In-band network telemetry high yield resource orchestration based on segment routing

Abstract

The present application relates to a kind of high-yield resource arrangement methods of in-band network telemetry based on segment routing, belong to communication technical field.The method solves the problem of path and telemetry data distribution in in-band network telemetry by constructing quadratic programming model with second constraint;Meanwhile, the method uses heuristic algorithm to solve the problem of long solving time of the quadratic programming model with second constraint by finding suboptimal solution, path selection and telemetry data distribution are carried out by comprehensively considering the path of service flow, and the arrangement of in-band network telemetry resource is realized.The present application is compared with three existing algorithms through experimental verification, and the results show that the present application has obvious advantages in overall gain of collected telemetry information, node coverage and telemetry data coverage.

Description

Technical Field

[0001] This invention belongs to the field of communication technology and relates to a method for orchestrating high-yield resources for in-band network telemetry based on segmented routing. Background Technology

[0002] With the rapid development of network scale, the increasing variety of services, and the year-on-year growth of bandwidth, network management methods are constantly emerging. However, the overall trend is towards remote, refined, and real-time capabilities, thus creating a stronger demand for network monitoring and measurement technologies. To achieve remote, refined management and to accurately and promptly locate and resolve network problems, it is generally expected that detection technologies should possess functions such as monitoring device status, proactively detecting network failures, fault location methods, and fault recovery. Traditional monitoring methods may not be effective in meeting current needs; therefore, it is necessary to research and implement more efficient network measurement technologies to ensure accurate monitoring and analysis of large-scale data flows.

[0003] The emergence of Software-defined Networking (SDN) and Programmable Data Plane (PDP), along with more open control plane and data plane programming capabilities, has reshaped traditional network measurement processes. Therefore, by leveraging the advantages of SDN and PDP, In-Band Network Telemetry (INT) has been proposed to achieve fine-grained and real-time network monitoring. Segment Routing (SR) is a novel source routing method aimed at solving traffic engineering problems, particularly in IP networks. The combination of INT and SR is considered an effective way to overcome the limitations of current network monitoring, such as limited coverage and high overhead.

[0004] In real-world physical networks, each physical node possesses various types of telemetry information, such as packet delay (one-way, round-trip, jitter), packet loss rate (i.e., packet loss rate), and bandwidth (including capacity, bottleneck bandwidth, and available bandwidth). INT only specifies a low-level device primitive, and how to achieve network-wide traffic monitoring remains undefined. SR inserts a series of SR tags into the header of each probe packet in the PDP switch at its ingress point to indicate the routing path, and each subsequent PDP switch can forward probe packets based solely on the correct SR tag. Specifically, INT provides rich telemetry data to the control plane for traffic engineering and fault recovery using SR, while the control plane also uses SR to adjust INT's data collection scheme for adaptive network monitoring. Against this backdrop, this invention proposes a heuristic algorithm for a high-yield resource orchestration method for in-band network telemetry based on segmented routing, abbreviated as SR-INTO. The SR-INTO algorithm considers both the gain from telemetry data and the bandwidth consumption of traffic flows on the links. To address the unprecedented challenges faced by today's NC&M systems, more powerful and adaptive network monitoring technologies are needed, capable of real-time visualization of network operations and timely and accurate detection or location of anomalies. How to set up a reasonable orchestration scheme to achieve full network monitoring is a problem that needs to be studied. Summary of the Invention

[0005] In view of this, the purpose of this invention is to provide a high-yield resource orchestration method for in-band network telemetry based on segmented routing, which orchestrates various telemetry information in the physical network in the context of INT and SR, while taking into account the gain brought by telemetry data and the bandwidth occupation of service flows on the link.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A high-yield resource orchestration method for in-band network telemetry based on segmented routing is proposed. This method solves the path and telemetry data allocation problem in in-band network telemetry by constructing a quadratic constraint quadratic programming model. At the same time, the method also uses a heuristic algorithm to find suboptimal solutions to solve the problem of excessive solution time of the quadratic constraint quadratic programming model. By comprehensively considering the path of the service flow, path selection and telemetry data allocation are performed to realize the orchestration of in-band network telemetry resources.

