Cloud computing server operation fault state monitoring method

Abstract

The invention relates to the technical field of digital data processing, and provides a cloud computing server operation fault state monitoring method, which comprises the following steps: acquiring a monitoring data curve of a monitoring element; obtaining coordinate points and a cluster of the coordinate points according to the monitoring data in the monitoring data curve; acquiring longitudinal point density and longitudinal distance characteristic indexes according to the clustering clusters of the coordinate points; acquiring a clustering segmentation function value according to the longitudinal point density and the longitudinal distance characteristic index; obtaining a clustering set by using a clustering segmentation function value; obtaining a master-slave feature discrimination value and a clustering deviation degree according to the clustering set; obtaining a deviation range interval and a master-slave discrimination range interval according to the master-slave feature discrimination value and the clustering deviation degree; and obtaining a dimensionality-reduced decision tree by using the deviation range interval and the master-slave discrimination range interval, and obtaining an operation fault monitoring result of the cloud server through the dimensionality-reduced decision tree. According to the method, the accuracy of the decision tree pruning algorithm is improved, and the accuracy of operation fault state monitoring is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of digital data processing, and in particular to a method for monitoring the operating fault status of a cloud computing server. Background Art

[0002] Cloud computing servers, also known as cloud servers, are efficient, shared computing resource service platforms. Due to the large number and variety of components in cloud servers, robustness is a key consideration during their construction. This ensures that failure of a single component will not cause the entire cloud server to fail. Therefore, in cloud server fault detection, simultaneous failure of multiple components, resulting in a cloud server failure, is often considered a combined failure.

[0003] When monitoring cloud computing server failures, due to the large number and variety of components in cloud servers, the resulting failure modes present complex and unusual conditions. This results in high computational complexity and low accuracy in fault status monitoring during cloud server failure analysis. Therefore, decision tree pruning algorithms are often used to reduce the dimensionality of the data.

[0004] Traditional decision tree dimensionality reduction algorithms are primarily divided into two approaches: pre-pruning and post-pruning. When using decision tree pruning algorithms to reduce data dimensionality, the decision to prune is typically made by estimating the fault detection performance of the decision tree before and after pruning. However, during this estimation process, to reduce computational complexity, not all possible fault combinations under a particular branch are calculated. This can lead to some fault combinations being mistakenly pruned, resulting in the loss of certain potential fault combinations and lower accuracy in the final fault state monitoring. Summary of the Invention

[0005] The present invention provides a method for monitoring the operating fault status of a cloud computing server to solve the problem of low accuracy of fault status monitoring. The technical solution adopted is as follows: An embodiment of the present invention provides a method for monitoring the operating fault status of a cloud computing server, the method comprising the following steps: Acquire monitoring information data of the cloud server, wherein the monitoring information data includes the number of monitoring elements, monitoring data of each monitoring element, and a monitoring data curve; Obtaining a differential vector between any two monitoring elements based on monitoring data curves of any two monitoring elements; obtaining the coordinate points of each monitoring element based on the differential vector between any two monitoring elements, and obtaining a clustering result of the coordinate points of each monitoring element based on the coordinate points; obtaining each sliding window of each cluster in the clustering result of the coordinate points of each monitoring element based on the shape of each cluster in the clustering result of the coordinate points of each monitoring element, and obtaining a clustering partition function value of each sliding window of each cluster in the clustering result of the coordinate points of each monitoring element based on the sliding window; obtaining a clustering set of the coordinate points of each monitoring element using the clustering partition function values ​​of all sliding windows of each cluster in the clustering result of the coordinate points of each monitoring element; Obtaining a cluster followability index of each cluster set of coordinate points of each monitoring element according to each cluster set of coordinate points of each monitoring element; obtaining a master-slave feature discrimination value of each cluster set of coordinate points of each monitoring element according to the cluster followability index; obtaining a cluster deviation of each cluster set of coordinate points of each monitoring element according to each cluster set of coordinate points of each monitoring element; obtaining a deviation range interval and a master-slave discrimination range interval of each cluster set of coordinate points of each monitoring element according to the master-slave feature discrimination value and the cluster deviation; A decision tree after dimensionality reduction is obtained by using the deviation range interval and the master-slave discrimination range interval, and an operation fault monitoring result of the cloud server is obtained through the decision tree after dimensionality reduction.

[0006] Preferably, the method for obtaining the differential vector between any two monitoring elements according to the monitoring data curves of any two monitoring elements is: Sampling the monitoring data curve of each monitoring element at a preset sampling frequency to obtain sampling data of each monitoring element, and using a sequence of the sampling data of each monitoring element in ascending time order as a sampling vector of each monitoring element; One of any two monitoring elements is used as the first target monitoring element, and the other monitoring element is used as the first marked monitoring element. The difference between the sampling data at the same acquisition time in the sampling vectors of the first target monitoring element and the first marked monitoring element is calculated, and a vector composed of all the differences in ascending time order is used as the differential vector of the first marked monitoring element relative to the first target monitoring element.

[0007] Preferably, the method of obtaining the coordinate point of each monitoring element according to the differential vector between any two monitoring elements and obtaining the clustering result of the coordinate point of each monitoring element according to the coordinate point is: For the differential vector between any two monitoring elements, one of the two monitoring elements is used as the second target monitoring element, the element at each acquisition moment in the differential vector is used as an abscissa, the element at each acquisition moment in the sampling vector of the second target monitoring element is used as a ordinate, and the data point determined by the abscissa and ordinate is used as a coordinate point of the second target monitoring element; All coordinate points of the second target monitoring element are used as input of a DBSCAN clustering algorithm, and the output of the DBSCAN clustering algorithm is used as a clustering result of all coordinate points of the second target monitoring element.

