A multi-modal content interactive library comprehensive display method

Abstract

The application discloses a kind of multi-modal content interactive library comprehensive display method, it is related to digital library field, first, this method is through node autonomous agent module Real-time acquisition CPU occupancy, I / O queue length and average response time and other operating performance parameters, and generates node heat index H by normalization calculation, realizes the quantitative description and dynamic evaluation of the running load state of each display node.The scheme avoids the problem of low resource utilization caused by manual setting refresh cycle and static task allocation in traditional library display system, can automatically execute scheduling strategy logic according to node performance change, so as to realize dynamic load balancing and resource adaptive scheduling of system among multiple nodes.Therefore, without increasing hardware resources, the system peak CPU occupancy is significantly reduced, with an average decrease of about 25%, and the real-time performance and response speed of node task allocation are improved, and the operating efficiency of the entire display system is improved.

Description

Technical Field

[0001] This invention relates to the field of digital library technology, specifically to a multimodal interactive library integrated display method. Background Technology

[0002] With the continuous advancement of digitalization and intelligentization in libraries, multimodal information display technology is increasingly widely used in cultural information dissemination and knowledge services. Modern libraries are gradually shifting from traditional static information display models to comprehensive display systems that integrate multimodal content such as images, videos, audio, and interactive feedback to improve information retrieval efficiency and user immersion. In this new display system, multiple display nodes are typically included, such as large-screen interactive walls, touch-screen guide tables, self-service borrowing and returning terminals, integrated document retrieval kiosks, and exhibition playback terminals. These nodes operate collaboratively within the same network environment to simultaneously present multiple types of multimodal interactive information content. Therefore, how to achieve dynamic content allocation, channel priority scheduling, and frame-level synchronization control among multiple nodes at the system level without changing the hardware structure has become a key technical problem that library multimodal display systems urgently need to solve.

[0003] Currently, in libraries and similar public cultural service scenarios, common display systems typically employ fixed task allocation and polling refresh mechanisms, with each display node operating independently and without coordination. When a display terminal loads highly complex multimodal content, such as high-definition video, 3D interactive models, or touch feedback tasks, its CPU utilization and I / O load will rapidly increase, leading to system lag. Meanwhile, other terminals may be under low load or idle, resulting in severely uneven resource allocation. Existing solutions often rely on manually configuring refresh cycles or using task migration strategies triggered by a single threshold, lacking quantitative evaluation and dynamic feedback mechanisms for node operating status, and failing to achieve adaptive scheduling and load balancing among nodes at the system level. Furthermore, traditional display systems often use static channel reconnection for node switching, which can easily lead to frame loss, screen flickering, or brief black screens during switching, affecting display continuity and user experience.

[0004] The root cause lies in the current lack of a bidirectional mapping mechanism between node performance status and task complexity in most display systems. The system cannot dynamically adjust task allocation and signal scheduling strategies based on the actual operating status of nodes. When nodes are under high load for extended periods without timely scheduling, it causes node cache refresh blockage, a decrease in channel refresh frequency, and a significant extension of interaction response time. This leads to abnormal effects such as delayed display, unresponsive interaction, and touch response drift, and in severe cases, may trigger system interruptions or task crashes. Furthermore, without a unified frame order alignment mechanism, the cumulative increase in frame time differences during cross-node task migration can cause display channel output asynchrony, resulting in visual misalignment and information discontinuity when users switch between different display terminals. These problems collectively limit the parallel operation capability and scalability of multimodal content in library integrated display systems, necessitating a quantifiable, self-learning, and frame-level synchronized dynamic content allocation method to address these issues. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a multimodal content interactive library integrated display method, which solves the problems mentioned in the background technology.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a multimodal content interactive library integrated display method, comprising the following steps:

[0007] S1. In the multiple display nodes of the library display system, the operating performance parameters of each display node are collected respectively, and the node autonomous agent module performs normalization calculation on the collected operating performance parameters, generates the node popularity index H, and executes the scheduling strategy logic.

[0008] S2. Based on the scheduling strategy logic, an execution priority switching mechanism is implemented to dynamically schedule the priority coefficient P of the computing node and perform a preliminary comparison and evaluation. Based on the preliminary comparison and evaluation results, a node resource release rate Rrel is generated. Based on the node resource release rate Rrel, the node autonomous agent module triggers the display channel scheduling module to adjust the channel priority table.

[0009] S3. Based on the output result of the generated node resource release rate Rrel, a prediction delay mechanism is triggered. The prediction delay mechanism is initiated by the central scheduling server of the library display system to start the delay prediction engine and calculate the node prediction delay Dpred.

[0010] S4. Perform a secondary comparison and evaluation based on the output of the predicted delay Dpred, and trigger the task migration mechanism based on the results of the secondary comparison and evaluation.

[0011] Preferably, S1 includes S11;

[0012] S11. The library display system is set up with display nodes in a dual-caching layer structure, including large-screen interactive wall, touch guide table, document retrieval all-in-one machine, self-service borrowing and returning terminal and theme exhibition playback terminal; and the content loading, eviction and refresh thread management is implemented between the dual-caching layer structure through the cache controller, and a cache status interface is provided to the channel scheduling module.

[0013] The dual-caching layer structure includes an L1 cache and an L2 cache.

[0014] Preferably, S1 further includes S12;

[0015] S12. Each display node is equipped with an autonomous agent module. The autonomous agent module collects the display node's operating status parameters according to the sampling window and uploads them to the central scheduling server.

[0016] The node operating status parameters include the CPU utilization Uc of the i-th display node. i The I / O queue length Io of the i-th display node i and the average response time Rt of the i-th display node i ;

[0017] The node popularity index H is calculated and displayed in the central scheduling server based on the node's operating parameters. The node popularity index H is calculated and output using the following algorithm formula: H i =(UC i +k1·Io i / Io max +k2·Rt i / Rt max H / 3; where H i This represents the node popularity index of the i-th displayed node. k1 and k2 represent the I / O weight calibration coefficient and response time weight calibration coefficient, respectively, and k1+k2=1. Their specific values ​​are set by the user. max Rt represents the historical maximum I / O queue length. max This represents the historical maximum average task response time.

[0018] Preferably, S1 further includes S13;

[0019] S13. Based on the node popularity index H and the CPU utilization rate Uc of the i-th display node. i The execution scheduling strategy logic is implemented, and an execution priority switching mechanism is established based on the result of the scheduling strategy logic.