[0008] The heuristic algorithm includes the following steps:

[0009] S1. First, calculate the link IDs traversed by each candidate path of a service flow. Then, combine the candidate paths of each service flow, calculate the overlap of link IDs in each combination, and select the candidate path combination with the smallest overlap. If multiple path combinations have the same overlap, select the combination with the smallest total path hop count as the routing path for each service flow.

[0010] S2. Arrange the telemetry information in all nodes of the underlying network in descending order of telemetry information gain; if the information gain is the same, arrange them in ascending order of load.

[0011] S3. Calculate the remaining telemetry data capacity for each service flow r, and sort the telemetry capacity of all service flows in descending order; allocate telemetry information for the service flow ranked first based on its tail node, and update the physical network resources.

[0012] S4. Repeat step S3 until all service flows can no longer collect telemetry information;

[0013] S5. For the sets of collected telemetry information and the sets of uncollected telemetry information obtained in steps S3 and S4, sort the collected sets in ascending order according to telemetry information gain, and sort the uncollected sets in descending order according to telemetry information gain. If the information gain is the same, sort them in ascending order according to their load.

[0014] S6. If the information gain of the first telemetry data in the already collected set is greater than the information gain of the first telemetry data in the uncollected set, then no replacement is performed and the process ends; otherwise, the telemetry information is replaced.

[0015] Furthermore, step S3 includes:

[0016] S31. For the service flow with the largest remaining capacity to carry telemetry data, at the tail node of the service flow, determine whether the telemetry data sorted in step S2 can be collected according to the constraints that the telemetry data should meet. If a telemetry data is collected, remove the telemetry data at the tail node and update the network status to proceed to step S4.

[0017] S32. If the remaining storage flow table entry capacity of the tail node is equal to 0 and the service flow has not yet collected any telemetry data at the tail node, and all telemetry data of the tail node has been collected, then delete the tail node on the service flow path and repeat step S31 with the next node as the tail node.

[0018] Otherwise, return to step S31 to select the next sorted service flow r+1 and collect telemetry data at its tail node.

[0019] In step S31, the constraints include: the remaining capacity of the service flow is greater than the size of the telemetry data to be collected; and the bandwidth capacity of the subsequent link does not exceed the capacity after adding the telemetry data to be collected.

[0020] Furthermore, in step S6, the method for changing the telemetry information is as follows:

[0021] First, calculate the FlowId of the service flow that collects the first telemetry data in the collected set. Then, based on the constraints, determine whether the telemetry data in the uncollected set can be replaced in order of sorting.

[0022] If the replacement conditions are met, the telemetry information is replaced, and the telemetry data in the collected set is sorted, and step S6 is repeated.

[0023] If no telemetry information meets the replacement conditions, the next sorted telemetry data in the collected set will be used as the first telemetry data in the collected set, and step S6 will be repeated.

[0024] During the replacement process, for telemetry data with the same gain that are not collected, the index of these telemetry data on the FlowId service flow path is calculated, sorted in descending order of the index, and the replacement is determined based on the sorting.

[0025] The constraints include: 1) the sum of the remaining capacity of the service flow and the size of the collected telemetry data is greater than or equal to the size of the uncollected telemetry data; 2) the bandwidth capacity of the subsequent link is not exceeded after the telemetry data is replaced; 3) the service flow collects other telemetry data at the node where the uncollected telemetry data is located or the remaining storage flow table capacity of that node is greater than 0.

[0026] Furthermore, the quadratic constraint quadratic programming model, with the optimization objective of maximizing overall information gain and minimizing total link bandwidth overhead, is expressed as:

[0027] max:Φ

[0028] Φ=α·Φ g -β·Φ c

[0029] In the formula, Φ g Φ represents the normalized total information gain. c This represents the normalized total link bandwidth cost, where α and β are both weighting coefficients.

