Intelligent operation and maintenance management system and method of data center

Abstract

The invention discloses an intelligent operation and maintenance management system and method for a data center. The method comprises the following steps: acquiring a network topology structure, performance data, power supply data and equipment state data from the data center; dividing the network into a plurality of sub-regions according to the network topology structure, and performing network resource configuration according to the performance data and the power supply data of the plurality of sub-regions; performing fault prediction on the plurality of sub-regions by using a fault prediction model constructed based on a deep learning algorithm to obtain predicted fault sub-regions and predicted fault types; and according to the predicted fault sub-region and the predicted fault type, locating a predicted fault node, and carrying out early warning operation on the predicted fault node. The invention relates to the technical field of network operation and maintenance management, and solves the technical problems of insufficient fault pre-judgment capability and insufficient intelligence and automation degree in an existing network operation and maintenance management system.

Description

Technical Field

[0001] The present invention belongs to the field of information technology, and relates to network management technology, and specifically to an intelligent operation and maintenance management system and method for a data center. Background Art

[0002] As the core hub of information technology, data centers gather a vast amount of hardware resources, including high-performance servers, large-capacity storage devices, and complex network devices, which together support the storage, processing, and transmission of data. In order to efficiently manage and maximize the use of these resources, the intelligent operation and maintenance management system of the data center came into being. Its core lies in the comprehensive intelligence, automation, and efficiency of operation and maintenance work through the integration of advanced IT operation and maintenance tools, automated processes, big data analysis, and machine learning technologies. The key to the intelligent operation and maintenance of data centers is that it not only needs to have the ability to monitor in real time, but also needs to be able to accurately capture every subtle change in the network environment of the data center, including key indicators such as CPU load, memory utilization, disk I / O performance, and network bandwidth usage, and can conduct in-depth analysis based on these data to provide accurate decision support for operation and maintenance personnel, ensuring the efficiency and accuracy of data center operation and maintenance work.

[0003] In the invention patent with publication number CN116170296B, an application was made for "a network automatic operation and maintenance management system and method". The system and method obtain network node status information and detect whether it is missing to determine whether the relevant network equipment has a power failure. If there is a power failure, it then determines whether the operation information of other related network devices is missing, thereby establishing a communication detection channel to troubleshoot and realize automatic fault node location when the network node is abnormal.

[0004] In the invention patent with publication number CN118509743A, an application was made for "an intelligent network operation and maintenance management system". The system collects performance parameters of network equipment, servers and databases through the data acquisition module, and transmits the data to the data analysis module for analysis and processing; if the temperature of the power distribution transmission cabinet is abnormal, it will be processed through the emergency processing module (including temperature monitoring auxiliary module, control module, control valve, cooling pipe and cable limit auxiliary component) to assist in cooling and limiting the cable.

[0005] Existing intelligent operation and maintenance management systems mainly focus on locating faults by checking the power status or monitoring temperature anomalies. However, there are many types of faults that may occur in data centers, such as hardware aging, software errors, configuration errors, network attacks and other types of faults. For example, hardware aging may cause the performance of the server to gradually decline, and there will be problems such as the long-term slow decrease in the read and write speed of the hard disk and the increase in the potential error rate of the memory chip. If you only rely on checking the power status and monitoring temperature anomalies, you will not be able to find them. In addition, power failures or temperature anomalies are usually not detected until the fault has occurred and developed to a certain extent, resulting in serious deficiencies in the timeliness of the system's fault handling, which increases the impact time and degree of damage of the fault on the data center business.

[0006] At the same time, although the operation and maintenance management system has achieved some automation functions, there are still a lot of manual operation links in actual operation. For example, daily inspection of equipment, handling of some simple faults, approval of configuration changes, etc. These manual operations are not only inefficient, but also prone to human errors. Especially in the case of large-scale data centers and a large number of devices, the workload of manual operations is huge, which increases the workload of operation and maintenance personnel and the probability of errors. Summary of the invention

[0007] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes an intelligent operation and maintenance management system and method for a data center, which are used to solve the technical problems of insufficient fault prediction capability, insufficient intelligence and automation in the existing network operation and maintenance management system.

[0008] To achieve the above-mentioned object, the first aspect of the present invention provides an intelligent operation and maintenance management system for a data center, comprising: a data acquisition module, a resource configuration module, a fault prediction module and a fault location module;

[0009] Data acquisition module: used to collect network topology, performance data, power supply data and equipment status data from the data center;

[0010] Resource configuration module: used to divide the network into several sub-areas according to the network topology structure; and to sort the importance of the performance data and power supply data of the several sub-areas, and configure network resources for the several sub-areas according to the importance;

[0011] Fault prediction module: used to use the fault prediction model to predict faults in several sub-areas, and obtain predicted fault sub-areas and predicted fault types; wherein the fault prediction model is built based on a deep learning algorithm and deployed in a data center;

[0012] Fault location module: used to locate the predicted fault node according to the predicted fault sub-area and predicted fault type, and perform early warning operations on the predicted fault node.

