If an anomaly occurs, service congestion is detected, triggering a switchover between the primary and backup UPF. The backup UPF is the standby UPF. Latency data includes the end-to-end latency stored in the UPF's service processing module.

[0027] Preferably, in some embodiments, the step S30, in which the service processing module of the UPF constructs a Gaussian model based on the latency data stored within forty-eight hours, includes:

[0028] Acquire latency data within 48 hours as a dataset;

[0029] Remove noise from the dataset to obtain a denoised dataset. Then, interpolate and fill in the missing values ​​in the denoised dataset to obtain a preprocessed dataset.

[0030] The latency data in the preprocessed dataset is decomposed according to the interface to obtain the interface latency corresponding to each interface.

[0031] Construct a Gaussian model based on the interface latency corresponding to each interface.

[0032] The step of decomposing the latency data in the preprocessed dataset according to the interface, and obtaining the interface latency corresponding to each interface, includes air interface latency, N3 interface latency, N6 interface latency and N9 interface latency.

[0033] Preferably, the step of constructing a Gaussian model based on the interface latency corresponding to each interface includes:

[0034] The Gaussian components are calculated using formula (1):

[0035] (1);

[0036] In the formula, Represents a time-delay dataset. The probability density function representing the dataset. Indicates the first The weight coefficients of each Gaussian component, where the sum of the weights of all Gaussian components is 1. This represents the probability density function of a univariate Gaussian distribution. Indicates the average delay. Indicates the time delay variance. Indicates the sequence number of the Gaussian component;

[0037] The optimal parameters in formula (1) are calculated iteratively using the expectation-maximization algorithm.

[0038] Among them, the weighted superposition of three independent Gaussian components calculated by formula (1) can accurately describe the probabilistic characteristics of the time delay data, so as to facilitate subsequent anomaly detection based on the 3σ principle. When, corresponding to the normal delay state, when At that time, corresponding to a slightly abnormal delay state, when At that time, it corresponds to a severe abnormal delay state.

[0039] Specifically, in some embodiments, the anomaly detection using the constructed Gaussian model with the 3σ principle includes:

[0040] The normal range of each Gaussian component is determined using the 3σ principle, and the time delay when an anomaly occurs is calculated using formula (2):

[0041] (2);

[0042] In the formula, This represents the end-to-end delay value for a single measurement. This represents the preset probability density threshold. Indicates the first The weighting coefficients of the Gaussian component, i.e., the weighting coefficients of the Gaussian component at the th ... The weight coefficients for this network state, where the sum of the weights of all Gaussian components is 1. Indicates the first Real-time end-to-end latency under various network conditions The probability density function of occurrence, Indicates the first Mean latency under various network conditions Indicates the first Delay variance under various network conditions This represents the index of the Gaussian component. The normal interval for each Gaussian component, based on the 3σ principle, is [...]. , Each Gaussian distribution This represents the latency distribution characteristics under a specific network state. It is calculated by weighted summation of the probability densities under three states. When the sum is less than a preset probability density threshold, it indicates the corresponding real-time end-to-end latency value. If the latency distribution does not conform to the normal network condition, it is determined to be an anomaly and a handover is triggered.

[0043] S40: Compile Keepalived that supports BFD (Bidirectional Forwarding Detection), configure a BFD instance for Keepalived, and bind a VRRP (Virtual Router Redundancy Protocol) instance to BFD.

[0044] S50: After establishing a BFD session, continuously send millisecond-level probe messages at preset time intervals. When a link failure or node failure is detected, immediately send a fault event to Keepalived to trigger VRRP state switching.

[0045] The preset time interval is 50ms. Link failure refers to port unreachability, and node failure refers to process being killed (forced termination). BFD is responsible for the rapid detection of link and node failures by periodically sending detection messages to monitor the connectivity of bidirectional forwarding paths. Keepalived is responsible for master / slave state switching and virtual IP (VIP) management, and triggers VRRP state machine changes based on BFD detection results.

