A data center network fault diagnosis method based on th strategy

Abstract

The present invention relates to the technical field of network fault diagnosis, and in particular to a data center network fault diagnosis method based on a th strategy, the method comprising: taking any node in a data center network structure as an initial point, traversing all nodes in the data center network structure, obtaining a first set and a second set by comparing the node with the initial point; obtaining a first set with the largest number of nodes based on the data center network structure, if the number of nodes in the first set with the largest number of nodes meets a preset condition, the first set with the largest number of nodes is a non-fault point set, and the second set is a fault point set; obtaining an uncertain set according to the fault point set and the non-fault point set, obtaining all fault nodes in the uncertain set based on the number of nodes in the uncertain set, reducing the error of fault node diagnosis, ensuring accuracy and improving efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of network fault diagnosis, and in particular to a data center network fault diagnosis method based on a TH strategy. Background Art

[0002] As a key infrastructure for cloud computing, data center networks have been built to provide increasingly popular online application services such as GFS, Bigtable, and Dryad; in distributed systems, data center networks build a bridge connecting data centers and a large number of servers. As the scale of data center networks gradually increases, failures of nodes and links in the network are inevitable, which has aroused people's attention to network reliability and availability. How to correctly identify and eliminate faulty nodes in the network when a fault has occurred in the network so that the network can still operate normally is the focus of network reliability research.

[0003] The key technology to solve this problem is fault diagnosis, that is, identifying the faulty nodes in the network. System-level fault diagnosis is to use the computing, communication and processing capabilities of the nodes in the system to implement system self-diagnosis. The basis of system-level fault diagnosis is the construction of the diagnostic model. Diagnostic degree is the maximum number of faulty processors that the system can guarantee to be diagnosed irreplaceably, and it plays an important role in measuring the reliability and fault tolerance of the network. The diagnostic degree problem is widely used in related problems such as the design and analysis of large-scale parallel computer systems and very large-scale integrated circuits. The traditional diagnostic degree assumes that all adjacent processors of any processor can fail at the same time. However, in a large multi-processor system, the probability that all adjacent processors of any processor fail at the same time is very small. For this reason, people have extended the traditional diagnostic degree concept and proposed conditional diagnostic degree, h-extra conditional diagnostic degree and t / h diagnostic degree. These diagnostic degrees under different conditions meet the needs of efficient fault diagnosis of data center networks under different conditions and can diagnose more faulty nodes.

[0004] However, network fault diagnosis algorithms based on the t / h strategy often miss faulty nodes, fail to detect all faulty nodes, and detect non-optimal paths, resulting in low efficiency and poor fault detection results. Summary of the invention

[0005] In order to solve the above-mentioned problems of poor fault detection effect and low efficiency, the purpose of the present invention is to provide a data center network fault diagnosis method based on th strategy, and the technical solution adopted is as follows:

[0006] An embodiment of the present invention provides a data center network fault diagnosis method based on th strategy, the method comprising the following steps:

[0007] Taking any node in the data center network structure as the initial point, traversing all nodes in the data center network structure, and obtaining a first set and a second set by comparing the node with the initial point;

[0008] Based on the data center network structure, a first set with the largest number of nodes is obtained. If the number of nodes in the first set with the largest number of nodes meets a preset condition, the first set with the largest number of nodes is a non-fault point set, and the second set is a fault point set;

[0009] An uncertain set is obtained according to the fault point set and the non-fault point set, and all fault nodes in the uncertain set are obtained based on the number of nodes in the uncertain set.

[0010] Preferably, the step of obtaining the first set and the second set by comparing the node with the initial point comprises:

[0011] Obtain the adjacent points of the traversed node, and use MM* to perform symptom comparison on the initial point, the traversed node, and the adjacent points. If the result is 0, the initial point, the traversed node, and the adjacent points are in the first set;

[0012] If the result is 1, the initial point, the traversed nodes, and the adjacent points are placed in the second set.

[0013] Preferably, the step of acquiring the first set with the largest number of nodes based on the data center network structure includes:

[0014] A plurality of first sets are obtained by traversing any node in the data center network structure, and the number of nodes in each first set is counted to select the first set including the largest number of nodes.

