Integrated intelligent control monitoring method and device for server room

Abstract

The invention discloses an integrated intelligent control monitoring method and device for a server room. The method comprises the following steps: constructing an equipment topological graph of the server room; constructing a message service queue of each equipment control instruction; wherein the control instruction comprises corresponding state change information; the message service queue comprises a message producer service queue and a message consumer service queue; inputting the equipment topological graph and the message service queue into a DeepWork operation and maintenance diagnosis model to obtain an operation and maintenance diagnosis result of each piece of equipment; wherein the DeepWork operation and maintenance diagnosis model comprises a graph embedding layer, a GraphSAGE network layer, a machine room equipment knowledge graph reasoning layer, a path sorting network layer and a fusion layer; and according to the operation and maintenance diagnosis result of each device and the machine room device knowledge graph, executing a predefined operation and maintenance script to control each device. Through the DeepWork operation and maintenance diagnosis model, the working efficiency and accuracy of server room operation and maintenance are further enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent operation and maintenance of server rooms, and in particular to an integrated intelligent control and monitoring method and device, and computing equipment for a server room. Background Art

[0002] As the scale of data centers continues to expand, the operation and maintenance of server rooms has become increasingly complex. The operation and maintenance of server rooms covers the continuous monitoring, maintenance and management of the software and hardware systems in the server room to ensure the stable operation, efficient performance and security of the system. For example, in terms of software, it includes monitoring of the running status of applications (such as response time, error rate, transaction completion rate, etc.), monitoring of the running status of middleware such as databases, message queues, caches, and monitoring of server operating systems (such as CPU usage, memory usage, disk I / O, etc.); in terms of hardware, it includes server monitoring (such as CPU temperature, fan speed, power module, etc.), network equipment monitoring (monitoring the running status of network equipment such as switches, routers, firewalls, etc.), as well as network traffic and bandwidth utilization.

[0003] Existing server room operation and maintenance methods include manual inspection methods, threshold-based monitoring methods, log analysis methods, and rule-based automated scripts. Among them, the manual inspection method requires maintenance personnel to regularly check the hardware equipment in the room, including servers, network equipment, air conditioning systems, etc. However, it is inefficient, prone to omissions, and difficult to detect faults in a timely manner. The threshold-based monitoring method monitors key performance indicators (such as CPU utilization, memory usage, disk space, etc.) by setting thresholds, and triggers an alarm when the preset threshold is reached. The disadvantage is that improper threshold setting will lead to false positives or false negatives, and lacks the ability to predict faults. The log analysis method analyzes the log files generated by the device to find abnormal behavior or error information. The disadvantage is that the log volume is huge, the efficiency of manual analysis is low, and it is difficult to find potential problems in real time. The rule-based automated script method uses predefined scripts to automatically perform common operation and maintenance operations, such as backup and restart services. The disadvantage is that it is not flexible enough and difficult to deal with complex or unexpected situations.

[0004] In summary, the existing server room operation and maintenance methods do not effectively manage the mutual influence and dependencies between different devices, and lack the ability to predict and handle faults in complex scenarios. Summary of the invention

[0005] In view of the above problems, the present invention is proposed to provide an integrated intelligent control and monitoring method and device, and computing equipment for a server room that overcomes the above-mentioned problem of low fault prediction capability in complex scenarios.

[0006] According to one aspect of the present invention, there is provided an integrated intelligent control and monitoring method for a server room, comprising:

[0007] Constructing a device topology diagram of the server room, wherein the device topology diagram includes the connection relationship between various servers, switches, and routers, as well as the ports, traffic, and operating performance parameters of various online services;

[0008] Constructing a message service queue for each device control instruction; wherein the control instruction includes corresponding state change information; the message service queue includes a message producer service queue and a message consumer service queue;

[0009] Input the device topology map and the message service queue into the DeepWork operation and maintenance diagnosis model to obtain the operation and maintenance diagnosis results of each device; wherein the DeepWork operation and maintenance diagnosis model includes a graph embedding layer, a GraphSAGE network layer, a computer room equipment knowledge graph reasoning layer, a path sorting network layer and a fusion layer;

[0010] According to the operation and maintenance diagnosis results of each device and the equipment knowledge graph of the computer room, a predefined operation and maintenance script is executed to control each device; wherein the operation and maintenance script includes a device dynamic control script and a service flow switching script.

[0011] In an optional manner, the connection relationship includes a physical connection relationship and a logical connection relationship; wherein the logical connection relationship includes a connection mode of a virtual switch and a virtual firewall;

[0012] The operating performance parameters include recording the CPU utilization of switches and routers, server memory utilization, server disk I / O performance indicators, and server port throughput.

[0013] In an optional manner, the graph embedding layer embeds the nodes in the device topology graph by using a DeepWalk algorithm to obtain an embedded representation of each device;

[0014] The graph embedding layer includes multi-hop relationships between the nodes.

[0015] In an optional manner, the GraphSAGE network layer obtains an embedded representation of each device by aggregating information of each device and corresponding neighboring nodes; wherein the GraphSAGE network layer includes a first GraphSAGE convolutional layer and a second GraphSAGE convolutional layer;

[0016] The first GraphSAGE convolutional layer inputs the initial embedding representation of each device, randomly samples a first preset number of nodes from the neighboring nodes of each device, aggregates the embedding representation of the neighboring nodes of each device through a first aggregation function, combines the embedding representation of the neighboring nodes with the initial embedding representation of the device, and then performs a nonlinear transformation;

[0017] The second GraphSAGE convolutional layer inputs the initial embedding representation of the device updated by the first GraphSAGE convolutional layer, randomly samples a second preset number of nodes from the neighbor nodes of each device, aggregates the embedding representation of the neighbor nodes of each device through a second aggregation function, combines the embedding representation of the neighbor nodes with the initial embedding representation of the device, and then performs a nonlinear transformation.

