Server network state dynamic adjustment method and device, equipment and storage medium

Abstract

The invention discloses a server network state dynamic adjustment method and device, equipment and a storage medium, and relates to the technical field of artificial intelligence, and the method comprises the steps: training a to-be-trained artificial intelligence model through the historical state data of a target to-be-adjusted server, so as to obtain a target artificial intelligence model; inputting the real-time network state data of the target to-be-adjusted server into the target artificial intelligence model to obtain a server adjustment strategy, and adjusting the network state of the target to-be-adjusted server based on the server adjustment strategy; and updating the target artificial intelligence model based on the adjustment feedback signal of the adjusted target to-be-adjusted server to obtain a new target artificial intelligence model, and then skipping to the step of inputting the real-time network state data of the target to-be-adjusted server into the target artificial intelligence model. Namely, the state of the server is automatically sensed by utilizing a training artificial intelligence model, a corresponding server adjustment strategy is generated, and the network performance of the server is intelligently adjusted.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to a server network state dynamic adjustment method and device, equipment and a storage medium. BACKGROUND

[0002] With the rapid development of cloud computing, big data, artificial intelligence and other applications, the requirements for server network performance, such as throughput, delay, packets per second (PPS, which is the abbreviation of data packets per second), are increasingly stringent. The current network configuration of the server, such as the network card parameters (interrupt aggregation, queue depth, RSS configuration (Really Simple Syndication)), operating system network stack parameters, and data processing unit (DPU, Data Processing Unit) offloading strategy, etc., are usually statically or semi-statically configured when the system is deployed.

[0003] However, the types of services carried by the server and the load conditions are dynamically changing. For example, some time periods may be dominated by large file transfers (requiring high throughput), while other time periods may be dominated by small packet high-frequency transactions (requiring low latency and high PPS). Static configuration is difficult to adapt to such dynamics, resulting in suboptimal network performance in some scenarios or the need for experienced operations personnel to manually optimize, which is time-consuming and labor-intensive and difficult to guarantee real-time and optimality. Although programmable network devices such as DPU provide rich configurability and hardware offloading capabilities, how to intelligently schedule and configure these resources according to real-time changing network traffic characteristics and application requirements is still a challenge. Some existing automatic optimization methods are mostly based on pre-set rules or thresholds, lack the ability to learn and predict complex correlations, and are difficult to achieve global and long-term optimal performance.

[0004] It can be seen that how to realize the dynamic adjustment of the server network state is a problem to be solved by those skilled in the art. SUMMARY

[0005] The purpose of the embodiments of the present application is to provide a server network state dynamic adjustment method, device, equipment and storage medium, which can generate a corresponding server adjustment strategy and intelligently adjust the server network performance. The specific scheme is as follows: In a first aspect, the present application discloses a server network state dynamic adjustment method, comprising: training a target artificial intelligence model using historical state data of a target server to be adjusted to obtain a target artificial intelligence model; inputting real-time network state data of the target server to be adjusted into the target artificial intelligence model to obtain a corresponding server adjustment strategy, and adjusting the network state of the target server to be adjusted based on the server adjustment strategy; updating the target artificial intelligence model based on the adjustment feedback signal of the adjusted target server to be adjusted to obtain a new target artificial intelligence model, and then jumping to the step of inputting the real-time network state data of the target server to be adjusted into the target artificial intelligence model.

[0006] Optionally, the target artificial intelligence model is obtained by training a to-be-trained artificial intelligence model using historical state data of the target server to be adjusted, including: collecting historical state data of the target server to be adjusted, and determining a to-be-trained artificial intelligence model from a plurality of to-be-selected artificial intelligence models based on a preset server network state adjustment target; training the to-be-trained artificial intelligence model using the historical state data in an offline state to obtain the target artificial intelligence model; deploying the target artificial intelligence model into a server state adjustment tool corresponding to the target server to be adjusted based on the preset server network state adjustment target.

[0007] Optionally, the target artificial intelligence model is obtained by training a to-be-trained artificial intelligence model using historical state data of the target server to be adjusted, including: determining the historical network parameter adjustment of the target server to be adjusted as an action space, and determining the historical network performance index as a reward signal; training a to-be-trained reinforcement learning model using the action space and the reward signal to learn a dynamic optimization strategy to obtain a target reinforcement learning model; training a to-be-trained supervised learning model using historical performance configuration data and corresponding working condition data of the target server to be adjusted to obtain a target supervised learning model; training a to-be-trained unsupervised learning model using historical traffic data of the target server to be adjusted to obtain a target unsupervised learning model; determining the target reinforcement learning model, and / or the target supervised learning model, and / or the target unsupervised learning model as the target artificial intelligence model.