[0008] Preferably, the method of obtaining each sliding window of each cluster in the clustering result of each monitoring element's coordinate point according to the shape of each cluster in the clustering result of each monitoring element's coordinate point, and obtaining the clustering partition function value of each sliding window of each cluster in the clustering result of each monitoring element's coordinate point according to the sliding window is: For each cluster in the clustering result of the coordinate points of each monitoring element, a minimum bounding rectangle is calculated for the shape of the cluster using the minimum bounding rectangle technique, the maximum height of the minimum bounding rectangle on the vertical axis is used as the vertical length of the window, the first preset parameters are used as the horizontal width and sliding step of the window respectively, and the result of each sliding of the window on the cluster is used as each sliding window of the cluster; For each sliding window of the cluster, the number of data points in the sliding window is used as the numerator, the horizontal width of the sliding window is used as the denominator, and the ratio of the numerator to the denominator is used as the vertical point density of the sliding window; The average height of the local cluster shape in the vertical axis direction within the sliding window is used as the local vertical distance difference of the sliding window, and the sum of the inverse of the ratio of the local vertical distance difference to the vertical length of the sliding window and a second preset parameter is used as the first component factor; the ratio of the variance of the vertical coordinate values ​​of all elements in the cluster to the variance of the vertical coordinate values ​​of all elements in the sliding window of the cluster is used as the second component factor; the product of the first component factor and the second component factor is used as the vertical distance characteristic index of the sliding window; The product of the longitudinal point density and the longitudinal distance characteristic index is used as the cluster segmentation function value of the sliding window.

[0009] Preferably, the method of obtaining the cluster set of the coordinate points of each monitoring element by using the clustering partition function values ​​of all sliding windows of each cluster cluster in the clustering result of the coordinate points of each monitoring element is: For each cluster in the clustering results of the coordinate points of each monitoring element, the minimum value of the clustering partitioning function of all sliding windows of the cluster is obtained, the sliding window corresponding to the minimum value is used as the target sliding window, the center point of the target sliding window is used as the longitudinal partitioning point, the cluster is partitioned along the longitudinal partitioning point in the longitudinal direction, and the partitioning results of the cluster are used as two cluster sets of the coordinate points of the monitoring element.

[0010] Preferably, the method of obtaining the cluster followability index of each cluster set of coordinate points of each monitoring element according to each cluster set of coordinate points of each monitoring element is: For each cluster of coordinate points of each monitoring element, a rectangular coordinate system is established with the center coordinate of the cluster set as the coordinate origin. 、 、 、 The straight lines are taken as each directrix; For each directrix, calculate the Euclidean distance between the projection point of each element in the cluster set on the directrix and the coordinate origin, and use the standard deviation of all the Euclidean distances as the standard deviation of the directrix; In the rectangular coordinate system The standard deviation of the directrix is The ratio of the standard deviations of the directrix is ​​used as the first multiplication factor; In the rectangular coordinate system 、 The maximum value of the standard deviation of the directrix is ​​used as the numerator, and the rectangular coordinate system 、 The minimum value of the standard deviation of the directrix is ​​used as the denominator, and the ratio of the numerator to the denominator is used as the second multiplication factor; The product of the first multiplication factor and the second multiplication factor is used as the clustering followability index of the cluster set.

[0011] Preferably, the method for obtaining the master-slave feature discrimination value of each cluster set of the coordinate points of each monitoring element according to the cluster followability index is: For each cluster set of the coordinate points of each monitoring element, the number of elements in the cluster set is used as the numerator, the area of ​​the minimum circumscribed rectangle of the cluster set is used as the denominator, and the ratio of the numerator to the denominator is used as the cluster density of the cluster set; The product of the cluster followability index and cluster density of the cluster set is used as the master-slave feature discrimination value of the cluster set.

[0012] Preferably, the method of obtaining the cluster deviation of each cluster set of coordinate points of each monitoring element according to each cluster set of coordinate points of each monitoring element is: Where, Indicates the The coordinate point of the monitoring element The cluster deviation of the cluster set, Indicates the The monitoring element corresponds to The number of sliding windows on the cluster set, Indicates the The monitoring element corresponds to The horizontal coordinate of the center point of the nth sliding window on the cluster set, Indicates the The monitoring element corresponds to The horizontal coordinate of the center point of the cluster set, Indicates the The monitoring element corresponds to The length of the horizontal axis range of the cluster set, Indicates the number of monitoring elements corresponding to the i-th The first cluster The clustering partition function value of the sliding window, Indicates the number of monitoring elements corresponding to the i-th The mean of the clustering partition function values ​​of all sliding windows on the cluster set.

[0013] Preferably, the method for obtaining the deviation range interval and the master-slave discrimination range interval of each cluster set of the coordinate point of each monitoring element according to the master-slave feature discrimination value and the cluster deviation degree is: Where, Indicates the The coordinate point of the monitoring element The deviation range of cluster sets, Indicates the The monitoring element corresponds to The horizontal coordinate of the center point of the cluster set, Indicates the The coordinate point of the monitoring element In the rectangular coordinate system of cluster sets The standard deviation of the directrix, Indicates the The coordinate point of the monitoring element The cluster deviation of the cluster set; Indicates the The coordinate point of the monitoring element The lower limit of the master-slave discrimination range of the cluster set, Indicates the The coordinate point of the monitoring element The upper limit of the master-slave discrimination range of the cluster set, Indicates the The coordinate point of the monitoring element The master-slave discrimination range interval of cluster sets, Indicates the The coordinate point of the monitoring element The master-slave feature discriminant value of a cluster set, Indicates the The mean of the master-slave feature discriminant values ​​of all cluster sets of the coordinate points of the monitoring elements.