[0046] Preferably, the triggering of the VRRP state switch in step 50 includes:

[0047] After receiving a failure event from BFD, Keepalived will switch the standby node by raising its VRRP priority to be higher than that of the master node. The standby node will then take over the master role and take over the virtual IP (VIP) and traffic forwarding.

[0048] The primary node automatically lowers its own VRRP priority by configuring a delay for preemption, while simultaneously raising the VRRP priority of the backup node, sending a VRRP announcement message, and then performing a primary / backup node switchover after the delay period.

[0049] S60: After receiving a fault event sent by BFD, Keepalived notifies the 5GC's control plane network element CPF (Control Plane Function) and data plane network element UPF to perform primary / backup node switching for signaling flow and primary / backup UPF switching for data flow.

[0050] When the 5GC control plane network element CPF performs a primary / standby switchover, it also updates the routing information of the 5GC's AMF (Access and Mobility Management Function) to ensure that the signaling flow of the 5GC control plane is switched to the node corresponding to the new primary role. When the 5GC data plane network element UPF performs a primary / standby switchover, it triggers GTP-U tunnel reconstruction. Combined with the session state pre-stored in Redis, it achieves a seamless switchover. Combined with the Keepalived VIP migration during VRRP state switching, it achieves a seamless data flow switchover. The 5GC industrial private network handover latency detection method utilizes a dual-probe measurement mechanism by deploying a first probe and a second probe on the N2 and N3 interfaces respectively. This allows for more precise timestamp settings on the control plane and data plane. Combined with 5G synchronization clock calibration to balance the clock deviation between the two probes, and by calculating multi-level latency using the differences in start and end timestamps of the wireless transmission latency, 5GC forwarding latency, and the processing start and end timestamps of each stage of the protocol stack composed of IP, UDP, and GTP-U, the method accurately calculates the network quality of the operating device, improving detection accuracy. Furthermore, by compiling keepalived to support bidirectional forwarding detection and coordinating with VRRP primary / backup node handover, the method achieves millimeter-level fault detection, improving fault detection efficiency, shortening the network fault detection cycle, increasing response speed, and enabling millisecond-level handover, thus enhancing handover efficiency.

[0051] Preferably, after step S60, that is, after Keepalived receives the fault event sent by BFD and notifies the 5GC control plane network element CPF and data plane network element UPF of the handover command, the 5GC industrial private network handover delay detection method further includes:

[0052] Configure a pre-set redundant path, configure one primary UPF and two backup UPFs in each service slice, classify the backup data according to the access frequency to obtain hot data, warm data and cold data, and store the hot data, warm data and cold data in the memory database, high-speed solid state drive (SSD) and hard disk drive (HDD) respectively.

[0053] This system configures one primary UPF and two backup UPFs in each service slice. The primary UPF handles daily data forwarding, while the backup UPFs are in standby mode. A dual-active data channel is established between the primary and backup UPFs via the 5G standard N9 interface, enabling real-time communication and status synchronization between the backup and primary UPFs. The primary UPF's GTP-U tunnel table stores the set of all active tunnel endpoints and is synchronized to the backup UPFs in real-time via the N9 interface, ensuring the backup UPFs always have access to the tunnel status of the primary UPF. Backup data is categorized into hot, warm, and cold data based on access frequency and stored on different storage devices to improve data retrieval efficiency. Hot data refers to backup data accessed frequently, such as user information and other time-sensitive data; warm data refers to backup data accessed at moderate frequencies, such as statistical data from UPF operation and other information with a certain timeliness; and cold data refers to backup data accessed infrequently, such as logs and tunnel information from UPF operation.

[0054] Specifically, storing hot data, warm data, and cold data in a memory database, a high-speed solid-state drive, and a hard disk drive, respectively, includes:

[0055] Hot data is stored in an in-memory database, and the LRU (Least Recently Used) algorithm is used to ensure that hot data is always kept in the in-memory database. Before system startup or peak business hours, the expected hot data is loaded into the in-memory database, and changes in the hot data are reflected in the in-memory database in a timely manner.

