Regional IP Anycast site division method based on delay optimization

Abstract

The present invention discloses a regional IP Anycast site division method based on delay optimization, and relates to the field of network technology. The method comprises the following steps: step one: analyzing the network topology structure and user distribution, using past IP Anycast site delay data, and combining with a machine learning model to predict network traffic and delay trends, and reducing the number of IP Anycast site access requests in the region in advance before the traffic peak arrives in the region, and sending the predicted data to step two; step two: adjusting the location and number of IP Anycast sites by obtaining the network congestion situation, evaluating the delay rate of the IP Anycast sites with a TTL step value of 1 around them, and sending the obtained evaluation data to step three; step three: transmitting the evaluation result of each IP Anycast site to the surrounding IP Anycast sites.

Description

Technical Field

[0001] The invention relates to the field of network technology, and in particular to a regional IP Anycast site division method based on delay optimization. Background Art

[0002] Anycast is a routing technology that allows multiple network devices to share the same IP address, which allows network traffic to be routed to the nearest active device. As the complexity of Internet services increases, simply relying on Anycast to achieve load balancing and failover is no longer sufficient to meet the needs. As a result, Anycast site division based on latency optimization has gradually become a trend. This optimization not only takes into account the geographical "closest" principle, but also combines factors such as network conditions, link quality, and bandwidth.

[0003] However, traditional Anycast site division is usually based on pre-defined rules and static configurations, which may be difficult to adapt to changes in network conditions in real time, such as burst traffic, link failures or congestion. In traditional methods, delay measurement may be based on fixed or periodic detection packets, which are not accurate enough due to insufficient detection frequency or limitations of the detection mechanism itself, and are difficult to reflect the actual network conditions. Although Anycast can provide multiple paths to the destination, in some cases, the routing selection may not be optimal, resulting in uneven distribution of traffic, and some sites may be overloaded while other sites are not fully utilized.

[0004] In view of the above technical defects, a solution is now proposed. Summary of the invention

[0005] The purpose of the present invention is to provide a regional IP Anycast site division method based on delay optimization to solve the above-mentioned technical defects. The present invention analyzes the network topology and user distribution, as well as historical data, combines a machine learning model, predicts network traffic and delay trends, and reduces the number of IP Anycast site access requests in the region in advance before the peak traffic in the region arrives, adjusts the location and number of sites considering network congestion, evaluates the delay rate of IP Anycast sites with a TTL step value of 1 around them, and transmits the IP Anycast site delay rates to each other. For the evaluation results of high-latency IP Anycast sites, the access request is transferred to the evaluation results of low-latency IP Anycast sites, the evaluation values ​​of IP Anycast sites are obtained in real time, the network performance is monitored and the IP Anycast site paths are continuously optimized.

[0006] The object of the present invention can be achieved by the following technical solution: A regional IP Anycast site division method based on delay optimization comprises the following steps:

[0007] Step 1: Analyze the network topology and user distribution, use past IP Anycast site delay data, and combine it with machine learning models to predict network traffic and delay trends. Before the peak traffic in the region arrives, reduce the number of IP Anycast site access requests in the region in advance, and send the predicted data to step 2.

[0008] Step 2: By obtaining the network congestion situation, the location and number of IP Anycast sites are adjusted, the delay rate of the surrounding IP Anycast sites with a TTL step value of 1 is evaluated, and the obtained evaluation data is sent to step 3;

[0009] Step 3: Transmit the evaluation result of each IP Anycast site to the surrounding IP Anycast sites with a TTL step value of 1. For the evaluation result of the high-latency IP Anycast site, transfer the access request to the evaluation result of the low-latency IP Anycast site to obtain the optimal IP Anycast site path, and transmit the optimal path data to step 4;

[0010] Step 4: Obtain the evaluation results of IP Anycast sites in real time, monitor network performance, and continuously optimize IP Anycast site paths.