[0013] Based on the above technical modules, the present invention realizes intelligent operation and maintenance management of the data center, including comprehensive perception of network status, efficient use of resources, early warning and rapid location of faults, thereby ensuring the stable operation of the data center, improving operation and maintenance efficiency, and reducing the risk of business interruption caused by faults and enterprise operating costs.

[0014] Furthermore, the network is divided into several sub-areas according to the network topology structure, including:

[0015] A1, the devices in the network topology are represented as nodes, and the connection links between the devices are represented as edges, thus obtaining a graph data structure;

[0016] A2, obtaining the hop distance between connected devices in the network topology, and setting the modularity function in the graph data structure according to the hop distance; wherein the hop distance is measured by the traceroute tool;

[0017] A3, clustering operation is performed on the graph data structure based on the clustering algorithm according to the modularity function, and the network is divided into several sub-areas.

[0018] By dividing the network into several sub-areas, we can more clearly understand the resource demand and usage of different areas, and allocate network resources in a targeted manner based on the business characteristics, importance and other factors of each sub-area, ensuring the rational use of resources, improving the efficiency of resource allocation and avoiding waste of resources.

[0019] Further, setting the modularity function in the graph data structure according to the hop count distance includes:

[0020] A2-1, according to the connection links between devices, the edge weights in the graph data structure are: Among them, w ij represents the edge weight between nodes i and j, B ij represents the bandwidth of the link between nodes i and j, U ij represents the utilization rate of the link between nodes i and j, R ij Represents the reliability of the link between nodes i and j, obtained through historical failure data statistics, α i (i=1,2,3) represents the weight coefficient, and the sum is 1;

[0021] A2-2, according to the hop distance between connected devices, the distance weight in the graph data structure is: Among them, d ij represents the distance weight between nodes i and j, h ij represents the hop distance between nodes i and j;

[0022] A2-3, based on edge weight and distance weight, the modularity function is: Wherein, c represents the divided sub-region.

[0023] When setting the modularity function, not only conventional factors such as bandwidth and utilization of the link between devices are considered, but also the reliability factor of the link obtained through historical fault data statistics and the hop distance factor between connected devices are combined. In this way, the modularity function is determined by combining multiple factors, making the basis for dividing sub-areas more comprehensive and objective, and being able to divide more accurately according to the actual situation of the network, avoiding the problem of unreasonable division caused by a single factor.

[0024] Furthermore, the clustering operation on the graph data structure based on the clustering algorithm according to the modularity function includes:

[0025] A3-1, initialize all nodes in the graph data structure into independent sub-regions, and the initial modularity value of each sub-region is 0;

[0026] A3-2, for each node i in the graph data structure, the modularity change value after node i moves to neighbor node j is calculated according to the modularity function;

[0027] A3-3, determine whether the modularity change value is greater than 0; if yes, move node i to the sub-region where neighbor node j is located; if no, keep the sub-region where node i is located unchanged;

[0028] A3-4, repeat A3-2 and A3-3 until the sub-area ownership of all nodes no longer changes, and several sub-areas are obtained.

[0029] The clustering algorithm based on modularity function is used to divide the sub-areas. The modularity change value after the node moves to the neighboring node is continuously calculated, and the sub-area affiliation of the node is dynamically adjusted according to the change value until the sub-area affiliation of all nodes no longer changes. This dynamic clustering method has high flexibility and can automatically adjust the sub-area division results according to the real-time status of the network, such as the increase or decrease of equipment, changes in links, etc., so that the divided sub-areas can always better adapt to the actual operation needs of the network, rather than a fixed division method that may gradually become unreasonable as the network develops.

[0030] Furthermore, the ranking of importance according to the performance data and power supply data of the plurality of sub-areas includes:

[0031] B1, collect performance data and power supply data of several sub-areas; the performance data includes bandwidth utilization, latency, packet loss rate, throughput, and number of connections; the power supply data includes voltage, current, power, frequency, number of power outages, and power outage time;

[0032] B2, standardize the performance data and power supply data according to the standardized formula to obtain a number of standard indicator data; wherein the number of standard indicator data includes the mth standard positive indicator of the nth sub-area and the mth standard negative indicator of the nth sub-region

[0033] B3, according to the formula Calculate the entropy value e of several standard indicator data m ; where N represents the number of sub-regions, p nm Represents the intermediate value used to calculate the entropy value;

[0034] B4, according to the entropy value e m Calculate the weights of several standard indicator data Where M represents the number of standard indicators;

[0035] B5, according to the formula The comprehensive score of each sub-region is calculated and ranked according to the importance.

[0036] By collecting rich performance data and power supply data, covering multiple key aspects of network operation and power security, the one-sidedness of relying on a single indicator to evaluate the importance of sub-regions is avoided, and the actual status of sub-regions in the network can be more comprehensively and accurately reflected. The weight of each standard indicator data is calculated by the entropy method, and its relative importance in the importance assessment can be determined according to the degree of dispersion of the data itself. The lower the entropy value, the smaller the degree of dispersion of the data, which means that the indicator provides more effective information and the greater the weight given.