[0015] Preferably, the preset condition is: the number of nodes in the first set with the largest number of nodes is greater than (Nth), N represents the number of all nodes in the data center network structure; t and h are constants in the t / h strategy respectively.

[0016] Preferably, the step of obtaining an uncertain set according to the fault point set and the non-fault point set includes:

[0017] Select nodes on the data center network structure that are neither in the fault point set nor in the non-fault point set, and all selected nodes constitute an uncertain set.

[0018] Preferably, the step of obtaining all faulty nodes in the uncertain set based on the number of nodes in the uncertain set includes:

[0019] If the number of nodes in the uncertain set is greater than h, all nodes in the uncertain set are fault-free nodes;

[0020] If the number of nodes in the fault point set is t-1, and the number of nodes in the uncertain set is greater than 0, randomly select a node a in the uncertain set to compare symptoms with the other two nodes b and node c. If the result is 1, then all nodes in the uncertain set except node a, node b and node c are fault-free nodes;

[0021] Select the adjacent point w of node b in the uncertain set, and compare the symptoms of node b with those of node a and node w. If the result is 0, node c is a faulty node, and all nodes in the uncertain set except node c are non-faulty nodes.

[0022] The present invention has the following beneficial effects: in the embodiment of the present invention, by traversing all nodes on the data center network structure, all nodes are preliminarily divided into two sets, namely a first set and a second set, according to the comparison results, and the first set with the largest number of nodes in all the first sets is selected. If the number of nodes in the selected first set meets the preset conditions, the first set selected at this time is recorded as a non-fault point set, and its corresponding second set is a fault point set, so as to complete the preliminary division of all nodes and reduce the time of the diagnosis and analysis process; further, an uncertain set on the data center network structure is obtained through the non-fault point set and the fault point set, and the number of nodes in the uncertain set is counted, and all fault nodes are obtained by analyzing and diagnosing again according to the number of nodes in the uncertain set, thereby improving the efficiency of the analysis process, reducing the error of fault node diagnosis, and having higher accuracy.

[0023] In addition, when diagnosing faulty nodes through the number of nodes in an uncertain set, the embodiments of the present invention can quickly and accurately locate the faulty nodes through symptom comparison results between nodes in the uncertain set, thereby improving the robustness and fault tolerance of the network. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. 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 creative work.

[0025] Figure 1 A flow chart of a data center network fault diagnosis method based on th strategy provided by an embodiment of the present invention;

[0026] Figure 2 A schematic diagram of a BCCC (3,2) structure provided by an embodiment of the present invention;

[0027] Figure 3 A schematic diagram of a BCCC (2,2) structure provided by an embodiment of the present invention;

[0028] Figure 4 A schematic diagram of the maximum connected branch of a BCCC (2,2) structure provided by one embodiment of the present invention;

[0029] Figure 5 A schematic diagram of the maximum connected branch of a BCCC (5, 2) structure provided by one embodiment of the present invention;

[0030] Figure 6 All fault nodes obtained by breadth-first search of a BCCC (5,2) structure provided by an embodiment of the present invention;

[0031] Figure 7 All fault nodes obtained by calling the UDiag(C,UC,C) method using a BCCC(5,2) structure provided by an embodiment of the present invention;

[0032] Figure 8 An average execution time corresponding to different options provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0033] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the specific implementation method, structure, features and effects of a data center network fault diagnosis method based on th strategy proposed by the present invention in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures, or characteristics in one or more embodiments may be combined in any suitable form.

[0034] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0035] The following is a detailed description of a specific solution of a data center network fault diagnosis method based on th strategy provided by the present invention in conjunction with the accompanying drawings.

[0036] See also Figure 1 , which shows a flow chart of a data center network fault diagnosis method based on th strategy provided by an embodiment of the present invention, the method comprising the following steps:

[0037] Step S100, taking any node in the data center network structure as an initial point, traversing all nodes in the data center network structure, and obtaining a first set and a second set by comparing the node with the initial point.