[0011] Preferably, the process of analyzing the network topology structure is as follows:

[0012] The network topology data is obtained through network mapping tools and data analysis. An ICMP request packet is sent to the target IP address. The ICMP request packet includes an ICMP header and a variable-length data portion. The path from the source to the target is tracked by detecting the connectivity between IP Anycast sites and measuring the round-trip time from the user device to the IP Anycast site. An ICMP packet is sent to the target address with the TTL value of the packet set to the current number of steps. The TTL value is incremented. When an intermediate router receives a packet with a TTL value of 0, the intermediate router sends an ICMP Time Exceeded message as a response. The response includes the IP address of the user device. When the IP Anycast site receives the ICMP packet, it will send an ICMP packet as a response, and record the number of steps and IP addresses along the way. A network topology map is constructed by performing DNS query domain name resolution and based on the data collected by the probe.

[0013] Preferably, in the user distribution situation, the process of indirectly obtaining the geographic location through the IP address is as follows:

[0014] By using the AP provided by a third-party service to convert IP addresses into geographic coordinates, the IP addresses are parsed, and the collected IP data is cleaned and verified. By checking for duplicate entries of the same IP address and geographic location, statistical methods are used to identify irregular data points, including extreme values, duplicate entries, and atypical patterns.

[0015] Preferably, in the user distribution situation, the process of visualizing the user distribution through map projection is as follows:

[0016] Create a user distribution heat map, use GIS software to perform spatial analysis, identify areas with high user density, combine network topology and user distribution data, and evaluate the predicted latency values ​​of Anycast sites in the area.

[0017] Preferably, the data collection and preprocessing process of the machine learning model is as follows:

[0018] Collect the delay time of reaching different IP Anycast sites at past time points, preprocess the collected data, remove outliers and missing values, standardize the collected data, and create a sliding window to capture the dependencies in the delay time series of the same IP Anycast site. The captured dependencies are represented by the time series x = [x1, x2, ..., x t ], where x t Represents the delay of the IP Anycast site at the tth time point, and converts the preprocessed data into a supervised learning form, that is, creates input-output pairs, predicts the delay at the next time point, and constructs the training samples as follows:

[0019] Input: x t-n+1:t =[x t-n+1 ,x t-n+2 ,...,x t ]

[0020] Output: y t+1 =x t+1

[0021] n is the size of the time window, which means that the data of the last n time points are used to predict the value of the next time point; x t+1 Refers to the observation value at a certain time point t+1 in the time series; y t+1 Refers to the target output of the model, that is, the delayed value at the predicted time point t+1, x t-n+1:t Refers to a continuous time series data starting from time point t-n+1 and ending at time point t; x t-n+1 refers to the first element, and x t Refers to the last element.

[0022] Preferably, the prediction process of the delay value predicted by the Anycast site is as follows:

[0023] Construct an RNN model and use the RNN unit. Its state update equation is expressed as:

[0024] h t =σ(W xh x t +W hh h t-1 +b h )

[0025] Among them, h t is the hidden state at time t; x t is the input at time t; W xh and W hh are the weight matrices from input to hidden state and from hidden state to hidden state; b h is the bias term of the hidden state; σ is the activation function, using tanh or ReLU;

[0026] Connect to the hidden state h t And the predicted value y is obtained by linear transformation t , the formula is:

[0027] y t =W hy h t +b y

[0028] Among them, W hy is the weight matrix from hidden state to output; b y is the output bias term;

[0029] The mean square error (MSE) is used as the loss function to measure the predicted value. and the true value y t The gap between:

[0030]

[0031] Where N is the number of samples;

[0032] The true value range of mean square error MSE is usually [0,100]. Compare the true value range of mean square error MSE to judge the performance of the model:

[0033] When the true value of the mean square error MSE is greater than or equal to 0 and less than or equal to 100, the model performs well;

[0034] When the true value of the mean square error MSE is greater than 100, the model performs well and needs to be optimized further;

[0035] When the true value of the mean square error MSE is less than 0, the model performs poorly and needs to be remodeled;

[0036] Get forecasts of future traffic peaks in a region based on collected historical data and current trends The maximum number of connections that the IP Anycast site can effectively handle reaches 80% of the maximum number of connections, which is used as the criterion for triggering the access bypass strategy to disperse the access requests to other IP Anycast sites that have not reached a high load level;

[0037] For the predicted value The number of connections is obtained to determine whether the traffic bypasses the strategy to distribute the load:

[0038] When the predicted value When the number of connections is greater than or equal to 80% of the maximum connection number threshold of the IP Anycast site area, the traffic bypass strategy is activated to distribute the load;

[0039] When the predicted value When the number of connections is less than 80% of the maximum connection number threshold of the IP Anycast site area, the traffic bypass strategy is not used to distribute the load and the routing mode of the existing user equipment is maintained.