[0020] The scheduling strategy logic calculates the average CPU utilization of the currently displayed node through the autonomous agent module in a sliding window of 15 seconds, and determines the CPU utilization threshold Ucth by combining it with historical running data. The CPU utilization threshold Ucth is the larger of the 95th percentile value of the node's historical CPU utilization Uc and the fixed upper limit of 0.80.

[0021] When the CPU utilization Uc of the i-th display node is detected i ≥CPU utilization threshold Ucth, and the node popularity index H of the i-th displayed node. i If the duration of the state above the historical median value is not less than 8 seconds, the current display node is judged to be under abnormal pressure.

[0022] When the determination is successful, a cache refresh status signal (Scache) is generated and reported to the central scheduling server. The cache refresh status signal (Scache) is a structured data packet, including the node identifier (IDnode), timestamp (Ts), and CPU utilization (Uc) of the i-th display node. i CPU utilization threshold Ucth, node popularity index H of the i-th displayed node i The cache control instruction Op and the cache hold time Thold are used to trigger the priority switching mechanism.

[0023] Conversely, if the condition is not met, the cache controller will automatically generate a recovery instruction to restore the L2 cache refresh.

[0024] Preferably, S2 includes S21;

[0025] S21. After triggering the priority switching mechanism, the cache refresh status signal Scache is transmitted to the autonomous proxy module, and the node identifier IDnode and the node popularity index H of the i-th displayed node are read from the structured data packet. i and cache control instructions (Op);

[0026] The channel scheduling module updates the task entry status according to the channel scheduling table of the corresponding node. When the buffer control instruction Op is set to PAUSE-L2, the refresh cycle parameter of the task entry associated with the L2 buffer is adjusted to 0.5 times its original value, and the signal channel proportion parameter is reduced by 20%.

[0027] In the updated channel scheduling table, each channel task entry includes three fields: task type, refresh cycle, and signal channel percentage.

[0028] The autonomous agent module is based on the updated channel scheduling table and uses an improved Sigmoid dynamic function to calculate the dynamic scheduling priority coefficient P of the node, thus smoothly controlling the priority switching process.

[0029] The dynamic scheduling priority coefficient P is calculated and output using the following algorithm formula: In the formula, P i Let represent the dynamic scheduling priority coefficient of the i-th display node, and let e represent the exponential function. This represents the sensitivity constant, with a value range of 0.2-1.0, which is set by the user. Hth represents the node heat threshold.

[0030] Preferably, S2 further includes S22;

[0031] S22. Based on the output of the dynamic scheduling priority coefficient P, a preliminary comparative evaluation is performed to determine the load status of the displayed nodes. Based on the preliminary comparative evaluation results, a resource protection mode is activated, and corresponding control operations are executed. The specific evaluation content is as follows:

[0032] When the dynamic scheduling priority coefficient P of the i-th display node i When the value is ≥0.7, it indicates that the display node is under high load. The channel scheduling module enters resource protection mode and performs the following corresponding control operations: the refresh frame rate of the priority channel is reduced to 0.6 times the original value; the cache refresh thread of the L2 cache is suspended, and only the cache update of the L1 cache is maintained to reduce disk and memory I / O pressure; the local cache controller is notified to calculate the node resource release rate Rrel of the i-th display node. i Its calculation formula is: Rrel i =ΔCPU i +ΔIO i , where ΔCPU i ΔIO represents the percentage decrease in average CPU utilization before and after scheduling of the i-th display node; i This represents the percentage decrease in the average I / O queue length of the i-th display node;

[0033] When the dynamic scheduling priority coefficient P of the i-th display node i When the value is less than 0.7, it indicates that the node is in a stable load state, and the channel scheduling module maintains the current channel priority table unchanged.

[0034] Preferably, S3 includes S31;

[0035] S31, when the node resource release rate Rrel of the i-th displayed node... i If the value is below 0.25 for three consecutive sampling periods, it indicates that local scheduling cannot solve the load problem. At this time, the startup delay prediction engine of the central scheduling server is activated to collect real-time running data from each display node.

[0036] The real-time operational data includes the average response time Rt of the i-th display node. i and the rate of change of response time dRt of the i-th display nodei / dt;

[0037] The extreme value normalization method is used to normalize the real-time running data and the node resource release rate Rrel, eliminating the difference between the real-time running data and the node resource release rate Rrel of the i-th display node. i The influence of the unit dimensions of all parameters.

[0038] Preferably, S3 further includes S32;

[0039] S32, Based on the normalized real-time running data and the node resource release rate Rrel of the i-th display node. i A feedback-based prediction delay model is used to calculate the node prediction delay Dpred and to quantitatively analyze the trend of node response time changes and resource release status.

[0040] The node prediction delay Dpred is calculated and output using the following feedback prediction delay model;

[0041] ; where, Dpred i Let d represent the node prediction delay of the i-th displayed node, d represent the integral function, and dt represent the time integral function. represents the time prediction gain, with a value range of 0.05-1, and u represents the resource feedback gain, with a value range of 0.05-1.

[0042] Preferably, S4 includes S41;

[0043] S41, Rt based on the average response time of the i-th historical display node i Set a delay threshold Dth to 1.5 times the value of the i-th display node, and set the node prediction delay Dpred of the i-th display node to be obtained in real time. i A secondary comparison and evaluation is performed with the delay threshold Dth, and the corresponding task migration mechanism is triggered based on the results of the secondary comparison and evaluation; the specific evaluation content is as follows;

[0044] When the node prediction delay of the i-th displayed node is Depred i When the latency threshold Dth is greater than or equal to the current display node, it is determined that the current display node will enter a latency overload prediction state in the future. At this time, the central scheduling server starts the content mirror pool call and triggers the task migration mechanism.

[0045] When the node prediction delay of the i-th displayed node is Depred i When the delay threshold Dth is less than the current delay threshold, it is determined that the current display node is in a stable delay state, and the central scheduling server maintains the current task allocation unchanged.

[0046] Preferably, S4 further includes S42;

[0047] S42. After the content mirror pool invocation mechanism is activated, the central low-degree server predicts the latency Dpred based on the i-th display node. i The latency threshold Dth and the node resource release rate Rrel of the i-th displayed node. i Calculate the migration priority index M of the i-th display node. i The specific calculation formula is: M i =Dpred / Dth×(1-Rrel)

[0048] The migration priority index M of all display nodes is obtained by traversing each display node, and the task migration mechanism is executed based on the migration priority index M of all display nodes. The task migration mechanism sorts all display nodes from high to low according to the migration priority index M, and then migrates the display tasks to the display node with the lowest load in order from high to low. Before the task migration, the channel synchronization control module of the central scheduling server reads the difference between the display buffer timestamp Tsrc of the migrating display node and the display buffer timestamp Tdst of the target display node. When the difference exceeds 30ms, the data frame alignment operation is performed.