[0014] Preferably, the method for obtaining the decision tree after dimensionality reduction by using the deviation range interval and the master-slave discrimination range interval is: Using the master-slave discrimination range interval of each cluster set of the coordinate points of each monitoring element as the master-slave discrimination range interval of the target cluster set, calculating the intersection interval between the master-slave discrimination range interval of the target cluster set and the deviation range interval of each of the remaining cluster sets, and calculating the ratio of the length of the intersection interval to the length of the master-slave discrimination range interval of the target cluster set. If the ratio is lower than a third preset parameter, the target cluster sets are considered to be independent. The ratio of the length of the union of the deviation range intervals of all independent cluster sets of the coordinate points of each monitoring element to the length of the union of the deviation range intervals of all cluster sets of the coordinate points of each monitoring element is used as the master-slave relationship measurement index between any two monitoring elements; A decision tree is constructed through all monitoring elements, and a decision tree pruning algorithm is used to obtain a reduced-dimensional decision tree based on the master-slave relationship metric index.

[0015] The beneficial effects of the present invention are as follows: the present invention converts time series data into cluster data by analyzing the correlation characteristics between the monitoring data of the monitoring elements, performing pairwise calculations between the monitoring data of the monitoring elements, and each cluster represents a certain data range, and the monitoring elements represented by the monitoring data may exhibit strong correlation characteristics. In principle, the distribution law of the coordinate points is characterized by the ratio of the variance of the coordinate points in the cluster, and the master-slave feature discrimination value is constructed in combination with the cluster density to characterize the strength of the correlation characteristics between the monitoring data of the monitoring elements represented by the coordinate points in each cluster. At the same time, according to the density distribution of the cluster set along the horizontal axis, the cluster deviation between the coordinate points of the cluster and the cluster geometric center is calculated, and the cluster deviation is used as the weight and combined with the Principle, the deviation range of each cluster set is obtained, compared with the traditional The principle takes into account the distribution of data, and the range is more accurate. Then, combined with the master-slave feature discrimination value of each cluster set, it is judged whether the cluster shows the characteristics of strong correlation between components, and finally the master-slave discrimination range interval is obtained. The decision tree pruning algorithm is used to complete the dimensionality reduction of the monitoring data of the monitoring component based on the deviation range interval and the master-slave discrimination range interval, and then the operation fault monitoring result of the cloud server is obtained. Compared with the traditional decision tree pruning algorithm, the present invention avoids the neglect of certain fault combinations due to pruning while fully pruning, which leads to incomplete consideration of the fault combination and the risk of fault monitoring failure. Therefore, the present invention improves the accuracy of operation fault status monitoring by improving the accuracy of the decision tree pruning algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] 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.

[0017] Figure 1 A flowchart of a method for monitoring a cloud computing server's operational fault status provided by one embodiment of the present invention; Figure 2 A schematic diagram of a sliding window on a cluster provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0019] See also Figure 1 , which shows a flow chart of a method for monitoring the operating fault status of a cloud computing server provided by one embodiment of the present invention, the method comprising the following steps: Step S001: Acquire monitoring data and monitoring data curves of monitoring elements of a cloud server.

[0020] The present invention uses a distributed cloud server as an example to monitor its operational failures. The monitoring data of each monitoring element of the distributed cloud server is obtained through the cloud server monitoring platform. The monitoring data of each monitoring element is the continuous monitoring data of each monitoring element of the previous day, and the number of monitoring elements of the distributed cloud server is obtained. In order to avoid the influence of different dimensions on the subsequent analysis results, the monitoring data of each monitoring element is z-score normalized to obtain the monitoring data of each monitoring element after normalization. The z-score normalization process is a well-known technology and the specific process is not repeated here. For the monitoring data of each monitoring element after normalization, the monitoring data curve of the monitoring element is drawn by MATLAB software. The horizontal axis of the monitoring data curve is time and the vertical axis is the size of the monitoring data.

[0021] At this point, the number of monitoring elements, monitoring data of the monitoring elements, and monitoring data curves are obtained.

[0022] Step S002: Obtain clusters of coordinate points according to the monitoring data curve, obtain longitudinal point density and longitudinal distance characteristic index according to the clusters of coordinate points, obtain cluster partitioning function values ​​according to the longitudinal point density and longitudinal distance characteristic index, and obtain cluster sets using the cluster partitioning function values.

[0023] In cloud servers, because different monitoring components form a functioning server through a specific structural combination, the monitoring data from each component often exhibits strong correlations. For example, when a router loses power, the router's load is always constant. If these two situations are considered a single fault combination, the simulation calculation for that fault combination will be invalid.

[0024] Furthermore, the number of monitoring elements is recorded as M. Since the monitoring data under different monitoring elements are likely to be correlated, in order to illustrate how to calculate the correlation of monitoring data between two monitoring elements, the first The data monitoring curve of the monitoring element and the The monitoring data curve of a monitoring element is used as an example to illustrate the following. .

[0025] The present invention expresses the correlation between two faults in a master-slave relationship. The fault monitoring parameters of each monitoring element are When The fault monitoring parameters of each monitoring element must be , indicating that when The fault monitoring parameters of each monitoring element are There is no need to The fault parameters of the monitoring element are monitored, which can be expressed as: The fault monitoring parameters of each monitoring element are When, The monitoring element is The main component of the monitoring component.

[0026] Respectively Monitoring element, The monitoring data curve of each monitoring element is sampled to obtain the Monitoring element, The sampling data of the monitoring element is sampled at a frequency of 10,000 Hz. Monitoring element, The sampling data of each monitoring element are arranged in ascending time order as the vectors of Monitoring element, The sampling vector of the monitoring element is Monitoring element, The sampling vectors of the monitoring elements are recorded as 、 ,in and The elements at the same position in represent the collected data at the same collection moment.

[0027] Calculate the The monitoring element is relative to the The differential vector of the monitoring elements : Where, Indicates the The monitoring element is relative to the The differential vector of the monitoring elements The gth value in and Respectively represent Monitoring element, The gth collected data in the sampling vector of the monitoring element; Indicates the The monitoring element is relative to the The differential vector of the monitoring element, G represents the The monitoring element is relative to the The number of values ​​in the difference vector of the monitoring elements, 、 、 Respectively represent The monitoring element is relative to the The first value, the second value, and the last value in the differential vector of the monitoring elements.