[0008] Optionally, before the real-time network state data of the target server to be adjusted is inputted into the target artificial intelligence model to obtain a corresponding server adjustment strategy, the method further includes: real-time monitoring and collecting the target server to be adjusted during operation by a monitoring agent tool pre-deployed on a to-be-monitored part of the target server to be adjusted to obtain the real-time network state data; The monitoring parts include a host operating system, a baseboard management controller, a network interface card and a data processing unit. The real-time network state data includes network traffic feature data, network device state data, system resource data and application layer performance indicators.

[0009] Optionally, the real-time network state data of the target server to be adjusted is input into a target artificial intelligence model to obtain a corresponding server adjustment strategy, and the network state of the target server to be adjusted is adjusted based on the server adjustment strategy, including: The real-time network state data of the target server to be adjusted is sequentially subjected to data cleaning, data normalization, data feature extraction and feature engineering to obtain a feature vector corresponding to the real-time network state data; The feature vector is input into the target artificial intelligence model to obtain a corresponding model output result; The model output result includes a current network state evaluation result and a performance trend prediction, or a network parameter adjustment strategy and a network parameter adjustment value; A corresponding server adjustment strategy is determined based on the model output result; According to the server adjustment strategy, the host network stack and the driver parameters of the target server to be adjusted are adjusted by using an operating system interface; According to the server adjustment strategy, the related network settings of the underlying hardware or firmware of the target server to be adjusted are adjusted through an interface of the baseboard management controller; According to the server adjustment strategy, the configuration of the network interface card / data processing unit is modified by using a management interface of the network interface card / data processing unit.

[0010] Optionally, a corresponding server adjustment strategy is determined based on the model output result, including: An adjustment evaluation result is obtained by using a preset decision engine to perform an evaluation operation on the model output result; the evaluation operation includes adjustment stability evaluation, adjustment amplitude evaluation and strategy constraint evaluation.

[0011] Optionally, the target artificial intelligence model is updated based on an adjustment feedback signal of the adjusted target server to be adjusted to obtain a new target artificial intelligence model, including: Network performance indicator change data of the adjusted target server to be adjusted is obtained to obtain an adjustment feedback signal; The target artificial intelligence model is subjected to reinforcement learning based on the adjustment feedback signal to obtain a new target artificial intelligence model.

[0012] In a second aspect, the present application discloses a server network state dynamic adjustment device, including: The model training module is configured to train the to-be-trained artificial intelligence model by using historical state data of the target server to be adjusted, so as to obtain a target artificial intelligence model. The server adjustment module is configured to input real-time network state data of the target server to be adjusted into the target artificial intelligence model, so as to obtain a corresponding server adjustment strategy, and adjust the network state of the target server to be adjusted based on the server adjustment strategy. The model updating module is configured to update the target artificial intelligence model based on an adjustment feedback signal of the adjusted target server to be adjusted, so as to obtain a new target artificial intelligence model, and then jump to the step of inputting the real-time network state data of the target server to be adjusted into the target artificial intelligence model.

[0013] In a third aspect, the present application discloses an electronic device, comprising: The memory is configured to save the computer program. The processor is configured to execute the computer program to implement the server network state dynamic adjustment method.

[0014] In a fourth aspect, the present application discloses a computer readable storage medium configured to save a computer program, and the computer program is executed by a processor to implement the server network state dynamic adjustment method.

[0015] In the present application, the to-be-trained artificial intelligence model is trained by using the historical state data of the target server to be adjusted, so as to obtain a target artificial intelligence model; the real-time network state data of the target server to be adjusted is input into the target artificial intelligence model, so as to obtain a corresponding server adjustment strategy, and the network state of the target server to be adjusted is adjusted based on the server adjustment strategy; the target artificial intelligence model is updated based on an adjustment feedback signal of the adjusted target server to be adjusted, so as to obtain a new target artificial intelligence model, and then the step of inputting the real-time network state data of the target server to be adjusted into the target artificial intelligence model is jumped to.

