Streaming media host load resource trend prejudgment and load balancing optimization method and system

Abstract

The invention discloses a streaming media host load resource trend pre-judgment and load balancing optimization method and system. The method comprises the following steps: comprehensively monitoring a plurality of dimension indexes of streaming media cluster nodes to form real-time collection data; intelligent diagnosis is executed through the integrated artificial intelligence model to realize anomaly detection of different dimensions; evaluating the health state of each node in real time and calculating a node load score to assist a load balancing decision, and predicting a load peak to perform resource pre-allocation in advance so as to realize predictive capacity expansion and contraction; when the system fails, fault nodes are automatically shielded, and loads are redistributed to healthy nodes; and hot contents are temporarily cached through edge nodes in synchronization, so that the streaming media session of the existing user is not interrupted. According to the invention, prejudgment of the size trend of resources occupied by host load change, host performance monitoring and automatic fault shielding are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of streaming media services, and particularly to a method and system for predicting the trend of load resources of a streaming media host and optimizing load balancing. Background Art

[0002] Streaming media services play an important role in real-time communication scenarios such as video conferencing, remote work, online education, monitoring, and real-time video monitoring. With the wide application of streaming media services, users have higher and higher requirements for the real-time performance, stability, and quality of experience (QoE) of streaming media. The network problems of streaming media services directly affect the user experience. Especially in application scenarios that require high bandwidth and low latency, such as high-definition video, live broadcast, and online education, any network-level problems and hardware resource problems may lead to negative experiences such as image quality degradation, freezing, and excessive latency. Traditional load balancing technologies mainly perform resource allocation and scheduling based on a set of fixed rules and preset parameters. For example, traffic is evenly distributed to each node, or a simple preset algorithm (such as round-robin, weighted round-robin) is used to select traffic processing nodes. These strategies are usually manually configured in advance, relying on historical data and experience to set hardware and bandwidth quotas, and require manual monitoring of traffic and manual adjustment of resources (such as adding servers, expanding bandwidth, etc.), and cannot adapt to the fluctuations of traffic and load in real time. Especially in scenarios where the load changes frequently and abnormally, the static method may result in unreasonable resource configuration (such as over-allocation or resource shortage). In addition, the biggest defect of this manual static configuration based on historical data and experience is that when the traffic suddenly increases or the load increases sharply, which is different from the history, the static load balancing configuration cannot respond in advance or even in real time, and requires manual monitoring and intervention, with slow reaction speed and slow processing speed, and cannot achieve early problem handling, resulting in system overload, increased latency, and even service interruption in scenarios that require real-time performance.

[0003] In the prior art, the load balancing of streaming media services mainly relies on manual intervention or an automated system based on simple rules, lacking the intelligent prediction ability of network and host load trends, resulting in problems such as low resource utilization rate, long response time, response lag, and slow fault recovery. Therefore, there is an urgent need for a management method for streaming media services that can intelligently predict load trends, automatically shield faulty devices, and optimize resource allocation. Summary of the Invention

[0004] The object of the present invention is to provide a method and system for predicting the trend of load resources of a streaming media host and optimizing load balancing. Through technologies such as data collection, comparison and analysis, load peak warning, resource bottleneck identification, anomaly detection, load scoring, dynamic adjustment, intelligent decision-making, and predictive scaling, the dynamic load balancing, fault self-healing, and resource optimization of streaming media services are realized, potential network problems, resource occupancy problems, and high-load problems are solved in advance, and the impact on the real-time usage process is reduced.