[0038] As a widely used data center network structure, BCCC has many excellent properties, such as using cheap commodity switches and only requiring two network interface card ports on commodity servers, etc., and the data center network BCCC has many ideal characteristics, such as high scalability, nearly equal-length parallel paths and small diameter; in BCCC, the basic unit is an element, and n servers connected to an n-port switch constitute an element; BCCCs of different dimensions are denoted as BCCC(n,k), k represents the dimension, and n represents the number of servers connected to each element through the switch; then for BCCC(n,0), it is simply constructed by one element and n switches.

[0039] For a BCCC (n, k), it is composed of n BCCC (n, k-1) and n k Elements are constructed to build a BCCC (n, k) that requires dual-port servers (k+1)n k+1 A-type n-port switches (k+1)n k and a B-type (k+1)-port switch n k+1 For details, please refer to Figure 2 , which shows a schematic diagram of a BCCC (3,2) structure.

[0040] The commonly used diagnostic model is generally the MM* model, which is a comparison-based model. Each node must test any pair of its adjacent nodes. The comparison task of system G is generally modeled as a multigraph, which is denoted as M(V(G), L), where L is a marked edge set. A marked edge (u, v)w∈L means that the results obtained by nodes u and v after receiving task w are compared at the same time, that is, uw, vw∈E(G). The set of all comparison results in M(V(G), L) is denoted as a comparison symptom of system G, represented by the symbol σ. If the result returned by comparing (x, y)z is a match, then σ(x, y)z=0; otherwise, if the returned result does not match, then σ(x, y)z=1. As shown in Table 1 below, it is all possible comparison results under the MM* model:

[0041] Table 1

[0042] Compare the states of node z The status of the compared nodes x, y Test results σ(i,y)z No trouble No faults 0 No trouble At least one fault exists 1 Fault Any 0 / 1

[0043] On the basis of the existing diagnostic model, it is necessary to select a diagnostic strategy to diagnose the data center network. The present disclosure uses the t / h diagnostic strategy; the t / h diagnostic strategy means that the system G generates a test symptom in the presence of a fault set F, wherein the number of faulty nodes does not exceed t, and all faulty nodes can be separated into a set F′ containing at most h fault-free nodes, and |F′|≤|F|+h; past practice has proved that when the number of misdiagnosed fault-free nodes is small, the t / h diagnostic strategy can greatly improve the self-diagnosis capability of the data center network, but the faulty node diagnosis based on the t / h diagnostic strategy has the problems of node diagnosis errors and high time complexity. Therefore, this embodiment optimizes the t / h diagnosis process so that it can quickly and accurately locate the faulty server and improve the robustness and fault tolerance of the data center network.

[0044] Specifically, take any node in the BCCC structure as the initial point, use the BFS breadth-first search algorithm to traverse the nodes, record the initial point as x, record the traversed node as y, select node y and an adjacent point z of the initial point x, and the adjacent point z is not the same node as node y; use the above-mentioned MM* model to compare the symptoms of these three nodes, if σ(y,z)x=0, put the nodes into the same set, and record this set as C; otherwise, if σ(y,z)x=1, put the nodes into another set, and record this set as UC; if the initial point x is a non-faulty node, then the corresponding set C contains all non-faulty nodes, and the set UC contains all faulty nodes; otherwise, if the initial point x is a faulty node, then the set C contains all faulty nodes, and the set UC contains all non-faulty nodes; in order to facilitate the subsequent diagnosis of the faulty node, the step of traversing this step to obtain the set is recorded as C-UC(u).

[0045] Step S200, based on the data center network structure, obtain the first set with the largest number of nodes. If the number of nodes in the first set with the largest number of nodes meets the preset conditions, the first set with the largest number of nodes is a non-fault point set, and the second set is a fault point set.

[0046] By traversing all nodes in the BCCC structure in step S100, two sets are obtained, namely set C and set UC; according to the BCCC construction structure, it can be obtained that when the fault set |F| of the faulty node is ≤(h+1)(k-1)+n-1, there is a largest connected branch in the BCCC structure that contains at least N-|F|-h nodes, and N represents the number of all nodes on the BCCC structure.

[0047] Call step C-UC(u) from any node on the BCCC structure. According to the traversal comparison of C-UC(u), the set C with the largest number of nodes can be obtained. If the number of nodes in the set C with the largest number of nodes is greater than (Nth), then it can be determined that set C is the largest connected branch in the BCCC structure, and set UC is the fault node set, so as to determine the fault point set and the non-fault point set.