[0028] It should be noted that and Elements at the same position represent data at the same acquisition moment. The difference vector represents the The monitoring data of the first monitoring element is relative to the The change of the monitoring data of each monitoring element, the more stable the change of the differential vector is within a period of time, the more stable the difference vector is within the period of time. The more likely a monitoring element is to belong to components.

[0029] Furthermore, the differential vector at a certain acquisition moment The values ​​in Axis coordinates, acquisition vector The sample data in Axis coordinates, you can get a coordinate point. The differential vector of all acquisition moments The values ​​in Axis coordinates, acquisition vector The sample data in Axis coordinates are obtained to obtain all coordinate points. All coordinate points are used as input to the DBSCAN (Density-Based Spatial Clustering of Applications with Noise, DBSCAN) clustering algorithm. The output of the DBSCAN clustering algorithm is used as the clustering result of all coordinate points. Each cluster in the clustering result contains multiple coordinate points. The DBSCAN clustering algorithm is a well-known technology, and the specific process is not repeated here.

[0030] The vertical coordinate range of each cluster in the clustering result is The data range of the monitoring data of each monitoring element, the horizontal axis range corresponding to each cluster is When the monitoring data range of a monitoring element is The fluctuation range of the monitoring data of each monitoring element. If the clustering result is relatively dense within a certain range of the horizontal axis, it means that the first A monitoring element is in the When the monitoring data of the monitoring element is within the horizontal axis range of the cluster, The monitoring data of each monitoring element must be within a fixed range. The monitoring data of the monitoring element is The monitoring data of each monitoring element has a strong follow-up feature. The more likely a monitoring element is to belong to the A monitoring element.

[0031] The clustering results obtained by the DBSCAN clustering algorithm are divided into a cluster as long as the density threshold of the coordinate points is met. This will cause the horizontal coordinate range of each cluster to be defined too large when classifying the master-slave relationship of the components. Therefore, the clustering results are divided as follows: Specifically, the minimum bounding rectangle technique is used to make the minimum bounding rectangle of each cluster in the clustering result, and the height of the minimum bounding rectangle of the kth cluster in the vertical axis direction is recorded as The minimum bounding rectangle technique is a well-known technique, and the specific process will not be described in detail. Taking the kth cluster in the clustering result as an example, the vertical length is set to , the horizontal width is The window size is , sliding from left to right in turn, traversing all elements of the kth cluster. In order to facilitate subsequent calculations, the window formed by the first sliding is recorded as the first sliding window, the window formed by the second sliding is recorded as the second sliding window, and so on. The schematic diagram of the sliding window on the cluster is as follows Figure 2 shown.

[0032] Calculate the first monitoring element corresponding to the The first cluster The vertical point density of a sliding window: Where, Indicates the number of monitoring elements corresponding to the i-th The first cluster The vertical point density of the sliding window, Indicates the number of monitoring elements corresponding to the i-th The first cluster The number of data points in the sliding window, Indicates the number of monitoring elements corresponding to the i-th The horizontal width of the sliding window for clusters.

[0033] The smaller the number of data points in the sliding window, the smaller the vertical density relative to the horizontal axis, and the smaller the vertical point density, that is, the lower the density of data points in the sliding window at this position, the more it needs to be separated as a separate cluster. In addition, the greater the vertical point density, the smaller the vertical point density within the cluster range. The monitoring data of the monitoring element is The stronger the following characteristics of the monitoring data of the monitoring element, the The more likely a monitoring element is to belong to the A monitoring element.

[0034] Furthermore, for the ith monitoring element corresponding to the The first cluster The average height of the local cluster shape in the vertical axis direction in the sliding window is recorded as the local vertical distance difference. The vertical length of the sliding window is used as the cluster vertical distance difference of the sliding window.

[0035] Calculate the first monitoring element corresponding to the The first cluster The vertical distance characteristic index of the sliding window: Where, Indicates the number of monitoring elements corresponding to the i-th The first cluster The vertical distance characteristic index of the sliding window, Indicates the number of monitoring elements corresponding to the i-th The first cluster The local vertical distance difference of the sliding window, Indicates the number of monitoring elements corresponding to the i-th The cluster vertical distance difference of the clusters, Indicates the number of monitoring elements corresponding to the i-th The variance of the ordinate values ​​of all elements in a cluster, Indicates the number of monitoring elements corresponding to the i-th The first cluster The variance of the ordinate values ​​of all elements in a sliding window.

[0036] The smaller the ratio of the local vertical distance difference to the cluster vertical distance difference, the smaller the height of the local cluster shape in the sliding window in the vertical axis direction is, which to a certain extent reflects that the cluster points in the local area of ​​the cluster cluster are denser in the vertical axis direction. At the same time, the i-th monitoring element corresponding to the i-th monitoring element is smaller. The first cluster The smaller the variance of the ordinate values ​​of all elements in a sliding window is relative to the variance of the ordinate values ​​of all elements in the cluster, that is, The larger it is, the smaller the dispersion of cluster points in the local area of ​​the cluster, that is, the higher the density, and the larger the vertical distance characteristic index.

[0037] Furthermore, the obtained vertical distance characteristic index is used as the weight of the vertical point density. The larger the value, the more clustered the points in the sliding window are. If the degree of clustering of the cluster points in the sliding window is lower, the more likely there is a point that can represent the first The monitoring element belongs to The more clusters of monitoring elements there are, the more they need to be separated into separate clusters.

[0038] Calculate the first monitoring element corresponding to the The first cluster The clustering partition function value of a sliding window: Where, Indicates the number of monitoring elements corresponding to the i-th The first cluster The clustering partition function value of the sliding window, Indicates the number of monitoring elements corresponding to the i-th The first cluster The vertical distance characteristic index of the sliding window, Indicates the number of monitoring elements corresponding to the i-th The first cluster The vertical point density of a sliding window.