[0005] The technical solution adopted by the present invention is as follows: A system for predicting the trend of load resources of a streaming media host and optimizing load balancing, which includes: A data collection and real-time monitoring module: used to comprehensively monitor several dimensional indicators of streaming media cluster nodes to form real-time collected data; the several dimensional indicators include server performance indicators, network status, and streaming media service indicators; An intelligent diagnosis and anomaly detection module: by analyzing the real-time collected data, intelligent diagnosis is performed through an integrated artificial intelligence model applicable to several different dimensional diagnosis tasks to achieve anomaly detection in different dimensions; the health status of each node is evaluated in real time through the loaded load scoring model to assist load balancing decisions, and at the same time, the load peak period is predicted to pre-allocate resources in advance to achieve predictive scaling; A fault automatic shielding and load balancing module: used to automatically shield faulty nodes in case of system failures and redistribute the load to healthy nodes; A fault self-healing module: used to synchronously cache popular content through edge nodes under the condition of automatically shielding faulty nodes in case of hardware failures, ensuring that the streaming media sessions of existing users are not interrupted.

[0006] Further, the streaming media service indicators include the bit rate of the video stream, the number of buffering times, and the user QoE (Quality of Experience) score.

[0007] Further, the server performance indicators include the CPU utilization rate, memory occupancy, disk I / O, network bandwidth, number of connections, and number of streaming media sessions of each node.

[0008] A method for predicting the trend of load resources of a streaming media host and optimizing load balancing, which includes the following steps: Step 1, data collection and real-time monitoring: comprehensively monitor several dimensional indicators of streaming media cluster nodes to form real-time collected data; the several dimensional indicators include server performance indicators, network status, and streaming media service indicators; Step 2, Intelligent Diagnosis and Anomaly Detection: By analyzing the real-time collected data, intelligent diagnosis is performed through an integrated artificial intelligence model to achieve anomaly detection in different dimensions; the health status of each node is evaluated in real time through the equipped load scoring model, and the node load score is calculated to assist in load balancing decisions. At the same time, the peak load period is predicted to pre-allocate resources in advance to achieve predictive scaling. Step 3, Automatic Fault Masking and Load Balancing: Automatically mask the faulty node in case of system failure and redistribute the load to the healthy nodes. Furthermore, in Step 3, the AI automatically analyzes and selects the optimal processing server in the current environment according to the different resource types relied on by different service requests.

[0009] Step 4, Fault Self-Healing: In case of hardware failure, while automatically masking the faulty node, popular content is temporarily cached through edge nodes to ensure that the streaming sessions of existing users are not interrupted.

[0010] Furthermore, Step 1 specifically includes the following steps: Step 1-1, Deploy the system monitoring tool as a data collection agent on each streaming service node, and collect and transmit monitoring data in real time based on the set collection frequency. Specifically, use open-source system monitoring tools (such as Prometheus, Telegraf, Collectd, etc.) as data collection agents.

[0011] Step 1-2, (Using MQTT) Transmit the collected data to the backend system to ensure low latency and high reliability of data transmission.

[0012] Step 1-3, Use a time series database (such as InfluxDB or TimescaleDB) to store the collected data to support efficient time series data queries.

[0013] Furthermore, Step 2 specifically includes the following steps: Step 2-1, Build and train an artificial intelligence model suitable for several different dimension diagnosis tasks. Step 2-2, Use the artificial intelligence model to analyze the real-time collected data for load prediction and anomaly detection. Step 2-3, Evaluate the health status of each node in real time through the equipped load scoring model, and calculate the real-time load score of each node.

[0014] Furthermore, Step 2-1 specifically includes the following steps: Step 2-1-1, First, clean, standardize, and normalize the collected data for use in artificial intelligence model training. Step 2-1-2: Jointly train the AI model by combining several diagnostic tasks of different dimensions to optimize the overall performance of the model; the diagnostic tasks include load prediction, anomaly detection, fault warning, etc.; Step 2-1-3: Use online learning to update the model in real time so that the model can maintain good adaptability as the system data continuously changes; Step 2-1-4: Package the AI model and related dependencies into a Docker container to ensure consistency and scalability in different environments; Step 2-1-5: Integrate the AI model into the intelligent diagnosis layer of the streaming media service platform, and the model obtains real-time performance data through the API for analysis and prediction.

[0015] Furthermore, in Step 2-1, the model is regularly evaluated, and the model parameters are adjusted according to the evaluation results to improve the prediction and diagnosis accuracy. The model is regularly trained, and incremental training is carried out in combination with newly collected data to ensure that the model can adapt to the changes in the streaming media load.