[0040] Preferably, the delay evaluation process of each IP Anycast site is as follows:

[0041] Obtain the load rates of the surrounding IP Anycast sites with a TTL step value of 1, as well as the load rates of the surrounding IP Anycast sites, and organize the load rates of the obtained IP Anycast sites with a TTL step value of 1 into a set, and use the sorting algorithm to obtain the load rate ranking G[G1, G2, G3...Gm] of the IP Anycast sites from large to small through the sorting algorithm for the delay rate data of the IP Anycast sites in the set, and G1≤G2≤G3≤G4≤...≤Gm, and assign values ​​to different rankings, recorded as As(m), the formula is:

[0042]

[0043] Where d is a fixed value, m is the sort value, and As(m) is the value assigned to each sort;

[0044] The IP Anycast site load within the obtained range is scored, denoted as Sc(G), and the scoring formula is:

[0045]

[0046] Wherein, Sc(G) is the delay rate score value of different IP Anycast sites.

[0047] Preferably, the process of judging whether to discard user equipment outside the area based on different IP Anycast site delay rate score values ​​is as follows:

[0048] Baseline Threshold Score S Ac The formula is:

[0049]

[0050] When the IP Anycast site latency score is greater than 60% of the baseline threshold score in the area Ac When there is a risk of high load on IP Anycast sites in the region, the latency score of IP Anycast sites in the range and the benchmark threshold score S are obtained. Ac 60% of the area is used as a comparison to determine whether to discard user devices outside the area:

[0051] When the IP Anycast site latency score is greater than the baseline threshold score by 60%S Ac When the user equipment outside the area is discarded, the traffic bypass strategy is activated to distribute the load, and IP Anycast sites outside the area are given priority;

[0052] When the IP Anycast site latency score is less than or equal to the baseline threshold score 60%S Ac When the IP Anycast site is connected, the user devices outside the area are not distinguished, the traffic detour strategy is not started to distribute the load, and the IP Anycast site is connected normally.

[0053] Preferably, the optimal IP Anycast site location acquisition process is as follows:

[0054] The delay rate score results of IP Anycast sites with a TTL step value of 1 around the IP Anycast site are obtained, and data packets are transmitted to the IP Anycast site with a high delay rate score result, and the TTL value of the data packet is recorded, as well as the TTL value and IP address along the way. The DNS query returns multiple IP addresses to the user device through polling, and obtains the lowest TTL value. The propagation method on this TTL value path is given priority, the DNS is recorded, and the request is redirected to the optimal IP Anycast site based on the lowest TTL value result.

[0055] Preferably, the process of continuously optimizing the IP Anycast site path is as follows:

[0056] Set the IP Anycast site visit KPIS, use automated tools to obtain the IP Anycast site visit KPIS, obtain the minimum access threshold E for each site, and try to establish a connection to verify the IP Anycast site service availability and determine whether the IP Anycast site is damaged:

[0057] When the IP Anycast site access count KPIS is greater than or equal to the minimum access threshold E, the IP Anycast site is not damaged and continues to respond to access traffic;

[0058] When the IP Anycast site access number KPIS is less than the minimum access threshold E, the IP Anycast site is damaged and cannot respond to the access traffic, and the IP Anycast site is marked.

[0059] The beneficial effects of the present invention are as follows:

[0060] (1) The present invention obtains network topology and user distribution information through historical data analysis and combines machine learning models to predict network traffic and delay trends in the region. The maximum number of connections effectively processed by the IP Anycast site is 80% as the threshold for judgment. By reducing the number of IP Anycast site access requests in the region in advance, the network congestion during the peak traffic period in the region is greatly alleviated. The mean square error MSE is used as the loss function to measure the predicted value. and the true value y t The difference between them is judged by MSE, so as to continue to optimize the model.