[0016] From the above technical scheme can be seen, the application trains an artificial intelligence model by using the historical state data of the target server to be adjusted, then monitors the real-time state of the target server to be adjusted by using the artificial intelligence model, outputs the corresponding server adjustment strategy in real time, and makes corresponding adjustment, then reversely updates the training artificial intelligence model according to the adjustment feedback signal of the target server to be adjusted after adjustment, and continues to monitor and adjust the server by using the updated model. In this way, the network configuration can be automatically and dynamically adjusted according to the real-time network load and system state, so that the network performance of the server always tends to be optimal, the average throughput is improved, the average delay is reduced, and the model can better adapt to diversified and dynamically changing business demands through automatic optimization, thereby improving the overall service quality and user experience. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present application, the drawings needed in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0018] Figure 1 A server network state dynamic adjustment method flow chart is disclosed in the present application. Figure 2 A server network state dynamic adjustment device structure schematic diagram is disclosed in the present application. Figure 3 An electronic equipment structure diagram is disclosed in the present application. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.

[0020] In the specification of the present application and the above drawings, the terms "include" and "have", as well as any variants of "include" and "have", are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units is not limited to the listed steps or units, but can include steps or units not listed.

[0021] In order to enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0022] Some existing server automatic tuning methods are mostly based on preset rules or thresholds, lack the learning and prediction ability of complex correlation, and are difficult to achieve global and long-term optimal performance. Therefore, the present application will introduce a kind of server network state dynamic adjustment method, which can solve the above problems.

[0023] Referring to Figure 1 The embodiment of the application discloses a kind of server network state dynamic adjustment method, comprising: Step S11: the historical state data of target server to be adjusted is used to train the artificial intelligence model to be trained to obtain target artificial intelligence model.

[0024] In the embodiment, before server monitoring, suitable artificial intelligence model is determined for subsequent server state monitoring operation. Therefore, the historical state data of target server to be adjusted is used to train the artificial intelligence model to be trained to obtain target artificial intelligence model, including: collecting the historical state data of target server to be adjusted, and determining the artificial intelligence model to be trained from several artificial intelligence models to be selected based on preset server network state adjustment target;In offline state, the historical state data is used to train the artificial intelligence model to be trained to obtain target artificial intelligence model;Target artificial intelligence model is deployed into the server state adjustment tool corresponding to target server to be adjusted based on preset server network state adjustment target. Specifically, first, corresponding model needs to be selected from several artificial intelligence models according to actual demand, including but not limited to reinforcement learning model (Reinforcement Learning, i.e. RL), supervised learning model (Supervised Learning) and unsupervised learning model (Unsupervised Learning). Then model training is carried out, wherein offline pre-training can be carried out, and online learning or fine tuning can be carried out according to actual running data after server deployment to adapt to specific environment and business drift.

[0025] In this embodiment, the historical state data of the target server to be adjusted is used to train the artificial intelligence model to be trained to obtain the target artificial intelligence model, including: determining the historical network parameter adjustment of the target server to be adjusted as an action space, determining the historical network performance index as a reward signal, training the reinforcement learning model to be trained using the action space and the reward signal to learn a dynamic optimization strategy to obtain a target reinforcement learning model; training the supervised learning model to be trained using the historical performance configuration data and the corresponding working condition data of the target server to be adjusted to obtain a target supervised learning model; training the unsupervised learning model to be trained using the historical traffic data of the target server to be adjusted to obtain a target unsupervised learning model; determining the target reinforcement learning model, and / or the target supervised learning model, and / or the target unsupervised learning model as the target artificial intelligence model. Specifically, for the reinforcement learning model, the network parameter adjustment can be regarded as an action space, and the network performance index (such as maximum throughput and minimum delay) can be regarded as a reward signal to train an AI (Artificial Intelligence) agent to learn an optimal dynamic optimization strategy. For the supervised learning model, the historical high-performance configuration data and the corresponding working condition data can be used to train a prediction model, including but not limited to a deep neural network (DNN) and a gradient boosting tree (GBT), which is used to predict the optimal parameter combination under a specific working condition. For the unsupervised learning model, it is used for traffic pattern recognition and anomaly detection to assist decision-making. The model can be pre-trained offline. After the training of the model is completed, a plurality of trained models can be selected according to the specific monitoring requirements of the server to form a comprehensive model, and then the target artificial intelligence model is obtained.

[0026] Step S12: inputting the real-time network state data of the target server to be adjusted into the target artificial intelligence model to obtain a corresponding server adjustment strategy, and adjusting the network state of the target server to be adjusted based on the server adjustment strategy.

