Two-stage regular network fault diagnosis method

Abstract

The invention relates to the technical field of fault diagnosis, in particular to a two-stage regular network fault diagnosis method, which comprises the following steps of: inputting a regular graph, a symptom list and t fault nodes, and initializing variables; traversing each node in the regular graph, and initializing an index; traversing the neighbor node pair, detecting a test result, and updating the global index of the neighbor node pair and the global index of the current node; sorting in a descending order based on the occurrence frequency of the nodes, and selecting the first t nodes as a basic fault node candidate set based on a sorting list; detecting whether the sorting list is non-empty, and if not, expanding candidates of the fault node; obtaining a threshold count, expanding candidates of fault nodes to obtain an expanded fault node candidate set, and combining the expanded fault node candidate set with the basic fault node candidate set to obtain a final fault node set; according to the method, in the hypercube network, the recognition accuracy rate and the recognition precision rate of the fault nodes both exceed 99%.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis, and in particular to a two-stage regular network fault diagnosis method. Background Art

[0002] With the rapid development of high-performance computing, cloud computing, and edge computing, the scale of multiprocessor systems continues to expand, with node counts reaching millions or even more. However, this increased system size not only leads to higher hardware failure rates but also makes them a prime target for cyber attackers. Attackers can disrupt the normal operation of processor nodes through malicious code injection, distributed denial of service (DDoS) attacks, side-channel attacks, and other means. They can even tamper with test results from diagnostic systems, resulting in false fault reports or missed detection of real faults. For example, in 2022, NVIDIA's GPU driver supply chain was attacked, causing multiple AI training servers to output erroneous results. Some nodes were mistakenly diagnosed as hardware failures due to memory overflow. In 2023, Microsoft Exchange Server was attacked by an APT, affecting over 60,000 servers worldwide. Some systems experienced processor overload and crashes due to system resource exhaustion. These incidents demonstrate that fault diagnosis in modern multiprocessor systems must simultaneously address hardware failures and malicious attacks, making improving system reliability and robustness a research focus in this field.

[0003] In recent years, researchers have proposed a variety of improved fault diagnosis metrics, such as conditional diagnosability, h-extra diagnosability, and r-component diagnosability. However, these methods still have limitations in ultra-large-scale systems, especially when multiple branch connectivity and node number conditions need to be met simultaneously, resulting in a significant decrease in diagnostic efficiency and accuracy. Summary of the Invention

[0004] The purpose of the present invention is to provide a two-stage regular network fault diagnosis method, which aims to solve the problem that the existing fault diagnosis methods in regular networks have insufficient diagnostic accuracy and cannot efficiently identify large-scale fault nodes.

[0005] To achieve the above object, the present invention provides a two-stage regular network fault diagnosis method, comprising the following steps:

[0006] Input a regular graph, a symptom list and t fault nodes, and initialize the variables;

[0007] Traverse each node in the regular graph, obtain a list of all neighbor nodes of the current node, and initialize the index;

[0008] Traversing the neighbor node pairs, detecting test results, and updating the global index of the neighbor node pairs and the global index of the current node;

[0009] Sort the nodes in descending order of frequency, generate a sorted list, and then select the first t nodes as the basic fault node candidate set based on the sorted list;

[0010] Check whether the sorted list is not empty, and if so, expand the candidates for the faulty node;

[0011] Recording the frequency of occurrence of the t-th node in the sorted list as a threshold count, expanding the candidates of the faulty nodes to obtain an expanded faulty node candidate set, and merging the basic faulty node candidate set and the expanded faulty node candidate set to obtain a final faulty node set;

[0012] The regular graph is subtracted from the final set of faulty nodes to obtain a set of fault-free nodes, and the final set of faulty nodes and the set of fault-free nodes are output.

[0013] The regular graph includes fault-free nodes and faulty nodes.

[0014] The initialization variables include initializing the fault node set, the non-fault node set, the basic fault node candidate set, and the extended fault node candidate set to be empty, and initializing the node occurrence frequency counter to 0.

[0015] Among them, the initialization index includes initializing the global index of the current node, that is, the global index of the current node is 0, and traversing each node in the regular graph; initializing the global index of the neighbor node pair of the current node, that is, the global index of the neighbor node pair is 0, and traversing all neighbor pairs of the current node.