[0039] The vertical point density and vertical distance feature index both represent the degree of clustering of cluster points within the sliding window. The lower the degree of clustering of the sliding window, the more likely it is that there is a feature that represents the first The monitoring element belongs to The clustering of monitoring elements, and the larger the value, the higher the The monitoring data of the monitoring element is The stronger the following characteristics of the monitoring data of each monitoring element, the The more likely a monitoring element is to belong to the A monitoring element.

[0040] For the i-th monitoring element corresponding to the A cluster is selected, and the sliding window corresponding to the minimum value of the cluster partition function of all sliding windows of the cluster is used as the target sliding window. The center point of the target sliding window is used as the vertical partition point, and the kth cluster is divided into two cluster sets along the vertical direction of the vertical partition point, thereby avoiding poor accuracy due to the excessive definition of the horizontal axis range of the cluster when classifying the master-slave relationship of the elements.

[0041] At this point, based on the above analysis, each cluster corresponding to the i-th monitoring element is segmented according to the same method as above to obtain all cluster sets corresponding to the i-th monitoring element.

[0042] Step S003: obtain the cluster followability index and cluster density according to the cluster set, obtain the master-slave feature discrimination value and cluster deviation according to the cluster followability index and cluster density, and obtain the deviation range interval and master-slave discrimination range interval according to the master-slave feature discrimination value and cluster deviation.

[0043] Based on the above analysis, the transformation of monitoring data from time series to cluster sets is completed. Next, the monitoring data range represented by each cluster set is analyzed. Is the first element subordinate to the components.

[0044] according to in principle, The data points fall within three standard deviations of the mean, and in order to avoid statistical errors in cluster element points, which may cause some data anomalies, the three standard deviations of the mean of each cluster set are used as the statistical feature range of the cluster set.

[0045] However, due to the different distributions of different cluster sets, in practice, the data of cluster sets are less likely to follow a normal distribution and more likely to deviate. Therefore, it is necessary to analyze the deviation of cluster sets.

[0046] Specifically, with the center coordinates of each cluster set as the center, the standard deviations in four directions are calculated to express the statistical feature range of the cluster. Specifically, a coordinate system is established with the center coordinates of the cluster set as the coordinate origin, and the standard deviations in four directions are calculated to express the statistical feature range of the cluster. , , , As each directrix, obtain the projection points of all cluster points in the cluster set on the directrix, calculate the Euclidean distance between each projection point on each directrix and the center coordinate, calculate the standard deviation of all Euclidean distances corresponding to each directrix, and use , , , The standard deviation of the straight line is recorded as 、 、 、 .

[0047] Furthermore, calculate the The monitoring element corresponds to Cluster following index of cluster sets: Where, Indicates the The monitoring element corresponds to The cluster following index of the cluster set, 、 、 、 Respectively represent The monitoring element corresponds to The cluster sets corresponding to , , , The standard deviation of the straight line is Indicates the The monitoring element corresponds to The cluster sets corresponding to , The maximum value of the standard deviation of the straight line is Indicates the The monitoring element corresponds to The cluster sets corresponding to , The straight line is the minimum value of the standard deviation of the directrix.

[0048] It is a cluster set along The standard deviation of the direction divided by the The standard deviation along the Standard deviation of direction The smaller it is, the closer the cluster is. The smaller the data range of the direction, the The monitoring element is The smaller the data difference of each monitoring element, the better the following effect. Standard deviation of direction The larger the value, the closer the cluster is to the The larger the data range of the direction, the The monitoring element is The larger the following range of a monitoring element, the better the following effect. The bigger it is, the better the following effect.

[0049] like Greater than , indicating that the cluster set is along The range in the direction is greater than that along The ratio of the two represents the difference in the range of the two. That is, the larger the ratio, the greater the range difference, which means that the points in the cluster are more likely to be along the The trend of direction arrangement represents the The data of each monitoring element has the following The trend is increasing with the increase of data of monitoring components. Represents the The data of the monitoring element and the The data of each monitoring element follows the trend of change. The larger the ratio, the stronger the trend. Component pair The better the component following effect.

[0050] Finally, the cluster followability index is obtained by multiplying the two. The larger the value, the Among the monitoring elements Within the data range represented by the cluster sets, the The monitoring element is The more monitoring elements follow, the better the effect.

[0051] Furthermore, based on the cluster followability index obtained above, the first The monitoring element corresponds to The master-slave feature discriminant value of a cluster set: Where, Indicates the The monitoring element corresponds to The cluster density of the cluster sets, Indicates the The monitoring element corresponds to The number of cluster points in a cluster set, Indicates the The monitoring element corresponds to The area of ​​the minimum bounding rectangle of a cluster set; Indicates the The monitoring element corresponds to The master-slave feature discriminant value of a cluster set, Indicates the The monitoring element corresponds to The cluster following index of a cluster set.

[0052] No. The monitoring element corresponds to Cluster following index of cluster sets The larger the The monitoring element corresponds to The cluster density of cluster sets The larger the The monitoring element is The better the following effect of each monitoring element, the better the following effect. The more likely a monitoring element is to belong to monitoring elements; the final master-slave feature discrimination value The larger the value, the The more likely a monitoring element is to belong to A monitoring element.

[0053] Furthermore, based on the above analysis, Each cluster set corresponding to the monitoring element is obtained by dividing the cluster clusters by setting a sliding window. It can be said that there are multiple sliding windows on the cluster set. The horizontal coordinate of the center point of each sliding window of the cluster set is obtained. The monitoring element corresponds to The horizontal coordinate of the center point of the nth sliding window of the cluster set is marked as ; At the same time, obtain the horizontal coordinate of the center point of each cluster set, and The monitoring element corresponds to The horizontal coordinate of the center point of the cluster set is marked as .