[0049] The data frame alignment operation records the timestamp and frame number information of the current output frame in real time by setting a local time stamp module in each display node. When the central scheduling server detects a migration request, it will simultaneously obtain the latest frame timestamp of the migration node and the target display node. If the frame time difference between the two display nodes is greater than or equal to 30ms, the target display node enters the pre-synchronization mode.

[0050] The pre-synchronization mode adjusts the local playback buffer rate and frame buffer queue depth frame by frame to gradually catch up with the timing of the output frames of the target display node to match that of the migration node.

[0051] After each adjustment, the frame time difference is re-detected. When the time difference is less than 30ms for two consecutive detection cycles, the frame order is determined to be aligned.

[0052] This invention provides a multimodal content interactive integrated library display method. It has the following beneficial effects:

[0053] (1) This method collects operational performance parameters such as CPU utilization, I / O queue length, and average response time in real time through a node autonomous agent module, and generates a node heat index H through normalization calculation, thereby realizing a quantitative description and dynamic evaluation of the operational load status of each display node. This scheme avoids the problem of low resource utilization caused by manually setting refresh cycles and static task allocation in traditional library display systems. It can automatically execute scheduling strategy logic according to changes in node performance, thereby realizing dynamic load balancing and adaptive resource scheduling among multiple nodes. Therefore, this invention significantly reduces the peak CPU utilization of the system by an average of about 25% without increasing hardware resources, and improves the real-time performance and response speed of node task allocation, thereby improving the overall operating efficiency of the display system.

[0054] (2) This method achieves intelligent adjustment of channel priority and cache refresh mechanism through the joint control of dynamic scheduling priority coefficient P and node resource release rate R_rel. When the node load exceeds the threshold, the system automatically enters resource protection mode, suspends L2 cache refresh, and only maintains L1 cache update; at the same time, according to the channel scheduling table update mechanism, the system can optimize the signal channel ratio and refresh cycle in real time without interrupting content output, and dynamically reduce the refresh frequency of low priority tasks. This mechanism effectively avoids the problems of display task congestion and cache accumulation in high-concurrency scenarios, significantly improves the operational stability and display continuity between nodes, reduces the overall frame rate fluctuation of the system to within ±3%, and makes the output of display content smoother and without jitter.

[0055] (3) This method achieves predictive control and frame-level synchronization of cross-node content allocation through the latency prediction engine and task migration mechanism of the central scheduling server. The system calculates the migration priority index M to respond in advance to potential latency risks. When task migration is triggered, the channel synchronization control module performs frame order alignment to achieve millisecond-level timing synchronization control with a time error of less than 30ms. This technology not only effectively prevents frame skipping, black screens, and display interruptions during task migration, but also significantly improves the time base consistency of multi-terminal collaborative display. Therefore, this invention can maintain high frame rate output and low latency interaction in complex multimodal content switching scenarios, making the system more fluid and stable in terms of user-side visual and operational experience. Attached Figure Description

[0056] Figure 1 This is a schematic diagram illustrating the steps of a multimodal content interactive library integrated display method according to the present invention.

[0057] Figure 2 This is to illustrate the internal module structure of the node. Detailed Implementation

[0058] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0059] Example 1

[0060] This invention provides a multimodal content interactive library integrated display method. Please refer to [link / reference]. Figure 1 This includes the following steps:

[0061] S1. In the multiple display nodes of the library display system, the operating performance parameters of each display node are collected respectively, and the node autonomous agent module performs normalization calculation on the collected operating performance parameters, generates the node popularity index H, and executes the scheduling strategy logic.

[0062] S2. Based on the scheduling strategy logic, an execution priority switching mechanism is implemented to dynamically schedule the priority coefficient P of the computing node and perform a preliminary comparison and evaluation. Based on the preliminary comparison and evaluation results, a node resource release rate Rrel is generated. Based on the node resource release rate Rrel, the node autonomous agent module triggers the display channel scheduling module to adjust the channel priority table.

[0063] S3. Based on the output of the node resource release rate Rrel, a delay prediction mechanism is triggered. The delay prediction mechanism starts the delay prediction engine through the central scheduling server of the library display system and calculates the node prediction delay Dpred.

[0064] S4. Perform a secondary comparison and evaluation based on the output of the predicted delay Dpred, and trigger the task migration mechanism based on the results of the secondary comparison and evaluation.

[0065] In this embodiment, the method ensures that the entire display system forms a continuous dynamic scheduling closed loop through steps S1 to S4, guaranteeing stable operation and continuous screen output for nodes under different load conditions. In S1, the system collects the operating performance parameters of each display node in real time and calculates the node heat index H. This aims to convert the CPU usage, I / O load, and response time of different nodes into a unified dimensionless heat index, enabling real-time visualization of node operating status. Through this real-time quantification process, the system can quickly identify nodes with abnormal loads, preventing overall performance degradation caused by task scheduling lag. Its physical meaning lies in mapping node operating pressure in numerical form, achieving continuous dynamic measurement of resource usage. In S2, the node autonomous agent module triggers a priority switching mechanism based on heat changes, calculating a dynamic scheduling priority coefficient P. When node load increases, it prioritizes high-frequency tasks by pausing L2 cache refresh and maintaining L1 fast channel operation. This implementation aims to avoid cache congestion and bandwidth conflicts when system pressure rises instantaneously, thereby maintaining stable output of real-time display tasks. Physically, this is equivalent to reducing I / O writes on non-critical paths, freeing up CPU and bus resources, and keeping the node's computing links unobstructed. In S3, when the node resource release rate Rrel remains consistently low, the central scheduling server activates the latency prediction engine to calculate the node's predicted latency Dpred, providing early warning of potential latency. This setup aims to detect potential response degradation trends in advance, preventing performance crashes from a time-based perspective. Its physical significance lies in feedforward modeling of the node response curve, transforming latency change trends into predictable signals, and achieving a shift from passive scheduling to active control. In S4, when the predicted latency Dpred exceeds the set latency threshold Dth, the system immediately triggers the task migration mechanism, performing frame sequence alignment through the channel synchronization module to ensure consistent output timing when switching tasks between different nodes. The physical significance of this process is to adjust the target node's playback rate and buffer depth to maintain the display frame time difference within 30ms, thereby eliminating screen jumps and latency misalignments caused by cross-node switching. In summary, this implementation method achieves stable and adaptive operation of the display system under high load and complex interactive scenarios through real-time closed-loop control of "heat quantification, dynamic scheduling, predicted latency, and frame alignment migration". While maintaining high refresh rate output, the system effectively reduces peak CPU and I / O usage, ensuring screen continuity and smooth operation.