[0027] In this embodiment, before the real-time network state data of the target server to be adjusted is input into the target artificial intelligence model to obtain the corresponding server adjustment strategy, the method further includes: performing real-time monitoring and collection on the target server to be adjusted during the running process by a monitoring agent tool pre-deployed on a to-be-monitored part of the target server to be adjusted, to obtain the real-time network state data; the to-be-monitored part includes a host operating system, a baseboard management controller, a network interface card, and a data processing unit; the real-time network state data includes network traffic feature data, network device state data, system resource data, and application layer performance indicators. Specifically, during the running process of the server, multi-dimensional data is collected in real time by the monitoring agent deployed on the host operating system, the BMC (Baseboard Management Controller), and the NIC (Network Interface Card) or DPU. It should be noted that the dimensional data includes but is not limited to: network traffic feature data, such as packet size distribution, protocol type (TCP / UDP / RoCE, etc.), connection number, flow duration, source / destination IP and port information; network device state data, such as NIC / DPU queue length, packet loss rate, error packet number, hardware offload engine, such as OVS (Open vSwitch) offload, encryption offload, RDMA (Remote Direct Memory Access) utilization and performance counters, interrupt frequency, buffer occupancy; system resource data, such as CPU (Central Processing Unit) utilization (overall and per core), memory bandwidth, PCIe (Peripheral Component Interconnect Express) bandwidth utilization, NUMA (Non-Uniform Memory Access) node memory access delay; application layer performance indicators, such as application response delay, transaction processing rate, etc., which can be obtained through application probes or standardized interfaces.

[0028] In this embodiment, the real-time network state data of the target server to be adjusted is input into the target artificial intelligence model to obtain a corresponding server adjustment strategy, and the network state of the target server to be adjusted is adjusted based on the server adjustment strategy, including: the real-time network state data of the target server to be adjusted is sequentially subjected to data cleaning, data normalization, data feature extraction and feature engineering to obtain a feature vector corresponding to the real-time network state data; the feature vector is input into the target artificial intelligence model to obtain a corresponding model output result; the model output result includes a current network state evaluation result and a performance trend prediction, or a network parameter adjustment strategy and a network parameter adjustment value; a corresponding server adjustment strategy is determined based on the model output result; the host network stack and the drive parameters of the target server to be adjusted are adjusted according to the server adjustment strategy and by using an operating system interface; the related network settings of the underlying hardware or firmware of the target server to be adjusted are adjusted according to the server adjustment strategy and by using an interface of a baseboard management controller; and the configuration of the network interface card / data processing unit is modified according to the server adjustment strategy and by using a management interface of the network interface card / data processing unit.

[0029] Specifically, the collected raw data is cleaned, normalized, feature extracted and feature engineered to construct a feature vector suitable for AI model input. For example, time series data is converted to statistical features, or graph neural networks are used to process network topology and traffic relationships. The real-time pre-processed feature vector is input into the trained AI model. Then, the AI model outputs an evaluation of the current network state, a prediction of future performance trends, or directly outputs recommended network parameter adjustment strategies / concrete parameter values. For example, adjusting the aggregation strength of interrupts, dynamically allocating net card queues to CPU cores, modifying the flow table offloading strategy of DPU, adjusting RoCE parameters, etc. Among them, the corresponding server adjustment strategy is determined based on the model output result, including: using a preset decision engine to evaluate the model output result to obtain an adjustment evaluation result; the evaluation operation includes adjustment stability evaluation, adjustment amplitude evaluation, and strategy constraint evaluation. That is, the adjustment suggestion output by the AI model is evaluated by a decision engine, considering the stability of the adjustment, the amplitude limit (to avoid violent shock) and the preset strategy constraint. After obtaining the corresponding server adjustment strategy, the host network stack and driver parameters can be adjusted through the operating system interface (such as ethtool, sysctl). Some underlying hardware or firmware related network settings can be adjusted through the BMC interface (if supported). The BMC interface includes but is not limited to IPMI (Intelligent Platform Management Interface), and the management interface of DPU / NIC (such as the API provided by NVIDIA DOCA SDK) is used to dynamically modify the configuration of DPU / NIC, such as hardware offload rules, queue configuration, QoS policy, etc.

[0030] Step S13: updating the target artificial intelligence model based on the adjustment feedback signal of the target server to be adjusted after adjustment to obtain a new target artificial intelligence model, and then jumping to the step of inputting the real-time network state data of the target server to be adjusted into the target artificial intelligence model.