[0016] Furthermore, Step 3 specifically includes the following steps: Step 3-1: Load balancing and request scheduling: The AI model uses a dynamic scheduling algorithm (such as weighted least connections or QoE-based hashing algorithm) to allocate requests according to the real-time load score and health status so as to preferentially select healthy nodes. Through precise load balancing, overload and resource waste are avoided.

[0017] Step 3-2: Session migration mechanism: Hot migrate the ongoing streaming media session to a low-load node. Achieve seamless migration of the streaming media session through technologies such as WebRTC SFU (Selective Forwarding Unit) to ensure the continuity of the user experience.

[0018] Step 3-3: Automatically shield faulty nodes: When the AI model detects that a node fails, automatically mark the faulty node as "risk status" and stop allocating new requests to the faulty node; at the same time, the system triggers the edge collaboration mechanism to preferentially direct user requests to the nearest healthy node to ensure uninterrupted service.

[0019] Furthermore, Step 4 specifically includes the following steps: Step 4-1: Monitor the hardware resources of each node through a monitoring tool to determine whether the use of disk resources exceeds the set threshold (such as the load of disk I / O exceeds 80%); if so, trigger an alarm and perform trend analysis on the historical data of disk I / O through machine learning algorithms to identify possible hardware fault trends; otherwise, execute Step 4-1; Step 4-2: Determine whether the alarm duration exceeds the set time (3 seconds); if so, the system will automatically mark the corresponding node as "risk status". Step 4-3: When the load balancer receives the node risk status, it stops allocating new requests to the corresponding node; at the same time, it transfers the requests to other healthy nodes through an algorithm. Step 4-4: Obtain the geographical location of the request and combine it with the health status of the CDN nodes to determine the optimal request routing.

[0020] Specifically, node selection and routing are performed according to the geographical location and load conditions, and the user requests are routed to the nearest CDN edge node with low load; the load status of the CDN nodes is monitored in real time to dynamically adjust the routing policy, avoid overloading of a certain node, and ensure fast response to requests.

[0021] Furthermore, in step 4-3, a temporary caching mechanism is enabled on the edge node synchronously, and the most frequently accessed popular content on the corresponding node is cached to the healthy edge node to ensure that the session is not interrupted.

[0022] The present invention adopts the above technical solutions. By collecting and analyzing various types of data (such as CPU, memory, bandwidth, latency, user behavior, etc.) of the streaming media system in real time, and using machine learning algorithms (such as time series analysis, regression model, LSTM, reinforcement learning, etc.) to predict the load trend, through intelligent data collection, anomaly detection, load scoring, predictive scaling, and other technologies, automatic resource allocation and pre-processing before problems occur are realized. For example, the AI system can predict the traffic changes in the next period of time, adjust the server resources in advance, and perform automatic repair or adjustment when anomalies occur. The present invention realizes the prediction of the trend of resource occupation by the host load change, the monitoring of the host performance, and the automatic shielding of faults. Brief Description of the Drawings

[0023] The following further describes the present invention in detail with reference to the drawings and specific embodiments; Figure 1 It is a schematic diagram of the architecture of the streaming media host load resource trend prediction and load balancing optimization system of the present invention. Specific Embodiments

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application.

[0025] As Figure 1 shown, the present invention discloses a streaming media host load resource trend prediction and load balancing optimization system, which includes: Data Acquisition and Real-time Monitoring Module: It is used to comprehensively monitor several dimensional indicators of the streaming media cluster nodes to form real-time collected data; the several dimensional indicators include server performance indicators, network status, and streaming media service indicators; Intelligent Diagnosis and Anomaly Detection Module: By analyzing the real-time collected data, it performs intelligent diagnosis through integrated artificial intelligence models applicable to several different dimensional diagnosis tasks to achieve anomaly detection in different dimensions; through the loaded load scoring model, it evaluates the health status of each node in real time to assist in load balancing decisions, and at the same time predicts the load peak period in advance to perform pre-allocation of resources to achieve predictive scaling; Fault Automatic Masking and Load Balancing Module: It is used to automatically mask faulty nodes in case of system failures and redistribute the load to healthy nodes; Fault Self-Healing Module: It is used to synchronously cache popular content through edge nodes when a hardware failure occurs under the condition of automatically masking faulty nodes, ensuring that the streaming media sessions of existing users are not interrupted.