[0061] (2) The present invention obtains the load rates of IP Anycast sites with a TTL step value of 1 around the current IP Anycast site, organizes them into a set, sorts and assigns values ​​to the load rates from large to small, scores the assigned IP Anycast sites, and compares them with the benchmark threshold scores, thereby judging whether to abandon user devices outside the area, and transferring access requests from high-latency IP Anycast site delay rate score values ​​to low-latency IP Anycast site delay rate score values, thereby obtaining the optimal IP Anycast site path and continuously optimizing the network performance of monitored IP Anycast sites. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] The present invention will be further described below in conjunction with the accompanying drawings;

[0063] Figure 1 It is a reference diagram of the method of the present invention. DETAILED DESCRIPTION

[0064] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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.

[0065] See also Figure 1 As shown, this embodiment provides a regional IP Anycast site division method based on delay optimization, including the following steps:

[0066] Step 1: Analyze the network topology and user distribution, use past IP Anycast site delay data, and combine it with machine learning models to predict network traffic and delay trends. Before the peak traffic in the region arrives, reduce the number of IP Anycast site access requests in the region in advance, and send the predicted data to step 2.

[0067] Obtain network topology data through network mapping tools and data analysis. Obtain network topology data. Deploy network probes through existing detectors. Probes refer to tools that collect network performance data. Send ICMP request packets to the target IP address. The ICMP request packet includes an ICMP header and a variable-length data portion. The ICMP header includes type, code, and checksum fields. Detect the connectivity between IP Anycast sites and measure the round-trip time from the user device to the IP Anycast site. Connectivity refers to whether the network connection from the user device to the IP Anycast site is working properly. The round-trip time refers to the time it takes for the user device's data packet to be sent from the user device to the IP Anycast site and then returned to the user device.

[0068] Trace the path from source to destination by sending an ICMP packet to the destination address. The TTL value of the packet is set to the current number of steps. The TTL value refers to the counter used to control the survival time of the data packet in the network. The number of steps refers to the increment of the TTL value after each request is sent. The intermediate router refers to the network device that forwards data packets between two or more networks. When the intermediate router receives a packet with a TTL value of 0, the intermediate router sends an ICMP Time Exceeded message as a response, which contains the IP address of the user device. Finally, when the IP Anycast site receives the ICMP packet, it will send an ICMP as a response, and record the number of steps and IP addresses along the way; DNS query refers to the process of sending a request to the IP Anycast site to find the IP address corresponding to a specific domain name. By performing DNS query domain name resolution and based on the data collected by the probe, nodes are created according to the IP address, the results of link discovery are drawn between nodes, and the graphical display effect is optimized. The generated topology map is compared with the known physical network map, and the network topology map is constructed using network management software or drawing tools;

[0069] Obtain geographic location indirectly through user's IP address, convert IP address into geographic coordinates by using AP provided by third-party service, parse IP address, clean and verify collected IP data, check duplicate entries of the same IP address and geographic location, extreme value refers to the value that deviates significantly from other observations in a set of data, duplicate entry refers to a set of records or data rows that are exactly the same in a data set, typical pattern refers to the data pattern that does not conform to common or expected behavior or trend, use statistical methods to identify data points that are not in line with the norm, and non-routine data points include extreme values, duplicate entries and atypical patterns, extreme values: data points with significantly different values ​​compared with other data, duplicate entries: a large number of repeated requests sent from the same IP address or geographic location in a short period of time, which indicates automated activity or error, atypical pattern: for example, abnormally high traffic from a geographic location in a specific time period, inconsistent data: for example, data sent from the same IP address or geographic location does not match other known information, ensure that all data is in a consistent format;

[0070] Use map projection to visualize user distribution and create user distribution heat maps. Use GIS software for spatial analysis. Use spatial clustering analysis tools to identify areas with significant high-density features, i.e., areas with high user density. Analyze users' online behavior patterns. Combine network topology and user distribution data to assess which areas need more Anycast sites. Use simulation tools to predict network performance under different configurations.