[0031] In this embodiment, the target artificial intelligence model is updated based on the adjustment feedback signal of the adjusted target server to be adjusted to obtain a new target artificial intelligence model, including: obtaining the network performance index change data of the adjusted target server to be adjusted to obtain the adjustment feedback signal; and performing reinforcement learning on the target artificial intelligence model based on the adjustment feedback signal to obtain a new target artificial intelligence model. Specifically, after adjustment is performed, the change of the network performance index is continuously monitored, and the change is input as a feedback signal to the AI model (especially for a reinforcement learning model), forming a closed-loop optimization. Then jump to the step of inputting the real-time network state data of the target server to be adjusted into the target artificial intelligence model. In this way, this process is repeated to realize continuous online learning and dynamic optimization. In this way, based on the continuous learning and online training capability of AI, the system performance can be further improved over time and with data accumulation. For deployment in a large-scale data center, federated learning technology can be used, that is, under the premise of protecting data privacy, the running data of multiple servers is collected and an AI model that is more generalized and powerful is trained together. Then, using transfer learning, the trained model is adapted to each server for server monitoring. At this time, feedback data from each server can be collected and sent to each server for training, which can expand the range of server update data, and the trained model can quickly adapt to new hardware platforms or business scenarios. In addition, single-server network optimization and data center-level network management and scheduling can be realized end-to-end and globally intelligent network optimization. Among them, the data center-level network can be combined with an SDN (Software Defined Network) controller, and the server can report its network status and optimization requirements to the controller, and the controller can perform global scheduling. In addition, explainable AI technology can be introduced to make the decision-making process and reasons of the AI model more transparent to the operation and maintenance personnel, enhance trust, and facilitate fault troubleshooting and manual intervention. In addition, considering that different servers correspond to different systems, the related interfaces and data formats of the model can be standardized, which facilitates the interoperation between servers, DPUs, operating systems, and AI optimization software of different manufacturers. In addition, in a server containing multiple CPUs, GPUs, DPUs, FPGAs (Field Programmable Gate Array), and other heterogeneous computing units, the AI model can be responsible for intelligent scheduling of network tasks among different processing units for allocation and cooperation to achieve overall optimization.

[0032] In operation, the AI model can be used to analyze long-term network performance data and device status data, predict potential hardware failures (such as network card, cable problems) or performance bottlenecks, and give early warnings or trigger proactive business migration, redundancy switching and other operations. When the network performance problem occurs, the AI model can assist in quickly locating the root cause of the problem (configuration problem, hardware problem or application problem), and try to automatically execute the recovery strategy.

[0033] Further, the present application will introduce the present scheme based on three practical application scenarios.

[0034] The first embodiment is a DPU network parameter dynamic tuning method based on reinforcement learning. For a server equipped with NVIDIA BlueField-3 DPU, run a variety of hybrid network applications (such as Web service, distributed storage access). First, during data collection, use DOCA Telemetry Service to collect detailed network flow information, hardware counters (such as RoCE, OVS offload statistics), Arm core utilization rate, etc. from BF3 DPU. At the same time, collect CPU load, preliminary delay data from the host OS (Operating System) Application layer feedback. The artificial intelligence model used is a lightweight reinforcement learning agent (such as a variant of PPO or DQN algorithm) deployed on the Arm core of BF3 DPU. State space (State): including the average length of DPU receive / send queue, the proportion of specific type of traffic (such as RDMA traffic, small packet traffic), the busy degree of CPU core (bound by network task), the packet loss rate in the recent period, etc. Action space (Action): including adjusting the interrupt aggregation parameters (rx-usecs, rx-frames) of BF3, adjusting the queue pair (QP, QueuePair) depth of RDMA operation, dynamically modifying the priority or behavior of OVS hardware offload flow table, adjusting the resource allocation ratio of different processing engines inside DPU. Reward function (Reward): designed to maximize a comprehensive performance indicator, for example: Reward=w1xThroughput-w2xLatency-w3xPacketLossRate, where w1, w2, w3 are weight factors, which can be adjusted according to business preferences. After the reinforcement learning agent selects an action, it directly executes the corresponding configuration change on BF3 DPU through DOCA API. In specific operation, the initial RL model is used or trained based on historical data. The data collection module continuously provides the current state to the AI processing module. The RL agent selects an action (network parameter adjustment) according to the current state. The configuration execution module applies the action. The data collection module observes the new state after the action is executed and obtains the reward. The RL agent updates its policy network and value network according to the (state, action, reward, new state) tuple. Repeat this process to achieve continuous online learning and dynamic optimization.