[0037] Furthermore, the standardized formula is Among them, x nm represents the mth positive indicator of the nth sub-region, y nm represents the mth negative indicator of the nth sub-region, and Indicates the standardized indicator data;

[0038] It should be noted that the positive indicators in the present invention include: bandwidth utilization, throughput, number of connections, voltage, current, power, and frequency; the negative indicators include: delay, packet loss rate, number of power outages, power outage time, CPU utilization, and memory utilization.

[0039] Furthermore, the configuring of network resources for the plurality of sub-areas according to the importance includes:

[0040] C1, normalize the comprehensive score of each sub-region to obtain the resource allocation weights P of several sub-regions n ;

[0041] C2, according to the formula Get the resource allocation of several sub-areas; where B represents the total network bandwidth, IPN represents the number of available IP addresses, C represents the number of CPU cores, M represents the amount of memory, and B n 、IPN n , C n 、M n represents the corresponding resource allocation amount of the nth sub-area, and β1, β2, β3, and β4 represent the allocation coefficients of the corresponding resources.

[0042] Furthermore, the fault prediction model is constructed based on a deep learning algorithm, including:

[0043] D1, records the time, type and sub-area of ​​historical faults, and collects the performance data and power supply data of the sub-area in the preset time period before and after the fault, as well as the device status data of the sub-area, to form a historical time series data set; the device status data includes CPU usage and memory usage;

[0044] D2, standardize the historical event sequence data set according to the standardization formula, and annotate the historical sequence data set according to the fault time point and fault type to obtain the preprocessed data set;

[0045] D3, build a time series model based on the deep learning algorithm, divide the preprocessed data set into a training set and a validation set according to a preset ratio, and input the time series model for training and validation to obtain the validation results;

[0046] D4, iterative training is performed based on the verification results and the optimization algorithm, and the model with the best verification accuracy is saved as the fault prediction model;

[0047] It should be noted that the input of the fault prediction model is performance data, power supply data and equipment status data from several sub-areas of the data center, and the output is the time, type and sub-area of ​​the fault.

[0048] Network failures are usually not isolated events, and the changes in data before and after them often follow certain rules. Through the time series model based on deep learning algorithms, the time sequence information of the data can be effectively utilized to capture the change pattern of the data over time, better predict the occurrence time of the failure, and discover potential failures in advance.

[0049] Further, locating the predicted fault node according to the predicted fault sub-area and the predicted fault type includes:

[0050] E1, collects performance data, power supply data and equipment status data of several nodes in the predicted fault sub-area, and screens out fault judgment data according to the predicted fault type; wherein the predicted fault type includes performance fault, power fault and equipment fault;

[0051] E2, traverse and compare the fault judgment data of several nodes to see whether they are equal to or greater than the corresponding preset fault threshold; if yes, mark it as a predicted fault node; if no, continue to traverse the next node.

[0052] According to the predicted fault sub-area and fault type, the performance data, power supply data and equipment status data of several nodes in the area are collected in a targeted manner, and the fault judgment data related to the predicted fault type is screened out, which reduces the amount of data processing and improves the efficiency of fault location. By comparing the predicted fault nodes with the preset fault threshold, possible fault nodes can be quickly marked, providing a clear troubleshooting direction for operation and maintenance personnel and speeding up the speed of fault location.

[0053] A second aspect of the present invention provides an intelligent operation and maintenance management method for a data center, comprising:

[0054] S1, collects network topology, performance data, power supply data, and equipment status data from the data center;

[0055] S2, dividing the network into several sub-areas according to the network topology structure, ranking the importance of the several sub-areas according to the performance data and power supply data of the several sub-areas, and allocating network resources for the several sub-areas according to the importance;

[0056] S3, using a fault prediction model to perform fault prediction on several sub-areas to obtain predicted fault sub-areas and predicted fault types; wherein the fault prediction model is constructed based on a deep learning algorithm and deployed in a data center;

[0057] S4, locating the predicted fault node according to the predicted fault sub-area and the predicted fault type, and performing a warning operation on the predicted fault node.

[0058] Compared with the prior art, the present invention has the following beneficial effects:

[0059] By dividing the network into several sub-areas, we can more clearly understand the resource requirements and usage of different areas, and allocate network resources in a targeted manner according to the importance of each sub-area, ensuring the rational use of resources, improving the efficiency of resource allocation, and avoiding resource waste;

[0060] The fault prediction model built based on the deep learning algorithm can predict the time, type and sub-area of ​​faults in advance based on the data collected by the data center, providing enough time for operation and maintenance personnel to prepare;

[0061] The fault location module quickly locates the predicted fault node according to the predicted fault type and predicted fault sub-area, providing clear troubleshooting directions for operation and maintenance personnel, speeding up fault location and thus reducing the risk of business interruption caused by faults;

[0062] The intelligent operation and maintenance management system can automatically complete operations such as network sub-area division, resource allocation, fault prediction and location, reduce manual intervention, improve operation and maintenance efficiency, reduce business interruptions caused by faults, and rationally utilize resources, thereby reducing enterprise network operating costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be 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.