[0018] In an optional manner, the computer room equipment knowledge graph reasoning layer embeds entities and relationships in the equipment topology graph through the TransE model; and performs relationship reasoning based on the embedded representation to obtain potential relationships between devices.

[0019] In an optional manner, the path sorting network layer optimizes the communication paths between devices based on the interaction frequency, data flow and importance between the devices.

[0020] In an optional manner, executing a predefined operation and maintenance script to control each device according to the operation and maintenance diagnosis result of each device and the equipment knowledge graph of the computer room further includes:

[0021] According to the operation and maintenance diagnosis results output by the DeepWork operation and maintenance diagnosis model, identify the equipment that needs to be controlled; wherein the operation and maintenance diagnosis results include the health status, performance indicators and potential faults of the equipment;

[0022] According to the equipment knowledge graph of the computer room, the corresponding equipment dynamic control script and service flow switching script are determined for the equipment that needs to be controlled, and added to the message service queue in the form of a message, which is executed by the message consumer service queue;

[0023] Among them, the device dynamic control script is used to adjust device configuration, restart services and resource allocation; the service flow switching script is used to transfer network traffic from potential faulty devices to other devices by adjusting load balancing.

[0024] In an optional manner, the calculation formula of the first GraphSAGE convolutional layer is:

[0025]

[0026] Among them, AttentionAgg1() is an aggregation function based on the first attention mechanism; is the set of sampled neighbor nodes N of node v k (v) The initial embedding representation of each neighbor node u;

[0027] The calculation formula of the second GraphSAGE convolutional layer is:

[0028]

[0029] Among them, AttentionAgg2() is an aggregation function based on the second attention mechanism; is the set of sampled neighbor nodes N of node v k (v) The initial embedding representation of each neighbor node u.

[0030] In an optional manner, the calculation formula of the attention weight of the first attention mechanism is:

[0031]

[0032] Among them, a1 is a learnable vector used to calculate the attention score; W1 is a learnable weight matrix used to transform node embedding; LeakyReLU is the activation function, and its function expression is u' is the neighbor node set N k Another neighbor node in (v);

[0033] The calculation formula of the attention weight of the second attention mechanism is:

[0034]

[0035] Among them, a2 is a learnable vector used to calculate the attention score; W2 is a learnable weight matrix used to transform node embedding; LReLU is the activation function, and its function expression is u' is the neighbor node set N k Another neighbor node in (v).

[0036] According to another aspect of the present invention, there is provided an integrated intelligent control and monitoring device for a server room, comprising:

[0037] A device topology construction module is used to construct a device topology of the server room, wherein the device topology includes the connection relationship between various servers, switches, and routers, as well as the ports, traffic, and operating performance parameters of various online services;

[0038] A consumer service queue construction module is used to construct a message service queue for each device control instruction; wherein the control instruction contains corresponding state change information; the message service queue includes a message producer service queue and a message consumer service queue;

[0039] An operation and maintenance diagnosis module, used to input the device topology map and the message service queue into the DeepWork operation and maintenance diagnosis model to obtain the operation and maintenance diagnosis results of each device; wherein the DeepWork operation and maintenance diagnosis model includes a graph embedding layer, a GraphSAGE network layer, a computer room equipment knowledge graph reasoning layer, a path sorting network layer and a fusion layer;

[0040] The equipment control module is used to execute predefined operation and maintenance scripts to control each device according to the operation and maintenance diagnosis results of each device and the equipment knowledge graph of the computer room; wherein the operation and maintenance scripts include equipment dynamic control scripts and service flow switching scripts.

[0041] According to another aspect of the present invention, there is provided a computing device, comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus;

[0042] According to the solution provided by the present invention, a device topology diagram of a server room is constructed, wherein the device topology diagram includes the connection relationship between each server, switch, and router, as well as the ports, traffic, and operating performance parameters of various online services; a message service queue for each device control instruction is constructed; wherein the control instruction contains corresponding state change information; the message service queue includes a message producer service queue and a message consumer service queue; the device topology diagram and the message service queue are input into the DeepWork operation and maintenance diagnosis model to obtain the operation and maintenance diagnosis results of each device; wherein the DeepWork operation and maintenance diagnosis model includes a graph embedding layer, a GraphSAGE network layer, a computer room equipment knowledge graph reasoning layer, a path sorting network layer, and a fusion layer; according to the operation and maintenance diagnosis results of each device and the computer room equipment knowledge graph, a predefined operation and maintenance script is executed to control each device; wherein the operation and maintenance script includes a device dynamic control script and a service flow cutting script. The present invention processes device control instructions through a message service queue, ensures the timeliness of the instructions, and facilitates the management and tracking of the state changes of the instructions. The DeepWalk algorithm is used in the graph embedding layer of the DeepWork operation and maintenance diagnosis model to map devices and their relationships into a low-dimensional space. The GraphSAGE network layer learns node features from local neighborhoods to enhance the ability to understand complex relationships between devices. The relationship reasoning is performed through the computer room equipment knowledge graph reasoning layer to mine implicit associations between devices, further enhancing the model's diagnostic capabilities. The dynamic control scripts for devices and service flow switching scripts are automatically executed based on the operation and maintenance diagnosis results and the computer room equipment knowledge graph, which can dynamically adjust the service load distribution according to real-time conditions to ensure service stability, significantly improving the efficiency and accuracy of server room operation and maintenance work.

