A concurrent and bandwidth dual threshold triggered video review pre-emption scheduling method and system

Abstract

The application discloses a concurrent and bandwidth double-threshold triggered video browsing preemption scheduling method and system, and belongs to the network scheduling field. The application is applied to multi-level networking platform cascade browsing. Firstly, service transmission data, priority information of each video browsing task and resource utilization data of each networking platform are acquired, and the platform resource data is constructed into a directed graph for processing based on a network model trained based on a graph attention mechanism, so that resource pressure scores of platform nodes are obtained. Then, in combination with the pressure scores and service quality information of the platform nodes of the video browsing task, the task path risk values of the video browsing tasks are determined. When a high-priority target video browsing task triggers a concurrency or bandwidth threshold, in combination with the priority, cross-level depth and task path risk value of a low-priority candidate video browsing task, a video browsing task to be scheduled is determined through Bayesian inference, and the resource of the video browsing task to be scheduled is released to execute the target video browsing task, so that the scheduling efficiency and user experience can be improved.

Description

Technical Field

[0001] This application belongs to the field of network scheduling, and in particular relates to a video access preemption scheduling method and system triggered by both concurrency and bandwidth thresholds. Background Technology

[0002] With the increasing prevalence and deeper application of public safety video surveillance networks, especially in complex scenarios involving multi-level cascading platforms, the demand for concurrent access to massive amounts of video resources has surged. To ensure the priority execution of critical access tasks during resource-constrained periods, a video access preemption scheduling method that dynamically adjusts resource allocation to guarantee the service quality for high-priority users has shown broad application prospects in areas such as emergency command and security for major events.

[0003] In existing technologies, some video access preemption scheduling methods employ a preemption scheduling approach based on static priorities and fixed resource thresholds. For example, when the number of concurrent channels or bandwidth reaches its limit, high-priority users are allowed to preempt resources occupied by low-priority users. These methods typically operate on a single platform or in a simple network environment, executing preemption decisions based on preset user levels, and to some extent solving the problem of high-priority tasks being blocked.

[0004] However, in complex scenarios involving cascading video retrieval across multiple networked platforms, the aforementioned existing technologies rely solely on the static priority of the retrieval task and the resource status of the local platform for decision-making. This fails to comprehensively assess the global and differentiated impact of the transmission path of the video retrieval task across multiple platforms on the entire cascading network. Consequently, the scheduling efficiency of high-priority retrieval tasks is low, and network congestion on multi-level networked platforms is difficult to fundamentally alleviate, affecting user experience. Therefore, the existing video retrieval preemption scheduling methods suffer from poor scheduling efficiency and user experience. Summary of the Invention

[0005] This application provides a video access preemption scheduling method, system, device, and computer storage medium triggered by both concurrency and bandwidth thresholds, which can improve scheduling efficiency and user experience.

[0006] Firstly, this application provides a video access preemption scheduling method triggered by both concurrency and bandwidth thresholds, applicable to cascaded access on multi-level network platforms. The method includes:

[0007] The service transmission data and priority information of each video retrieval task being executed in the multi-level network platform are obtained, as well as the resource utilization data of each network platform. The service transmission data includes cross-level depth information and the quality of service information of each video retrieval task on each path of the network platform.

[0008] Using each network platform as a node, the cascading relationship between network platforms as edges, and the resource utilization data of the network platforms as node attribute features, a directed graph of network platforms is constructed. The directed graph of network platforms is then input into a trained network model to obtain the resource pressure score of each node. The network model is trained based on the historical directed graph of network platforms through a graph attention mechanism.

[0009] Based on the service quality information of each video retrieval task on each network platform and the resource pressure score of the corresponding node, the task path risk value of each video retrieval task is determined.

[0010] Obtain the target priority information of the target video retrieval task. If the target priority information is high priority and the number of concurrent paths of all currently executing video retrieval tasks is greater than the concurrent path threshold or the total bandwidth requirement of all currently executing video retrieval tasks and the target video retrieval task is greater than the bandwidth requirement threshold, then the video retrieval task with low priority information is identified as a candidate video retrieval task.

[0011] Based on the priority information, cross-level depth information, and task path risk value of the candidate video retrieval tasks, the video retrieval tasks to be scheduled are determined through Bayesian inference, and the resources occupied by the video retrieval tasks to be scheduled are released to be used for the execution of the target video retrieval task.

[0012] In one feasible implementation, before determining the video retrieval task to be scheduled based on the priority information, cross-level depth information, and task path risk value of the candidate video retrieval tasks through Bayesian inference, the method further includes:

[0013] Obtain the historical behavior logs of users who are tasked with accessing candidate videos;

[0014] Adjust the priority sub-levels in the priority information of candidate video retrieval tasks based on user historical behavior logs.