[0066] Example 2

[0067] Please see Figure 1 and Figure 2 Specifically: S1 includes S11;

[0068] S11. The library display system is set up with display nodes in a dual-caching layer structure, including large-screen interactive wall, touch guide table, document retrieval all-in-one machine, self-service borrowing and returning terminal and theme exhibition playback terminal; and the content loading, eviction and refresh thread management is implemented between the dual-caching layer structure through the cache controller, and a cache status interface is provided to the channel scheduling module.

[0069] The dual-caching layer structure includes an L1 cache and an L2 cache;

[0070] The L1 cache is used to store real-time display content and output it with a high refresh rate.

[0071] The L2 cache is used to store static and infrequently accessed content and output it with a low refresh cycle.

[0072] S1 also includes S12;

[0073] S12. Each display node is equipped with an autonomous agent module. The autonomous agent module collects the display node's running status parameters according to the sampling window and uploads them to the central scheduling server.

[0074] Node running status parameters include the CPU utilization Uc of the i-th node. i The I / O queue length Io of the i-th display node i and the average response time Rt of the i-th display node i ;

[0075] Where: the CPU utilization rate Uci of the i-th display node is obtained from the operating system performance counter;

[0076] The I / O queue length Io of the i-th node is displayed. i Obtained from the hardware driver layer interface;

[0077] The average response time Rt of the i-th display node i Real-time statistics are generated by the display management middleware;

[0078] The node popularity index H is calculated and displayed in the central scheduling server based on the node's operating parameters. The node popularity index H is calculated and output using the following algorithm formula: H i =(UC i +k1·Io i / Io max +k2·Rt i / Rt max H / 3; where H i This represents the node popularity index of the i-th displayed node. k1 and k2 represent the I / O weight calibration coefficient and response time weight calibration coefficient, respectively, and k1+k2=1. Their specific values ​​are set by the user. maxRt represents the historical maximum I / O queue length. max This represents the historical maximum average task response time;

[0079] To avoid dimensional differences between different indicators, this formula uses a proportional normalization method to normalize Io. i 、Rt i The values ​​are converted to a range of 0-1. The I / O and response time dimensions are balanced using coefficients k1 and k2, so that their impact on the overall heat index is comparable to that of the CPU utilization rate. The sum of the three indicators is divided by 3 to form a stable average comprehensive value, which is used for direct comparison between different nodes. The node heat index H output is a dimensionless result with a value range of 0-1.

[0080] S1 also includes S13;

[0081] S13. Based on the node popularity index H and the CPU utilization rate Uc of the i-th display node. i The execution scheduling strategy logic is implemented, and an execution priority switching mechanism is established based on the result of the scheduling strategy logic.

[0082] The scheduling strategy logic calculates the average CPU utilization of the currently displayed node through the autonomous agent module in a sliding window of 15 seconds, and determines the CPU utilization threshold Ucth by combining it with historical running data. The CPU utilization threshold Ucth is the larger of the 95th percentile value of the node's historical CPU utilization Uc and the fixed upper limit of 0.80.

[0083] When the CPU utilization Uc of the i-th display node is detected i ≥CPU utilization threshold Ucth, and the node popularity index H of the i-th displayed node. i If the duration of the state value above the historical median is not less than 8 seconds, the current display node is judged to be under abnormal pressure.

[0084] When the condition is met, a cache refresh status signal (Scache) is generated and reported to the central scheduling server. The cache refresh status signal (Scache) is a structured data packet, including the node identifier (IDnode), timestamp (Ts), and the CPU utilization (Uc) of the i-th display node. i CPU utilization threshold Ucth, node popularity index H of the i-th displayed node i The cache control instruction Op and the cache retention time Thold are used to trigger the priority switching mechanism.

[0085] Conversely, if the condition is not met, the cache controller will automatically generate a recovery instruction to restore the L2 cache refresh.

[0086] In this embodiment, in S11, the method uniformly constructs a dual-buffer layer structure of L1 buffer + L2 buffer for high-load display nodes such as the large-screen interactive wall and touch guide table. This is to physically separate "content that must be refreshed in real time" from "content that can be refreshed more slowly": once the node pressure increases, only the L2 buffer needs to be activated first to pause the refresh of static content, without interrupting the real-time display of the L1 buffer. In this way, even with 4K video playback and frequent touch operations during peak hours, the entire node will not be "crushed." The real effect is that the CPU and I / O are not overwhelmed by low-value refresh tasks. In S12, the CPU utilization Uc is uniformly collected through the autonomous proxy module. i I / O queue length (Io) i and average response time Rt i The node heat index H is calculated on the central scheduling server side. The essential purpose is to compress three different dimensions and units of "stress signals" into a dimensionless index of 0-1, avoiding misjudgments caused by looking only at CPU or only at response time. For example, high CPU but high I / O and Rt. i Under normal circumstances, nodes are not considered truly abnormal, thus achieving an "objective score" of the overall operational status of nodes in a physical sense. In S13, a 15-second sliding window and a CPU utilization threshold Ucth are introduced, requiring Uc... i The cache refresh status signal Scache is only generated when the node's heat index (Hi) remains consistently high and above the historical median for 8 seconds. The purpose is to filter out instantaneous spikes and avoid frequent strategy switching due to short-term access peaks of a few seconds. Actions such as pausing L2 cache refresh and adjusting channel priority are only triggered when a node is indeed in a "continuously high-pressure" state. The real physical meaning is to ensure that the system only tightens resources "before it really can't hold up," while maintaining sufficient throughput during normal fluctuations. This improves the overall system's operational stability, cache utilization efficiency, and the continuity of multimodal display.

[0087] Example 3

[0088] Please see Figure 1 Specifically: S2 includes S21;

[0089] S21. After triggering the priority switching mechanism, the cache refresh status signal Scache is transmitted to the autonomous proxy module, and the node identifier IDnode and the node popularity index H of the i-th displayed node are read from the structured data packet. i and cache control instructions (Op);

[0090] The channel scheduling module updates the task entry status according to the channel scheduling table of the corresponding node. When the buffer control instruction Op is set to PAUSE-L2, the refresh cycle parameter of the task entry associated with the L2 buffer is adjusted to 0.5 times its original value, and the signal channel proportion parameter is reduced by 20%.