[0016] The specific method of expanding the candidates for the faulty node is as follows:

[0017] Starting from the t+1th node in the sorted list, judgment is performed. If the frequency of occurrence of the node is the same as the threshold count, the node can be regarded as an extended fault node and added to the extended fault node candidate set until the frequency of occurrence of subsequent nodes is different from the threshold count.

[0018] The present invention provides a two-stage regular network fault diagnosis method, which inputs a regular graph, a symptom list and t fault nodes, and initializes variables; traverses each node in the regular graph, obtains a list of all neighbor nodes of the current node, and initializes the index; traverses the neighbor node pairs, detects test results, updates the global index of the neighbor node pairs and the global index of the current node; sorts the nodes in descending order based on their frequency of occurrence, generates a sorted list, and then selects the first t nodes as a basic fault node candidate set based on the sorted list; detects whether the sorted list is non-empty, and if not, expands the candidate set of the fault node; records the frequency of the t-th node in the sorted list as a threshold count, and then expands the candidate set of the fault node. Expand the candidates of the faulty nodes to obtain an extended faulty node candidate set, merge the basic faulty node candidate set and the extended faulty node candidate set to obtain a final faulty node set; subtract the regular graph from the final faulty node set to obtain a fault-free node set, output the final faulty node set and the fault-free node set. In this method, in a hypercube network, the algorithm has an accuracy (Accuracy) and precision (Precision) of more than 99% for identifying faulty nodes. The algorithm has low time complexity and is suitable for large-scale networks. It can be extended to other Internet topologies (such as star networks, cube networks, etc.) and actual network systems (such as power system networks). BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0020] Figure 1 Schematic diagram of test results under the MM* model;

[0021] Figure 2 Schematic diagram of the flow of the two-stage regular network fault diagnosis algorithm based on the MM* model of the present invention;

[0022] Figure 3 This is a diagram of the network topology hypercube network Q4 used in the present invention;

[0023] Figure 4 Schematic diagram of random node failures in the hypercube network Q4;

[0024] Figure 5 A schematic diagram of a basic fault node candidate set base_nodes and an extended fault node candidate set extended_nodes obtained by the present invention;

[0025] Figure 6 Schematic diagram of the faulty node set M and the non-faulty node set H finally diagnosed by the present invention, wherein (A) is a schematic diagram of the faulty node set M, and (B) is a schematic diagram of the non-faulty node set H.

[0026] Figure 7 The present invention provides a flow chart of a two-stage regular network fault diagnosis method. DETAILED DESCRIPTION

[0027] The following describes embodiments of the present invention in detail, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and are not to be construed as limiting the present invention.

[0028] See also Figures 1 to 7 ,The present invention provides a two-stage regular network fault diagnosis method, comprising the following steps;

[0029] S1 inputs a regular graph, a symptom list, and t fault nodes, and initializes variables;

[0030] In the embodiment of the present invention, the input regular graph G is a hypercube network Q n For example, the hypercube network Q n The topology of is defined as follows.

[0031] An n-dimensional hypercube network Q n It is a 2 n nodes and n2 n-1 An n-dimensional regular network consisting of edges. Among them, the node set V(Q n )={s1s2…s n ∣s i ∈{0,1},i=1,2,…,n};edge set E(Q n )={(s,y)|s=s1s2…s n ,y1y2…y n , like Figure 3 As shown in the figure, the structure of the hypercube network Q4 is given. Each node in the hypercube network Q4 has 4 neighbors, where 4 is the regularity of the hypercube. Figure 4 Given is a schematic diagram of the probability of random node failure in the hypercube network Q4.

[0032] The specific method of identifying the faulty node is: nEach node w in the MM* model is diagnosed, that is, the node w is required to send the same task to any two of its adjacent neighbor nodes u and v, and then compare their feedback. The test node w compares the feedback results of the tested nodes u and v, and uses the symbol σ((u,v) w ) indicates that when nodes w, u and v are all fault-free, the feedback results of nodes u and v are the same, and the test result of test node w is σ((u,v) w )=0; when the test node w is fault-free and at least one of the tested nodes u and v is faulty, the feedback results of nodes u and v are different. In this case, the test result displayed by the test node w is σ((u,v) w )=1; when the test node w is faulty, no matter whether the tested node is fault-free or faulty, the test result of the test node w is unreliable, that is, σ((u,v) w )=0orσ((u,v) w )=1.