[0071] The delay time of reaching different IP Anycast sites at past time points is collected, and the collected data is preprocessed to remove outliers and missing values. The outliers are identified using the quartile range. Any value lower than Q1-1.5*IQR or higher than Q3+1.5*IQR can be regarded as an outlier. For missing data with a large proportion of missing values, the gaps can be filled by interpolation. For data sets with a small proportion of missing values, the records containing missing values ​​can be directly deleted. The collected data is standardized and the data of each feature is transformed into a fixed interval, usually between [0, 1]. A sliding window is created to capture the dependencies in the delay time series of the same IP Anycast site, and the captured dependencies are represented by a time series x=[x1,x2,...,x t ], where x t Represents the delay of the IP Anycast site at the tth time point, and converts the preprocessed data into a supervised learning form, that is, creates input-output pairs, predicts the delay at the next time point, and constructs the training samples as follows:

[0072] Input: x t-n+1:t =[x t-n+1 ,x t-n+2 ,...,x t ]

[0073] Output: y t+1 =x t+1

[0074] n is the size of the time window, which means that the data of the last n time points are used to predict the value of the next time point; x t+1 Refers to the observation value at a certain time point t+1 in the time series; y t+1 Refers to the target output of the model, that is, the delayed value at the predicted time point t+1, x t-n+1:t Refers to a continuous time series data starting from time point t-n+1 and ending at time point t; x t-n+1 refers to the first element, and x t Refers to the last element;

[0075] Construct an RNN model and use the RNN unit. Its state update equation is expressed as:

[0076] h t =σ(W xh x t +W hh h t-1 +b h )

[0077] Among them, h t is the hidden state at time t; xt is the input at time t; W xh and W hh are the weight matrices from input to hidden state and from hidden state to hidden state; b h is the bias term of the hidden state; σ is the activation function, using tanh or ReLU;

[0078] Connect to the hidden state h t And the predicted value y is obtained by linear transformation t , the formula is:

[0079] y t =W hy h t +b y

[0080] Among them, W hy is the weight matrix from hidden state to output; b y is the output bias term;

[0081] The mean square error (MSE) is used as the loss function to measure the predicted value. and the true value y t The gap between:

[0082]

[0083] Where N is the number of samples; the true value of the mean square error MSE is usually in the range of [0,100]. The MSE difference between the predicted value and the true value is judged. When the true value of the mean square error MSE is greater than or equal to 0 and less than or equal to 100, the model performs well.

[0084] When the true value of the mean square error MSE is greater than 100, the model performs well and needs to be optimized further;

[0085] When the true value of the mean square error MSE is less than 0, the model performs poorly and needs to be remodeled;

[0086] Get forecasts of future traffic peaks in a region based on collected historical data and current trends The software obtains the maximum connection capacity data of the IP Anycast site through the number of connections of the IP Anycast site. When the number of connections of the IP Anycast site reaches 80% of its designed maximum capacity, it is the data peak, which is used as the standard for triggering the access bypass strategy. That is, when it reaches 40,000 connections out of 50,000 connections, the performance may begin to decline, and the access requests will be dispersed to other IP Anycast sites that have not reached the high load level;

[0087] For the predicted value The number of connections is obtained to determine whether the traffic bypasses the strategy to distribute the load:

[0088] When the predicted value When the number of connections is greater than or equal to 80% of the maximum connection number threshold of the IP Anycast site area, the traffic bypass strategy is activated to distribute the load;

[0089] When the predicted value When the number of connections is less than 80% of the maximum connection number threshold of the IP Anycast site area, the traffic bypass strategy is not used to distribute the load, and the routing mode of the existing user equipment is maintained;

[0090] Step 2: By obtaining the network congestion situation, the location and number of IP Anycast sites are adjusted, the delay rate of the surrounding IP Anycast sites with a TTL step value of 1 is evaluated, and the obtained evaluation data is sent to step 3;

[0091] Enable the traffic monitoring function on the user device. The traffic monitoring function is used to monitor the network bandwidth usage, detect abnormal traffic patterns and locate network fault points. Configure the device to send traffic records to the traffic collector. Collect traffic monitoring records from network devices through the traffic collector. Use the traffic analysis tool to parse and analyze the collected IP Anycast site traffic data. Calculate the total upload and download traffic and load rate of the site traffic through the site traffic data, evaluate the utilization of the IP Anycast site bandwidth, obtain the load rate of nearby IP Anycast sites, and the load rate of the surrounding IP Anycast sites, and organize the load rates near the obtained nearby IP Anycast sites into a set. Use the sorting algorithm to obtain the IP Anycast site delay rate data in the set. Obtain the load rate ranking G[G1, G2, G3...Gm] from large to small, and G1≤G2≤G3≤G4≤...≤Gm, and assign values ​​to different rankings, recorded as As(m), the formula is:

[0092]

[0093] Where d is a fixed value, m is the sort value, and As(m) is the value assigned to each sort;

[0094] The IP Anycast site load within the obtained range is scored, denoted as Sc(G), and the scoring formula is:

[0095]

[0096] Among them, Sc(G) is the delay rate score value of different IP Anycast sites;

[0097] Get the benchmark IP Anycast site latency threshold score S Ac , the formula is:

[0098]

[0099] When the IP Anycast site latency score is greater than 60% of the baseline threshold score in the area Ac When there is a risk of high load on IP Anycast sites in the region, the latency score of IP Anycast sites in the range and the benchmark threshold score S are obtained. Ac 60% of the area is used as a comparison to determine whether to discard user devices outside the area:

[0100] When the IP Anycast site latency score is greater than the baseline threshold score by 60%S Ac When the user equipment outside the area is discarded, the traffic bypass strategy is activated to distribute the load, and IP Anycast sites outside the area are given priority;

[0101] When the IP Anycast site latency score is less than or equal to the baseline threshold score 60%S Ac When the user equipment outside the area is not distinguished, the traffic detour strategy is not activated to distribute the load, and the IP Anycast site is connected normally;

[0102] Step 3: Transmit the evaluation result of each IP Anycast site to nearby IP Anycast sites, transfer the access request of the evaluation result of the high-latency IP Anycast site to the evaluation result of the low-latency IP Anycast site, obtain the optimal IP Anycast site path, and transmit the optimal path data to step 4;

[0103] The obtained nearby IP Anycast site delay rate score results, IP Anycast site delay rate score and benchmark threshold score S Ac 60% is considered as high latency, and the data packet is transmitted to the IP Anycast site with a high latency rate score, and the TTL value of the data packet is recorded, as well as the TTL value and IP address along the way. The DNS query returns multiple IP addresses to the user device through polling. Polling means returning multiple IP addresses for the same domain name query and obtaining the lowest TTL value. The propagation method on the path of this TTL value is given priority, and the DNS is recorded. According to the result of the lowest TTL value, the request is redirected to the optimal IP Anycast site.

[0104] Step 4: Obtain the evaluation results of IP Anycast sites in real time, monitor network performance, and continuously optimize IP Anycast site paths;

[0105] Set the IP Anycast site access KPIS and use automated tools to obtain the IP Anycast site access KPIS. When some nodes fail, the surrounding nodes can still handle additional access requests without being overloaded. This critical value is set as the minimum access threshold E, and try to establish a connection to verify the IP Anycast site service availability and determine whether the IP Anycast site is damaged:

[0106] When the IP Anycast site access count KPIS is greater than or equal to the minimum access threshold E, the IP Anycast site is not damaged and continues to respond to access traffic;

[0107] When the IP Anycast site access number KPIS is less than the minimum access threshold E, the IP Anycast site is damaged and cannot respond to the access traffic, and the IP Anycast site is marked.

[0108] The above formulas are obtained by collecting a large amount of data for software simulation and selecting a formula that is close to the actual value. The coefficients in the formula are set by technical personnel in this field according to actual conditions. The above are only preferred specific implementations of the present invention, but the protection scope of the present invention is not limited to this. Any technical personnel familiar with the technical field within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, make equivalent replacement or change, which should be covered within the protection scope of the present invention.