[0030] The constraints of this model are expressed as follows:

[0031]

[0032]

[0033]

[0034]

[0035]

[0036]

[0037]

[0038] In the formula, P r f represents the set of candidate paths for the r-th business flow; r,k The variable represents a binary decision variable, with a value of 1 if the r-th business flow uses the k-th candidate path, and 0 otherwise; R represents the set of all businesses. This represents a binary decision variable; if the r-th service flow collects the m-th telemetry data at node v, the value is 1; otherwise, it is 0. V represents the set of nodes, M... v This represents the set of INT metadata types that can be collected on node v; Indicates whether the k-th candidate path of the r-th business flow passes through node v; δ r,v This represents a binary decision variable; if the r-th service flow collects telemetry data at node v, the value is 1, otherwise it is 0; s v,m Indicates the length of the INT metadata m of node v; β v This indicates the maximum number of service flows that node v can support for collecting telemetry data at that node. Indicates whether the k-th candidate path of the r-th service flow passes through link e; u s v′ v s Both u and t represent nodes; u,m B represents the detection period for collecting telemetry data m at node u; e(u,v) This indicates that the link e(u,v) can be used for the bandwidth capacity of data allocation in this model.

[0039] The beneficial effects of this invention are as follows: This invention solves the path and telemetry data allocation problem in in-band network telemetry by constructing a quadratic constraint quadratic programming model. This quadratic constraint quadratic programming model can adopt different telemetry data acquisition schemes for multiple service flows in the network according to specified paths to achieve network-wide monitoring. Furthermore, it proposes an efficient heuristic algorithm to address the problem of excessively long solution time when using the quadratic programming quadratic constraint model in large-scale networks and with a large number of service flows. By comprehensively considering the path sum of service flows for path selection and telemetry data allocation, its performance is close to that of the quadratic programming quadratic constraint model in small topologies, and in large topologies, its performance also shows improvement over the other three comparative algorithms in terms of total benefit, number of monitored nodes, and average number of telemetry data points monitored per node.

[0040] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0041] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein:

[0042] Figure 1 This is a schematic diagram of the overall framework of the method described in the embodiments of the present invention;

[0043] Figure 2 This is the N6S8 topology diagram used in the method described in the embodiments of the present invention;

[0044] Figure 3 This is the NSFNET topology diagram used in the method described in the embodiments of the present invention;

[0045] Figure 4 This is a USB topology diagram used in the method described in the embodiments of the present invention;

[0046] Figure 5 This is a comparison chart of total revenue when the number of service flows is 2 to 12 based on the N6S8 topology.

[0047] Figure 6 This is a comparison chart of total revenue when the number of service flows based on the NSFNET topology is 4 to 24.

[0048] Figure 7 This is a comparison chart of the number of monitoring nodes when the number of service flows is 4 to 24 based on the NSFNET topology.

[0049] Figure 8 This is a comparison chart showing the average number of telemetry data points monitored by monitoring nodes when the number of service flows is 4 to 24 based on the NSFNET topology.

[0050] Figure 9 This is a comparison chart of total revenue when the number of service flows is 4 to 24 based on the USB topology diagram;

[0051] Figure 10 This is a comparison chart of the number of monitoring nodes when the number of service flows is 4 to 24 based on the USB topology map;

[0052] Figure 11 This is a comparison chart showing the average number of telemetry data points monitored per node when the number of service flows is 4 to 24 based on the USB topology. Detailed Implementation

[0053] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0054] This invention proposes a high-yield resource orchestration method for in-band network telemetry based on segmented routing. Under the same physical network and number of service flows, compared with the three comparative methods Sel-INT, G-PR, and G-BDP, this method has higher total revenue, number of monitoring nodes, and average number of telemetry data points monitored per node.

[0055] like Figure 1 As shown, the method is described in detail below:

[0056] Step 1: First, calculate the link IDs traversed by each candidate path of a service flow. Then, combine the candidate paths of each service flow, calculate the overlap of link IDs in each combination, and select the candidate path combination with the smallest overlap. If multiple path combinations have the same overlap, select the combination with the smallest total path hop count as the routing path for each service flow.