[0064] Figure 1 A technical flow chart of an intelligent operation and maintenance management system for a data center provided by the present invention;

[0065] Figure 2 A flow chart of the method for dividing a plurality of network sub-areas provided by the present invention;

[0066] Figure 3 A flow chart of a method for ranking the importance of a plurality of sub-regions provided by the present invention;

[0067] Figure 4 A flow chart of the method for determining and predicting fault nodes provided by the present invention. DETAILED DESCRIPTION

[0068] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. 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 creative work are within the scope of protection of the present invention.

[0069] See also Figure 1-Figure 4 The first aspect of the present invention provides an intelligent operation and maintenance management system for a data center, including:

[0070] Data acquisition module: used to collect network topology, performance data, power supply data and equipment status data from the data center;

[0071] Resource allocation module: divide the network into several sub-areas according to the network topology structure; and rank the importance of the several sub-areas according to the performance data and power supply data, and allocate network resources for the several sub-areas according to the importance;

[0072] Fault prediction module: used to use the fault prediction model to predict faults in several sub-areas, and obtain predicted fault sub-areas and predicted fault types; the fault prediction model is built based on a deep learning algorithm and deployed in the data center;

[0073] Fault location module: used to locate the predicted fault node according to the predicted fault sub-area and predicted fault type, and perform early warning operations on the predicted fault node;

[0074] It should be noted that in the system, the data acquisition module, resource configuration module, fault prediction module and fault location module are in communication connection, and the data acquisition module is in communication connection with the resource configuration module, fault prediction module and fault location module respectively.

[0075] In this embodiment, the data center is the data source of the entire intelligent operation and maintenance management system. Various data such as network topology, performance data, power supply data, and equipment status data are collected from the data center through the data acquisition module. These data provide the basis for subsequent resource configuration, fault prediction and location operations.

[0076] In order to ensure the reasonable allocation of network resources, in this embodiment, the network topology structure obtained by the data acquisition module is converted into a graph data structure, and the sub-areas are divided by using a clustering algorithm based on graph theory and combining the logical distance between devices.

[0077] In graph-theory-based clustering algorithms, traditional modularity functions focus on the difference between edge density within a region and edge density in a random situation. However, in actual network topologies, especially enterprise networks, the distance between nodes has an important impact on network performance, management, and fault propagation. For example, in a widely distributed enterprise network, even if two nodes are connected in the graph structure, if the distance between them is far (for example, there are many network hops), it may be more reasonable to divide them into different regions for management and fault location.

[0078] Therefore, when combining distance data, weight values ​​are assigned according to the distance between nodes, so that the connection density and distance factors can be comprehensively considered when calculating modularity. In this way, during the clustering process, the algorithm tends to divide nodes that are close and closely connected into the same area. For nodes that are far away, even if they are connected, due to the influence of distance weights, their contribution to being divided into the same area in the modularity calculation will be reduced, so that the divided area takes into account the connection relationship while also taking into account the physical or logical distance between nodes.

[0079] Specifically, firstly, the devices in the network topology are represented as nodes, and the connection links between the devices are represented as edges to obtain a graph data structure; then, the edge weights in the graph data structure are obtained according to the connection links between the devices: Among them, w ij represents the edge weight between nodes i and j, B ij represents the bandwidth of the link between nodes i and j, U ij represents the utilization rate of the link between nodes i and j, R ij Represents the reliability of the link between nodes i and j, obtained through historical failure data statistics, α i (i=1,2,3) represents the weight coefficient obtained based on the expert evaluation method, and the sum is 1; then the traceroute tool is used to measure and obtain the hop distance between the connected devices in the network topology structure, and the distance weight in the graph data structure is obtained according to the hop distance: Among them, d ij represents the distance weight between nodes i and j, h ij represents the hop distance between nodes i and j; the modularity function based on edge weight and distance weight is: Wherein, c represents the divided sub-area;

[0080] Finally, the graph data structure is clustered based on the clustering algorithm according to the modularity function, and the network is divided into several sub-areas. The specific process is as follows:

[0081] First, all nodes in the graph data structure are initialized as independent sub-regions, and the initial modularity value of each sub-region is 0;

[0082] Then, for each node i in the graph data structure, the modularity change value after node i moves to neighbor node j is calculated according to the modularity function;

[0083] Next, determine whether the modularity change value is greater than 0; if yes, move node i to the sub-region where neighbor node j is located; if no, keep the sub-region where node i is located unchanged;

[0084] Repeat the above operation until the sub-region ownership of all nodes does not change, and the regional division result of the network topology is obtained. At this point, each region contains several nodes and the links connecting them, forming a relatively independent sub-network. Each region can also be numbered or named for subsequent management and analysis.