[0048] Step S300, obtaining an uncertain set according to the fault point set and the non-fault point set, and obtaining all fault nodes in the uncertain set based on the number of nodes in the uncertain set.

[0049] It can be seen from the t / h diagnosis strategy that at most h fault-free nodes will be diagnosed as faulty nodes; construct an uncertain set U, which is composed of all nodes on the BCCC structure that do not belong to set C and set UC, that is, U = NC-UC; count the number of nodes in the uncertain set U, if the number of nodes in the uncertain set U is greater than h, then all nodes in the uncertain set U are fault-free nodes; if the number of nodes in set UC is t-1, and the number of nodes in the uncertain set U is greater than 0, then call the UDiag method, which means that a node a is arbitrarily selected in the uncertain set U to test the other two nodes b and c. If the diagnosis result is 1, it means that one of the nodes a, b and c is a faulty node, and at this time, in the uncertain set U, all nodes except node a, b and c are fault-free nodes.

[0050] Furthermore, node b is used to test node a and node b's neighboring node w in the uncertain set U. If the diagnosis result is 0, it means that node c is a faulty node. At this time, except for node c, all other nodes in the uncertain set U are non-faulty nodes. Similarly, different nodes are used for testing until all node symptoms are tested out. In other words, all faulty nodes are diagnosed under the condition that at most h non-faulty nodes are diagnosed as faulty nodes.

[0051] As a preferred example, firstly, a BCCC structure is recursively constructed by Python, and then the BCCC structure is implemented as a graph using the NetworkX package. Figure 3 , which shows a schematic diagram of a BCCC(2,2) structure, on which an initial point is selected to perform the C-UC(u) step; since |F|≤(h+1)(k-1)+n is specified, four fault nodes are randomly generated, namely 1110, 1100, 0012 and 1011; the C-UC(u) step is called N times to obtain the maximum connected branch, and the nodes are placed in the set C through breadth-first search. The maximum connected branch is as follows Figure 4As shown, the set C = 0000, 0001, 0002, 0010, 0101, 1002, 0011, 0100, 0102, 1000, 1001, 0111, 0110, 1102, 1010, 1101, 0112, 1012 and 1112; the corresponding set UC = 1110, 1100, 0012, 1011. At this time, all faulty nodes are diagnosed and there are no undiagnosed nodes.

[0052] As another preferred example, the BCCC (5,2) structure is selected, and h=4. Then (h+1)(k-1)+n=10, and the randomly generated fault nodes are 0000, 0001, 0011, 0002, 0012, 0021, 0031, 0022, 0032, and 0040. Figure 5 is the largest connected branch obtained by breadth-first search of the structure, Figure 6 The structure calls the C-UC(u) step N times and obtains the faulty node through breadth-first search. At this time, except for the 0000 node, all other faulty nodes can be diagnosed, satisfying |F|=(h+1)(k-1)+n and |U|>0. Calling the UDiag(C,UC,C) method can diagnose the faulty node 0000, as shown in Figure 7 , which shows all the faulty nodes. At this time, all the faulty nodes have been diagnosed and there are no undiagnosed nodes.

[0053] It should be noted that in the t / h-Diag method, the C-UC(u) step is first called N times. Each time the C-UC(u) step traverses all neighbors of node u to obtain the largest connected component, that is, traverses (N+k-1) times. Therefore, the N C-UC(u) steps take at most O(N(N+k-1)) time; secondly, the time complexity of the UDiag(C,UC,U) method is O(2|U|); finally, the other steps of the t / h-Diag method take at most O(N). Therefore, the total time complexity of the fault diagnosis method in this embodiment is O(N(N+k-1))+O(2|U|)+O(N)=O(N(N+k-1)+2|U|+N)=O(N(N+k-1)).