[0033] Symptom List SL: The results of the MM* model test on the hypercube network Q n The test results of each node in the symptom set are called symptom sets. In this application, the symptom list SL is used to store the test results, such as SL(w_index,(u,v)_index)=0, SL(w_index,(b,i)_index)=1), where w_index and (u,v)_index are test indexes.

[0034] t faulty nodes: Hypercube network Q n The actual number of faulty nodes in is compared with the final output fault point set M.

[0035] Initialize the fault node set M, the fault-free node set H, the basic fault node candidate set base_nodes, and the extended fault node candidate set extended_nodes to be empty; wherein, the fault node set M = base_nodes ∪ extended_nodes, the fault-free node set H = V(Q n )-M;

[0036] Initialize the node appearance frequency counter node_appearance_counts to 0, and use it to count the frequencies of nodes u and v appearing in the neighbor pair when the test result SL(w_index,(u,v)_index)=1 under the MM* model.

[0037] S2 traverses each node in the regular graph, obtains a list of all neighbor nodes of the current node, and initializes the index;

[0038] In the embodiment of the present invention, traversing the hypercube network Q n For each node w in , get the list of all neighbor nodes of the current node w, and initialize the index (including the current node index w_index and the current node neighbor pair index (u,v)_index);

[0039] Initialize the index w_index of the current node w, that is, w_index = 0, and traverse the hypercube network Q n Each node in

[0040] Initialize the neighbor pair (u, v)_index of the current node w, that is, (u, v)_index = 0, and traverse all neighbor pairs of the current node w.

[0041] S3 traverses the neighbor node pair, detects the test results, and updates the global index of the neighbor node pair and the global index of the current node;

[0042] In the embodiment of the present invention, all neighbor pairs (u, v) of the current node w are traversed, and the test results are checked. If the test result SL(w_index, (u, v)_index) = 1 under the MM* model, the frequencies of occurrence of nodes u and v are both increased by 1, the counters node_appearance_counts of nodes u and v are updated respectively, the index (u, v)_index is updated, and the next pair of neighbor nodes of the current node is tested until all neighbor pairs of the current node are traversed; the index w_index is updated, and all neighbor pairs of the next node are tested until all the neighbor pairs of the hypercube network Q are traversed. n Each node in .

[0043] S4 sorts the nodes in descending order of frequency of occurrence to generate a sorted list, and then selects the first t nodes as a candidate set of basic fault nodes based on the sorted list;

[0044] In the embodiment of the present invention, the hypercube network Q is counted according to the node appearance frequency in the node counter node_appearance_counts. n Sort each node in descending order to obtain the sorted_nodes list. Select the first t nodes with the highest node occurrence frequency from sorted_nodes as the basic fault node candidate set base_nodes.

[0045] S5 detects whether the sorted list is not empty, and if so, expands the candidates for the faulty node;

[0046] In the embodiment of the present invention, it is determined whether the sorted_nodes list is not empty. If it is not empty, the fault node is expanded.

[0047] S6 records the frequency of occurrence of the t-th node in the sorted list as a threshold count, expands the candidates of the faulty nodes to obtain an expanded faulty node candidate set, and merges the basic faulty node candidate set and the expanded faulty node candidate set to obtain a final faulty node set;

[0048] In the embodiment of the present invention, a threshold count is obtained, that is, the frequency of occurrence of the t-th node in the sorted_nodes list, and is recorded as the threshold count.

[0049] The specific method of obtaining a basic fault node candidate set base_nodes and an extended fault node candidate set extended_nodes is:

[0050] Traversing the hypercube network Q n For each node w in the network, all neighbor pairs of the current node w are tested. If there is a neighbor pair (u, v), and the test result SL(w_index,(u,v)_index)=1 under the MM* model, then the frequencies of nodes u and v appearing in the neighbor pair are increased by 1, that is, the counters node_appearance_counts of nodes u and v are updated respectively, until all neighbor pairs of the current node are traversed, and then all neighbor pairs of the next node are traversed, thus traversing the entire hypercube network Q. n All nodes in (wherein, all neighbor pairs of node w are explained: taking the hypercube network Q4 as an example, its regularity is 4, then node w has 4 neighbors, set as a, b, c, d, then the neighbor pairs of node w are {(a, b), (a, c), (a, d), (b, c), (b, d), (c, d)}). Then sort the nodes in descending order according to their frequency of occurrence from high to low to obtain a sorted list sorted_nodes, and add the first t nodes in the sorted list to the basic fault node candidate set base_nodes. In order to prevent misdiagnosis, it is necessary to further determine whether there is an extended node, that is, starting from the t+1th node in the sorted list, whether the frequency of subsequent nodes is the same as that of the tth node (that is, whether the frequency of subsequent nodes is the same as the threshold count). If the same, these nodes are added to the extended fault node candidate set extended_nodes.

[0051] Starting from the t+1th node in the sorted_nodes list, the evaluation is performed. If the frequency of this node is equal to the threshold count, it is considered an extended fault node and added to extended_nodes. This continues until the frequency of subsequent nodes is not equal to the threshold count. The basic fault node candidate set and the extended fault node candidate set are merged to obtain the final fault node set.

[0052] S7: Subtract the regular graph from the final set of faulty nodes to obtain a set of non-faulty nodes, and output the final set of faulty nodes and the set of non-faulty nodes.

[0053] In the embodiment of the present invention, the hypercube network Q n The difference between the final set of faulty nodes and the set of non-faulty nodes is obtained, that is, H = V (Q n )-M, output the final result, the faulty node set M and the non-faulty node set H.

[0054] The specific method to obtain the final fault point set M and the fault-free point set H is:

[0055] The fault point set M is obtained by combining the basic fault node candidate set base_nodes with the extended fault node candidate set extended_nodes; the fault-free point set H is the entire hypercube network Q n The nodes in minus the set of fault points are obtained.

[0056] The above disclosure is merely a preferred embodiment of a two-stage regular network fault diagnosis method of the present invention. It is certainly not intended to limit the scope of the present invention. A person skilled in the art will understand that implementing all or part of the processes of the above embodiment and making equivalent changes according to the claims of the present invention still fall within the scope of the invention.

Claims

1. A two-stage regular network fault diagnosis method, characterized in that: The following steps are included: Input a regular graph, a symptom list and t fault nodes, and initialize the variables; Traverse each node in the regular graph, obtain a list of all neighbor nodes of the current node, and initialize the index; Traversing the neighbor node pairs, detecting test results, and updating the global index of the neighbor node pairs and the global index of the current node; Sort the nodes in descending order of frequency, generate a sorted list, and then select the first t nodes as the basic fault node candidate set based on the sorted list; Check whether the sorted list is not empty, and if so, expand the candidates for the faulty node; Recording the frequency of occurrence of the t-th node in the sorted list as a threshold count, expanding the candidates of the faulty nodes to obtain an expanded faulty node candidate set, and merging the basic faulty node candidate set and the expanded faulty node candidate set to obtain a final faulty node set; The regular graph is subtracted from the final set of faulty nodes to obtain a set of fault-free nodes, and the final set of faulty nodes and the set of fault-free nodes are output.

2. The two-stage regular network fault diagnosis method according to claim 1, characterized in that ； The regular graph includes non-faulty nodes and faulty nodes.

3. The two-stage regular network fault diagnosis method according to claim 1, characterized in that ； The initialization variables include initializing the fault node set, the non-fault node set, the basic fault node candidate set, and the extended fault node candidate set to be empty, and initializing the node occurrence frequency counter to 0.

4. The two-stage regular network fault diagnosis method according to claim 1, characterized in that ； The initialization index includes initializing the global index of the current node, that is, the global index of the current node is 0, and traversing each node in the regular graph; initializing the global index of the neighbor node pair of the current node, that is, the global index of the neighbor node pair is 0, and traversing all neighbor pairs of the current node.

5. The two-stage regular network fault diagnosis method according to claim 1, It is characterized by: The specific method of expanding the candidates of the faulty node is as follows: Starting from the t+1th node in the sorted list, judgment is performed. If the frequency of occurrence of the node is the same as the threshold count, the node can be regarded as an extended fault node and added to the extended fault node candidate set until the frequency of occurrence of subsequent nodes is different from the threshold count.

Images