[0054] Calculate the The monitoring element corresponds to Cluster deviation of cluster sets: Where, Indicates the The monitoring element corresponds to The cluster deviation of the cluster set, Indicates the The monitoring element corresponds to The number of sliding windows on the cluster set, Indicates the The monitoring element corresponds to The horizontal coordinate of the center point of the nth sliding window of the cluster set, Indicates the The monitoring element corresponds to The horizontal coordinate of the center point of the cluster set, Indicates the The monitoring element corresponds to The length of the horizontal axis range of the cluster set, Indicates the number of monitoring elements corresponding to the i-th The first cluster set The clustering partition function value of the sliding window, Indicates the number of monitoring elements corresponding to the i-th The mean of the clustering partition function values ​​of all sliding windows of a cluster set.

[0055] The cluster partition function value represents the cluster points along the cluster The density in the axis direction represents the density at the position of the sliding window; Represents the distance between the center of the sliding window and the center of the cluster set, divided by The purpose is to convert distance into proportion; It is the density of each sliding window position relative to the density of the entire cluster set. The larger its value, the greater the density of the sliding window position. The bigger, and The larger the value is, the more cluster deviation The bigger.

[0056] A positive value indicates that the dense area of ​​cluster points in the cluster is biased towards the center. Positive direction; negative value indicates that the dense area of ​​cluster points in the cluster is biased towards the center. Negative direction. The larger the value, the more serious the deviation. When dividing the data range of the cluster, the longer the distance should be divided in the direction with high coordinate point density to make the data division range more reasonable.

[0057] Furthermore, according to The deviation range interval and master-slave discrimination range interval of each cluster set corresponding to the monitoring element: Where, Indicates the The coordinate point of the monitoring element The deviation range of cluster sets, Indicates the The monitoring element corresponds to The horizontal coordinate of the center point of the cluster set, Indicates the The coordinate point of the monitoring element In the rectangular coordinate system of cluster sets The standard deviation of the directrix, Indicates the The coordinate point of the monitoring element The cluster deviation of the cluster set; Indicates the The coordinate point of the monitoring element The lower limit of the master-slave discrimination range of the cluster set, Indicates the The coordinate point of the monitoring element The upper limit of the master-slave discrimination range of the cluster set, Indicates the The coordinate point of the monitoring element The master-slave discrimination range interval of cluster sets, Indicates the The coordinate point of the monitoring element The master-slave feature discriminant value of a cluster set, Indicates the The mean of the master-slave feature discriminant values ​​of all cluster sets of the coordinate points of the monitoring elements.

[0058] In the first formula here, multiplying by 3 is based on in principle, The data points fall within three standard deviations of the mean. or The cluster set range is shifted left or right according to the cluster deviation, and the resulting interval range is finally used as the range of the mth cluster set.

[0059] Is the parameter The standardized mathematical method will not be described here; this formula is to multiply the interval of the first formula by Obtained, because the The larger the master-slave feature discrimination value of a cluster set is, the more likely it is that the jth element belongs to the interval of the i-th element.

[0060] At this point, the deviation range interval and the master-slave discrimination range interval of the cluster set are obtained.

[0061] Step S004: obtaining a decision tree after dimensionality reduction by using the deviation range interval and the master-slave discrimination range interval, and obtaining the operation fault monitoring result of the cloud server through the decision tree after dimensionality reduction.

[0062] The master-slave discrimination range interval of each cluster set of the coordinate point of each monitoring element is used as the master-slave discrimination range interval of the target cluster set, and the intersection interval between the master-slave discrimination range interval of the target cluster set and the deviation range interval of each other cluster set is calculated. The ratio of the length of the intersection interval to the length of the master-slave discrimination range interval of the target cluster set is calculated. If the ratio is lower than 10%, the target cluster set is considered to be independent.

[0063] The ratio of the length of the union of the deviation range intervals of all independent cluster sets of the coordinate points of each monitoring element to the length of the union of the deviation range intervals of all cluster sets is used as the master-slave relationship measurement index between any two monitoring elements.

[0064] Furthermore, the mean value of all monitoring data of each monitoring element is used as the judgment value of each monitoring element, and the sequence composed of the judgment values ​​of all monitoring elements in ascending order of numerical values ​​is used as the judgment value sequence, and the value of the zth element in the judgment value sequence is recorded as A decision tree is constructed by judging the value sequence, and each node in the decision tree represents a monitoring element.

[0065] Specifically, a monitoring element corresponding to the minimum element value in the judgment value sequence is used as a decision node, that is, A corresponding monitoring element is used as a decision node; the value sequence greater than and less than or equal to A monitoring element corresponding to the element value of is used as the first opportunity node of the decision node, and the value greater than and less than or equal to A monitoring element corresponding to the element value of is used as the second opportunity node of the decision node, and the value in the judgment value sequence greater than and less than or equal to A monitoring element corresponding to the element value of is used as the third chance node of the decision node, and so on; for any chance node, let’s take the u-th chance node as an example, and judge the value sequence greater than and less than or equal to A monitoring element corresponding to the element value of is used as the first result node of the u-th chance node, and the value in the judgment value sequence greater than and less than or equal to A monitoring element corresponding to the element value of is used as the second result node of the u-th chance node, and the value in the judgment value sequence greater than and less than or equal to The monitoring element corresponding to the element value of is used as the third result node of the u-th chance node; and so on. The decision tree constructed in the above manner has three layers, where the first layer is the decision layer, the second layer is the chance layer, and the third layer is the result layer, and each node represents a monitoring element.

[0066] Furthermore, the master-slave relationship metric index between the two monitoring elements represented by each pair of adjacent nodes in the decision tree is calculated. If the master-slave relationship metric index is higher than 90%, it indicates that the two monitoring elements represented by the adjacent nodes have a strong subordinate relationship, and the branches between the adjacent nodes need to be pruned. Pruning all nodes in the decision tree yields a pruned decision tree, which is the decision tree after dimensionality reduction.