[0043] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0045] Figure 1 A schematic diagram showing a flow chart of an integrated intelligent control and monitoring method for a server room according to an embodiment of the present invention;

[0046] Figure 2A schematic diagram showing a framework of an integrated intelligent control and monitoring device for a server room according to an embodiment of the present invention;

[0047] Figure 3a to Figure 3b A schematic diagram of a server room according to an embodiment of the present invention is shown;

[0048] Figure 4 A module schematic diagram of an integrated intelligent control and monitoring device for a server room according to an embodiment of the present invention is shown;

[0049] Figure 5 A schematic diagram of the structure of a computing device according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0050] The exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present invention and to enable the scope of the present invention to be fully communicated to those skilled in the art.

[0051] Figure 1 The flowchart of the integrated intelligent control and monitoring method of the server room according to the embodiment of the present invention is shown. Figure 1 As shown, the following steps are included:

[0052] Step S101, constructing a device topology diagram of the server room, wherein the device topology diagram includes the connection relationship between various servers, switches, and routers, as well as the ports, traffic, and operating performance parameters of various online services.

[0053] In this embodiment, Figure 3a to Figure 3b As shown, the connection relationship includes a physical connection relationship and a logical connection relationship; wherein the logical connection relationship includes a connection mode of a virtual switch and a virtual firewall;

[0054] The operating performance parameters include recording the CPU utilization of switches and routers, server memory utilization, server disk I / O performance indicators, and server port throughput.

[0055] Specifically, we use network scanning tools (such as Nmap, SolarWinds, etc.) and monitoring tools (such as Zabbix, Prometheus, Grafana, etc.) to detect the server, switch and router information (such as IP address, MAC address, port configuration, connection status, etc.) in the computer room, and the physical or logical connection relationship between storage devices (port number, VLAN configuration, routing protocol, etc.). We use service monitoring tools to collect the port numbers, traffic data and operating performance parameters (such as CPU usage, memory usage, response time, etc.) of various online services.

[0056] Create a topology map through API functions such as Grafana or Neo4j, and draw nodes of servers, switches, routers, and various services in the topology map. Label each node with its name, IP address, MAC address, and various service information. Use straight lines, curves, or arrows to represent the physical or logical connection relationship between devices, and label the port number, VLAN configuration, routing protocol, traffic data, and operating performance parameters on the connection line. The layout of the topology map can be constructed according to the actual layout of the computer room and the network architecture. For example, divide the topology map into different layers such as the core layer, aggregation layer, and access layer, and use different colors, shapes, and sizes to represent different types of devices and connection relationships.

[0057] For example, to create the above nodes and relationships in Neo4j via a Cypher query:

[0058] CREATE(s1:Server{name:'Server1',IP:'192.168.1.1',MAC:'00-1A-2B-3C-4D-5E'})

[0059] CREATE(sw1:Switch{name:'Switch1',IP:'192.168.1.10',MAC:'01-23-45-67-89-AB'})

[0060] CREATE(r1:Router{name:'Router1',IP:'192.168.1.254',MAC:'02-CD-EF-GH-IJ-KL'})

[0061] CREATE(svc1:Service{name:'WebService',port:80,responseTime:50})

[0062] CREATE(s1)-[:CONNECTED_TO{port:'Eth1',VLAN:10,routingProtocol:'N / A',traffic:'100Mbps'}]->(sw1)

[0063] CREATE(sw1)-[:CONNECTED_TO{port:'Gi0 / 1',VLAN:10,routingProtocol:'OSPF',traffic:'200Mbps'}]->(r1)

[0064] CREATE(svc1)-[:DEPLOYED_ON{port:'internal port',traffic:'5Mbps',cpuUsage:10,memoryUsage:20}]->(s1)

[0065] Step S102, constructing a message service queue for each device control instruction; wherein the control instruction includes corresponding state change information; and the message service queue includes a message producer service queue and a message consumer service queue.

[0066] In order to manage and control the above devices and services, a message service queue is introduced to process the control instructions of devices and services. The message queue can centrally manage and track the sending and receiving of control instructions, which is convenient for problem troubleshooting and performance tuning.

[0067] Specifically, parameters such as the number of queues, persistence strategy, and security settings are configured through message queue systems such as RabbitMQ and Kafka. The message producer service is implemented in the system that needs to send control instructions, and the control instructions are encapsulated into messages and sent to the message queue. The message producer handles possible sending failures and implements the message consumer service in the system that needs to receive and execute control instructions. The control instructions are received from the message queue, the instruction content is parsed, and the corresponding control operations are performed according to the instruction type and the target device. By monitoring the operating status of the message queue system (including the sending, receiving, and execution of control instructions, as well as any exceptions or errors), the stability and reliability of the system are further improved.

[0068] Step S103, input the device topology map and the message service queue into the DeepWork operation and maintenance diagnosis model to obtain the operation and maintenance diagnosis results of each device; wherein the DeepWork operation and maintenance diagnosis model includes a graph embedding layer, a GraphSAGE network layer, a computer room equipment knowledge graph reasoning layer, a path sorting network layer and a fusion layer.

[0069] In this embodiment, the complex device topology diagram and message service queue are converted into a low-dimensional vector representation through the graph embedding layer, which reduces the complexity of data processing and improves the operating efficiency of the model. The computer room equipment knowledge graph reasoning layer can deeply understand the relationship and operation logic between devices through the pre-built knowledge graph, and can more accurately locate potential faults in computer room operation and maintenance. By sorting the paths of potential faults through the path sorting network layer, the equipment and links that are most likely to have problems can be quickly located, greatly reducing the troubleshooting time. By integrating diagnostic information from different levels through the fusion layer, a more comprehensive operation and maintenance diagnosis result can be formed, which improves the accuracy and reliability of the diagnosis.