[0057] Step 2: Sort the telemetry information within all nodes in descending order of telemetry information gain. If the information gain is the same, sort them in ascending order of their load.

[0058] Step 3: Sort the telemetry capabilities of all service flows in descending order. First, allocate telemetry information to the service flow with the highest ranking based on its tail node and update the physical network resources.

[0059] Step 4: Repeat step 3 until all service flows can no longer collect telemetry information;

[0060] Step 5: For the collected telemetry information sets and uncollected telemetry information sets obtained in Steps 3 and 4, sort the collected sets in ascending order according to telemetry information gain, and sort the uncollected sets in descending order according to telemetry information gain. If the information gain is the same, sort them in ascending order according to their load.

[0061] Step 6: If the gain of the first telemetry information in the collected set is greater than the gain of the first telemetry information in the uncollected set, then no exchange is performed; otherwise, the telemetry information is replaced.

[0062] Step 7: Repeat step 6 until all uncollected telemetry data no longer meet the replacement criteria, then end the entire process.

[0063] Furthermore, in this embodiment, step 3 also includes the following steps:

[0064] Step 3.1: First, allow the service flow with the largest remaining capacity to carry telemetry data to collect data from the tail node on its path. If the remaining storage flow table capacity of the node is equal to 0 and the service flow has not yet collected any telemetry data from the node, and all telemetry data from the node has been collected, then delete the tail node on the service flow path; otherwise, proceed to step 3.2.

[0065] Step 3.2: Based on the results of sorting the telemetry data in Step 2, the service flow first collects the telemetry data ranked first. The criteria for determining whether it can be collected are: 1) whether the remaining capacity of the service flow is greater than the size of the telemetry data; 2) whether adding the telemetry data exceeds the bandwidth capacity of subsequent links.

[0066] If the collection conditions are not met, determine whether subsequent telemetry data meets the collection conditions. If a telemetry data is collected, remove the telemetry data at that node and update the network status to proceed to step 4. If all telemetry information at that node does not meet the collection conditions, allow the second-ranked service flow to collect telemetry data at its tail node and repeat step 3.

[0067] Furthermore, in this embodiment, the data replacement method in step 6 is specifically as follows:

[0068] First, calculate the FlowId of the first telemetry data m1 collected in the collected set. Then, exchange the first telemetry data in the uncollected set. If multiple telemetry data have the same gain, calculate the index of each telemetry data on the FlowId service flow path, sort them in descending order of index, and replace the telemetry data that is later in the path first to reduce the impact on link bandwidth.

[0069] Data exchange must meet the following conditions: 1) The remaining capacity of the service flow + the size of the collected telemetry data ≥ the size of the uncollected telemetry data; 2) The bandwidth capacity of the subsequent link is not exceeded after the telemetry data is replaced; 3) The service flow has collected other telemetry data at the node where the uncollected telemetry data is located or the remaining storage flow table capacity of that node is greater than 0.

[0070] If the above conditions are not met, the subsequent telemetry data in the uncollected set is used for judgment until all uncollected telemetry data no longer meet the replacement conditions. Then, the information gain of the next telemetry data m2 in the collected set is compared with the information gain of the first telemetry data in the uncollected set. If the information gain of telemetry data m2 is greater than the information gain of the first telemetry data in the uncollected set, the entire process ends. Otherwise, step 6 is repeated with telemetry data m2 as the first telemetry data in the collected set. If the replacement conditions are met, the telemetry information is changed and the sorting of the collected set is updated, and step 6 is repeated.

[0071] Example 2

[0072] To address the issue of path and telemetry data allocation in in-band network telemetry, this embodiment proposes a quadratic constraint quadratic programming model. This model can adopt different telemetry data acquisition schemes for multiple service flows in the network according to specified paths to achieve full network monitoring.

[0073] Specifically, the above optimization model is described as follows:

[0074] (s r ,d r ,c r ) represents the business model, where s r Let d represent the source node of the r-th business flow. r c represents the destination node of the r-th business flow. r Let R represent the remaining capacity of the r-th service flow that can be used to collect telemetry data. Let R denote the set of all services.

