A node anomaly detection method in a constellation network based on component modules

Abstract

The application discloses a kind of node anomaly detection methods in constellation network based on component module, it includes the following steps S1, calculate the g component detectability of constellation network under mutual test model;S2, the topology of constellation network is constructed and is tested between nodes based on mutual test model to generate syndrome;S3, the weighted constellation generated by syndrome is generated by edge deletion operation Subgraph consisting of multiple components;S4, obtain all component modules of subgraph by breadth-first search algorithm;S5, the node number of each component module is calculated and component module is sorted according to node number descending order;S6, all nodes of the largest component module are detected as no anomaly and join no anomaly node set;S7, detect the remaining nodes in constellation network except no anomaly node set, finally determine no anomaly node set and anomaly node set.The application utilizes the network fault tolerance characteristics of constellation network, and the detection effect is better in large-scale processing machine system.

Description

Technical Field

[0001] This invention relates to the fields of processor anomaly detection and network reliability, and more particularly to a method for detecting node anomalies in a star map network based on component modules. Background Technology

[0002] With the increasing speed of information processing and the growing demand for high-performance computing, multiprocessor systems are entering the era of electronic computing. Simultaneously, as network scale continues to expand, network reliability faces severe challenges. In industrial and real-time systems, large-scale anomalies can cause the entire network to disconnect, resulting in numerous component failures. Therefore, the network's anomaly detection capability is closely related to the number of components the network can accommodate in practical applications. Currently, research on large-scale network component detection is still limited. Starmap, as a general-purpose interconnect network, has a topology suitable for large-scale processor systems and offers high communication efficiency and a certain degree of fault tolerance. Summary of the Invention

[0003] The purpose of this invention is to provide a node anomaly detection method in a star map network based on component modules. By combining a system-level mutual testing model with the fault-tolerant characteristics of the star map, the method detects abnormal nodes in the topology by finding the largest non-abnormal component in the network as the detector.

[0004] The technical solution adopted in this invention is:

[0005] A method for detecting node anomalies in a star map network based on component modules, comprising the following steps:

[0006] S1 computes the star map network S n The detectability ct of component g under the test model g (S n );

[0007] S2 constructs a star map network S n The topology is then used to generate syndrome δ by testing between nodes based on a mutual testing model.

[0008] S3 is from S n The weighted star graph generated from the syndrome δ is used to generate a subgraph δ(S) composed of multiple components through edge deletion operations. n );

[0009] S4 obtains the subgraph δ(S) using a breadth-first search algorithm. n All component modules Com1, Com2, ..., Com k ;

[0010] S5 calculates each component module Com1, Com2, ..., Com k The number of nodes N1, N2, ..., Nk The component modules are then sorted in descending order of the number of nodes, and the component module with the most nodes is defined as the largest component module. largest ;

[0011] S6 will be the largest component module Com largest All nodes are detected as anomalies and added to the set of anomaly-free nodes S. Ff ;

[0012] S7 detection star map network S n Except for the set of no abnormal nodes S Ff The remaining nodes, after which the set of nodes without anomalies, S, is finally determined. Ff and the abnormal node set S F .

[0013] Furthermore, the specific method for step S1 is as follows:

[0014] S101, the detectability ct of component g is calculated using the construction method. g (S n The upper bound of ).

[0015] S102, Prove the detectability ct of component g through indivisibility. g (S n The lower bound of ) is obtained, and thus when n≥4, the 2-component detection degree is ct2(S). n ) = 2n-3 and the detection degree of 3-components is ct3(S) n ) = 3n - 6;

[0016] Furthermore, the specific method for step S2 is as follows:

[0017] S201, generated by the edge connection rules of the star map network. n Randomly select CT g (S n ) abnormal nodes;

[0018] S202 is tested using a system-level mutual test model, targeting the StarMap network S. n Each node i in the network sends the same test information to its neighboring node j, and then obtains the information returned by node j; the mutual test results between nodes i and j are obtained and denoted by the symbol δ(i,j); the star map network S is obtained. n The test results of all nodes are used to generate the syndrome δ;

[0019] Furthermore, the specific method for step S3 is as follows:

[0020] S301, detection by StarMap Network S n The syndrome δ yielded all node test results;

[0021] S302, Check the star map network S n For each edge ij, delete the edge with a test result δ(i,j) = 1, thus generating a subgraph δ(S) composed of multiple components. n );

[0022] Furthermore, the specific method for step S4 is as follows:

[0023] S401, in subgraph δ(S n In the process of selecting any node j as the starting test node, starting from this node, visit its adjacent nodes in turn, and continue to visit along each path until all nodes with a connected path to node j have been visited. At this point, a component module C is obtained.