[0070] Specifically, Figure 2 As shown in Figure 1, graph embedding algorithms such as DeepWalk and Node2Vec are used to convert the device topology graph into node vectors in a low-dimensional vector space, where each node vector contains information about the device and its neighboring devices. The implicit representation of the node vector is further learned through the GraphSAGE network layer, where the GraphSAGE network layer can use the node's own attributes and the information of neighboring nodes to generate a richer and more accurate node representation. The computer room equipment knowledge graph (including equipment type, function, failure mode and other information) is introduced into the computer room equipment knowledge graph reasoning layer. Fault reasoning is performed on the node representation generated by the GraphSAGE network layer through reasoning algorithms such as path query and rule reasoning. Then, the path sorting network layer generates potential fault paths based on the results of the knowledge graph reasoning layer. The HITS path sorting algorithm is used to sort the potential fault paths to determine the devices and links that are most likely to have problems. The diagnostic information of the graph embedding layer, GraphSAGE network layer, computer room equipment knowledge graph reasoning layer and path sorting network layer are integrated through the fusion layer. Finally, the final operation and maintenance diagnosis results shown in Table 1 are output through the fusion algorithm.

[0071] Table 1: Operation and maintenance diagnosis results of some equipment

[0072] Device ID Device Type Failure probability Potential failure paths D1 switch 0.8 D1→D2→D3→Message Service Queue 1 D2 server 0.5 D2→D4→Message Service Queue 2 D3 router 0.6 D3 → External Network D4 Storage devices 0.3 D4 → D1

[0073] Through the above table, operation and maintenance personnel can quickly understand which devices have a higher risk of failure (D1 has the highest risk of failure) and take corresponding measures to prevent or repair them.

[0074] In an optional manner, the graph embedding layer embeds the nodes in the device topology graph by using a DeepWalk algorithm to obtain an embedded representation of each device;

[0075] The graph embedding layer includes multi-hop relationships between the nodes.

[0076] In this embodiment, the DeepWalk algorithm encodes graph structure information into the embedded representation of nodes by simulating the path of random walks of nodes in the graph. Even in high-dimensional and complex device topology graphs, the complex relationships between nodes can be reflected through low-dimensional embedding vectors. After converting the device topology graph into an embedded representation, the computational complexity of subsequent clustering is significantly reduced. When generating random walk paths, the DeepWalk algorithm can include multi-hop relationships between nodes, so that the embedded representation more comprehensively reflects the dependencies and associations between devices, and can more accurately diagnose potential faults.

[0077] Specifically, a certain number of random walk paths are generated for each node in the graph embedding layer, and the length and number of the paths can be adjusted according to actual needs. Using the DeepWalk algorithm, the node sequence in the random walk path is used as input, and the embedded representation of the node is learned through a model such as the Skip-Gram model. This model attempts to maximize the co-occurrence probability between a node and its neighboring nodes (nodes that appear in the random walk path) to learn the association information between nodes. The learned embedded representation is input as a feature into the DeepWork operation and maintenance diagnosis model for fault prediction.

[0078] As shown in Table 2, this is a device topology graph containing four nodes (D1, D2, D3, D4) and connection relationships. After graph embedding using the DeepWalk algorithm, the embedded representation of each device is obtained.

[0079] Table 2

[0080] Device ID Embedded representation (simplified) D1 [0.1,0.8] D2 [0.5,0.4] D3 [0.2,0.7] D4 [0.6,0.3]

[0081] In Table 2, D1 and D3 are closely connected in the topology graph, and their embedded representations are closer in the vector space, indicating that D1 and D3 have similar failure modes.

[0082] In an optional manner, the GraphSAGE network layer obtains an embedded representation of each device by aggregating information of each device and corresponding neighboring nodes; wherein the GraphSAGE network layer includes a first GraphSAGE convolutional layer and a second GraphSAGE convolutional layer;

[0083] The first GraphSAGE convolutional layer inputs the initial embedding representation of each device, randomly samples a first preset number of nodes from the neighboring nodes of each device, aggregates the embedding representation of the neighboring nodes of each device through a first aggregation function, combines the embedding representation of the neighboring nodes with the initial embedding representation of the device, and then performs a nonlinear transformation;

[0084] The second GraphSAGE convolutional layer inputs the initial embedding representation of the device updated by the first GraphSAGE convolutional layer, randomly samples a second preset number of nodes from the neighbor nodes of each device, aggregates the embedding representation of the neighbor nodes of each device through a second aggregation function, combines the embedding representation of the neighbor nodes with the initial embedding representation of the device, and then performs a nonlinear transformation.

[0085] In this embodiment, the GraphSAGE network layer not only generates embedded representations for nodes, but also generates embedded representations for unseen nodes, which has stronger generalization ability when processing new data. The multi-layer convolutional layer of GraphSAGE can aggregate the information of neighbor nodes layer by layer, thereby capturing the multi-hop relationship between nodes, so that the GraphSAGE network layer can learn richer node embedding representations.

[0086] Specifically, the initial embedding representation of each device is input into the first GraphSAGE convolutional layer. A first preset number of nodes are randomly sampled from the neighbor nodes of each device as a sampling node set. The embedding representations in the sampling node set are aggregated to the target node through a first aggregation function such as mean aggregation and maximum pooling. The aggregated neighbor node embedding representation is combined with the initial embedding representation of the device (such as splicing), and then the updated embedding representation is obtained through a nonlinear transformation (such as a fully connected layer plus an activation function).