[0026] Furthermore, the streaming media service indicators include the bit rate of the video stream, the number of buffering times, and the user QoE (Quality of Experience) score.

[0027] Furthermore, the server performance indicators include the CPU utilization rate, memory occupancy, disk I / O, network bandwidth, number of connections, and number of streaming media sessions of each node.

[0028] Method for Predicting the Trend of Streaming Media Host Load Resources and Optimizing Load Balancing, which includes the following steps: Step 1, Data Acquisition and Real-time Monitoring: Comprehensively monitor several dimensional indicators of the streaming media cluster nodes to form real-time collected data; the several dimensional indicators include server performance indicators, network status, and streaming media service indicators; Step 2, Intelligent Diagnosis and Anomaly Detection: By analyzing the real-time collected data, perform intelligent diagnosis through integrated artificial intelligence models to achieve anomaly detection in different dimensions; through the loaded load scoring model, evaluate the health status of each node in real time and calculate the node load score to assist in load balancing decisions, and at the same time predict the load peak period in advance to perform pre-allocation of resources to achieve predictive scaling; Step 3, Fault Automatic Masking and Load Balancing: Automatically mask faulty nodes in case of system failures and redistribute the load to healthy nodes; Furthermore, in Step 3, the AI automatically analyzes and selects the optimal processing server in the current environment according to the different resource types relied on by different service requests.

[0029] Step 4, Fault Self-Healing: When a hardware failure occurs, synchronously cache popular content through edge nodes under the condition of automatically masking faulty nodes, ensuring that the streaming media sessions of existing users are not interrupted.

[0030] Furthermore, step 1 specifically includes the following steps: Step 1-1, deploy a system monitoring tool as a data collection agent on each streaming service node, and collect and transmit monitoring data in real time based on the set collection frequency; Specifically, use open-source system monitoring tools (such as Prometheus, Telegraf, Collectd, etc.) as data collection agents.

[0031] Step 1-2, (using MQTT) transmit the collected data to the backend system to ensure low latency and high reliability of data transmission.

[0032] Step 1-3, use a time series database (such as InfluxDB or TimescaleDB) to store the collected data to support efficient time series data queries.

[0033] Furthermore, step 2 specifically includes the following steps: Step 2-1, build and train an artificial intelligence model suitable for several different dimension diagnosis tasks; Step 2-2, use the artificial intelligence model to analyze the real-time collected data for load prediction and anomaly detection; Step 2-3, use the loaded load scoring model to evaluate the health status of each node in real time and calculate the real-time load score of each node.

[0034] Furthermore, step 2-1 specifically includes the following steps: Step 2-1-1, first clean, standardize and normalize the collected data for use in artificial intelligence model training; Step 2-1-2, jointly train the artificial intelligence model in combination with several different dimension diagnosis tasks to optimize the overall performance of the model; the diagnosis tasks include load prediction, anomaly detection, fault warning, etc.; Step 2-1-3, use online learning to update the model in real time so that the model maintains good adaptability as the system data changes continuously; Step 2-1-4, package the artificial intelligence model and related dependencies into a Docker container to ensure consistency and scalability in different environments; Step 2-1-5, integrate the artificial intelligence model into the intelligent diagnosis layer of the streaming service platform, and the model obtains real-time performance data through the API for analysis and prediction.

[0035] Furthermore, in step 2-1, the artificial intelligence model is regularly evaluated, and the model parameters are adjusted according to the evaluation results to improve the prediction and diagnosis accuracy. The model is regularly trained, and incremental training is carried out in combination with newly collected data to ensure that the model can adapt to changes in streaming media load.