[0091] In the updated channel scheduling table, each channel task entry includes three fields: task type, refresh cycle, and signal channel percentage.

[0092] The autonomous agent module is based on the updated channel scheduling table and uses an improved Sigmoid dynamic function to calculate the dynamic scheduling priority coefficient P of the node, thus smoothly controlling the priority switching process.

[0093] The dynamic scheduling priority coefficient P is calculated and output using the following algorithm formula: In the formula, P i Let represent the dynamic scheduling priority coefficient of the i-th display node, and let e represent the exponential function. This represents the sensitivity adjustment constant, with a value range of 0.2-1.0, which is set by the user and is used to smooth the conversion rate. Hth represents the node heat threshold.

[0094] This formula is essentially a typical application of the Sigmoid function / logistic function. Its original form comes from modeling the probability of binary events in mathematics and statistics. This function is often used in: logistic regression in statistics; activation functions in neural networks; and soft thresholding in cybernetics to "smooth" "hard thresholds".

[0095] Its mathematical characteristics are: the independent variable takes the value of a real number, the function value changes continuously in the interval (0,1), it approaches 0 when the independent variable is very small, and approaches 1 when the independent variable is very large, and the middle part is approximately S-shaped;

[0096] The problem to be solved in this application is: how to smoothly map the difference between the node comprehensive popularity index H and the node popularity threshold Hth to a "scheduling intensity coefficient" dynamic scheduling priority coefficient P, so as to control whether and to what extent the refresh frequency of low priority tasks is compressed.

[0097] Dimensional Consistency Analysis: Node Comprehensive Popularity Index H and Node Popularity Threshold Hth: These are calculated by normalizing CPU utilization Uc, I / O queue length Io, and response time Rt, and are defined as dimensionless (0-1).

[0098] Node popularity threshold Hth: A threshold within the same index system as the node comprehensive popularity index H, also dimensionless; therefore, the difference (Hth) i -Hth) is dimensionless;

[0099] The sensitivity coefficient is used to adjust the "slope" of the exponential function, and its dimension is dimensionless.

[0100] The independent variable of the exponential function e must be a dimensionless quantity. Since the independent variable is already dimensionless, the dynamic scheduling priority coefficient P of the output result is defined as a dimensionless value in the interval 0-1.

[0101] The node popularity threshold Hth is determined by taking the lower limit of a fixed safe value from the historical node comprehensive popularity index H sequence, with a value of 0.6, to prevent underestimation of the popularity threshold.

[0102] S2 also includes S22;

[0103] S22. Based on the output of the dynamic scheduling priority coefficient P, a preliminary comparative evaluation is performed to determine the load status of the displayed nodes. Based on the preliminary comparative evaluation results, a resource protection mode is activated, and corresponding control operations are executed. The specific evaluation content is as follows:

[0104] When the dynamic scheduling priority coefficient P of the i-th display node i When the value is ≥0.7, it indicates that the display node is under high load. The channel scheduling module enters resource protection mode and performs the following corresponding control operations: the refresh frame rate of the priority channel is reduced to 0.6 times the original value to reduce the proportion of image and interactive tasks in the total signal bandwidth; the cache refresh thread of the L2 buffer is suspended, and only the cache update of the L1 buffer is maintained to reduce disk and memory I / O pressure; the local cache controller is notified to calculate the node resource release rate Rrel of the i-th display node. i Its calculation formula is: Rrel i =ΔCPU i +ΔIO i , where ΔCPU i ΔIO represents the percentage decrease in average CPU utilization before and after scheduling of the i-th display node; i This represents the percentage decrease in the average I / O queue length of the i-th display node;

[0105] When the dynamic scheduling priority coefficient P of the i-th display node i When the value is less than 0.7, it indicates that the node is in a stable load state, and the channel scheduling module maintains the current channel priority table unchanged.

[0106] In this embodiment, step S2 of the method, through steps S21 and S22, truly translates "detecting abnormal node pressure" into "how to adjust the channel and reduce pressure." After triggering the priority switch, the cache refresh status signal Scache is used to accurately locate which node (node ​​identifier IDnode) is causing the problem, as well as the current node heat index Hi and cache control instruction Op. This is to ensure that the channel scheduling module only adjusts the truly overloaded display nodes and only the task entries related to the L2 cache. First, the refresh cycle of tasks associated with the L2 cache is halved, and the signal channel ratio is reduced by 20%. This is equivalent to actively "squeezing out" secondary slow content when bandwidth is insufficient, so that the critical real-time display is not dragged down. Subsequently, by improving the dynamic scheduling priority coefficient P calculated by the Sigmoid function, the "whether the heat exceeds the node heat threshold Hth" is transformed into a continuous intensity signal between 0 and 1. Only when the dynamic scheduling priority coefficient P ≥ 0.7 does it enter the resource protection mode, uniformly reducing the refresh frame rate of priority channels, pausing L2 buffer refresh, and calculating the node resource release rate Rreli to numerically confirm how much CPU and I / O this rate limiting actually releases. Without this hierarchical logic, if the CPU utilization Uci suddenly increases during peak hours, the system will either remain completely still or react too abruptly, easily leading to frequent screen flickering or all channels being degraded together. Through the coordination of the dynamic scheduling priority coefficient P and the node resource release rate Rreli, this implementation achieves: no erratic changes during slight fluctuations, tightening only when under sustained high pressure, and after tightening, it is quantifiable whether "the pressure has been reduced." The real physical effect is that, without sacrificing the core display smoothness, low-priority content is selectively sacrificed, significantly reducing the probability of single-node overload and improving the stability and controllability of the entire library multimodal display system in high-concurrency scenarios.

[0107] Example 4

[0108] Please see Figure 1 Specifically: S3 includes S31;

[0109] S31, when the node resource release rate Rrel of the i-th displayed node... i If the value is below 0.25 for three consecutive sampling periods, it indicates that local scheduling cannot solve the load problem. At this time, the startup delay prediction engine of the central scheduling server is activated to collect real-time running data from each display node.

[0110] Real-time operational data includes the average response time Rt of the i-th display node. i and the rate of change of response time dRt of the i-th display node i / dt;

[0111] The extreme value normalization method is used to normalize the real-time running data and the node resource release rate Rrel, eliminating the difference between the real-time running data and the node resource release rate Rrel of the i-th display node. i The influence of the unit dimensions of all parameters.

[0112] S3 also includes S32;

[0113] S32, Based on the normalized real-time running data and the node resource release rate Rrel of the i-th display node. i A feedback-based prediction delay model is used to calculate the node prediction delay Dpred and to quantitatively analyze the trend of node response time changes and resource release status.