[0015] In one feasible implementation, the method further includes:

[0016] Obtain the task progress information for the candidate video retrieval task;

[0017] Based on the priority information, cross-level depth information, and task path risk value of the candidate video retrieval tasks, the video retrieval tasks to be scheduled are determined through Bayesian inference, including:

[0018] Based on the priority information, cross-level depth information, task path risk value, and task progress information of the candidate video retrieval tasks, the video retrieval tasks to be scheduled are determined through Bayesian inference.

[0019] In one feasible implementation, based on the priority information, cross-level depth information, task path risk value, and task progress information of candidate video retrieval tasks, the video retrieval tasks to be scheduled are determined through Bayesian inference, including:

[0020] Based on the priority sub-levels in the priority information, the prior probability of task scheduling for candidate video retrieval tasks is determined from the preset mapping relationship between levels and probabilities. The lower the priority sub-level, the higher the corresponding prior probability of task scheduling.

[0021] The cross-level depth information, task path risk value, and task progress information of the candidate video retrieval task are quantified and then multiplied to obtain the task state conditional probability. The deeper the cross-level depth information, the greater the probability value of the task state conditional probability. The greater the risk value of the task path risk value, the greater the probability value of the task state conditional probability. The lower the progress information of the task progress, the greater the probability value of the task state conditional probability.

[0022] By calculating the product of the prior probability of task scheduling and the conditional probability of task state, the task scheduling score of the candidate video retrieval task is obtained, and the candidate video retrieval task with the highest task scheduling score is determined as the video retrieval task to be scheduled.

[0023] In one feasible implementation, based on the service quality information of each video retrieval task on each network platform along each path and the resource pressure score of the corresponding node, the task path risk value of each video retrieval task is determined, including:

[0024] The service quality information of the video retrieval task on each network platform is converted into a service quality score for the video retrieval task on each network platform.

[0025] The service quality score and resource pressure score corresponding to each network platform of the video retrieval task are weighted and summed to obtain the initial path risk value of each video retrieval task.

[0026] Based on the distribution dispersion of resource pressure scores across all networked platforms along the video retrieval task path, the initial path risk value is adjusted to obtain the task path risk value for each video retrieval task.

[0027] In one feasible implementation, the number of concurrent paths for all video retrieval tasks being executed includes the total number of concurrent channels for all video retrieval tasks being executed in the multi-level network platform.

[0028] In one feasible implementation, the total bandwidth requirement of all ongoing video retrieval tasks and the target video retrieval task includes the sum of the total bandwidth utilization of all ongoing video retrieval tasks and the estimated bandwidth requirement of the target video retrieval task.

[0029] Secondly, this application provides a video access preemption scheduling system triggered by both concurrency and bandwidth thresholds, applicable to cascaded access on multi-level networked platforms. The system includes:

[0030] The acquisition module is used to acquire service transmission data and priority information for each video retrieval task being executed in the multi-level network platform, as well as resource utilization data for each network platform. The service transmission data includes cross-level depth information and service quality information for each video retrieval task on each path of the network platform.

[0031] The module is used to construct a directed graph of network platforms, with each network platform as a node, the cascading relationship between network platforms as edges, and the resource utilization data of the network platforms as node attribute features. The directed graph of network platforms is then input into the trained network model to obtain the resource pressure score of each node. The network model is trained based on the historical directed graph of network platforms through a graph attention mechanism.

[0032] The determination module is used to determine the task path risk value of each video retrieval task based on the service quality information of each network platform and the resource pressure score of the corresponding node for each video retrieval task.

[0033] The determination module is also used to obtain the target priority information of the target video retrieval task. When the target priority information is high priority and the number of concurrent paths of all currently executing video retrieval tasks is greater than the number of concurrent paths threshold or the total bandwidth requirement of all currently executing video retrieval tasks and the target video retrieval task is greater than the bandwidth requirement threshold, the video retrieval task with low priority information is determined as a candidate video retrieval task.

[0034] The scheduling module is used to determine the video retrieval tasks to be scheduled based on the priority information, cross-level depth information and task path risk value of the candidate video retrieval tasks through Bayesian inference, and release the resources occupied by the video retrieval tasks to be scheduled so that the target video retrieval task can be executed.

[0035] Thirdly, this application provides an electronic device, the device including: a processor, and a memory storing computer program instructions; the processor reads and executes the computer program instructions to implement the video access preemption scheduling method with concurrent and bandwidth dual threshold triggering as described in any embodiment of the first aspect.

[0036] Fourthly, this application provides a computer-readable storage medium storing computer program instructions, which, when executed by a processor, implement the video access preemption scheduling method with concurrent and bandwidth dual threshold triggering as described in any embodiment of the first aspect.