[0036] Furthermore, step 3 specifically includes the following steps: Step 3-1, load balancing and request scheduling: The artificial intelligence model distributes requests according to the real-time load score and health status, using a dynamic scheduling algorithm (such as weighted least connection or QoE-based hashing algorithm) to preferentially select healthy nodes. Through precise load balancing, overload and resource waste are avoided.

[0037] Step 3-2, session migration mechanism: Migrate ongoing streaming media sessions to low-load nodes hot. Achieve seamless migration of streaming media sessions through technologies such as WebRTC SFU (Selective Forwarding Unit) to ensure the continuity of the user experience.

[0038] Step 3-3, automatically shield faulty nodes: When the artificial intelligence model detects that a node fails, automatically mark the faulty node as "risk status" and stop allocating new requests to the faulty node; at the same time, the system triggers the edge collaboration mechanism to preferentially direct user requests to the nearest healthy node to ensure uninterrupted service.

[0039] Furthermore, step 4 specifically includes the following steps: Step 4-1, monitor the hardware resources of each node through a monitoring tool to determine whether the use of hard disk resources exceeds the set threshold (such as the load of disk I / O exceeds 80%); if so, trigger an alarm and perform trend analysis on the historical data of disk I / O through machine learning algorithms to identify possible hardware failure trends; otherwise, execute step 4-1; Step 4-2, determine whether the alarm duration exceeds the set time (3 seconds); if so, the system will automatically mark the corresponding node as "risk status"; Step 4-3, when the load balancer receives the node risk status, stop allocating new requests to the corresponding node; at the same time, transfer the requests to other healthy nodes through an algorithm; Step 4-4, obtain the geographical location of the request and combine it with the health status of the CDN node to determine the optimal request routing.

[0040] Specifically, perform node selection and routing according to the geographical location and load conditions, route user requests to the nearest CDN edge node with low load; monitor the load status of the CDN node in real time and dynamically adjust the routing strategy to avoid overload of a certain node and ensure fast response to requests.

[0041] Furthermore, in step 4-3, a temporary caching mechanism is enabled on the edge nodes synchronously, and the most frequently accessed popular content on the corresponding nodes is cached on healthy edge nodes to ensure that the session is not interrupted.

[0042] The specific working principle of the present invention will be described in detail below:

[0043] Data collection: Use open-source system monitoring tools (such as Prometheus, Telegraf, Collectd, etc.) as data collection agents, which are deployed on each streaming media service node to collect and transmit monitoring data in real time. The collected data needs to be standardized and a common data format (such as JSON or Prometheus Metric format) is adopted for subsequent data analysis and model processing. Set an appropriate collection frequency according to the load situation, usually between 1 second and 5 seconds, to ensure that the real-time load situation can be reflected.

[0044] Data transmission: Use MQTT to transmit the collected data to the backend system to ensure low latency and high reliability of data transmission.

[0045] Data storage: Use a time series database (such as InfluxDB or TimescaleDB) to store the collected data to support efficient time series data queries.

[0046] (2) Intelligent diagnosis layer: It includes DeepSeek model training and use, model evaluation and optimization.

[0047] DeepSeek model training and use: Based on DeepSeek, deep learning modeling is carried out. Deep neural networks (such as LSTM, GRU, CNN, etc.) are used to process time series data for load prediction and anomaly detection. First, the collected data is cleaned, standardized, and normalized for model training. Combine multiple tasks (such as load prediction, anomaly detection, fault warning, etc.) for joint training to optimize the overall performance of the model. Use online learning to update the model in real time. As the system data changes continuously, the model can also maintain good adaptability. Package the model and related dependencies into a Docker container to ensure consistency and scalability in different environments. Integrate the model into the intelligent diagnosis layer of the streaming media service platform, and the model obtains real-time performance data through the API for analysis and prediction.

[0048] Model evaluation and optimization: Use evaluation metrics such as accuracy, recall rate, and F1 score to evaluate the model regularly, and adjust the model parameters according to the evaluation results to improve the prediction and diagnosis accuracy. Regularly train the model and perform incremental training in combination with newly collected data to ensure that the model can adapt to the changes in the streaming media load.