Claims

1. A regional IP Anycast site division method based on delay optimization, characterized in that: The steps of the method include: Step 1: Analyze the network topology and user distribution, use the past IP Anycast site delay data, and combine the machine learning model to process the delay data to predict network traffic and delay trends. Before the peak traffic in the region arrives, reduce the number of IP Anycast site access requests in the region in advance, and send the predicted data to step 2; Step 2: By obtaining the network congestion situation, the location and number of IP Anycast sites are adjusted, the delay rate of the surrounding IP Anycast sites with a TTL step value of 1 is evaluated, and the obtained evaluation data is sent to step 3; Step 3: Transmit the evaluation result of each IP Anycast site to the surrounding IP Anycast sites with a TTL step value of 1. For the evaluation result of the high-latency IP Anycast site, transfer the access request to the evaluation result of the low-latency IP Anycast site to obtain the optimal IP Anycast site path, and transmit the optimal path data to step 4; Step 4: Obtain the evaluation results of IP Anycast sites in real time, monitor network performance, and continuously optimize IP Anycast site paths; Wherein, in the step 1, analyzing the network topology structure includes: obtaining network topology data through network mapping tools and data analysis, sending an ICMP request packet to the target IP address, the ICMP request packet includes an ICMP header and a variable-length data portion, tracing the path from the source to the target by detecting the connectivity between IP Anycast sites and measuring the round-trip time from the user device to the IP Anycast site, and sending an ICMP packet to the target address, the TTL value of the packet is set to the current number of steps, and the TTL value is incremented. When an intermediate router receives a packet with a TTL value of 0, the intermediate router sends an ICMP TimeExceeded message as a response, and the IP address of the user device is included in the response. Finally, when the IP Anycast site receives the ICMP packet, it will send an ICMP packet as a response, and record the number of steps and IP addresses along the way, and build a network topology map by performing DNS query domain name resolution and based on the data collected by the probe; In the step 1, the user distribution analysis includes: Obtaining geographic location indirectly through IP addresses: By using the AP provided by a third-party service to convert IP addresses into geographic coordinates, resolving IP addresses, cleaning and verifying collected IP data, and checking for duplicate entries of the same IP address and geographic location, using statistical methods to identify irregular data points, which include extreme values, duplicate entries, and atypical patterns; Convert IP addresses to geographic coordinates and visualize user distribution through map projection: Create user distribution heat maps, use GIS software for spatial analysis, identify areas with high user density, combine network topology and user distribution data, and evaluate the predicted latency values ​​of Anycast sites in the area; In the step 1, the machine learning model processes the delay data, including: collecting the delay time of reaching different IP Anycast sites at past time points, preprocessing the collected data, removing outliers and missing values, standardizing the collected data, creating a sliding window to capture the dependencies in the delay time series of the same IP Anycast site, and representing the captured dependencies through a time series x=[x1,x2,...,x t ], where x t Represents the delay of the IP Anycast site at the tth time point, and converts the preprocessed data into a supervised learning form, that is, creates input-output pairs, predicts the delay at the next time point, and constructs the training samples as follows: Input: x t-n+1:t =[x t-n+1 ,x t-n+2 ,...,x t ] Output: y t+1 =x t+1 n is the size of the time window, which means that the data of the last n time points are used to predict the value of the next time point; x t+1 Refers to the observation value at a certain time point t+1 in the time series; y t+1 Refers to the target output of the model, that is, the delayed value at the predicted time point t+1, x t-n+1:t Refers to a continuous time series data starting from time point t-n+1 and ending at time point t; x t-n+1 refers to the first element, and x t Refers to the last element.

2. The method for regional IP Anycast site division based on delay optimization according to claim 1, characterized in that: The prediction process of the delay value predicted by the Anycast site is as follows: Construct an RNN model and use the RNN unit. Its state update equation is expressed as: h t =σ(W xh x t +W hh h t-1 +b h ) Among them, h t is the hidden state at time t; x t is the input at time t; W xh and W hh are the weight matrices from input to hidden state and from hidden state to hidden state; b h is the bias term of the hidden state; σ is the activation function, using tanh or ReLU; Connect to the hidden state h t And the predicted value y is obtained by linear transformation t , the formula is: y t =W hy h t +b y Among them, W hy is the weight matrix from hidden state to output; b y is the output bias term; The mean square error (MSE) is used as the loss function to measure the predicted value. and the true value y t The gap between: Where N is the number of samples; The true value range of mean square error MSE is usually [0,100]. Compare the true value range of mean square error MSE to judge the performance of the model: When the true value of the mean square error MSE is greater than or equal to 0 and less than or equal to 100, the model performs well; When the true value of the mean square error MSE is greater than 100, the model performs well and needs to be optimized further; When the true value of the mean square error MSE is less than 0, the model performs poorly and needs to be remodeled; Get forecasts of future traffic peaks in a region based on collected historical data and current trends The maximum number of connections that the IP Anycast site can effectively handle reaches 80% of the maximum number of connections, which is used as the criterion for triggering the access bypass strategy to disperse the access requests to other IP Anycast sites that do not reach the high latency level; For the predicted value The number of connections is obtained to determine whether the traffic bypasses the strategy to disperse the delay: When the predicted value When the number of connections is greater than or equal to 80% of the maximum connection number threshold in the IP Anycast site area, the traffic bypass strategy is activated to disperse the delay; When the predicted value When the number of connections is less than 80% of the maximum connection number threshold of the IP Anycast site area, no detour strategy is used to disperse the delay of the traffic, and the routing method of the existing user equipment is maintained.