[0075] The physical network model is represented by an undirected graph G(V,E), where V represents the set of nodes and E represents the set of links. Let B... e(u,v) Let g represent the bandwidth capacity available for SR-INTO for link e(u,v)∈E, in Mbps, where u represents a node. v,m This indicates that INT metadata m∈M is collected on node v∈V. v The resulting information gain is expressed in units. Let s v,m This represents the length of the INT metadata m of node v, in bytes, where m ∈ M. v M v This represents the set of INT metadata types that can be collected on node v. Let P... r Let represent the set of candidate paths for the r-th business flow. (This is a boolean variable) indicates whether the k-th candidate path of the r-th traffic flow passes through link e. Let β v This indicates the maximum number of service flows that node v can support for collecting telemetry data at that node. Let t v,mThis represents the detection period for collecting telemetry data m at node v. Let... (A boolean variable) indicates whether the k-th candidate path of the r-th business flow passes through node v. Let It represents the set of all links (including link (u,v)) preceding link (u,v)∈E on the k-th candidate path of the r-th traffic flow.

[0076] The normalized total information gain brought about by the SR-INTO collection scheme is calculated using the following formula (1):

[0077]

[0078] In the formula, Let represent a binary decision variable. If the r-th service flow collects the m-th telemetry data at node v, then The value is 1 if it is 1, otherwise it is 0.

[0079] The normalized total bandwidth overhead of the SR-INTO collection scheme on the link is calculated using the following formula (2):

[0080]

[0081] In the formula, u s v′ v s Both represent nodes, P r This represents the set of candidate paths for the r-th business flow. Indicates at node u s The collection of INT metadata types that can be collected; f r,k Let f represent a binary decision variable. If the r-th business flow uses the k-th candidate path, then f r,k The value is 1 if it is 1, otherwise it is 0.

[0082] Combining the above equation, we can obtain equation (3):

[0083] Φ=α·Φ g -β·Φ c (3)

[0084] In the formula, α and β are the equilibrium Φ g and Φ c Non-negative weights for importance.

[0085] Based on the above, this embodiment constructs the following quadratic constrained quadratic programming model, whose optimization objective is to maximize the overall information gain and minimize the total link bandwidth overhead:

[0086] max:Φ (4)

[0087] The constraints are as follows:

[0088]

[0089]

[0090]

[0091]

[0092]

[0093]

[0094]

[0095] δ r,v Let δ represent a binary decision variable. If the r-th service flow collects telemetry data at node v, then δ r,v The value is 1 if it is 1, otherwise it is 0.

[0096] The above constraints are respectively: 1) Equation (5) represents the candidate path selection constraint: each service flow can only select one of the candidate paths; 2) Equation (6) represents the node telemetry data selection constraint: each type of telemetry data on each node can only be collected once; 3) Equation (7) represents the telemetry data collection constraint: the service flow can only collect telemetry data on the nodes on its routing path; 4) Equation (8) represents the node collection constraint: if the service flow collects one or more telemetry data on the node, it means that the service flow collects telemetry data on the node; 5) Equation (9) represents the service flow carrying telemetry data constraint: the telemetry data collected by the service flow does not exceed its available capacity; 6) Equation (10) represents the node resource constraint: the number of service flows collecting telemetry data on the node cannot exceed its processing capacity; 7) Equation (11) represents the link bandwidth capacity constraint: the bandwidth used for SR-INTO will not exceed the available bandwidth capacity on each link.

[0097] Example 3

[0098] In-band network telemetry orchestration is an NP-hard problem, and the solution method takes a long time when there are many physical network models or traffic flows. In order to overcome this drawback, this embodiment uses the heuristic method described in Embodiment 1 to find a suboptimal solution to achieve in-band network telemetry orchestration.

[0099] The heuristic method (i.e., the SR-INTO algorithm) is as follows: Figure 1 As shown, in this embodiment, Visual Studio 2022 is used as the simulation software to implement and verify the SR-INTO algorithm.

[0100] When the physical network uses the NSFNET network topology, the physical network contains 14 nodes and 22 links, such as... Figure 3As shown in Table 1, the specific parameter settings are as follows.