[0049] (3)Build the load balancing layer: including the construction of three subsystems: a dynamic scheduling system, a session migration system, and an edge collaboration and request routing system Build the dynamic scheduling system: Obtain the real-time load scores of each node (such as CPU load, latency, QoE score, etc.) from the intelligent diagnosis layer through an interface. Based on the load scores, set and call the load balancing algorithm based on QoE-based hashing. According to the algorithm results, dynamically allocate streaming media requests to nodes with higher scores and lighter loads. After each request allocation, feedback the current traffic load status through the load score interface in the first step for adjustment in the next round of scheduling decisions.

[0050] Build the session migration system: Establish a load balancing system, set up to monitor the load conditions of each node in real time through an interface. Once the load of a certain node exceeds the set threshold, call session migration. Select a suitable low-load node for session migration according to the health status and load score of the node. Seamlessly transfer the state of the session through hot migration technology to avoid interruption of the user experience. The migrated session continues to execute on the new node, and feedback the migration result to ensure the real-time nature of load balancing.

[0051] Build the edge collaboration and request routing system: Obtain the geographical location of the request, and combine it with the health status of the CDN nodes to determine the optimal request routing. Select and route nodes according to the geographical location and load conditions, and route user requests to the nearest and lightly loaded CDN edge nodes. Monitor the load status of CDN nodes in real time and dynamically adjust the routing strategy to avoid overloading of a certain node and ensure fast response to requests.

[0052] (4)Fault self-healing and energy efficiency optimization: including three subsystems: hardware fault detection, hardware fault switching, and fault recovery

[0053] Build the hardware fault detection system: Monitor the hardware resources of each node through monitoring tools, especially key resources such as disk I / O, memory, CPU, and network. Conduct detailed monitoring of disk I / O, set thresholds (such as when the load of disk I / O exceeds 80%), and trigger an alarm. Perform trend analysis on the historical data of disk I / O through machine learning algorithms to identify possible hardware fault trends. For example, analyze disk I / O data through LSTM (Long Short-Term Memory Network) to predict whether a hardware fault will occur.

[0054] Build a hardware failure switching system: When the disk I / O abnormally increases and lasts for more than 3 seconds, the system will automatically mark this node as "risk status". When the load balancer receives the node risk status, it will stop allocating new requests to this node. The requests will be transferred to other healthy nodes through an algorithm. Enable a temporary caching mechanism on the edge node to cache the most frequently accessed popular content on this node to a healthy edge node to ensure that the session is not interrupted.

[0055] Build a failure recovery system: When the disk I / O problem of the node is resolved (for example, the disk I / O returns to the normal range), the system will automatically detect the recovery situation. When the node returns to normal, the load balancer will gradually restore the traffic to this node while performing load balancing to avoid system crashes caused by suddenly restored traffic. When recovering the node, resynchronize the cached content to this node to ensure content consistency.

[0056] The dataset construction for training the DeepSeek model of the present invention is achieved by collecting historical monitoring data, fault logs, and user feedback, and annotating the normal / abnormal status. The DeepSeek model adopts multi-task learning to simultaneously train load prediction (regression task) and anomaly classification (binary classification task) to improve the generalization ability. The DeepSeek model can learn online and continuously fine-tune the model with real-time data after deployment to adapt to hardware upgrades or business changes.

[0057] The present invention realizes a low-latency data pipeline, uses Apache Kafka or Pulsar to achieve high-throughput monitoring data streams, and ensures that the data latency is <100 ms. The second-level metric collection is realized through Prometheus + VictoriaMetrics, and Grafana is used for real-time visualization.

[0058] The present invention has fault tolerance and elasticity. When the DeepSeek service is abnormal, it will automatically degrade to a weighted round-robin policy based on response time. Use Docker / Kubernetes to limit the resource occupation of a single node and achieve resource isolation to avoid cascading failures.