[0037] This application presents a video access preemption scheduling method, system, device, and computer storage medium triggered by both concurrency and bandwidth thresholds. Applied to multi-level networked platform cascaded access scenarios, it first constructs a directed graph of the network platforms and calculates the resource pressure score of each network platform node using a graph attention model. Then, combining the actual service quality information of each video access task at each level of platform with the resource pressure score, a task path risk value comprehensively reflects its global impact on the entire cascaded network is generated. Finally, during preemption decisions, not only the static priority of the task but also its cross-level depth and task path risk value are considered. Bayesian inference is used to select the low-priority task with the greatest impact on network bottlenecks and the lowest efficiency for release. Therefore, this application can achieve accurate identification of preemption targets and optimized reallocation of network resources, thereby improving the scheduling efficiency of high-priority access tasks and effectively alleviating network congestion in complex multi-level cascaded network environments, thus improving user experience.

[0038] Furthermore, by introducing task progress information for candidate video retrieval tasks, in multi-level networked platform cascading retrieval scenarios, the global impact on the entire cascading network is assessed not only based on the task's static priority, cross-level depth, and task path risk value, but also by incorporating task completion progress as a new key dimension. When making preemption decisions, integrating task progress information into the judgment criteria through Bayesian inference tends to retain low-priority tasks that are about to complete, while prioritizing tasks with lower progress. This avoids resource waste and impact on user experience caused by preempting tasks that are about to complete. Therefore, this application can further optimize user service experience and improve scheduling rationality in resource-constrained scenarios while ensuring high-priority tasks are handled effectively. Attached Figure Description

[0039] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a flowchart illustrating a video access preemption scheduling method triggered by both concurrency and bandwidth thresholds, provided in one embodiment of this application.

[0041] Figure 2 This is a flowchart illustrating a method for determining a video retrieval task to be scheduled using Bayesian inference, provided in one embodiment of this application.

[0042] Figure 3 This is a flowchart illustrating a method for determining the risk value of a video retrieval task according to an embodiment of this application;

[0043] Figure 4 This is a schematic diagram of the structure of a video access preemption scheduling system triggered by concurrency and bandwidth dual thresholds provided in one embodiment of this application;

[0044] Figure 5 This is a schematic diagram of the hardware structure of an electronic device provided in one embodiment of this application. Detailed Implementation

[0045] The features and exemplary embodiments of various aspects of this application will be described in detail below. To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain this application and not to limit it. For those skilled in the art, this application can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this application by illustrating examples.

[0046] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes said element.

[0047] In complex scenarios involving cascading video retrieval across multiple networked platforms, existing technologies rely solely on the static priority of retrieval tasks and the resource status of the local platform for decision-making. This fails to comprehensively assess the global and differentiated impact of the transmission path of video retrieval tasks across multiple platforms on the entire cascading network. Consequently, high-priority retrieval task scheduling is inefficient, and network congestion on multi-level networked platforms remains difficult to alleviate, impacting user experience. Therefore, existing video retrieval preemption scheduling methods suffer from poor scheduling efficiency and user experience.

[0048] To address the problems of existing technologies, embodiments of this application provide a video access preemption scheduling method, system, device, and computer storage medium triggered by both concurrency and bandwidth thresholds. The video access preemption scheduling method triggered by both concurrency and bandwidth thresholds provided in this application embodiment will be described below first.

[0049] Figure 1This diagram illustrates a flowchart of a video access preemption scheduling method based on concurrency and bandwidth dual-threshold triggering, according to an embodiment of this application. This method is applied to cascaded access on multi-level network platforms, such as... Figure 1 As shown, the method includes steps S110 to S160.

[0050] The application scenario of multi-level networked platform cascading access refers to the request and transmission of video resources that often needs to traverse multiple independent video management platforms interconnected through cascading protocols. In this scenario, the multi-level networked platform refers to multiple video surveillance management units with hierarchical or parallel relationships, which together form a large-scale distributed monitoring network, such as a cascading system composed of provincial, municipal, and county-level video platforms.

[0051] S110: Obtain service transmission data and priority information for each video retrieval task being executed in the multi-level network platform, as well as resource utilization data for each network platform. The service transmission data includes cross-level depth information and service quality information for each video retrieval task on each path of the network platform.

[0052] A video retrieval task refers to a user-initiated operation of real-time previewing or historical video playback. Service transmission data refers to a dataset describing the transmission characteristics of a video retrieval task in a cascaded network. Among them, cross-level depth information is a numerical value that quantifies the number of network platform hops a task traverses from its initiator to its target device; Quality of Service information is a set of indicators reflecting the health of network transmission of the video stream on each network platform along the way, such as network transmission latency that causes image delays and packet loss rate that causes image stuttering.