[0114] The node prediction delay Dpred is calculated using the following feedback-based prediction delay model;

[0115] ; where, Dpred i Let d represent the node prediction delay of the i-th display node, d represent the integral function, and dt represent the time integral function. represents the time prediction gain, with a value range of 0.05-1, and u represents the resource feedback gain, with a value range of 0.05-1, used for stability control.

[0116] In this embodiment, method S3, through S31 and S32, forms a complete prediction closed loop from "detecting anomalies" to "early intervention". When the resource release rate Rrel of the i-th display node... i When the latency is below 0.25 for three consecutive sampling periods, the system determines that local scheduling has failed. This means that even if nodes pause L2 cache flushing and reduce the frame rate, they cannot effectively release resources, indicating that the overload problem has become persistent and cumulative. At this point, the central scheduling server activates the latency prediction engine to achieve centralized analysis across nodes. The latency prediction engine collects the average response time Rt of each node. i and its rate of change dRt i / dt, combined with resource release rate Rrel iExtreme value normalization is performed to map parameters from different dimensions to the same scale, avoiding weight imbalances caused by differences in units. For example, when the response time unit is milliseconds and the resource release rate is a proportion, unnormalized processing will cause latency characteristics to dominate the entire judgment, leading to misjudgment of migration timing. Subsequently, a feedback-based prediction latency model is used to calculate the node prediction latency Dpred. The cumulative change trend of response time is captured by the integral term, and the effect of the previous scheduling and resource release is corrected by the feedback gain term, realizing dynamic self-correction of latency trends. The physical significance of this design is that the system no longer reacts solely based on the current load, but combines time series changes to predict future latency risks, shifting from passive response to proactive defense. When the model detects that Dpred is about to approach the latency threshold, it can trigger task migration or resource reallocation in advance, avoiding frame accumulation or display stuttering at nodes. Through this mechanism, the system significantly improves the latency prediction accuracy and response speed in multi-node display environments, reduces erroneous and delayed migrations, and achieves real-time stable output of the multimodal display system in complex interactive scenarios.

[0117] Example 5

[0118] Please see Figure 1 Specifically: S4 includes S41;

[0119] S41, Rt based on the average response time of the i-th historical display node i Set a delay threshold Dth to 1.5 times the value of the i-th display node, and set the node prediction delay Dpred of the i-th display node to be obtained in real time. i A secondary comparison and evaluation is performed with the delay threshold Dth, and the corresponding task migration mechanism is triggered based on the results of the secondary comparison and evaluation; the specific evaluation content is as follows;

[0120] When the node prediction delay of the i-th displayed node is Depred i When the latency threshold Dth is greater than or equal to the current display node, it is determined that the current display node will enter a latency overload prediction state in the future. At this time, the central scheduling server starts the content mirror pool call and triggers the task migration mechanism.

[0121] When the node prediction delay of the i-th displayed node is Depred i When the delay threshold Dth is less than the current delay threshold, it is determined that the current display node is in a stable delay state, and the central scheduling server maintains the current task allocation unchanged.

[0122] S4 also includes S42;

[0123] S42. After the content mirror pool invocation mechanism is activated, the central low-degree server predicts the latency Dpred based on the i-th display node. i The latency threshold Dth and the node resource release rate Rrel of the i-th displayed node.i Calculate the migration priority index M of the i-th display node. i The specific calculation formula is: M i =Dpred / Dth×(1-Rrel); where Dpred / Dth reflects latency pressure, and (1-Rrel) smoothly adjusts its weight within the 0-1 range to avoid abrupt scheduling jumps (such as hard switching based solely on threshold judgment); (1-Rrel) represents the directional correction of resource occupancy, and Rrel represents the proportion of resources released by the system after the most recent scheduling. The larger Rrel is, the more resources the node has successfully released after scheduling, reducing load pressure. Therefore, its contribution to "migration urgency" should be inverse: more resources mean less need to migrate; thus, the inverse function form (1-R_rel) is introduced. When the resource release rate is high (e.g., 0.8), (1-Rrel) = 0.2, making the migration priority index M of the i-th node... i The value decreases; when the resource release rate is low (e.g., 0.1), (1-Rrel)=0.9, making the migration priority index M of the i-th display node smaller. i Get bigger;

[0124] The migration priority index M of all display nodes is obtained by traversing each display node, and the task migration mechanism is executed based on the migration priority index M of all display nodes. The task migration mechanism sorts all display nodes from high to low according to the migration priority index M, and then migrates the display tasks to the display node with the lowest load in order from high to low. Before the task migration, the channel synchronization control module of the central scheduling server reads the difference between the display buffer timestamp Tsrc of the migrating display node and the display buffer timestamp Tdst of the target display node. When the difference exceeds 30ms, the data frame alignment operation is performed.

[0125] The data frame alignment operation records the timestamp and frame number information of the current output frame in real time by setting a local time stamp module in each display node. When the central scheduling server detects a migration request, it will simultaneously obtain the latest frame timestamp of the migration node and the target display node. If the frame time difference between the two display nodes is greater than or equal to 30ms, the target display node enters the pre-synchronization mode.

[0126] The pre-synchronization mode adjusts the local playback buffer rate and frame buffer queue depth frame by frame to gradually catch up with the timing of the output frames of the target display node to match that of the migration node.

[0127] After each adjustment, the frame time difference is re-detected. When the time difference is less than 30ms for two consecutive detection cycles, the frame order is determined to be aligned.

[0128] This method dynamically adjusts the buffer rate and compares it with the time base marker to ensure that the display output remains continuous with millisecond-level precision during task migration, effectively avoiding common frame skipping and display interruption problems when switching across nodes;

[0129] Taking a common display system with a refresh rate of 60 frames per second (fps) as an example, the single frame period is about 16.67 milliseconds. It is generally believed that the difference between two frames (i.e., about 30-35 milliseconds) is the maximum synchronization error that can be perceived by the naked eye. Therefore, the threshold should be controlled within 30ms as the optimal frame period.

[0130] When the time difference between the output frames of the migrating node and the target node exceeds 30ms, and if the target node is "lagging behind", the output speed must be increased to catch up with the source node. The playback engine speed should be adjusted to 1.02 to 1.05 times the original speed, that is, the number of frames played per second should be slightly increased by 2% to 5% (for example, from 3 frames to 5 frames) until the time difference returns to within the threshold.