[0035] The second embodiment is a server global network configuration recommendation based on supervised learning. For a large number of homogeneous or heterogeneous servers in a data center, a central management platform is used for monitoring. During technical data collection, network, system, and application data are collected by agents on each server and aggregated to a central data lake. In implementation, a large amount of historical data is first collected, including performance under different configurations (benchmark results or real business KPIs). The data is labeled, for example, at a certain known "optimal" performance point, the working conditions and configurations at that time are recorded. One or more supervised learning models are trained using these data, such as gradient boosting decision tree XGBoost or multilayer perceptron (MLP, Multilayer Perceptron). The model can predict a set of "recommended" network parameter combinations according to the input real-time working condition features. Then, a lightweight inference engine on each server loads the trained model. Real-time local working condition data is obtained, input into the model, and the recommended configuration is obtained. Finally, the recommended configuration from the model is compared with the current configuration. If the difference reaches a certain threshold or the expected benefit is significant, the configuration is updated. The update operation can be performed by the server management software to issue instructions to the OS, BMC, or DPU. To ensure stability, gray-scale update or A / B testing can be used to gradually apply the new configuration. In this way, global data can be used for training to learn more extensive patterns. This is suitable for scenarios that require collaborative optimization across multiple servers.

[0036] The third embodiment is AI-assisted NUMA-aware network I / O (Input / Output) path optimization. For multi-path CPU servers, network I / O performance is significantly affected by NUMA affinity. Specifically, the AI model considers NUMA-related factors such as CPU core and network card queue binding, memory allocation strategy, and interrupt affinity when deciding network parameters. The data collection module collects the load of each NUMA node, cross-node memory access, and the NUMA node information connected by the PCIe bus of the network card. The AI model predicts the performance gain under different NUMA binding strategies. The configuration execution module adjusts the irqbalance configuration, the EAL (Environment Abstraction Layer) core mask of the DPDK application, or adjusts the NUMA affinity of application threads and memory through tools such as numactl, and maps the CPU core of the network card RSS queue. This deep optimization for the NUMA architecture characteristics of the server can further tap the potential of the hardware.

[0037] The above three embodiments mainly focus on network performance (throughput, delay, PPS). In addition, the optimization target of the scalable AI model, such as guaranteeing network performance while considering server energy optimization (for example, dynamically adjusting CPU frequency, shutting down idle cores or network ports according to network load), security (such as dynamically adjusting firewall policy or traffic shaping policy to respond to abnormal traffic or attacks based on AI analysis of network traffic patterns).

[0038] It can be seen that, in the embodiment, the historical state data of the target server to be adjusted is used to train an artificial intelligence model to be trained to obtain a target artificial intelligence model; real-time network state data of the target server to be adjusted is input into the target artificial intelligence model to obtain a corresponding server adjustment strategy, and the network state of the target server to be adjusted is adjusted based on the server adjustment strategy; the target artificial intelligence model is updated based on the adjustment feedback signal of the adjusted target server to be adjusted to obtain a new target artificial intelligence model, and then jumping to the step of inputting the real-time network state data of the target server to be adjusted into the target artificial intelligence model.

[0039] As can be seen from the above technical solutions, the present application trains an artificial intelligence model by using the historical state data of the target server to be adjusted, then monitors the real-time state of the target server to be adjusted using this artificial intelligence model, and outputs the corresponding server adjustment strategy in real time and makes corresponding adjustment, and then updates the training artificial intelligence model in reverse according to the adjustment feedback signal of the adjusted target server to be adjusted, and continues to monitor and adjust the server using the updated model. In this way, the network configuration can be automatically and dynamically adjusted according to the real-time network load and system state, so that the server network performance always tends to be optimal, the average throughput is improved, the average delay is reduced, and through automatic optimization, the model can better adapt to diversified and dynamically changing business demands, improving overall service quality and user experience.

[0040] Reference Figure 2 The embodiments of the present application also correspondingly disclose a server network state dynamic adjustment device, comprising: The model training module 11 is configured to train an artificial intelligence model to be trained using the historical state data of the target server to be adjusted to obtain a target artificial intelligence model; The server adjustment module 12 is configured to input real-time network state data of the target server to be adjusted into the target artificial intelligence model to obtain a corresponding server adjustment strategy, and adjust the network state of the target server to be adjusted based on the server adjustment strategy; The model updating module 13 is configured to update the target artificial intelligence model based on the adjustment feedback signal of the target server to be adjusted to obtain a new target artificial intelligence model, and then jump to the step of inputting the real-time network state data of the target server to be adjusted into the target artificial intelligence model.

[0041] In this embodiment, an artificial intelligence model is trained by using the historical state data of the target server to be adjusted, and then the real-time state of the target server to be adjusted is monitored by using the artificial intelligence model, the corresponding server adjustment strategy is output in real time, and corresponding adjustment is performed, then the training artificial intelligence model is updated reversely according to the adjustment feedback signal of the target server to be adjusted after adjustment, and the server is continuously monitored and adjusted by using the updated model. In this way, the network configuration can be automatically and dynamically adjusted according to the real-time network load and system state, so that the network performance of the server always tends to be optimal, the average throughput is improved, the average delay is reduced, and the model can better adapt to diversified and dynamically changing business demands through automatic optimization, thereby improving the overall service quality and user experience.