[0024] S402, after obtaining a component, determine the subgraph δ(S) n If no node has been visited, select an unvisited node as the starting node and repeat step S402 until all component modules Com1, Com2, ..., Com are obtained. k ;

[0025] Furthermore, the specific method for step S7 is as follows:

[0026] S701, if the size of the abnormal node set Size(s) is less than the detection rate ct g (S n If the nodes are consistent, then all remaining nodes are added to the set S of nodes without anomalies. Ff ;

[0027] S702, if the size of the abnormal node set Size(s) is less than the detection rate ct g (S n If there is a discrepancy, and the number of remaining undetected abnormal nodes is less than the size of the undetected components, repeat step S6 until all nodes have been detected; otherwise, if there are still isolated nodes remaining, add them to the abnormal node set S. F .

[0028] This invention employs the above technical solution. First, it proves the component detection rate of the star graph network. Then, it constructs the topology of the star graph network and performs tests between topology nodes based on a mutual testing model to generate syndromes. The weighted star graph generated from the syndromes generates a subgraph composed of multiple components through edge deletion. Then, it obtains all component modules of the subgraph using a breadth-first search algorithm, calculates the number of nodes in each component module, and sorts the component modules in descending order of the number of nodes. All nodes of the largest component module are detected as anomalies and added to the anomaly-free node set. The remaining nodes in the star graph network, excluding the anomaly-free node set, are then detected, ultimately determining the anomaly-free node set and the anomaly node set. In summary, this invention utilizes the star graph network S... n The study investigated the fault tolerance characteristics of networks and proposed a component anomaly detection method for multiprocessor networks, which showed good detection performance in large-scale processor systems. The node anomaly detection method for star map networks based on component modules provided in this invention is even more efficient. Attached Figure Description

[0029] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments;

[0030] Figure 1 This is a 4-dimensional star diagram (S4) of the network topology used in this invention.

[0031] Figure 2 This is a flowchart illustrating a multiprocessor network anomaly detection method based on star maps according to the present invention.

[0032] Figure 3 This is a schematic diagram showing the results of inter-node testing of the system-level mutual testing model used in this invention.

[0033] Figure 4 For a subgraph δ(S) of the present invention n ) Schematic diagram;

[0034] Figure 5 To obtain the subgraph δ(S) of this invention n The largest component module Com largest Schematic diagram. Implementation

[0035] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings.

[0036] like Figures 1 to 5As shown in the first example, this invention discloses a method for detecting anomalies in multiprocessor network components based on a star graph. In multiprocessor networks, as the network scale increases, network reliability faces severe challenges, requiring timely and rapid detection of potential anomalies. In a multiprocessor network with a star graph topology, individual processors are considered nodes in the graph, and communication connections between processors are considered edges between nodes. Processors exhibiting anomalies are considered anomalous nodes, while normal processors are considered normal nodes. To ensure stable and normal system operation, this invention proposes testing processor nodes using a system-level mutual testing model and detecting the processor system using a component detection algorithm to identify anomalous nodes within the system. Figure 1 As shown, the topology of the present invention is presented below.

[0037] An n-dimensional star map S n It is an (n-1)-dimensional regular network consisting of n! nodes and (n-1)n! / 2 edges. For example... Figure 1 As shown. Its nodes can be represented by a sequence of length n. That is, for S n For any point v = x1x2...x n |x i ∈{1, 2,...,n}, i=1, 2,...n, and x i ≠x j For i≠j, there are (n-1) neighbors.

[0038] In this invention, we divide the processor into three states: no abnormal nodes, abnormal nodes, and undetected nodes.

[0039] This invention proposes a method for anomaly detection of multiprocessor network components based on star graphs. First, a star graph network topology is constructed, and tests are performed between processor nodes using a system-level mutual testing model to generate a syndrome δ. Then, a subgraph δ(S) is generated based on the syndrome δ. n ); Obtain network δ(S) n The largest component module Com largest The nodes within the set are then identified as anomalies. Taking advantage of the star chart's fault-tolerant nature, the adjacent nodes of this anomaly-free maximal component are then checked. If the size of the anomaly node set Slze(s) is equal to the detection degree T, then all remaining nodes are added to the anomaly-free node set S. Ff Otherwise, if the number of remaining undetected abnormal nodes is less than the number of the first large component node that has not been operated after sorting, repeat the above steps until the anomaly detection of the entire multiprocessor network is completed.

[0040] like Figure 2The flowchart shown illustrates a method for detecting abnormal nodes in a multiprocessor network based on a star map, comprising the following steps:

[0041] S1 computes the star map network S n The detectability ct of component t under the test model. g (S n );

[0042] S2 constructs a star map network S n The topology is then used to generate syndrome δ by testing between nodes based on a mutual testing model.