[0059] Optimization scenario example: When a certain live channel suddenly becomes popular, DeepSeek detects that the load of the central node exceeds the threshold and automatically triggers: Redirect 50% of the new user requests to the standby area cluster. Enable the pre-generated low-bitrate version of the video to reduce the bandwidth pressure. When the disk IO of a certain node abnormally increases, DeepSeek marks it as "risk status" within 3 seconds: The load balancer stops allocating new requests to this node. Start the temporary storage of popular content on the edge node to ensure that the connected users do not experience stream interruption.

[0060] The present invention adopts the above technical solutions. By collecting and analyzing various types of data of the streaming media system in real time (such as CPU, memory, bandwidth, latency, user behavior, etc.), and using machine learning algorithms (such as time series analysis, regression model, LSTM, reinforcement learning, etc.) to predict the load trend, through intelligent data collection, anomaly detection, load scoring, predictive scaling, and other technologies, automatic resource allocation and pre-processing before problems occur are realized. For example, the AI system can predict the traffic changes in the future period, adjust the server resources in advance, and perform automatic repair or adjustment when anomalies occur. The present invention realizes the prediction of the trend of the resource occupied by the host load change, the monitoring of the host performance, and the automatic shielding of faults.

[0061] Obviously, the described embodiments are part of the embodiments of this application, rather than all embodiments. Without conflict, the embodiments in this application and the features in the embodiments can be combined with each other. Usually, the components of the embodiments of this application described and illustrated in the drawings here can be arranged and designed in various different configurations. Therefore, the detailed description of the embodiments of this application is not intended to limit the scope of this application claimed, but merely represents the selected embodiments of this application. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of this application.

Claims

1. A streaming media host load resource trend prediction and load balancing optimization system, characterized in that: It includes: Data collection and real-time monitoring module: used to comprehensively monitor several dimension indicators of the streaming media cluster nodes to form real-time collected data; The several dimension indicators include server performance indicators, network status, and streaming media service indicators; Intelligent diagnosis and anomaly detection module: By analyzing the real-time collected data, perform intelligent diagnosis through an integrated artificial intelligence model applicable to several different dimension diagnosis tasks to achieve anomaly detection in different dimensions; Real-time evaluate the health status of each node through the loaded load scoring model to assist in load balancing decisions; At the same time, predict the peak load period and pre-allocate resources in advance to achieve predictive scaling; Fault automatic shielding and load balancing module: used to automatically shield faulty nodes in case of system failures and redistribute the load to healthy nodes; Fault self-healing module: used to synchronously cache popular content through edge nodes temporarily in case of hardware failures under the condition of automatically shielding faulty nodes to ensure that the streaming media sessions of existing users are not interrupted.

2. The streaming media host load resource trend prediction and load balancing optimization system according to claim 1, wherein: The streaming media service indicators include the bit rate of the video stream, the number of buffering times, and the user QoE score; The server performance indicators include the CPU utilization rate, memory occupancy, disk I / O, network bandwidth, connection number, and streaming media session number of each node.

3. Method for predicting the trend of load resources of a streaming media host and optimizing load balancing, using the system for predicting the trend of load resources of a streaming media host and optimizing load balancing according to claim 1 or 2, characterized in that: The method includes the following steps: Step 1, Data collection and real-time monitoring: Comprehensively monitor several dimension indicators of the streaming media cluster nodes to form real-time collected data; The several dimension indicators include server performance indicators, network status, and streaming media service indicators; Step 2, Intelligent diagnosis and anomaly detection: By analyzing the real-time collected data, perform intelligent diagnosis through an integrated artificial intelligence model to achieve anomaly detection in different dimensions; Real-time evaluate the health status of each node through the loaded load scoring model and calculate the node load score to assist in load balancing decisions, and at the same time, predict the peak load period and pre-allocate resources in advance to achieve predictive scaling; Step 3, Fault automatic shielding and load balancing: Automatically shield faulty nodes in case of system failures and redistribute the load to healthy nodes; Step 4, Fault self-healing: In case of hardware failures, synchronously cache popular content through edge nodes temporarily under the condition of automatically shielding faulty nodes to ensure that the streaming media sessions of existing users are not interrupted.