3. The method for regional IP Anycast site division based on delay optimization according to claim 2 is characterized in that: The delay of each IP Anycast site is evaluated as follows: Obtain the delay rate of the IP Anycast site with a TTL step value of 1 around the current IP Anycast site, and organize the delay rates of the IP Anycast sites with a TTL step value of 1 around the obtained IP Anycast site into a set, and use the sorting algorithm to obtain the IP Anycast site delay rate ranking G[G1, G2, G3...Gm] from large to small for the IP Anycast site, and G1≤G2≤G3≤G4≤...≤Gm, and assign values ​​to different rankings, recorded as As(m), the formula is: Where d is a fixed value, m is the sort value, and As(m) is the value assigned to each sort; The latency rate of the IP Anycast site in the acquired range is scored, denoted as Sc(G), and the scoring formula is: Wherein, Sc(G) is the delay rate score value of different IP Anycast sites.

4. The method for regional IP Anycast site division based on delay optimization according to claim 3 is characterized in that: The process of judging whether to discard user equipment outside the area based on different IP Anycast site delay rate score values ​​is as follows: Get the benchmark IP Anycast site latency threshold score S Ac , the formula is: When the IP Anycast site latency score is greater than the baseline threshold score S in the area Ac When the latency rate of IP Anycast sites in the region is 60%, there is a risk of high load on the IP Anycast sites in the region. By obtaining the latency rate score of IP Anycast sites in the region and the benchmark threshold score S Ac 60% of the area is used as a comparison to determine whether to discard user devices outside the area: When the IP Anycast site delay rate score is greater than the baseline threshold score S Ac When the traffic volume reaches 60%, the user devices outside the area are discarded, and the traffic bypass strategy is activated to distribute the load, giving priority to IP Anycast sites outside the area. When the IP Anycast site latency score is less than or equal to the baseline threshold score S Ac When the traffic volume is 60%, the user devices outside the area are not distinguished, the traffic bypass strategy is not started to distribute the load, and the IP Anycast site is connected normally.

5. The method for regional IP Anycast site division based on delay optimization according to claim 4 is characterized in that: The optimal IP Anycast site location acquisition process is as follows: For the current IP Anycast site, the delay rate score results of the surrounding IP Anycast sites with a TTL step value of 1 are obtained, and the data packet is transmitted to the IP Anycast site with a high IP Anycast site delay rate score result, and the TTL value of the data packet is recorded, as well as the TTL value and IP address along the way. The DNS query returns multiple IP addresses to the user device through polling, and obtains the lowest TTL value. The propagation method on this TTL value path is given priority, the DNS is recorded, and the request is redirected to the optimal IP Anycast site based on the lowest TTL value result.

6. The method for regional IP Anycast site division based on delay optimization according to claim 5, characterized in that: The process of continuously optimizing the IP Anycast site path is as follows: Set the IP Anycast site visit KPIS, use automated tools to obtain the IP Anycast site visit KPIS, obtain the minimum access threshold E for each site, and try to establish a connection to verify the IP Anycast site service availability and determine whether the IP Anycast site is damaged: When the IP Anycast site access count KPIS is greater than or equal to the minimum access threshold E, the IP Anycast site is not damaged and continues to respond to access traffic; When the IP Anycast site access number KPIS is less than the minimum access threshold E, the IP Anycast site is damaged and cannot respond to the access traffic, and the IP Anycast site is marked.

Images