[0042] Further, the embodiment of the application further discloses an electronic device, Figure 3 is an electronic device structure diagram according to an example embodiment, the contents in the figure cannot be considered as any limitation on the use range of the application. The electronic device can specifically include at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25 and a communication bus 26. The memory 22 is used to store a computer program, the computer program is loaded and executed by the processor 21 to realize the related steps in the server network state dynamic adjustment method disclosed in any of the preceding embodiments. In addition, the electronic device in the embodiment can be an electronic computer.

[0043] In this embodiment, the power supply 23 is used to provide working voltage for each hardware device on the electronic device; the communication interface 24 can create a data transmission channel between the electronic device and the external device, and the communication protocol followed by the communication interface 24 can be any communication protocol applicable to the technical solution of the application, which is not limited here; the input / output interface 25 is used to obtain external input data or output data to the outside, and the specific interface type can be selected according to the specific application needs, which is not limited here.

[0044] In addition, the memory 22 as a carrier for resource storage can be a read-only memory, a random access memory, a magnetic disk or an optical disk, etc., and the resources stored thereon can include an operating system 221, a computer program 222, etc., and the storage mode can be temporary storage or permanent storage.

[0045] The operating system 221 is configured to manage and control each hardware device on the electronic device and the computer program 222, which can be Windows Server, Netware, Unix, Linux, etc. The computer program 222 can further include computer programs capable of performing other specific work in addition to the computer program capable of performing the server network state dynamic adjustment method disclosed by any one of the foregoing embodiments and executed by the electronic device.

[0046] Further, the present application also discloses a computer readable storage medium for storing a computer program; wherein the computer program is executed by a processor to implement the server network state dynamic adjustment method disclosed above. The specific steps of the method can refer to the corresponding content disclosed in the foregoing embodiments, which will not be repeated here.

[0047] Further, the present application also discloses a computer program product comprising computer programs / instructions; wherein the computer programs / instructions are executed by a processor to implement the alarm aggregation method disclosed above. The specific steps of the method can refer to the corresponding content disclosed in the foregoing embodiments, which will not be repeated here.

[0048] The embodiments in the specification are described in a progressive manner, and each embodiment focuses on the difference from other embodiments. The same or similar parts of each embodiment can be referred to each other. For the device disclosed by the embodiments, since it corresponds to the method disclosed by the embodiments, the description is relatively simple, and the related parts can refer to the method part.

[0049] The skilled person can further realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized by electronic hardware, computer software or a combination of both. In order to clearly show the interchangeability of hardware and software, the components and steps of each example have been described in the above description. Whether the functions are realized by hardware or software depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0050] The steps of the method or algorithm described in combination with the embodiments disclosed herein can be directly implemented by hardware, software modules executed by a processor, or a combination of both. The software modules can be placed in random access memory (RAM), memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the art.

[0051] Finally, it needs to be pointed out that in this document, relational terms such as first and second and the like can only be intended to distinguish one entity or operation from another entity or operation without necessarily requiring or implying any actual such relationship or order between such entities or operations. Moreover, the terms "comprising", "including", or any other variant thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without more limitations, an element defined by the statement "comprising a" does not exclude the existence of additional identical elements in the process, method, article, or apparatus including the stated element.

[0052] The above detailed description of the technical solutions provided by the present application has been provided, and the principles and implementation manners of the present application have been described by applying specific examples. The above description of the examples is only for the purpose of helping to understand the method of the present application and its core idea; at the same time, for those skilled in the art, according to the idea of the present application, the specific implementation manner and application range will be changed, and the above description of the content of the specification should not be understood as a limitation of the present application.

Claims

1. A method for dynamically adjusting the network status of a server, characterized in that, include: The target artificial intelligence model is obtained by training the target artificial intelligence model using historical state data of the target server to be adjusted. The real-time network status data of the target server to be adjusted is input into the target artificial intelligence model to obtain the corresponding server adjustment strategy, and the network status of the target server to be adjusted is adjusted based on the server adjustment strategy. The target AI model is updated based on the adjustment feedback signal of the target server to be adjusted to obtain a new target AI model. Then, the process jumps to the step of inputting the real-time network status data of the target server to be adjusted into the target AI model.