4. The method for predicting the trend of load resources of a streaming media host and optimizing load balancing according to claim 3, wherein: Step 1 specifically includes the following steps: Step 1-1, Deploy the system monitoring tool as a data collection agent on each streaming media service node, and collect and transmit monitoring data in real time based on the set collection frequency; Step 1-2, Transmit the collected data to the backend system to ensure low latency and high reliability of data transmission; Step 1-3, Use a time series database to store the collected data to support efficient time series data queries.

5. The method for predicting the trend of load resources of a streaming media host and optimizing load balancing according to claim 3, wherein: Step 2 specifically includes the following steps: Step 2-1, Build and train an artificial intelligence model applicable to several different dimension diagnosis tasks; Step 2-2, Use the artificial intelligence model to analyze the real-time collected data for load prediction and anomaly detection; Step 2-3, Real-time evaluate the health status of each node through the loaded load scoring model and calculate the real-time load score of each node.

6. The method for predicting the trend of load resources of a streaming media host and optimizing load balancing according to claim 5, characterized in that: Step 2-1 specifically includes the following steps: Step 2-1-1: First, clean, standardize, and normalize the collected data for use in artificial intelligence model training; Step 2-1-2: Jointly train the artificial intelligence model in combination with several different-dimensional diagnostic tasks to optimize the overall performance of the model; the diagnostic tasks include load prediction, anomaly detection, fault warning, etc.; Step 2-1-3: Use online learning to update the model in real time so that the model maintains good adaptability as the system data continues to change; Step 2-1-4: Package the artificial intelligence model and related dependencies into a Docker container to ensure consistency and scalability in different environments; Step 2-1-5: Integrate the artificial intelligence model into the intelligent diagnosis layer of the streaming media service platform, and the model obtains real-time performance data through the API for analysis and prediction.

7. The method for predicting the trend of load resources of a streaming media host and optimizing load balancing according to claim 6, characterized in that: In Step 2-1, regularly evaluate the artificial intelligence model, and adjust the model parameters according to the evaluation results to improve the prediction and diagnosis accuracy; regularly train the model, and perform incremental training in combination with newly collected data to ensure that the model can adapt to the changes in the streaming media load.

8. The method for predicting the trend of load resources of a streaming media host and optimizing load balancing according to claim 3, wherein: Step 3 specifically includes the following steps: Step 3-1: Load balancing and request scheduling: The artificial intelligence model uses a dynamic scheduling algorithm to allocate requests according to the real-time load score and health status to preferentially select healthy nodes; Step 3-2: Session migration mechanism: Hot migrate the ongoing streaming media session to a low-load node; Step 3-3: Automatically shield faulty nodes: When the artificial intelligence detects that a node fails, automatically mark the faulty node as "risk status" and stop allocating new requests to the faulty node; at the same time, the system triggers the edge collaboration mechanism to preferentially direct user requests to the nearest healthy node.

9. The method for predicting the trend of load resources of a streaming media host and optimizing load balancing according to claim 3, wherein: Step 4 specifically includes the following steps: Step 4-1: Monitor the hardware resources of each node through a monitoring tool to determine whether the use of disk resources exceeds the set threshold; if so, trigger an alarm and perform trend analysis on the historical data of disk I / O through a machine learning algorithm to identify possible hardware fault trends; otherwise, execute Step 4-1; Step 4-2: Determine whether the alarm duration exceeds the set time; if so, the system will automatically mark the corresponding node as "risk status"; Step 4-3: When the load balancer receives the node risk status, stop allocating new requests to the corresponding node; at the same time, transfer the requests to other healthy nodes through an algorithm; Step 4-4: Obtain the geographical location of the request and combine it with the health status of the CDN node to determine the optimal request routing.

10. The method for predicting the trend of load resources of a streaming media host and optimizing load balancing according to claim 9, wherein: In Step 4-3, enable a temporary caching mechanism on the edge node synchronously, and cache the most frequently accessed popular content on the corresponding node to a healthy edge node to ensure that the session is not interrupted.

Images