[0087] Then, the updated device embedding representation of the first GraphSAGE convolutional layer is input into the second GraphSAGE convolutional layer. A second preset number of nodes are randomly sampled from the neighbor nodes of each device as a sampling node set. The embedding representations in the sampling node set are aggregated to the target node through a second aggregation function (which may be different from the first layer). The aggregated neighbor node embedding representation is combined with the updated device embedding representation of the first layer, and the final embedding representation is obtained again through a nonlinear transformation.

[0088] For example, for a device topology graph containing four device nodes (D1, D2, D3, D4) and several edges, in the first GraphSAGE convolutional layer, two nodes (such as D2 and D3) are randomly sampled from its neighboring nodes for each device node (such as D1), and then the embedding representations of these two neighboring nodes are aggregated to D1 through the mean aggregation function, and concatenated and nonlinearly transformed with the initial embedding representation of D1 to obtain the embedding representation of D1 after the first layer update. In the second GraphSAGE convolutional layer, D1 is again sampled from its neighboring nodes (which may include new neighbors or previously sampled neighbors), and similar aggregation, combination and transformation operations are performed to finally obtain the embedding representation of D1 after being processed by two GraphSAGE network layers, which will more comprehensively reflect the relationship between D1 and its neighboring nodes.

[0089] In an optional manner, the calculation formula of the first GraphSAGE convolutional layer is:

[0090]

[0091] Among them, AttentionAgg1() is an aggregation function based on the first attention mechanism; is the set of sampled neighbor nodes N of node v k (v) The initial embedding representation of each neighbor node u;

[0092] The calculation formula of the second GraphSAGE convolutional layer is:

[0093]

[0094] Among them, AttentionAgg2() is an aggregation function based on the second attention mechanism; is the set of sampled neighbor nodes N of node v k (v) The initial embedding representation of each neighbor node u.

[0095] In this embodiment, GraphSAGE not only considers the characteristics of the node itself, but also the characteristics of its surrounding neighbor nodes, so that richer node representations can be learned.

[0096] For example, for an undirected graph containing four nodes A, B, C, and D, each node has an initial feature vector h 0 .

[0097] In the first GraphSAGE convolutional layer, we sample node A’s neighbor nodes B and C, and use the attention mechanism’s aggregation function to aggregate the feature vectors of the two neighbor nodes.

[0098]

[0099] According to the feature vector of node A itself Get the first layer representation of A

[0100] In the second GraphSAGE convolutional layer, we sample A’s neighbor nodes again, still taking B and C as an example, and aggregate the feature vectors of these two neighbor nodes in the first layer.

[0101]

[0102] According to the feature vector of node A in the first layer Get the final representation vector of A.

[0103] In an optional manner, the calculation formula of the attention weight of the first attention mechanism is:

[0104]

[0105] Among them, a1 is a learnable vector used to calculate the attention score; W1 is a learnable weight matrix used to transform node embedding; LeakyReLU is the activation function, and its function expression is u' is the neighbor node set N k Another neighbor node in (v);

[0106] The calculation formula of the attention weight of the second attention mechanism is:

[0107]

[0108] Among them, a2 is a learnable vector used to calculate the attention score; W2 is a learnable weight matrix used to transform node embedding; LReLU is the activation function, and its function expression is u' is the neighbor node set N k Another neighbor node in (v).

[0109] In this embodiment, the attention mechanism is used to enable the model to dynamically assign weights according to the interactions between nodes, so as to better capture the important connections between nodes in the graph, thereby learning richer and more discriminative node representations.

[0110] For example, for an undirected graph containing four nodes A, B, C, and D, each node has an initial eigenvector h 0 .

[0111] In the first GraphSAGE convolutional layer, we sample node A’s neighbor nodes B and C and calculate the attention score. and

[0112]

[0113]

[0114] Aggregate the feature vectors of nodes B and C according to the attention scores:

[0115]

[0116] In the second GraphSAGE convolutional layer, we still sample node A’s neighbor nodes B and C and calculate the attention scores. and

[0117]

[0118]

[0119] Aggregate the feature vectors of nodes B and C according to the attention scores:

[0120]

[0121] In an optional manner, the computer room equipment knowledge graph reasoning layer embeds entities and relationships in the equipment topology graph through the TransE model; and performs relationship reasoning based on the embedded representation to obtain potential relationships between devices.

[0122] In this embodiment, the TransE model uses a knowledge graph embedding method to represent entities and relationships in the graph and perform relational reasoning to discover potential relationships between entities. The TransE model is used to map entities and relationships to a low-dimensional vector space. The goal of TransE is to minimize the following loss function:

[0123]

[0124] Among them, (h, r, t) is a true triple (head entity, relation, tail entity), (h', r', t') is a negative sample triple, T is the set of all true triplets, C is the set of all negative sample triplets, γ is the boundary value, d is the distance metric function (usually Euclidean distance or Manhattan distance), and n is an exponent.

[0125] Using the generated entity and relation embeddings, we can predict unknown relations or entities by querying similarity vectors. For example, if we know that there is a relation R between entities A and B, we can find the relation R with h A +Rh B The closest vectors predict other entities related to entity A.

[0126] For example, the entities (devices) are: Server1, Server2, Switch1, Router1

[0127] The relationship is: connects_to

[0128] Entities and relationships are encoded as:

[0129] Entity ID: Server1→1, Server2→2, Switch1→3, Router1→4

[0130] Relationship ID: connects_to → 1

[0131] Embedding means:

[0132] Use the TransE model to generate embedding vectors for each entity and relation:

[0133] h1(Server1), h2(Server2), h3(Switch1), h4(Router1), r1(connects_to)

[0134] Relational Reasoning:

[0135] Assume that Server1 connects to Switch1, which is a triple of (h1, r1, h3). If you want to know whether Server2 is also connected to Switch1, that is (h2, r1, vector to be found). Then use the TransE model to predict entities related to Server2 by finding the vector closest to h2+r1-h3. If h2+r1-h3 is closest to h3, it can be inferred that Server2 is also connected to Switch1. By comparing the predicted results with the actual connection relationship, the accuracy of the model can be evaluated.