[0043] S3 is from S n The weighted star graph generated from the syndrome δ is used to generate a subgraph δ(S) composed of multiple components through edge deletion operations. n );

[0044] S4 obtains the subgraph δ(S) using a breadth-first search algorithm. n All component modules Com1, Com2, ..., Com k ;

[0045] S5 calculates each component module Com1, Com2, ..., Com k The number of nodes N1, N2, ..., N k And sort the component modules in descending order by the number of nodes;

[0046] S6 will be the largest component module Com largest All nodes are detected as anomalies and added to the set of anomaly-free nodes S. Ff ;

[0047] S7 detection star map network S n Except for the set of no abnormal nodes S Ff The remaining nodes, after which the set of nodes without anomalies, S, is finally determined. Ff and the abnormal node set S F .

[0048] Furthermore, the specific method for step S1 is as follows:

[0049] S101, the detectability ct of component g is calculated using the construction method. g (S n The upper bound of ).

[0050] S102, Prove the detectability ct of component g through indivisibility. g (S n The lower bound of ) is obtained, and thus when n≥4, the 2-component detection degree is ct2(S). n ) = 2n-3 and the detection degree of 3-components is ct3(S) n) = 3n - 6;

[0051] like Figure 3 As shown, the specific method of step S2 is as follows:

[0052] S201, generated by the edge connection rules of the star map network. n Randomly select CT g (S n There are ) abnormal nodes such that t≤2n-3;

[0053] S202 is tested using a system-level mutual test model, targeting the StarMap network S. n Each node i in the network sends the same test information to its neighboring node j, and then obtains the information returned by node j; the mutual test results between nodes i and j are obtained and denoted by the symbol δ(i,j); the star map network S is obtained. n The test results of all nodes are used to generate the syndrome δ;

[0054] like Figure 4 As shown, the specific method for step S3 is as follows:

[0055] S301, detection by StarMap Network S n The syndrome δ yielded all node test results;

[0056] S302, Check the star map network S n For each edge ij, delete the edge with a test result δ(i,j) = 1, thus generating a subgraph δ(S) composed of multiple components. n );

[0057] Furthermore, the specific method for step S4 is as follows:

[0058] S401, in subgraph δ(S n In the process of selecting any node j as the starting test node, starting from this node, visit its adjacent nodes in turn, and continue to visit along each path until all nodes with a connected path to node j have been visited. At this point, a component module C is obtained.

[0059] S402, after obtaining a component, determine the subgraph δ(S) n If no node has been visited, select an unvisited node as the starting node and repeat step 4-2 until all component modules Com1, Com2, ..., Com are obtained. k ;

[0060] like Figure 5 As shown, the specific method for step S7 is as follows:

[0061] S701, if the size of the abnormal node set Size(s) is less than the detection rate ct g (S n If the nodes are consistent, then all remaining nodes are added to the set S of nodes without anomalies. Ff ;

[0062] S702, if the size of the abnormal node set Size(s) is less than the detection rate ct g (S n If there is a discrepancy, and the number of remaining undetected abnormal nodes is less than the size of the undetected components, repeat step S6 until all nodes have been detected; otherwise, if there are still isolated nodes remaining, add them to the abnormal node set S. F .

[0063] In addition, the performance of the star map component anomaly detection algorithm is measured through index analysis. The specific method is as follows:

[0064] (1) Give the definition of true positive rate, which is the ratio of the number of nodes correctly detected as anomalous nodes to the total number of true anomalous nodes;

[0065] (1.1) Assume the total number of nodes is m, the number of nodes correctly detected as abnormal is p1, the number of nodes correctly detected as normal is p2, the number of nodes misdiagnosed as abnormal is q1, and the number of nodes misdiagnosed as normal is q2.

[0066] (1.2) Based on the definition of true positive rate, the true positive rate of the algorithm is obtained.

[0067] (1.3) It can be concluded that the true positive rate of the 4-dimensional star map network remains between 0.8 and 1. When the number of abnormal nodes does not exceed 5, all abnormal nodes can be detected. As the number of abnormal nodes increases, the TPR is still above 0.8.

[0068] (2) The true negative rate is defined as the ratio of the number of nodes that are correctly detected as non-abnormal nodes to the total number of true non-abnormal nodes.

[0069] (2.1) Based on the definition of true negative rate, the true positive rate of the algorithm is obtained.

[0070] (2.2) It can be observed that as the number of abnormal nodes increases, the true negative rate of the algorithm decreases.

[0071] (3) The definition of accuracy is given, which is the ratio of the sum of the number of correctly detected nodes to the sum of the number of all nodes.