[0085] In this embodiment, the reliability of the connection link based on historical fault data statistics can be obtained in the following way: collect the fault records of the link in the past period of time (such as the past year), including the number of faults, the duration of the faults, etc. For example, if a link has 5 faults in the past year, the average duration of each fault is 30 minutes, and the total time (in minutes) of a year is about 525600 minutes, then the unreliability of this link can be obtained in the following way:

[0086] Assume that the total link failure time is Tf and the total observation time is Tt, then the link reliability R = 1-Tf / Tt. Substituting the above values, the reliability of this link is R = 1-150 / 525600≈0.9997.

[0087] An example is given to illustrate the above method of dividing the network into sub-areas:

[0088] Suppose there is a simple network topology with 5 nodes, labeled A, B, C, D, and E. The connection relationship between them is as follows: A is connected to B, and the edge weight ω AB =0,6, the hop distance between A and B is h AB =1, then the distance weight d AB =1 / (1+1)=0.5; B is connected to C, ω BC =0.4,h BC =1, then the distance weight d BC =0.5; C is connected to D, ω CD =0.3,h CD =1, then the distance weight d CD =0.5; D is connected to E, ω DE =0.5,h DE =1, then the distance weight d DE =0.5; A is connected to E, ω AE =0.2, h AE =2, then the distance weight d AE =1 / 3;

[0089] First, each node is regarded as an independent sub-region, and the modularity value of each sub-region is 0;

[0090] First iteration:

[0091] Starting from node A: Consider moving A to the subregion where B is located:

[0092] The original modularity is 0 (A is a separate sub-region). If A is moved to the sub-region where B is located, the new sub-region is {A, B}. The modularity is recalculated and the modularity is Q_AB≈0.13. At this time, ΔQ=0.13, so A is moved to the sub-region where B is located.

[0093] Next, consider node B: consider moving B to the sub-area where C is located:

[0094] The modularity of the current sub-region {A, B} is Q_AB≈0.13; if B is moved to the sub-region where C is located, the new sub-region is {A, B, C}, and the modularity is recalculated to obtain Q_ABC≈0.19. At this time, ΔQ=0.06>0, so B is moved to the sub-region where C is located, and the sub-region becomes {A, B, C};

[0095] Consider node C: Consider moving C to the sub-region where D is located:

[0096] The modularity of the current sub-region {A, B, C} is Q_ABC≈0.19. If C moves to the sub-region where D is located, the new sub-region is {A, B, C, D}. The modularity is recalculated and Q_ABCD≈0.17 is obtained. At this time, ΔQ=-0.02<0, so C remains in the current sub-region {A, B, C}.

[0097] Consider node D: Consider moving D to the subregion where E is located:

[0098] When D is alone, the modularity is 0; if D is moved to the subregion where E is located, the new subregion is {D, E}, and the modularity is recalculated to obtain Q_DE≈0.09. At this time, ΔQ=0.09>0, so D is moved to the subregion where E is located, and the subregion becomes {D, E}.

[0099] Second iteration:

[0100] After the first round of iteration, the sub-regions are {A, B, C} and {D, E}. These two sub-regions are regarded as new nodes. Consider moving {A, B, C} to the sub-region where {D, E} is located. Then:

[0101] The modularity of the current sub-region {A,B,C} is Q_ABC≈0.19. If {A,B,C} moves to the sub-region where {D,E} is located, the new sub-region is {A,B,C,D,E}, and the modularity is recalculated to obtain Q_ABCDE≈0.14. At this time, the modularity change value ΔQ==-0.05<0, so {A,B,C} remains in the current sub-region, and the final clustering results are sub-regions {A,B,C} and {D,E}.

[0102] After obtaining the sub-areas of the network, they are ranked according to their importance based on their performance data and power supply data, and then network resources are configured for each sub-area so that each sub-area can obtain network resources that match its actual performance status and power supply conditions, thereby ensuring the efficient and stable operation of the entire network system and avoiding waste of resources or operational problems in some areas due to insufficient resources.

[0103] In this embodiment, the actual conditions of the sub-regions in terms of operating performance and power supply are collected to reflect various factors such as the load pressure of network traffic in each sub-region, the smoothness of data transmission, the stability and reliability of power supply, and the data is standardized to eliminate the influence of differences between different indicators. The amount of information is determined by calculating the entropy value of the standard data, and the weight of each indicator is derived based on this. The comprehensive score of each sub-region is then obtained through weighted calculation, thereby accurately obtaining the importance of the sub-region.

[0104] Specifically, we first collect performance data (such as bandwidth utilization, latency, packet loss rate, throughput, number of connections, etc.) and power supply data (such as voltage, current, power, frequency, number of power outages, and power outage duration) of several sub-areas; then, we combine the performance data and power supply data according to the standardized formula Standardization is performed to obtain several standard indicator data; among them, x nm represents the mth positive indicator of the nth sub-region, y nm Indicates the mth negative indicator of the nth sub-area, and the standard indicator data includes the standard positive indicator and standard negative indicators Then, according to the formula Calculate the entropy value e of several standard indicator data m ; where N represents the number of sub-regions, p nm Represents the intermediate value used to calculate the entropy value; according to the entropy value e m Calculate the weights of several standard indicator data Where M represents the number of standard indicators; finally, according to the formula Calculate the comprehensive score of each sub-area and sort them according to the size of the summary score to get the importance ranking;

[0105] It should be noted that the positive indicators in the present invention include: bandwidth utilization, throughput, number of connections, voltage, current, power, and frequency; the negative indicators include: delay, packet loss rate, number of power outages, power outage time, CPU utilization, and memory utilization.