[0131] If the target node is "ahead", the output speed must be slowed down or the buffer temporarily expanded to wait for the source node to catch up; pause the rendering of new frames for 1 to 2 frame cycles (about 16–33ms), or temporarily reduce the playback speed to 0.95x to allow the source node to catch up with the timeline.

[0132] In this embodiment, method S4 achieves complete closed-loop control of "prediction, decision-making, migration, and synchronization" through S41 and S42. When the node prediction delay Dpred... i When the latency threshold Dth is reached or exceeded (based on the historical average response time Rt) iThe 1.5x threshold setting indicates that the node is highly likely to experience display latency or frame backlog in the future. This 1.5x threshold design takes into account the non-linear growth characteristics of display tasks: if set too low, frequent system migrations will cause scheduling jitter; if set too high, migration lag will cause frame interruptions. Therefore, this threshold strikes a balance between response stability and scheduling sensitivity. At this time, the central scheduling server calls the content mirror pool to immediately prepare a mirror copy of the corresponding display task, ensuring that the master data does not need to be reloaded during migration execution, physically reducing I / O operation time. Subsequently, the migration priority index M is used to sort the migration urgency of each node. The reason for introducing the (1-Rrel) term is that a high resource release rate Rrel indicates that the node has recovered well, and continuing migration at this time would waste resources; therefore, the inverse function can automatically reduce the migration weight of the node. When Rrel is low, it indicates insufficient resource release, and the migration priority should be increased. This design makes the scheduling logic no longer a simple "threshold trigger," but a comprehensive balance control based on latency pressure and resource recovery, ensuring that migration actions are both necessary and restrained. Before migration, the channel synchronization control module detects the timestamp difference (Tsrc-Tdst) between the migration node and the target node's display buffer. If it exceeds 30ms, the frame alignment process begins. The 30ms threshold is set because human vision is most sensitive to frame differences of 30-35ms; exceeding this threshold results in noticeable frame skipping. The system dynamically corrects the output timing by adjusting the playback rate (1.02-1.05 times or 0.95 times) and buffer depth, ensuring the frame time difference remains less than 30ms for two consecutive cycles, thus maintaining screen continuity during task migration. Through this implementation, the system achieves end-to-end control from "delay prediction to on-demand migration to dynamic frame alignment." Its purpose is to complete load transfer before nodes are fully overloaded and ensure seamless screen transitions during migration. Physically, this mechanism transforms task migration from a traditional "sudden fix" to a "gradual adjustment," preventing flickering, stuttering, or interruptions during node switching. The beneficial effects are a reduction in overall display system response latency of approximately 30% to 40%, improved image stability to millisecond-level precision, and virtually imperceptible user transitions between multiple terminals, significantly enhancing the smoothness and robustness of the multimodal display system.

[0133] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A multimodal interactive library integrated display method, characterized in that: Includes the following steps: S1. In the multiple display nodes of the library display system, the operating performance parameters of each display node are collected respectively, and the node autonomous agent module performs normalization calculation on the collected operating performance parameters, generates the node popularity index H, and executes the scheduling strategy logic. S2. Based on the scheduling strategy logic, an execution priority switching mechanism is implemented to dynamically schedule the priority coefficient P of the computing node and perform a preliminary comparison and evaluation. Based on the preliminary comparison and evaluation results, a node resource release rate Rrel is generated. Based on the node resource release rate Rrel, the node autonomous agent module triggers the display channel scheduling module to adjust the channel priority table. S3. Based on the output result of the generated node resource release rate Rrel, a prediction delay mechanism is triggered. The prediction delay mechanism is initiated by the central scheduling server of the library display system to start the delay prediction engine and calculate the node prediction delay Dpred. S4. Perform a secondary comparison and evaluation based on the output of the predicted delay Dpred, and trigger the task migration mechanism based on the results of the secondary comparison and evaluation.

2. The multimodal content interactive library integrated display method according to claim 1, characterized in that: S1 includes S11; S11. The library display system is set up with display nodes in a dual-caching layer structure, including large-screen interactive wall, touch guide table, document retrieval all-in-one machine, self-service borrowing and returning terminal and theme exhibition playback terminal; and the content loading, eviction and refresh thread management is implemented between the dual-caching layer structure through the cache controller, and a cache status interface is provided to the channel scheduling module. The dual-caching layer structure includes an L1 cache and an L2 cache.

3. The multimodal content interactive library integrated display method according to claim 1, characterized in that: S1 further includes S12; S12. Each display node is equipped with an autonomous agent module. The autonomous agent module collects the display node's operating status parameters according to the sampling window and uploads them to the central scheduling server. The node operating status parameters include the CPU utilization Uc of the i-th display node. i The I / O queue length Io of the i-th display node i and the average response time Rt of the i-th display node i ; The node popularity index H is calculated and displayed in the central scheduling server based on the node's operating parameters. The node popularity index H is calculated and output using the following algorithm formula: H i =(UC i +k1·Io i / Io max +k2·Rt i / Rt max H / 3; where H i This represents the node popularity index of the i-th displayed node. k1 and k2 represent the I / O weight calibration coefficient and response time weight calibration coefficient, respectively, and k1+k2=1. Their specific values ​​are set by the user. max Rt represents the historical maximum I / O queue length. max This represents the historical maximum average task response time.

4. The multimodal content interactive library integrated display method according to claim 3, characterized in that: S1 also includes S13; S13. Based on the node popularity index H and the CPU utilization rate Uc of the i-th display node. i The execution scheduling strategy logic is implemented, and an execution priority switching mechanism is established based on the result of the scheduling strategy logic. The scheduling strategy logic calculates the average CPU utilization of the currently displayed node through the autonomous agent module in a sliding window of 15 seconds, and determines the CPU utilization threshold Ucth by combining it with historical running data. The CPU utilization threshold Ucth is the larger of the 95th percentile value of the node's historical CPU utilization Uc and the fixed upper limit of 0.

80. When the CPU utilization Uc of the i-th display node is detected i ≥CPU utilization threshold Ucth, and the node popularity index H of the i-th displayed node. i If the duration of the state above the historical median value is not less than 8 seconds, the current display node is judged to be under abnormal pressure. When the determination is successful, a cache refresh status signal (Scache) is generated and reported to the central scheduling server. The cache refresh status signal (Scache) is a structured data packet, including the node identifier (IDnode), timestamp (Ts), and CPU utilization (Uc) of the i-th display node. i CPU utilization threshold Ucth, node popularity index H of the i-th displayed node i The cache control instruction Op and the cache hold time Thold are used to trigger the priority switching mechanism. Conversely, if the condition is not met, the cache controller will automatically generate a recovery instruction to restore the L2 cache refresh.