[0056] Frequently accessed warm data is stored in high-speed solid-state drives, while other warm data is stored in ordinary hard drives. An efficient index structure is built to speed up the query speed of warm data. Prefetching technology is used to predict the data that may be needed based on access patterns and load it into the cache in advance.

[0057] Cold data is compressed and stored on a mechanical hard drive to reduce storage space usage. Processing cold data requires batch processing to improve efficiency.

[0058] By combining LRU algorithm and prefetching technology, backup accuracy and recovery efficiency can be improved, and storage costs can be reduced.

[0059] Specifically, after storing the hot data, warm data, and cold data in a memory database, a high-speed solid-state drive, and a hard disk drive, respectively, the 5GC industrial private network switching latency detection method further includes:

[0060] When the 5GC forwarding latency of the primary UPF is greater than the preset low latency value and the duration reaches the preset detection period threshold, the primary / backup UPF switchover is triggered.

[0061] The 5GC forwarding latency, as defined in step S20, can be calculated by the UPF based on the enqueue and dequeue times in the GTP-U tunnel forwarding logs stored in the primary and backup Redis of the UPF. The 5GC forwarding latency is determined by whether it exceeds a preset low latency value and whether the duration of this exceedance reaches a preset detection period threshold. This determines whether the 5GC forwarding latency exceeding the preset low latency value is a non-instantaneous fluctuation, thus indicating a potential UPF fault. If a UPF fault exists, a primary / backup UPF switchover is required. The preset low latency value can be 5ms to meet the low latency requirements of industrial private networks. The preset detection period threshold can be 3 periods, meaning the duration of the forwarding latency exceeding the preset low latency value must reach 3 detection periods to avoid false triggering of the switchover due to instantaneous fluctuations. Therefore, when the primary UPF's forwarding latency exceeds the preset low latency value and the duration reaches the preset detection period threshold, a UPF fault is detected, triggering a primary / backup UPF switchover. Specifically:

[0062] When the 5GC forwarding latency of the primary UPF is greater than 5ms and lasts for 3 consecutive detection cycles, the primary / backup UPF switchover is triggered.

[0063] Preferably, triggering the primary / backup UPF switch includes:

[0064] Determine the optimal backup UPF;

[0065] Select the optimal backup UPF and switch to the primary UPF.

[0066] Specifically, determining the optimal backup UPF involves:

[0067] The path from the primary UPF to the backup UPF is calculated using formula (3), and the backup UPF with the smallest path is selected as the optimal backup UPF.

[0068] (3);

[0069] In the formula, This indicates the path from the primary UPF to the backup UPF. Indicates the main UPF up to the th Round-trip latency of a backup UPF Indicates the first The current load of each standby UPF, including CPU utilization or bandwidth utilization. This represents the load sensitivity coefficient, which is used to adjust the weight of load rate in decision-making. The load sensitivity coefficient ranges from 0.7 to 1.0, making the system more sensitive to load changes and ensuring low latency requirements.

[0070] When the primary UPF fails, an emergency switchover strategy is adopted to restore services, and the backup UPF immediately takes over the services, reducing terminal time. When the primary UPF is running normally, a regular synchronization strategy is adopted, combining "pull mode" and "push mode". The backup UPF periodically requests change logs from the primary UPF, and the primary UPF promptly sends change logs to the backup UPF when its state changes, ensuring that the primary and backup UPF states are synchronized.

[0071] During primary / standby UPF switching, the standby UPF is selected based on the optimal path algorithm, combined with emergency switching strategies and regular synchronization strategies, ensuring a seamless experience for users and meeting the low latency requirements of industrial private networks. By designing pre-configured redundant paths, employing an optimized Redis storage mechanism, and using differential incremental synchronization technology, seamless UPF switching is guaranteed.