[0136] In an optional manner, the path sorting network layer optimizes the communication paths between devices based on the interaction frequency, data flow and importance between the devices.

[0137] In this embodiment, by selecting the best communication path, communication delay can be reduced and data transmission speed can be increased. By allocating data streams, overload of certain paths can be avoided, and when problems occur in the main path, the backup path can be quickly switched to.

[0138] For example, the interaction frequency, data flow size and importance information between server A and server B, the interaction information between server A and switch X and router Y, and the interaction information between server B and switch X and router Y are collected.

[0139] A path ranking network model is constructed, which evaluates the quality of each path based on factors such as the interaction frequency between devices, data flow size and importance.

[0140] Calculate the path score from server A to server B through switch X, and the path score from server A to server B through router Y.

[0141] Assume that the path from server A to server B through switch X has a score of 90 points, while the path from server A to server B through router Y has a score of 75 points. Therefore, the path from server A to server B through switch X is selected as the optimal path. Optionally, check the interaction between devices regularly, such as once a day or a week. If it is found that the data flow between server A and server B has increased significantly, and the path from server A to server B through switch X begins to be congested, the path score needs to be re-evaluated. If the path from server A to server B through router Y now has a higher score, then switch to this path as the new optimal path.

[0142] Step S104, executing a predefined operation and maintenance script to control each device according to the operation and maintenance diagnosis results of each device and the equipment knowledge graph of the computer room; wherein the operation and maintenance script includes a device dynamic control script and a service flow switching script.

[0143] In this embodiment, the operation and maintenance script can automatically perform equipment control tasks, reducing the need for manual intervention. Based on real-time diagnosis results, the operation and maintenance script can respond and take action quickly. By combining the knowledge graph of computer room equipment, the operation and maintenance script can identify problems more accurately.

[0144] For example, if the CPU utilization of server A is continuously higher than 80%, server A and its associated services and connected devices can be found in the equipment knowledge graph of the computer room, the key business services carried by server A can be identified, and the main switch B and storage device C connected to it can be determined.

[0145] Select a device dynamic control script to adjust the load on server A. For example, lower the priority of non-critical processes on server A and transfer part of the workload to a standby server D. Alternatively, execute a customized device dynamic control script to adjust the workload on server A.

[0146] In an optional manner, executing a predefined operation and maintenance script to control each device according to the operation and maintenance diagnosis result of each device and the equipment knowledge graph of the computer room further includes:

[0147] According to the operation and maintenance diagnosis results output by the DeepWork operation and maintenance diagnosis model, identify the equipment that needs to be controlled; wherein the operation and maintenance diagnosis results include the health status, performance indicators and potential faults of the equipment;

[0148] According to the equipment knowledge graph of the computer room, the corresponding equipment dynamic control script and service flow switching script are determined for the equipment that needs to be controlled, and added to the message service queue in the form of a message, which is executed by the message consumer service queue;

[0149] Among them, the device dynamic control script is used to adjust device configuration, restart services and resource allocation; the service flow switching script is used to transfer network traffic from potential faulty devices to other devices by adjusting load balancing.

[0150] In this embodiment, the most appropriate control script can be selected to solve the problem based on the equipment knowledge graph in the computer room. The service flow switching script can quickly transfer network traffic from potential faulty equipment to healthy equipment, reducing the impact of the fault on the business.

[0151] According to the solution provided by the present invention, a device topology diagram of a server room is constructed, wherein the device topology diagram includes the connection relationship between each server, switch, and router, as well as the ports, traffic, and operating performance parameters of various online services; a message service queue for each device control instruction is constructed; wherein the control instruction contains corresponding state change information; the message service queue includes a message producer service queue and a message consumer service queue; the device topology diagram and the message service queue are input into the DeepWork operation and maintenance diagnosis model to obtain the operation and maintenance diagnosis results of each device; wherein the DeepWork operation and maintenance diagnosis model includes a graph embedding layer, a GraphSAGE network layer, a computer room equipment knowledge graph reasoning layer, a path sorting network layer, and a fusion layer; according to the operation and maintenance diagnosis results of each device and the computer room equipment knowledge graph, a predefined operation and maintenance script is executed to control each device; wherein the operation and maintenance script includes a device dynamic control script and a service flow cutting script. The present invention processes device control instructions through a message service queue, ensures the timeliness of the instructions, and facilitates the management and tracking of the state changes of the instructions. The DeepWalk algorithm is used in the graph embedding layer of the DeepWork operation and maintenance diagnosis model to map devices and their relationships into a low-dimensional space. The GraphSAGE network layer learns node features from local neighborhoods to enhance the ability to understand complex relationships between devices. The relationship reasoning is performed through the computer room equipment knowledge graph reasoning layer to mine implicit associations between devices, further enhancing the model's diagnostic capabilities. The dynamic control scripts for devices and service flow switching scripts are automatically executed based on the operation and maintenance diagnosis results and the computer room equipment knowledge graph, which can dynamically adjust the service load distribution according to real-time conditions to ensure service stability, significantly improving the efficiency and accuracy of server room operation and maintenance work.

[0152] Figure 4 The integrated intelligent control and monitoring device for a server room of an embodiment of the present invention comprises: a device topology diagram construction module 410 , a consumer service queue construction module 420 , an operation and maintenance diagnosis module 430 and a device control module 440 .