[0072] (3.1) Based on the definition of accuracy, the accuracy of the algorithm is calculated to be...

[0073] (3.2) When the number of outliers in the 4-dimensional star graph network is less than 20% of the total number of nodes, the accuracy is always 1.

[0074] This invention employs the above technical solution, firstly constructing a star map network topology, and then generating a syndrome δ by testing between processor nodes through a system-level mutual testing model; then, generating a subgraph δ(S) based on the syndrome δ. n ); Obtain network δ(S) n The largest component module Com largest And detect the nodes within it as anomalies; combining the fault-tolerant characteristics of the star map, use this anomaly-free maximal component to detect its adjacent nodes. If the size of the anomaly node set Size(s) is less than the detection degree ct g (S n If the nodes are consistent, then all remaining nodes are added to the set S of nodes without anomalies. Ff Otherwise, if the size of the abnormal node set Size(s) is less than the detection rate ct g (S n If there is a discrepancy, and the number of remaining undetected abnormal nodes is less than the size of the undetected components, repeat the above steps until all nodes have been detected; otherwise, if there are still isolated nodes remaining, add them to the abnormal node set S. F .

[0075] The effective effects of this invention are: 1. It studies the network topology of star diagrams, which is suitable for multiprocessor networks, and its fault-tolerant characteristics, thereby proving its component detection capability; 2. It utilizes star diagram networks S... n This paper studies a component anomaly detection method for multiprocessor networks, considering the network topology and fault tolerance characteristics. Analysis and measurement of indicators show that the node anomaly detection algorithm based on component modules in star-map networks has better detection performance and higher efficiency. Obviously, the described embodiments are only some, not all, of the embodiments in this application. Without conflict, the embodiments and features in the embodiments of this application can be combined with each other. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the detailed description of the embodiments of this application is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

Claims

1. A method for detecting node anomalies in a star map network based on component modules, characterized in that: It includes the following steps: S1, Calculate the star map network Under the mutual testing model Component inspection First, calculate using the construction method. Component detectability The upper bound is then proven through indistinguishability. Component detectability The lower bound, and thus the current At that time, the detection rate of 2-components was And the 3-component detection rate is ; S2, Constructing a star map network The topology is used, and inter-node testing is performed using a mutual testing model to generate syndromes. ; S3, by syndrome The generated weighted star graph, after deleting edges containing test results, produces a subgraph composed of multiple components. ; S4, obtain the subgraph using a breadth-first search algorithm. All component modules , This represents the total number of component modules; S5, calculate each component module separately. Number of nodes The component modules are then sorted in descending order of the number of nodes, and the component module with the most nodes is defined as the largest component module. ; S6, the largest component module All nodes are detected as anomalies and added to the set of anomaly-free nodes. ; S7, detecting star map networks Except for the set of nodes without anomalies The remaining nodes, if the size of the abnormal node set... With detection degree If they are consistent, then add all remaining nodes to the set of nodes without anomalies. If the size of the abnormal node set With detection degree If there is a discrepancy, if the number of remaining undetected anomalous nodes is less than the size of the undetected components, repeat step S6 until all nodes are detected. If the number of remaining undetected anomalous nodes is not less than the size of the undetected components and there are still isolated nodes remaining, add the remaining isolated nodes to the anomalous node set. ; Finally, the set of nodes without anomalies was determined. and abnormal node set .

2. The node anomaly detection method in a star map network based on component modules according to claim 1, characterized in that: The specific steps of step S2 are as follows: S201 is generated using the edge connection rules of the star map network. Random selection One abnormal node, S202 is tested using a system-level mutual test model, targeting the StarMap network. Each node in to its adjacent nodes Send the same test information, then obtain the node. The returned information; the node is obtained. and The mutual test results, and using symbols Indicates; obtaining star map network The test results of all nodes generate a syndrome. .

3. The node anomaly detection method in a star map network based on component modules according to claim 1, characterized in that: The specific steps of step S3 are as follows: S301, detection by StarMap Network syndrome All node test results obtained; S302, Check the star map network Each edge Delete the existing test results The edges are used to generate a subgraph consisting of multiple components. .

4. The node anomaly detection method in a star map network based on component modules according to claim 1, characterized in that: The specific steps of step S4 are as follows: S401, in the subgraph Choose any node As the starting test node; starting from the starting test node, visit its adjacent nodes in sequence, following each path downwards, until all nodes are connected. All nodes with connected paths were visited, resulting in a component module. ; S402, a component module is obtained. Then, determine the subgraph. Check if any nodes have been unvisited; if not, select an unvisited node as the starting node; repeat step S402 until all component modules are obtained. .

Images