[0072] The primary UPF generates change logs in real time based on business needs and sends them to its in-memory database. The UPF's in-memory database uses a change log queue for incremental synchronization. This queue records session state changes in an append-only manner to ensure atomicity and sequentiality of operations. It employs a circular buffer structure to guarantee high write performance and automatically clean up expired log entries, saving storage space. Each log entry includes the change type, session identifier, changed fields, and timestamp information. Upon receiving logs, the UPF's in-memory database adds them to the change log queue. The change log queue periodically stores logs to disk to prevent data loss. The backup UPF subscribes to the change log queue from its in-memory database. When there are changes in the change log queue, the backup UPF is notified, enabling it to update its local database with the latest data promptly.

[0073] Redis Sentinel technology is used to optimize the master-slave design of Redis, monitoring the operational status of the Redis master-slave cluster. When the master node fails, it automatically performs fault detection and master-slave failover, ensuring uninterrupted cluster service and providing clients with the latest cluster topology information. To meet low latency requirements, the following Redis Sentinel configuration parameters can be set: down-after-milliseconds (master node failure detection time), failover-timeout (failover wait time, i.e., master-slave failover wait time), and parallel-syncs (the number of replicas that can be reconfigured to use the new master server simultaneously after failover) can be set to 2.

[0074] In summary, the 5GC industrial private network handover latency detection method of the present invention utilizes a dual-probe measurement mechanism by deploying a first probe and a second probe on the N2 and N3 interfaces respectively. This allows for more precise timestamp settings on the control plane and data plane. Combined with 5G synchronization clock calibration to balance the clock deviation between the two probes, and by calculating multi-level latency using the differences in start and end timestamps of the wireless transmission latency, 5GC forwarding latency, and the processing start and end timestamps of each stage of the protocol stack composed of IP, UDP, and GTP-U, the method accurately calculates the network quality of the operating device, improving detection accuracy. Furthermore, by compiling keepalived to support bidirectional forwarding detection and coordinating with VRRP primary / backup node handover, the method achieves millimeter-level fault detection, improving fault detection efficiency, shortening the network fault detection cycle, increasing response speed, and enabling millisecond-level handover, thus improving handover efficiency.

[0075] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for detecting 5GC industrial private network handover latency, characterized in that, The method comprises the following steps: Calibrating the clock offset between the first probe arranged at the N2 interface and the second probe arranged at the N3 interface through the 5G synchronization clock; Estimating the wireless transmission delay according to the reference signal receiving power and the reference signal receiving quality in the air interface measurement report combined with the path loss model by the 5G base station, calculating the 5GC forwarding delay according to the enqueuing time and dequeuing time in the GTP-U tunnel forwarding log stored in the primary Redis and the standby Redis of the UPF, obtaining the difference between the start and end timestamps of the processing of each stage of the protocol stack composed of IP, UDP and GTP-U by the eBPF probe arranged in the kernel of the SMF or the UPF, adding the wireless transmission delay, the 5GC forwarding delay and the difference between the start and end timestamps of the processing of each stage of the protocol stack to obtain the end-to-end delay and save it; Constructing a Gaussian model according to the delay data stored within 48 hours by the service processing module of the UPF, and performing abnormality detection by using the Gaussian model obtained by construction and adopting the 3σ principle, and triggering the switching of the primary UPF and the standby UPF when an abnormality occurs; Compiling Keepalived supporting BFD, and configuring BFD instances for Keepalived, and binding VRRP instances with BFD; After establishing a BFD session, continuously sending millisecond-level probe packets at a preset time interval, and when detecting a link fault or a node fault, sending a fault event to Keepalived to trigger VRRP state switching; After receiving the fault event sent by BFD, Keepalived notifies switching instructions to the control plane network element CPF and the data plane network element UPF of the 5GC to switch the signaling flow and the data flow. The method of constructing a Gaussian model according to the delay data stored within 48 hours by the service processing module of the UPF comprises: Obtaining the delay data within 48 hours as a data set; Removing noise in the data set to obtain a denoised data set, and interpolating and completing the missing values in the denoised data set to obtain a preprocessed data set; Decomposing the delay data in the preprocessed data set according to the interfaces to obtain the interface delay corresponding to each interface; Constructing a Gaussian model according to the interface delay corresponding to each interface.