[0067] Based on the above analysis, a decision tree pruning algorithm is used to generate a reduced-dimensionality decision tree based on the master-slave relationship metric index. This decision tree pruning algorithm is well-known, and the specific process is not detailed here. In the reduced-dimensionality decision tree, leaf nodes are traversed back along the decision tree to obtain multiple nodes. Each traversal path is considered a node combination, known as a fault combination. Each fault combination is evaluated and classified by technicians into a fault mode, and a maintenance strategy is developed for each fault mode.

[0068] When performing fault detection on a running cloud server, if the various detection data of the cloud server simultaneously meet the abnormal monitoring data range corresponding to each node of one of the fault combinations, it is determined that the cloud server has the fault mode of the fault combination and maintenance is performed according to the formulated strategy.

[0069] Thus, a method for monitoring the operating fault status of a cloud computing server is completed.

[0070] The various embodiments in this specification are described in a progressive manner. Similar parts between the various embodiments can be referred to in conjunction with each other. Each embodiment focuses on the differences from other embodiments. The above description is merely 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 shall be included in the scope of protection of the present invention.

Claims

1. A method for monitoring the operating fault status of a cloud computing server, characterized in that: The method comprises the following steps: Acquire monitoring information data of the cloud server, wherein the monitoring information data includes the number of monitoring elements, monitoring data of each monitoring element, and a monitoring data curve; Obtain the differential vector between any two monitoring elements according to the monitoring data curves of any two monitoring elements; obtain the coordinate point of each monitoring element according to the differential vector between any two monitoring elements, and obtain the clustering result of the coordinate point of each monitoring element according to the coordinate point; obtain each sliding window of each cluster in the clustering result of the coordinate point of each monitoring element according to the shape of each cluster in the clustering result of the coordinate point of each monitoring element, and obtain the clustering partition function value of each sliding window of each cluster in the clustering result of the coordinate point of each monitoring element according to the sliding window; obtain the clustering set of the coordinate point of each monitoring element by using the clustering partition function values ​​of all sliding windows of each cluster in the clustering result of the coordinate point of each monitoring element; According to each cluster set of the coordinate points of each monitoring element, a cluster followability index of each cluster set of the coordinate points of each monitoring element is obtained; according to the cluster followability index, a master-slave feature discrimination value of each cluster set of the coordinate points of each monitoring element is obtained; according to each cluster set of the coordinate points of each monitoring element, a cluster deviation degree of each cluster set of the coordinate points of each monitoring element is obtained; according to the master-slave feature discrimination value and the cluster deviation degree, a deviation range interval and a master-slave discrimination range interval of each cluster set of the coordinate points of each monitoring element are obtained; The deviation range interval and the master-slave discrimination range interval are used to obtain a decision tree after dimensionality reduction, and the operation fault monitoring result of the cloud server is obtained through the decision tree after dimensionality reduction.

2. A method for monitoring the operating fault status of a cloud computing server according to claim 1, characterized in that: The method for obtaining the differential vector between any two monitoring elements according to the monitoring data curves of any two monitoring elements is: Sampling the monitoring data curve of each monitoring element at a preset sampling frequency to obtain sampling data of each monitoring element, and using a sequence composed of the sampling data of each monitoring element in ascending time order as a sampling vector of each monitoring element; One of any two monitoring elements is taken as the first target monitoring element, and the other monitoring element is taken as the first marked monitoring element. The difference between the sampling data at the same acquisition time in the sampling vectors of the first target monitoring element and the first marked monitoring element is calculated, and a vector composed of all the differences in ascending time order is taken as the differential vector of the first marked monitoring element relative to the first target monitoring element.

3. A method for monitoring the operating fault status of a cloud computing server according to claim 1, characterized in that: The method of obtaining the coordinate point of each monitoring element according to the differential vector between any two monitoring elements and obtaining the clustering result of the coordinate point of each monitoring element according to the coordinate point is: For a differential vector between any two monitoring elements, one of the two monitoring elements is used as a second target monitoring element, an element at each acquisition time in the differential vector is used as a horizontal coordinate, an element at each acquisition time in the sampling vector of the second target monitoring element is used as a vertical coordinate, and a data point determined by the horizontal and vertical coordinates is used as a coordinate point of the second target monitoring element; All coordinate points of the second target monitoring element are used as input of the DBSCAN clustering algorithm, and the output of the DBSCAN clustering algorithm is used as the clustering result of all coordinate points of the second target monitoring element.

4. A method for monitoring the operating fault status of a cloud computing server according to claim 1, characterized in that: The method of obtaining each sliding window of each cluster in the clustering result of each monitoring element's coordinate point according to the shape of each cluster in the clustering result of each monitoring element's coordinate point, and obtaining the clustering partition function value of each sliding window of each cluster in the clustering result of each monitoring element's coordinate point according to the sliding window is: For each cluster in the clustering result of the coordinate point of each monitoring element, a minimum bounding rectangle is made for the shape of the cluster using the minimum bounding rectangle technique, the maximum height of the minimum bounding rectangle on the vertical axis is used as the vertical length of the window, the first preset parameters are used as the horizontal width and the sliding step of the window respectively, and the result of each sliding of the window on the cluster is used as each sliding window of the cluster; For each sliding window of the cluster, the number of data points in the sliding window is used as the numerator, the horizontal width of the sliding window is used as the denominator, and the ratio of the numerator to the denominator is used as the vertical point density of the sliding window; The average height of the local cluster shape in the longitudinal direction in the sliding window is used as the local longitudinal distance difference of the sliding window, and the inverse of the ratio of the local longitudinal distance difference to the longitudinal length of the sliding window and the sum of the second preset parameter are used as the first component factor; the ratio of the variance of the longitudinal coordinate values ​​of all elements in the cluster to the variance of the longitudinal coordinate values ​​of all elements in the sliding window of the cluster is used as the second component factor; the product of the first component factor and the second component factor is used as the longitudinal distance characteristic index of the sliding window; The product of the longitudinal point density and the longitudinal distance characteristic index is used as the clustering segmentation function value of the sliding window.