[0106] When allocating resources for the network, the importance of each sub-region is first considered, and the comprehensive score of each sub-region is normalized to determine the resource allocation weight of each sub-region. Then use the total network bandwidth B, available IP address number IPN, CPU core number C, memory quantity M and other network resources as the allocation base, combined with the resource allocation weight of each sub-area, according to the formula Allocate corresponding resources to each sub-area; among them, B n 、IPN n , C n 、M n represents the corresponding resource allocation amount of the nth sub-area, and β1, β2, β3, and β4 represent the allocation coefficients of the corresponding resources. For example, for sub-areas with high importance, more network bandwidth, a suitable number of available IP addresses, sufficient CPU cores and memory will be allocated accordingly to meet their high-load network traffic processing needs and complex business operation requirements. For sub-areas with relatively low importance, resource allocation will be reasonably reduced, but their basic operation needs can still be guaranteed, thereby improving the overall utilization efficiency of network resources and reducing unnecessary resource idleness and waste.

[0107] In addition, a dynamic monitoring mechanism can be established during the resource allocation process. Because the network operation status changes dynamically, the performance data and power supply data of the sub-area may change over time, so their importance will also change accordingly. By monitoring the data of the sub-area in real time or regularly, when it is found that the data changes exceed a certain threshold, the resource allocation weight and resource allocation amount are adjusted in time, so that the network resources can always match the actual needs of the sub-area, further enhancing the adaptability and stability of the network system, and ensuring the continuity and high quality of network services.

[0108] The intelligent operation and maintenance management system of the data center of the present invention also includes a fault prediction module and a fault location module. By receiving various data from the data center in real time, monitoring the operating status of each sub-area, and combining the pre-deployed fault prediction model, it is determined which sub-areas may have fault problems, and then the fault location process is quickly started according to the prediction results, the specific nodes where the fault may occur are accurately found, and early warning signals are issued in time, so that the operation and maintenance personnel can do a good job of troubleshooting in advance.

[0109] In the fault prediction module, the fault prediction model is built based on a deep learning algorithm. It uses the rich information in historical data and, after data processing and model training, becomes an important tool to ensure the stable operation of the data center. Specifically, the construction of the fault prediction model includes the following steps:

[0110] D1, data collection and organization: Collect historical operation and maintenance records of the data center, record in detail the time, type and sub-area information of historical faults, and collect performance data and power supply data of the sub-area in the preset time period before and after the fault, as well as the equipment status data of the sub-area; for example, data from 10 minutes before the fault to 5 minutes after the fault can be collected to ensure that key status changes before and after the fault are covered; form a historical time series data set that reflects the changes in the operating status of the fault sub-area.

[0111] D2, data preprocessing: First, the historical time series data set is standardized according to the standardization formula to unify the dimension and range of the data; then, the historical series data set is labeled according to the fault time point and fault type, and each data point is matched with the corresponding network status information, so that the model can learn the characteristics of the data under different fault modes during the training process;

[0112] D3, model construction and preliminary training: First, build a time series model based on a deep learning algorithm. For example, according to the characteristics of the data and the needs of fault prediction, you can choose a suitable deep learning architecture, such as a long short-term memory network (LSTM) or a gated recurrent unit (GRU), to handle long-term dependencies in time series data; then, divide the preprocessed data set into a training set and a validation set according to a preset ratio (such as 80% training set, 20% validation set); input the training set into the time series model for training. During the training process, the model continuously adjusts its own parameters to minimize the difference between the prediction results and the labeled data. At the same time, use the validation set to conduct a preliminary evaluation of the training effect of the model and obtain the validation results, including evaluation indicators such as accuracy, recall, and F1 value, to understand the generalization ability of the model on unseen data. D4, model optimization and determination: Iterative training is performed based on the validation results and the optimization algorithm. If the validation results do not meet the expected accuracy requirements, use an optimization algorithm (such as stochastic gradient descent, Adam, etc.) to adjust the parameters of the model, such as learning rate, weight initialization, etc. And continuously improve the performance of the model through multiple iterative training.

[0113] Finally, during the iterative training process, the accuracy of each verification is continuously compared, and the model with the best verification accuracy is saved as the fault prediction model. This model can accurately output the time, type and sub-area of ​​the fault occurrence based on the input performance data, power supply data and equipment status data from several sub-areas of the data center, providing reliable prediction capabilities for the fault prediction module and achieving early warning of potential faults in the data center.