2. The 5GC industrial private network handover latency detection method of claim 1, wherein, The method of calibrating the clock offset between the first probe arranged at the N2 interface and the second probe arranged at the N3 interface through the 5G synchronization clock comprises: Monitoring the switching preparation time and the resource allocation time by the first probe arranged at the N2 interface; Monitoring the data forwarding time between the source UPF and the target UPF by the second probe arranged at the N3 interface; Calibrating the clock offset between the first probe and the second probe through the 5G synchronization clock.

3. The 5GC industrial private network handover latency detection method of claim 2, wherein, The method of monitoring the switching preparation time and the resource allocation time by the first probe arranged at the N2 interface comprises: Embedding a 10ns precision timestamp in the signaling by the first probe arranged at the N2 interface, and collecting the timestamps of the switching preparation signaling event and the resource allocation signaling event, respectively.

4. The 5GC industrial private network handover latency detection method of claim 2, wherein, The data forwarding time between the source UPF and the target UPF is monitored through the second probe arranged at the N3 interface, and specifically: The second probe arranged at the N3 interface inserts a time-stamped OAM detection packet into the data stream sent by the source UPF to the target UPF, and collects the time stamp of the source UPF completing data stream encapsulation and sending to the target UPF and the time stamp of the target UPF successfully receiving the data stream, respectively.

5. The 5GC industrial private network handover latency detection method of claim 1, wherein, The VRRP state switching is triggered, including: After the Keepalived receives the fault event sent by the BFD, the backup node raises its VRRP priority to be greater than the VRRP priority of the master node to perform master-slave node switching, and the backup node preoccupies the master role and takes over the virtual IP and traffic forwarding; The master node automatically reduces its VRRP priority by configuring a delay preemption, and raises the VRRP priority of the backup node, sends a VRRP announcement packet, and performs master-slave node switching after a delay time.

6. The 5GC industrial private network handover latency detection method of claim 1, wherein, After the Keepalived receives the fault event sent by the BFD, the notification switching instruction is sent to the control plane network element CPF and the data plane network element UPF of the 5GC, and the 5GC industrial private network switching delay detection method further includes: A preset redundant path is set, a master UPF and two backup UPFs are configured in each service slice, backup data is classified according to the access frequency, hot data, warm data and cold data are obtained, and the hot data, warm data and cold data are stored in the memory database, the high-speed solid-state disk and the mechanical hard disk, respectively.

7. The 5GC industrial private network handover latency detection method of claim 6, wherein, After the hot data, warm data and cold data are stored in the memory database, high-speed solid-state disk and mechanical hard disk, respectively, the 5GC industrial private network switching delay detection method further includes: When the 5GC forwarding delay of the master UPF is greater than the preset low delay value and the duration reaches the preset detection period threshold, the master-slave UPF switching is triggered.

8. The 5GC industrial private network handover latency detection method of claim 7, wherein, The master-slave UPF switching is triggered, including: Determine the optimal backup UPF; Select the optimal backup UPF and the master UPF for switching.

9. The 5GC industrial private network handover latency detection method of claim 8, wherein, The optimal backup UPF is determined, specifically: The following formula is used to calculate the path from the master UPF to the backup UPF, and the backup UPF with the smallest path is selected as the optimal backup UPF; In the formula, This indicates the path from the primary UPF to the backup UPF. Indicates the main UPF up to the th Round-trip latency of a backup UPF Indicates the first The current load of each backup UPF This represents the load sensitivity coefficient.

Images