5. The multimodal content interactive library integrated display method according to claim 4, characterized in that: S2 includes S21; S21. After triggering the priority switching mechanism, the cache refresh status signal Scache is transmitted to the autonomous proxy module, and the node identifier IDnode and the node popularity index H of the i-th displayed node are read from the structured data packet. i and cache control instructions (Op); The channel scheduling module updates the task entry status according to the channel scheduling table of the corresponding node. When the buffer control instruction Op is set to PAUSE-L2, the refresh cycle parameter of the task entry associated with the L2 buffer is adjusted to 0.5 times its original value, and the signal channel proportion parameter is reduced by 20%. In the updated channel scheduling table, each channel task entry includes three fields: task type, refresh cycle, and signal channel percentage. The autonomous agent module is based on the updated channel scheduling table and uses an improved Sigmoid dynamic function to calculate the dynamic scheduling priority coefficient P of the node, thus smoothly controlling the priority switching process. The dynamic scheduling priority coefficient P is calculated and output using the following algorithm formula: In the formula, P i Let represent the dynamic scheduling priority coefficient of the i-th display node, and let e represent the exponential function. This represents the sensitivity constant, with a value range of 0.2-1.0, which is set by the user. Hth represents the node heat threshold.

6. The multimodal content interactive library integrated display method according to claim 5, characterized in that: S2 further includes S22; S22. Based on the output of the dynamic scheduling priority coefficient P, a preliminary comparative evaluation is performed to determine the load status of the displayed nodes. Based on the preliminary comparative evaluation results, a resource protection mode is activated, and corresponding control operations are executed. The specific evaluation content is as follows: When the dynamic scheduling priority coefficient P of the i-th display node i When the value is ≥0.7, it indicates that the display node is under high load. The channel scheduling module enters resource protection mode and performs the following corresponding control operations: the refresh frame rate of the priority channel is reduced to 0.6 times the original value; the cache refresh thread of the L2 cache is suspended, and only the cache update of the L1 cache is maintained to reduce disk and memory I / O pressure; the local cache controller is notified to calculate the node resource release rate Rrel of the i-th display node. i Its calculation formula is: Rrel i =ΔCPU i +ΔIO i , where ΔCPU i ΔIO represents the percentage decrease in average CPU utilization before and after scheduling of the i-th display node; i This represents the percentage decrease in the average I / O queue length of the i-th display node; When the dynamic scheduling priority coefficient P of the i-th display node i When the value is less than 0.7, it indicates that the node is in a stable load state, and the channel scheduling module maintains the current channel priority table unchanged.

7. The multimodal content interactive library integrated display method according to claim 6, characterized in that: S3 includes S31; S31, when the node resource release rate Rrel of the i-th displayed node... i If the value is below 0.25 for three consecutive sampling periods, it indicates that local scheduling cannot solve the load problem. At this time, the startup delay prediction engine of the central scheduling server is activated to collect real-time running data from each display node. The real-time operational data includes the average response time Rt of the i-th display node. i and the rate of change of response time dRt of the i-th display node i / dt; The extreme value normalization method is used to normalize the real-time running data and the node resource release rate Rrel, eliminating the difference between the real-time running data and the node resource release rate Rrel of the i-th display node. i The influence of the unit dimensions of all parameters.

8. The multimodal content interactive library integrated display method according to claim 7, characterized in that: S3 further includes S32; S32, Based on the normalized real-time running data and the node resource release rate Rrel of the i-th display node. i A feedback-based prediction delay model is used to calculate the node prediction delay Dpred and to quantitatively analyze the trend of node response time changes and resource release status. The node prediction delay Dpred is calculated and output using the following feedback prediction delay model; ; where, Dpred i Let d represent the node prediction delay of the i-th displayed node, d represent the integral function, and dt represent the time integral function. represents the time prediction gain, with a value range of 0.05-1, and u represents the resource feedback gain, with a value range of 0.05-1.

9. A multimodal content interactive library integrated display method according to claim 8, characterized in that: S4 includes S41; S41, Rt based on the average response time of the i-th historical display node i Set a delay threshold Dth to 1.5 times the value of the i-th display node, and set the node prediction delay Dpred of the i-th display node to be obtained in real time. i A secondary comparison and evaluation is performed with the delay threshold Dth, and the corresponding task migration mechanism is triggered based on the results of the secondary comparison and evaluation; the specific evaluation content is as follows; When the node prediction delay of the i-th displayed node is Depred i When the latency threshold Dth is greater than or equal to the current display node, it is determined that the current display node will enter a latency overload prediction state in the future. At this time, the central scheduling server starts the content mirror pool call and triggers the task migration mechanism. When the node prediction delay of the i-th displayed node is Depred i When the delay threshold Dth is less than the current delay threshold, it is determined that the current display node is in a stable delay state, and the central scheduling server maintains the current task allocation unchanged.

10. A multimodal content interactive library integrated display method according to claim 9, characterized in that: S4 also includes S42; S42. After the content mirror pool invocation mechanism is activated, the central low-degree server predicts the latency Dpred based on the i-th display node. i The latency threshold Dth and the node resource release rate Rrel of the i-th displayed node. i Calculate the migration priority index M of the i-th display node. i The specific calculation formula is: M i =Dpred / Dth×(1-Rrel) The migration priority index M of all display nodes is obtained by traversing each display node, and the task migration mechanism is executed based on the migration priority index M of all display nodes. The task migration mechanism sorts all display nodes from high to low according to the migration priority index M, and then migrates the display tasks to the display node with the lowest load in order from high to low. Before the task migration, the channel synchronization control module of the central scheduling server reads the difference between the display buffer timestamp Tsrc of the migrating display node and the display buffer timestamp Tdst of the target display node. When the difference exceeds 30ms, the data frame alignment operation is performed. The data frame alignment operation records the timestamp and frame number information of the current output frame in real time by setting a local time stamp module in each display node. When the central scheduling server detects a migration request, it will simultaneously obtain the latest frame timestamps of the migration node and the target display node; if the frame time difference between the two display nodes is greater than or equal to 30ms, the target display node will enter pre-synchronization mode. The pre-synchronization mode adjusts the local playback buffer rate and frame buffer queue depth frame by frame to gradually catch up with the timing of the output frames of the target display node to match that of the migration node. After each adjustment, the frame time difference is re-detected. When the time difference is less than 30ms for two consecutive detection cycles, the frame order is determined to be aligned.

Images