[0114] After receiving the predicted fault sub-area and the corresponding predicted fault type information, the next step is to determine the fault node based on the output of the fault prediction module:

[0115] First, the performance data, power supply data, and equipment status data of several nodes in the predicted fault sub-area are collected; then, the large amount of collected data is screened in a targeted manner according to the given predicted fault type to obtain fault judgment data that is closely related to the fault type. Specifically, if the predicted fault type is a performance fault, then the screened fault judgment data may mainly focus on indicators related to node performance, such as latency, throughput, and other data; if it is a power fault, the fault judgment data will focus on indicators related to power supply, such as voltage, current, power, and other data; and for equipment faults, the fault judgment data will focus on indicators related to equipment status, such as CPU usage, memory usage, and other conditions;

[0116] Then, the collected fault judgment data of several nodes are traversed, and the fault judgment data of each node is compared with the preset corresponding fault threshold in turn, and the nodes whose fault judgment data are greater than or equal to the preset fault threshold are screened out, marked as predicted fault nodes, and early warning signals are sent in time;

[0117] For example, for the delay indicator of a node (as part of the performance fault judgment data), if its actual delay value is equal to or greater than the delay preset fault threshold set for the performance fault, it meets the judgment condition of the faulty node and the node is marked as a predicted faulty node.

[0118] A second aspect of the present invention provides an intelligent operation and maintenance management method for a data center, comprising:

[0119] S1, collects network topology, performance data, power supply data, and equipment status data from the data center;

[0120] S2, dividing the network into several sub-areas according to the network topology structure, and configuring network resources for the several sub-areas according to the performance data and power supply data of the several sub-areas;

[0121] S3, using the fault prediction model to predict faults in several sub-areas, and obtaining predicted fault sub-areas and predicted fault types; wherein the fault prediction model is built based on a deep learning algorithm and deployed in a data center;

[0122] S4, locating the predicted fault node according to the predicted fault sub-area and the predicted fault type, and performing a warning operation on the predicted fault node.

[0123] Part of the data in the above formula is calculated by removing the dimension and taking its numerical value. The formula is a formula closest to the actual situation obtained by software simulation of a large amount of collected data; the preset parameters and preset thresholds in the formula are set by technical personnel in this field according to actual conditions or obtained through simulation of a large amount of data.

[0124] Working principle of the present invention:

[0125] In the data acquisition module, network topology, performance data, power supply data, and equipment status data are collected from the data center to provide basic data support for subsequent module operations;

[0126] In the resource configuration module, the network topology is converted into a graph data structure, and then the hop distance between the connected devices in the network topology is obtained. The modularity function in the graph data structure is set according to the hop distance. The modularity function then performs clustering operations on the graph data structure based on the clustering algorithm to divide the network into several sub-areas. After that, the comprehensive score of each sub-area is calculated based on the performance data and power supply data of several sub-areas, and the resource configuration weight is normalized to perform resource configuration for each sub-area.

[0127] In the fault prediction module, the fault prediction model built based on the deep learning algorithm predicts the time, type and sub-area of ​​the fault according to the relevant data of several sub-areas input;

[0128] In the fault location module, the performance data, power supply data and equipment status data of several nodes in the predicted fault sub-area are first collected, and the predicted fault nodes are obtained according to the predicted fault type and the preset fault threshold. The predicted fault nodes are then warned, providing a clear troubleshooting direction for the operation and maintenance personnel.

[0129] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. An intelligent operation and maintenance management system for a data center, characterized in that: include: Data acquisition module: used to collect network topology, performance data, power supply data and equipment status data from the data center; Resource configuration module: used to divide the network into several sub-areas according to the network topology structure; and to sort the importance of the performance data and power supply data of the several sub-areas, and configure network resources for the several sub-areas according to the importance; Fault prediction module: used to perform fault prediction on several sub-areas using a fault prediction model to obtain predicted fault sub-areas and predicted fault types; wherein the fault prediction model is constructed based on a deep learning algorithm; Fault location module: used to locate the predicted fault node according to the predicted fault sub-area and predicted fault type, and perform early warning operations on the predicted fault node.

2. The intelligent operation and maintenance management system of a data center according to claim 1, characterized in that: The network is divided into several sub-areas according to the network topology structure, including: A1, the devices in the network topology are represented as nodes, and the connection links between the devices are represented as edges, thus obtaining a graph data structure; A2, obtaining the hop distance between connected devices in the network topology, and setting the modularity function in the graph data structure according to the hop distance; A3, clustering operation is performed on the graph data structure based on the clustering algorithm according to the modularity function, and the network is divided into several sub-areas.

3. The intelligent operation and maintenance management system of a data center according to claim 2, characterized in that: The step of setting the modularity function in the graph data structure according to the hop count distance includes: A2-1, according to the connection links between devices, the edge weights in the graph data structure are: Among them, w ij represents the edge weight between nodes i and j, B ij represents the bandwidth of the link between nodes i and j, U ij represents the utilization rate of the link between nodes i and j, R ij Represents the reliability of the link between nodes i and j, obtained through historical failure data statistics, α i (i=1,2,3) represents the weight coefficient, and the sum is 1; A2-2, according to the hop distance between connected devices, the distance weight in the graph data structure is: Among them, d ij represents the distance weight between nodes i and j, h ij represents the hop distance between nodes i and j; A2-3, based on edge weight and distance weight, the modularity function is: Wherein, c represents the divided sub-region.