[0153] The device topology diagram construction module 410 is used to construct a device topology diagram of the server room, wherein the device topology diagram includes the connection relationship between various servers, switches, and routers, as well as the ports, traffic, and operating performance parameters of various online services;

[0154] The consumer service queue construction module 420 is used to construct a message service queue for each device control instruction; wherein the control instruction includes corresponding state change information; the message service queue includes a message producer service queue and a message consumer service queue;

[0155] The operation and maintenance diagnosis module 430 is used to input the device topology map and the message service queue into the DeepWork operation and maintenance diagnosis model to obtain the operation and maintenance diagnosis results of each device; wherein the DeepWork operation and maintenance diagnosis model includes a graph embedding layer, a GraphSAGE network layer, a computer room equipment knowledge graph reasoning layer, a path sorting network layer and a fusion layer;

[0156] The device control module 440 is used to execute predefined operation and maintenance scripts to control each device according to the operation and maintenance diagnosis results of each device and the equipment knowledge graph of the computer room; wherein the operation and maintenance scripts include device dynamic control scripts and service flow switching scripts.

[0157] Figure 5 The schematic diagram of the structure of the computing device embodiment of the present invention is shown. The specific embodiment of the present invention does not limit the specific implementation of the computing device.

[0158] like Figure 5 As shown, the computing device may include: a processor (processor) 502 , a communications interface (Communications Interface) 504 , a memory (memory) 506 , and a communication bus 508 .

[0159] The processor 502, the communication interface 504, and the memory 506 communicate with each other via the communication bus 508. The communication interface 504 is used to communicate with other devices such as a client or other server network elements. The processor 502 is used to execute the program 510, which can specifically execute the relevant steps in the embodiment of the integrated intelligent control and monitoring method of the server room.

[0160] Specifically, the program 510 may include program codes, which include computer operation instructions.

[0161] The processor 502 may be a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present invention. The one or more processors included in the computing device may be processors of the same type, such as one or more CPUs; or processors of different types, such as one or more CPUs and one or more ASICs.

[0162] The memory 506 is used to store the program 510. The memory 506 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0163] According to the solution provided by the present invention, a device topology diagram of a server room is constructed, wherein the device topology diagram includes the connection relationship between each server, switch, and router, as well as the ports, traffic, and operating performance parameters of various online services; a message service queue for each device control instruction is constructed; wherein the control instruction contains corresponding state change information; the message service queue includes a message producer service queue and a message consumer service queue; the device topology diagram and the message service queue are input into the DeepWork operation and maintenance diagnosis model to obtain the operation and maintenance diagnosis results of each device; wherein the DeepWork operation and maintenance diagnosis model includes a graph embedding layer, a GraphSAGE network layer, a computer room equipment knowledge graph reasoning layer, a path sorting network layer, and a fusion layer; according to the operation and maintenance diagnosis results of each device and the computer room equipment knowledge graph, a predefined operation and maintenance script is executed to control each device; wherein the operation and maintenance script includes a device dynamic control script and a service flow cutting script. The present invention processes device control instructions through a message service queue, ensures the timeliness of the instructions, and facilitates the management and tracking of the state changes of the instructions. The DeepWalk algorithm is used in the graph embedding layer of the DeepWork operation and maintenance diagnosis model to map devices and their relationships into a low-dimensional space. The GraphSAGE network layer learns node features from local neighborhoods to enhance the ability to understand complex relationships between devices. The relationship reasoning is performed through the computer room equipment knowledge graph reasoning layer to mine implicit associations between devices, further enhancing the model's diagnostic capabilities. The dynamic control scripts for devices and service flow switching scripts are automatically executed based on the operation and maintenance diagnosis results and the computer room equipment knowledge graph, which can dynamically adjust the service load distribution according to real-time conditions to ensure service stability, significantly improving the efficiency and accuracy of server room operation and maintenance work.

[0164] Those skilled in the art will appreciate that the modules in the devices in the embodiments can be adaptively changed and set in one or more devices different from the embodiments. The modules or units or components in the embodiments can be combined into one module or unit or component, and in addition they can be divided into multiple submodules or subunits or subcomponents. Except that at least some of such features and / or processes or units are mutually exclusive, all features disclosed in this specification (including the accompanying claims, abstracts and drawings) and all processes or units of any method or device disclosed in this manner can be combined in any combination. Unless otherwise explicitly stated, each feature disclosed in this specification (including the accompanying claims, abstracts and drawings) can be replaced by an alternative feature that provides the same, equivalent or similar purpose. In addition, those skilled in the art will appreciate that although some embodiments herein include certain features included in other embodiments but not other features, the combination of features of different embodiments means being within the scope of the present invention and forming different embodiments. For example, in the following claims, any one of the claimed embodiments can be used in any combination. The present invention can be implemented by hardware including several different elements and by appropriately programmed computers. In a unit claim that lists several means, several of these means may be embodied by the same hardware item. The steps in the above embodiments should not be understood as limiting the execution order unless otherwise specified.

Claims

1. An integrated intelligent control and monitoring method for a server room, characterized in that: include: Constructing a device topology diagram of the server room, wherein the device topology diagram includes the connection relationship between various servers, switches, and routers, as well as the ports, traffic, and operating performance parameters of various online services; Constructing a message service queue for each device control instruction; wherein the control instruction includes corresponding state change information; the message service queue includes a message producer service queue and a message consumer service queue; Input the device topology map and the message service queue into the DeepWork operation and maintenance diagnosis model to obtain the operation and maintenance diagnosis results of each device; wherein the DeepWork operation and maintenance diagnosis model includes a graph embedding layer, a GraphSAGE network layer, a computer room equipment knowledge graph reasoning layer, a path sorting network layer and a fusion layer; According to the operation and maintenance diagnosis results of each device and the equipment knowledge graph of the computer room, a predefined operation and maintenance script is executed to control each device; wherein the operation and maintenance script includes a device dynamic control script and a service flow switching script.