[0054] In order to more intuitively reflect the effect of this embodiment, the execution experiment results of this method are given below. The experimental platform is Intel Core i7-11800H, 2.3GHz, 16GB DRAM, 64-bit Windows and x64 processor; the programming language used is Python, and BCCC(n,k) is used as the target network, where k=2, 3≤n≤7, 1≤h≤3. In addition, (h+1)(k-1)+n fault nodes are randomly selected in BCCC(n,k), and the average time of a single processing is calculated by executing the algorithm 500 times, as shown in FIG. Figure 8 As shown, it is the average execution time of the t / h-Diag method in the simulation experiment, when k=2, 3≤n≤6 is selected. Figure 8 The average execution time of h = 1, h = 2 and h = 3 is shown respectively. It can be seen from the figure that when more normal nodes are allowed to be faulty nodes, the average execution time will increase. The simulation results are consistent with the time complexity of the t / h-Diag method, and the simulation results show that the execution time of the t / h-Diag method is similar to that of n k+1 The time complexity of the t / h-Diag method is O(N(N+k-1)), and N=(k+1)n k+1 .

[0055] See Table 2, which summarizes the number of faulty nodes diagnosed by the t / h-Diag method in the simulation experiment:

[0056] Table 2

[0057]

[0058]

[0059] It can be seen from Table 2 that the t / h-Diag method can effectively diagnose all faulty nodes, ensuring the accuracy and effectiveness of faulty node diagnosis.

[0060] Further, referring to Table 3, which shows one of the experimental results of 500 simulation experiments using the t / h-Diag method, the order of the diagnosed fault nodes is the order obtained using the t / h-Diag method, specifically:

[0061] Table 3

[0062]

[0063] In summary, the embodiment of the present invention takes any node in the data center network structure as the initial point, traverses all nodes in the data center network structure, and compares each node with the initial point to obtain two sets; obtains the maximum connected branch according to the corresponding nodes in each of the two sets to obtain the fault point set and the fault-free point set therein, and then obtains the uncertain set according to all nodes on the data center network structure and the nodes in the fault point set and the fault-free point set, determines all fault nodes based on the number of nodes contained in the uncertain set, and improves the efficiency of fault node diagnosis and the accuracy of fault node identification.

[0064] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0065] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0066] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A data center network fault diagnosis method based on th strategy, characterized in that: The method comprises the following steps: Taking any node in the data center network structure as the initial point, traversing all nodes in the data center network structure, and obtaining a first set and a second set by comparing the node with the initial point; Based on the data center network structure, a first set with the largest number of nodes is obtained. If the number of nodes in the first set with the largest number of nodes meets a preset condition, the first set with the largest number of nodes is a non-fault point set, and the second set is a fault point set; Acquire an uncertain set according to the fault point set and the non-fault point set, and acquire all fault nodes in the uncertain set based on the number of nodes in the uncertain set; The step of obtaining the first set and the second set by comparing the node with the initial point comprises: Obtain the adjacent points of the traversed node, and use MM* to perform symptom comparison on the initial point, the traversed node, and the adjacent points. If the result is 0, the initial point, the traversed node, and the adjacent points are in the first set; If the result is 1, the initial point, the traversed nodes and the adjacent points are placed in the second set; The step of acquiring the first set with the largest number of nodes based on the data center network structure includes: Traversing any node in the data center network structure to obtain multiple first sets, counting the number of nodes in each first set, and selecting the first set including the largest number of nodes; The preset condition is: the number of nodes in the first set with the largest number of nodes is greater than , Indicates the number of all nodes in the data center network structure; and They are the constants in the t / h strategy respectively; The step of obtaining all faulty nodes in the uncertain set based on the number of nodes in the uncertain set includes: If the number of nodes in the uncertain set is greater than h, all nodes in the uncertain set are fault-free nodes; If the number of nodes in the fault point set is t-1, and the number of nodes in the uncertain set is greater than 0, randomly select a node a in the uncertain set to compare symptoms with the other two nodes b and node c. If the result is 1, then all nodes in the uncertain set except node a, node b and node c are fault-free nodes; Select the adjacent point w of node b in the uncertain set, and compare the symptoms of node b with those of node a and node w. If the result is 0, node c is a faulty node, and all nodes in the uncertain set except node c are non-faulty nodes.

2. According to the th strategy-based data center network fault diagnosis method of claim 1, it is characterized in that: The step of obtaining an uncertain set according to the fault point set and the non-fault point set comprises: Select nodes on the data center network structure that are neither in the fault point set nor in the non-fault point set, and all selected nodes constitute an uncertain set.

Images