[0101] Table 1 NSFNET Physical Network Parameter Settings

[0102]

[0103] When the physical network uses a USB network topology, the physical network contains 24 nodes and 43 links, such as... Figure 4 As shown in Table 2, the specific parameter settings are as follows.

[0104] Table 2 USB Physical Network Parameter Settings

[0105]

[0106] The service flow includes three attributes: source node, destination node, and remaining capacity that can be used for SR-INTO to collect telemetry data. The specific parameters are set such that the source node and destination node are randomly generated, and the remaining capacity that can be used for SR-INTO to collect telemetry data is in the range of [100, 1000] bytes.

[0107] The methods selected in this embodiment for comparison with the present invention include Sel-INT, G-PR, and G-BDP. Sel-INT proposes a selective in-band network telemetry collection scheme, G-PR proposes a path-ordering-based greedy algorithm, and G-BDP proposes a reward-ordering-based greedy algorithm.

[0108] First, a performance analysis is performed on the algorithm described in Example 1. Considering, Figure 2 The topology of N6S8 shown has α >> β, with maximizing the overall information gain as the main objective. Figure 5 The performance comparison of the algorithms is shown. It can be seen that when the number of traffic flows is 2, 4, and 6, QCQP, SR-INTO, and G-BDP have similar performance. However, when the number of traffic flows increases, SR-INTO, G-PR, and G-BDP cannot follow the trend of QCQP as well as SR-INTO. Therefore, for QCQP, which has become a thorny problem in large-scale in-band network telemetry orchestration, the results of SR-INTO can be used to approximate an accurate solution.

[0109] For simulations based on NSFNET and USB topologies, this embodiment does not consider QCQP because of their high time complexity. However, in the NSFNET topology, the case of α >> β is still considered, and... Figures 6-8 The simulation results are displayed. Figure 6It can be seen that the SR-INTO solution achieves a 24%, 6%, and 1.3% improvement in total gain compared to the other three comparison algorithms. Under resource constraints, compared to the other three methods, such as... Figure 7 and Figure 8 As shown, SR-INTO can monitor more nodes and telemetry data.

[0110] In USB network topology, the case of α >> β is also considered, by... Figure 9 It can be seen that the SR-INTO solution consistently improves the overall performance by 24.9%, 12%, and 1% compared to the other three comparison algorithms. From... Figure 10 , Figure 11 It can also be seen that compared to the other three methods, SR-INTO can monitor more nodes and telemetry data when resources are limited.

[0111] Furthermore, as can be seen from Table 3 below, the running time of SR-INTO is not significantly different from that of the other three comparison algorithms.

[0112] Table 3 Comparison of average running time (s) of various algorithms in different network topologies when the number of traffic flows ranges from 4 to 24.

[0113]

[0114] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for orchestrating high-yield resources in in-band network telemetry based on segmented routing, characterized in that: This method solves the problem of path and telemetry data allocation in in-band network telemetry by constructing a quadratic constraint quadratic programming model. At the same time, the method also uses a heuristic algorithm to find suboptimal solutions to solve the problem of excessive solution time of the quadratic constraint quadratic programming model. By comprehensively considering the path of the service flow, the method performs path selection and telemetry data allocation, thereby realizing the orchestration of in-band network telemetry resources. The heuristic algorithm includes the following steps: S1. First, calculate the link IDs traversed by each candidate path of a service flow. Then, combine the candidate paths of each service flow, calculate the overlap of link IDs in each combination, and select the candidate path combination with the smallest overlap. If multiple path combinations have the same overlap, select the combination with the smallest total path hop count as the routing path for each service flow. S2. Arrange the telemetry information in all nodes of the underlying network in descending order of telemetry information gain; if the information gain is the same, arrange them in ascending order of load. S3. Calculate the remaining telemetry data capacity for each service flow r, and sort the telemetry capacity of all service flows in descending order; allocate telemetry information for the service flow ranked first based on its tail node, and update the physical network resources. S4. Repeat step S3 until all service flows can no longer collect telemetry information; S5. For the sets of collected telemetry information and the sets of uncollected telemetry information obtained in steps S3 and S4, sort the collected sets in ascending order according to telemetry information gain, and sort the uncollected sets in descending order according to telemetry information gain. If the information gain is the same, sort them in ascending order according to their load. S6. If the information gain of the first telemetry data in the already collected set is greater than the information gain of the first telemetry data in the uncollected set, then no replacement is performed and the process ends; otherwise, the telemetry information is replaced.