2. The integrated intelligent control and monitoring method for a server room according to claim 1, characterized in that: The connection relationship includes a physical connection relationship and a logical connection relationship; wherein the logical connection relationship includes a connection mode of a virtual switch and a virtual firewall; The operating performance parameters include recording the CPU utilization of switches and routers, server memory utilization, server disk I / O performance indicators, and server port throughput.

3. The integrated intelligent control and monitoring method for a server room according to claim 1 or 2, characterized in that: The graph embedding layer embeds the nodes in the device topology graph through the DeepWalk algorithm to obtain an embedded representation of each device; The graph embedding layer includes multi-hop relationships between the nodes.

4. The integrated intelligent control and monitoring method for a server room according to claim 3, characterized in that: The GraphSAGE network layer obtains the embedded representation of each device by aggregating the information of each device and the corresponding neighboring nodes; wherein the GraphSAGE network layer includes a first GraphSAGE convolutional layer and a second GraphSAGE convolutional layer; The first GraphSAGE convolutional layer inputs the initial embedding representation of each device, randomly samples a first preset number of nodes from the neighboring nodes of each device, aggregates the embedding representation of the neighboring nodes of each device through a first aggregation function, combines the embedding representation of the neighboring nodes with the initial embedding representation of the device, and then performs a nonlinear transformation; The second GraphSAGE convolutional layer inputs the initial embedding representation of the device updated by the first GraphSAGE convolutional layer, randomly samples a second preset number of nodes from the neighbor nodes of each device, aggregates the embedding representation of the neighbor nodes of each device through a second aggregation function, combines the embedding representation of the neighbor nodes with the initial embedding representation of the device, and then performs a nonlinear transformation.

5. The integrated intelligent control and monitoring method for a server room according to claim 1, characterized in that: The computer room equipment knowledge graph reasoning layer embeds entities and relationships in the equipment topology graph through the TransE model; and performs relationship reasoning based on the embedded representation to obtain potential relationships between devices.

6. The integrated intelligent control and monitoring method for a server room according to claim 1, characterized in that: The path sorting network layer optimizes the communication paths between devices based on the interaction frequency, data flow and importance between devices.

7. The integrated intelligent control and monitoring method for a server room according to claim 1, characterized in that: The executing of the predefined operation and maintenance script to control each device according to the operation and maintenance diagnosis result of each device and the equipment knowledge graph of the computer room further includes: According to the operation and maintenance diagnosis results output by the DeepWork operation and maintenance diagnosis model, identify the equipment that needs to be controlled; wherein the operation and maintenance diagnosis results include the health status, performance indicators and potential faults of the equipment; According to the equipment knowledge graph of the computer room, the corresponding equipment dynamic control script and service flow switching script are determined for the equipment that needs to be controlled, and added to the message service queue in the form of a message, which is executed by the message consumer service queue; Among them, the device dynamic control script is used to adjust device configuration, restart services and resource allocation; the service flow switching script is used to transfer network traffic from potential faulty devices to other devices by adjusting load balancing.

8. The integrated intelligent control and monitoring method for a server room according to claim 4, characterized in that: The calculation formula of the first GraphSAGE convolutional layer is: Among them, AttentionAgg1() is an aggregation function based on the first attention mechanism; is the set of sampled neighbor nodes N of node v k (v) The initial embedding representation of each neighbor node u; The calculation formula of the second GraphSAGE convolutional layer is: Among them, AttentionAgg2() is an aggregation function based on the second attention mechanism; is the set of sampled neighbor nodes N of node v k (v) The initial embedding representation of each neighbor node u.

9. The integrated intelligent control and monitoring method for a server room according to claim 8, characterized in that: The calculation formula of the attention weight of the first attention mechanism is: Among them, a1 is a learnable vector used to calculate the attention score; W1 is a learnable weight matrix used to transform node embedding; LeakyReLU is the activation function, and its function expression is u′ is the set of neighbor nodes N k Another neighbor node in (v); The calculation formula of the attention weight of the second attention mechanism is: Among them, a2 is a learnable vector used to calculate the attention score; W2 is a learnable weight matrix used to transform node embedding; LReLU is the activation function, and its function expression is u′ is the set of neighbor nodes N k Another neighbor node in (v).

10. An integrated intelligent control and monitoring device for a server room, characterized in that: The integrated intelligent control and monitoring method according to any one of claims 1 to 9 is implemented, comprising: A device topology construction module is used to construct a device topology of the server room, wherein the device topology includes the connection relationship between various servers, switches, and routers, as well as the ports, traffic, and operating performance parameters of various online services; A consumer service queue construction module is used to construct a message service queue for each device control instruction; wherein the control instruction contains corresponding state change information; the message service queue includes a message producer service queue and a message consumer service queue; An operation and maintenance diagnosis module, used to input the device topology map and the message service queue into the DeepWork operation and maintenance diagnosis model to obtain the operation and maintenance diagnosis results of each device; wherein the DeepWork operation and maintenance diagnosis model includes a graph embedding layer, a GraphSAGE network layer, a computer room equipment knowledge graph reasoning layer, a path sorting network layer and a fusion layer; The equipment control module is used to execute predefined operation and maintenance scripts to control each device according to the operation and maintenance diagnosis results of each device and the equipment knowledge graph of the computer room; wherein the operation and maintenance scripts include equipment dynamic control scripts and service flow switching scripts.

Images