4. The intelligent operation and maintenance management system of a data center according to claim 2, characterized in that: The clustering operation on the graph data structure based on the clustering algorithm according to the modularity function includes: A3-1, initialize all nodes in the graph data structure into independent sub-regions, and the initial modularity value of each sub-region is 0; A3-2, for each node i in the graph data structure, the modularity change value after node i moves to neighbor node j is calculated according to the modularity function; A3-3, determine whether the modularity change value is greater than 0; if yes, move node i to the sub-region where neighbor node j is located; if no, keep the sub-region where node i is located unchanged; A3-4, repeat A3-2 and A3-3 until the sub-area ownership of all nodes no longer changes, and several sub-areas are obtained.

5. The intelligent operation and maintenance management system of a data center according to claim 1, characterized in that: The importance of ranking is based on the performance data and power supply data of several sub-areas, including: B1, collect performance data and power supply data of several sub-areas; the performance data includes bandwidth utilization, latency, packet loss rate, throughput, and number of connections; the power supply data includes voltage, current, power, frequency, number of power outages, and power outage time; B2, standardize the performance data and power supply data according to the standardized formula to obtain a number of standard indicator data; wherein the number of standard indicator data includes the mth standard positive indicator of the nth sub-area and the mth standard negative indicator of the nth sub-region B3, according to the formula Calculate the entropy value e of several standard indicator data m ; where N represents the number of sub-regions, p nm Represents the intermediate value used to calculate the entropy value; B4, according to the entropy value e m Calculate the weights of several standard indicator data Where M represents the number of standard indicators; B5, according to the formula The comprehensive score of each sub-region is calculated and ranked according to the importance.

6. The intelligent operation and maintenance management system of a data center according to claim 5, characterized in that: The standardized formula is Among them, x nm represents the mth positive indicator of the nth sub-region, y nm represents the mth negative indicator of the nth sub-region, and Indicates the standardized indicator data.

7. The intelligent operation and maintenance management system of a data center according to claim 1, characterized in that: The network resource configuration for the plurality of sub-areas according to the importance includes: C1, normalize the comprehensive score of each sub-region to obtain the resource allocation weights P of several sub-regions n ; C2, according to the formula Get the resource allocation of several sub-areas; where B represents the total network bandwidth, IPN represents the number of available IP addresses, C represents the number of CPU cores, M represents the amount of memory, and B n 、IPN n , C n 、M n represents the corresponding resource allocation amount of the nth sub-area, and β1, β2, β3, and β4 represent the allocation coefficients of the corresponding resources.

8. The intelligent operation and maintenance management system of a data center according to claim 1, characterized in that: The fault prediction model is built based on a deep learning algorithm and includes: D1, records the time, type and sub-area of ​​historical faults, and collects the performance data and power supply data of the sub-area in the preset time period before and after the fault, as well as the device status data of the sub-area, to form a historical time series data set; the device status data includes CPU usage and memory usage; D2, standardize the historical event sequence data set according to the standardization formula, and annotate the historical sequence data set according to the fault time point and fault type to obtain the preprocessed data set; D3, build a time series model based on the deep learning algorithm, divide the preprocessed data set into a training set and a validation set according to a preset ratio, and input the time series model for training and validation to obtain the validation results; D4, iterative training is performed based on the verification results and the optimization algorithm, and the model with the best verification accuracy is saved as the fault prediction model.

9. The intelligent operation and maintenance management system of a data center according to claim 1, characterized in that: The locating the predicted fault node according to the predicted fault sub-area and the predicted fault type includes: E1, collects performance data, power supply data and equipment status data of several nodes in the predicted fault sub-area, and screens out fault judgment data according to the predicted fault type; wherein the predicted fault type includes performance fault, power fault and equipment fault; E2, traverse and compare the fault judgment data of several nodes to see whether they are equal to or greater than the corresponding preset fault threshold; if yes, mark it as a predicted fault node; if no, continue to traverse the next node.

10. An intelligent operation and maintenance management method for a data center, applied to an intelligent operation and maintenance management system for a data center according to any one of claims 1 to 9, characterized in that: include: S1, collects network topology, performance data, power supply data, and equipment status data from the data center; S2, dividing the network into several sub-areas according to the network topology structure, ranking the importance of the several sub-areas according to the performance data and power supply data of the several sub-areas, and allocating network resources for the several sub-areas according to the importance; S3, using a fault prediction model to predict faults in several sub-areas to obtain predicted fault sub-areas and predicted fault types; wherein the fault prediction model is constructed based on a deep learning algorithm and deployed in a data center; S4, locating the predicted fault node according to the predicted fault sub-area and the predicted fault type, and performing a warning operation on the predicted fault node.

Images