2. The method for dynamically adjusting the server network status according to claim 1, characterized in that, The step of training the AI ​​model using historical state data of the target server to be adjusted, in order to obtain the target AI model, includes: Collect historical state data of the target server to be adjusted, and determine the artificial intelligence model to be trained from several artificial intelligence models to be selected based on the preset server network state adjustment target; In offline mode, the historical state data is used to train the artificial intelligence model to be trained in order to obtain the target artificial intelligence model; Based on the preset server network status adjustment target, the target artificial intelligence model is deployed to the server status adjustment tool corresponding to the target server to be adjusted.

3. The method for dynamically adjusting server network status according to claim 1, characterized in that, The step of training the AI ​​model using historical state data of the target server to be adjusted, in order to obtain the target AI model, includes: The historical network parameter adjustments of the target server to be adjusted are determined as the action space, and the historical network performance indicators are determined as reward signals. The reinforcement learning model to be trained is trained using the action space and the reward signal to learn a dynamic optimization strategy in order to obtain the target reinforcement learning model; The historical performance configuration data and corresponding operating condition data of the target server to be adjusted are used to train the supervised learning model to obtain the target supervised learning model. The unsupervised learning model to be trained is trained using the historical traffic data of the target server to be adjusted, so as to obtain the target unsupervised learning model; The target reinforcement learning model, and / or the target supervised learning model, and / or the target unsupervised learning model, are determined as the target artificial intelligence model.

4. The method for dynamically adjusting the server network status according to claim 1, characterized in that, Before inputting the real-time network status data of the target server to be adjusted into the target artificial intelligence model to obtain the corresponding server adjustment strategy, the method further includes: The monitoring agent tool is pre-deployed on the monitored part of the target server to be adjusted to perform real-time monitoring and collection of data during the operation of the target server to be adjusted, so as to obtain the real-time network status data. The monitored components include the host operating system, baseboard management controller, network interface card, and data processing unit. The real-time network status data includes network traffic characteristic data, network device status data, system resource data, and application layer performance indicators.

5. The method for dynamically adjusting server network status according to claim 1, characterized in that, The step of inputting the real-time network status data of the target server to be adjusted into the target artificial intelligence model to obtain the corresponding server adjustment strategy, and adjusting the network status of the target server to be adjusted based on the server adjustment strategy, includes: The real-time network status data of the target server to be adjusted is sequentially subjected to data cleaning, data normalization, data feature extraction and feature engineering to obtain the feature vector corresponding to the real-time network status data. The feature vector is input into the target artificial intelligence model to obtain the corresponding model output result; The model output includes current network status assessment results and performance trend predictions, or network parameter adjustment strategies and network parameter adjustment values; Based on the output of the model, determine the corresponding server adjustment strategy; According to the server adjustment strategy, the host network stack and driver parameters of the target server to be adjusted are adjusted using the operating system interface; According to the server adjustment strategy, the relevant network settings of the underlying hardware or firmware of the target server to be adjusted are adjusted through the interface of the baseboard management controller. According to the server adjustment strategy, the configuration of the network interface card / data processing unit is modified using the management interface of the network interface card / data processing unit.

6. The method for dynamically adjusting server network status according to claim 5, characterized in that, The step of determining the corresponding server adjustment strategy based on the model output includes: The model output is evaluated using a preset decision engine to obtain an adjustment evaluation result; the evaluation operation includes an evaluation of the server stability after adjustment, an evaluation of the adjustment magnitude, and an evaluation of policy constraints.

7. The method for dynamically adjusting the server network status according to any one of claims 1 to 6, characterized in that, The step of updating the target artificial intelligence model based on the adjusted feedback signal of the target server to be adjusted, to obtain a new target artificial intelligence model, includes: Obtain the network performance index change data of the target server to be adjusted after adjustment, so as to obtain the adjustment feedback signal; The target artificial intelligence model is subjected to reinforcement learning based on the adjustment feedback signal to obtain a new target artificial intelligence model.

8. A server network status dynamic adjustment device, characterized in that, include: The model training module is used to train the AI ​​model to be trained using the historical state data of the target server to be adjusted, so as to obtain the target AI model. The server adjustment module is used to input the real-time network status data of the target server to be adjusted into the target artificial intelligence model to obtain the corresponding server adjustment strategy, and adjust the network status of the target server to be adjusted based on the server adjustment strategy. The model update module is used to update the target artificial intelligence model based on the adjustment feedback signal of the target server to be adjusted, so as to obtain a new target artificial intelligence model, and then jump to the step of inputting the real-time network status data of the target server to be adjusted into the target artificial intelligence model.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the steps of the server network state dynamic adjustment method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the server network state dynamic adjustment method as described in any one of claims 1 to 7.

Images