2. The method according to claim 1, characterized in that: The quadratic constrained quadratic programming model aims to maximize overall information gain and minimize total link bandwidth overhead, and is expressed as follows: max: Φ Φ=a·Φ g -b·F c In the formula, Φ g Φ represents the normalized total information gain. c This represents the normalized total link bandwidth cost, where α and β are both weighting coefficients. The constraints of this model are expressed as follows: In the formula, P r f represents the set of candidate paths for the r-th business flow; r，k The variable represents a binary decision variable, with a value of 1 if the r-th business flow uses the k-th candidate path, and 0 otherwise; R represents the set of all businesses. This represents a binary decision variable; if the r-th service flow collects the m-th telemetry data at node v, the value is 1; otherwise, it is 0. V represents the set of nodes, M... v This represents the set of INT metadata types that can be collected on node v; Indicates whether the k-th candidate path of the r-th business flow passes through node v; δ r，v This represents a binary decision variable; if the r-th service flow collects telemetry data at node v, the value is 1, otherwise it is 0; s v，m Indicates the length of the INT metadata m of node v; β v This indicates the maximum number of service flows that node v can support for collecting telemetry data at that node. Indicates whether the k-th candidate path of the r-th service flow passes through link e; u s v′ v s Both u and t represent nodes; u，m B represents the detection period for collecting telemetry data m at node u; e(u，v) This indicates the bandwidth capacity that link e(u, v) can be used to allocate data in this model.

3. The method according to claim 1, characterized in that: Step S3 includes: S31. For the service flow with the largest remaining capacity to carry telemetry data, at the tail node of the service flow, determine whether the telemetry data sorted in step S2 can be collected according to the constraints that the telemetry data should meet. If a telemetry data is collected, remove the telemetry data at the tail node and update the network status to proceed to step S4. S32. If the remaining storage flow table entry capacity of the tail node is equal to 0 and the service flow has not yet collected any telemetry data at the tail node, and all telemetry data of the tail node has been collected, then delete the tail node on the service flow path and repeat step S31 with the next node as the tail node. Otherwise, return to step S31 to select the next sorted service flow r+1 and collect telemetry data at its tail node.

4. The method according to claim 3, characterized in that: The constraints include: the remaining capacity of the service flow is greater than the size of the telemetry data to be collected; and the bandwidth capacity of the subsequent links does not exceed the capacity after adding the telemetry data to be collected.

5. The method according to claim 1, characterized in that: In step S6, the method for changing the telemetry information is as follows: First, calculate the FlowId of the service flow that collects the first telemetry data in the collected set. Then, based on the constraints, determine whether the telemetry data in the uncollected set can be replaced in order of sorting. If the replacement conditions are met, the telemetry information is replaced, the telemetry data in the collected set is reordered, and step S6 is repeated. If no telemetry information meets the replacement conditions, the next sorted telemetry data in the collected set will be used as the first telemetry data in the collected set, and step S6 will be repeated. For telemetry data with the same gain that are not collected, calculate the index of these telemetry data on the FlowId service flow path, sort them in descending order of the index, and determine whether they can be replaced based on the sorting.

6. The method according to claim 5, characterized in that: The constraints include: 1) the sum of the remaining capacity of the service flow and the size of the collected telemetry data is greater than or equal to the size of the uncollected telemetry data; 2) the bandwidth capacity of the subsequent link is not exceeded after the telemetry data is replaced; 3) the service flow collects other telemetry data at the node where the uncollected telemetry data is located or the remaining storage flow table capacity of that node is greater than 0.

Images