5. The method for monitoring the operating fault status of a cloud computing server according to claim 1, characterized in that: The method for obtaining the clustering set of the coordinate points of each monitoring element by using the clustering partition function values ​​of all sliding windows of each cluster cluster in the clustering result of the coordinate points of each monitoring element is: For each clustering cluster in the clustering results of the coordinate points of each monitoring element, the minimum value of the clustering partitioning function of all sliding windows of the clustering cluster is obtained, the sliding window corresponding to the minimum value is used as the target sliding window, the center point of the target sliding window is used as the longitudinal partitioning point, the clustering cluster is partitioned in the longitudinal direction along the longitudinal partitioning point, and the partitioning results of the clustering cluster are used as two clustering sets of the coordinate points of the monitoring element.

6. A method for monitoring the operating fault status of a cloud computing server according to claim 1, characterized in that: The method for obtaining the clustering followability index of each clustering set of coordinate points of each monitoring element according to each clustering set of coordinate points of each monitoring element is: For each cluster of coordinate points of each monitoring element, a rectangular coordinate system is established with the center coordinate of the cluster set as the coordinate origin. , , , The straight lines are taken as each directrix; For each directrix, the Euclidean distance between the projection point of each element in the cluster set on the directrix and the coordinate origin is calculated, and the standard deviation of all the Euclidean distances is taken as the standard deviation of the directrix; In the rectangular coordinate system The standard deviation of the directrix is The ratio of the standard deviations of the directrix is ​​used as the first multiplication factor; In the rectangular coordinate system , The maximum value of the standard deviation of the directrix is ​​taken as the numerator, and the rectangular coordinate system , The minimum value of the standard deviation of the directrix is ​​used as the denominator, and the ratio of the numerator to the denominator is used as the second multiplication factor; The product of the first multiplication factor and the second multiplication factor is used as the clustering followability index of the cluster set.

7. A method for monitoring the operating fault status of a cloud computing server according to claim 1, characterized in that: The method for obtaining the master-slave feature discrimination value of each cluster set of the coordinate point of each monitoring element according to the cluster followability index is: For each cluster set of the coordinate points of each monitoring element, the number of elements in the cluster set is taken as the numerator, the area of ​​the minimum circumscribed rectangle of the shape of the cluster set is taken as the denominator, and the ratio of the numerator to the denominator is taken as the cluster density of the cluster set; The product of the cluster followability index and the cluster density of the cluster set is used as the master-slave feature discrimination value of the cluster set.

8. The method for monitoring the operating fault status of a cloud computing server according to claim 1, characterized in that: The method for obtaining the cluster deviation of each cluster set of coordinate points of each monitoring element according to each cluster set of coordinate points of each monitoring element is: In the formula, Indicates The coordinate point of the monitoring element The cluster deviation of the cluster set is Indicates The monitoring element corresponds to The number of sliding windows on the cluster set, Indicates The monitoring element corresponds to The horizontal coordinate of the center point of the nth sliding window on the cluster set, Indicates The monitoring element corresponds to The horizontal coordinate of the center point of the cluster set, Indicates The monitoring element corresponds to The length of the horizontal axis range of the cluster sets, Indicates the first monitoring element corresponding to The cluster set The clustering partition function value of the sliding window, Indicates the first monitoring element corresponding to The mean of the clustering partition function values ​​of all sliding windows on the cluster set.

9. A method for monitoring the operating fault status of a cloud computing server according to claim 1, characterized in that: The method for obtaining the deviation range interval and the master-slave discrimination range interval of each cluster set of the coordinate point of each monitoring element according to the master-slave feature discrimination value and the cluster deviation degree is: In the formula, Indicates The coordinate point of the monitoring element The deviation range of cluster sets is Indicates The monitoring element corresponds to The horizontal coordinate of the center point of the cluster set, Indicates The coordinate point of the monitoring element In the rectangular coordinate system of cluster sets The standard deviation of the directrix, Indicates The coordinate point of the monitoring element The cluster deviation of the cluster sets; Indicates The coordinate point of the monitoring element The lower limit of the master-slave discrimination range of cluster sets, Indicates The coordinate point of the monitoring element The upper limit of the master-slave discrimination range of cluster sets, Indicates The coordinate point of the monitoring element The master-slave discrimination range interval of cluster sets, Indicates The coordinate point of the monitoring element The master-slave feature discriminant value of the cluster set, Indicates The mean of the master-slave feature discrimination values ​​of all cluster sets of the coordinate points of the monitoring elements.

10. A method for monitoring the operating fault status of a cloud computing server according to claim 1, characterized in that: The method of obtaining the decision tree after dimension reduction by using the deviation range interval and the master-slave discrimination range interval is: Taking the master-slave discrimination range interval of each cluster set of the coordinate point of each monitoring element as the master-slave discrimination range interval of the target cluster set, calculating the intersection interval between the master-slave discrimination range interval of the target cluster set and the deviation range interval of each other cluster set, and calculating the ratio of the length of the intersection interval to the length of the master-slave discrimination range interval of the target cluster set, if the ratio is lower than a third preset parameter, it is considered that the target cluster set is independent; The ratio of the length of the union interval of the deviation range intervals of all independent cluster sets of the coordinate points of each monitoring element to the length of the union interval of the deviation range intervals of all cluster sets of the coordinate points of each monitoring element is used as the master-slave relationship measurement index between any two monitoring elements; A decision tree is constructed through all monitoring elements, and a decision tree pruning algorithm is used to obtain a reduced-dimensional decision tree based on the master-slave relationship metric index.

Images