[0053] Priority information is used to distinguish the importance level of different video retrieval tasks. Priorities are divided into two main categories: high priority and low priority. The low priority category can be further divided into several sub-levels. For example, emergency command tasks belong to the high priority category, while case investigation and routine inspection tasks belong to the low priority category, but the sub-level of case investigation is higher than that of routine inspection. Resource utilization data refers to a set of parameters describing the hardware and network load status of each networked platform. This can include the platform's CPU utilization, memory usage, and real-time bandwidth throughput of the network egress.

[0054] In the signaling scheduling service of a multi-level network platform, all ongoing video retrieval tasks are monitored. First, the initial signaling message of a video retrieval task is parsed to extract identifiers containing user roles and service types. Based on a pre-defined priority mapping table, the priority category and specific sub-level of the task are queried and determined to obtain priority information. Second, a globally unique session identifier is generated for each cross-platform video retrieval task, and a distributed tracing process is initiated. This process embeds tracing information into custom fields of the signaling message. As the signaling is forwarded in the cascaded network, cross-level depth information is incrementally recorded at each network platform node it flows through. Simultaneously, network probe technology is used to measure the network transmission latency and packet loss rate of the task's video stream in the current platform's transmission segment in real time, thereby obtaining the service quality information of that node. Finally, priority information associated with the same session identifier, progressively accumulated cross-level depth information, and service quality information collected at each node are correlated in real time with resource utilization data such as memory usage and real-time bandwidth throughput periodically reported by each network platform. Data collection for all video retrieval tasks being executed in the multi-level network platform is completed in the above manner.

[0055] For example, monitoring and data collection are performed on all ongoing video retrieval tasks within a cascaded platform comprised of provincial, municipal, and county-level video platforms. For instance, multiple video retrieval tasks may exist simultaneously, including several high-priority emergency dispatches and several low-priority routine inspections. Taking a video retrieval task for emergency case tracking as an example, this task is initiated by the provincial emergency command center and needs to cross municipal and county-level platforms. When this task is established, the signaling dispatch service of the provincial platform parses its signaling and, based on a pre-set priority mapping table, determines its priority as high-priority. Subsequently, a unique session identifier is assigned to the task, and a distributed tracing process is started, at which point its cross-level depth information is initially 0.

[0056] When the request for this task flows through the municipal platform, its cross-level depth information is updated to 1. The municipal platform measures and records the link service quality information from the provincial to the municipal platform, specifically network latency of 30 milliseconds and packet loss rate of 0.01%. Simultaneously, the municipal platform also reports its own resource utilization data, such as memory utilization of 60% and real-time bandwidth throughput of 80% at the network egress point. As the request continues to the district / county platform, the task's cross-level depth information is updated to 2. The district / county platform records the link service quality information from the municipal to the district / county platform and its resource utilization data, specifically network latency of 40 milliseconds and packet loss rate of 0.1%, and reports its memory utilization of 90% and real-time bandwidth throughput of 85%. Finally, by performing the above data collection for each running video access task, the priority information, cross-level depth information, service quality information, and resource utilization data of all running video access tasks in the cascading platform are integrated in real time using their respective unique session identifiers to form a complete dataset containing the current status of all video access tasks in the cascading platform.

[0057] S120: Using each network platform as a node, the cascading relationship between network platforms as edges, and the resource utilization data of the network platforms as node attribute features, a directed graph of network platforms is constructed. The directed graph of network platforms is then input into the trained network model to obtain the resource pressure score of each node. The network model is trained based on the historical directed graph of network platforms through a graph attention mechanism.

[0058] A directed graph of network platforms is a data model used to describe the topology and state of multi-level network platforms. Each network platform is abstracted as a node in the graph, the directional cascading relationships between platforms are abstracted as edges, and the real-time resource utilization data of each platform serves as the attribute feature of the corresponding node. Resource pressure score is a quantitative value used to characterize the overall pressure level of a network platform node in the entire cascading network environment. A historical directed graph of network platforms refers to a set of directed graphs of network platforms recording resource utilization data of each platform at different points in the past, which is used as training data to train the network model. Graph attention mechanism is a processing method that assigns different importance weights to the neighbors of a node when processing graph data, allowing the model to focus on the neighbor node information that has the greatest impact on the current node. A trained network model is a graph neural network model based on the graph attention mechanism, learned from a large number of historical network platforms, capable of receiving the directed graph of network platforms and outputting the resource pressure score of each node.

[0059] First, based on pre-configured or automatically discovered inter-platform topology relationships via cascading protocols, a directed graph data structure is constructed for all networked platforms and their connections. Then, real-time resource utilization data collected from each networked platform is used as corresponding node attribute features to populate the directed graph. Next, this directed graph containing real-time states is input into a trained network model for inference. This network model consists of multiple stacked graph attention layers. In the first layer, the model receives the original node attribute features, aggregates neighbor information through a graph attention mechanism, and generates a preliminary embedding representation of the node. In each subsequent layer, the model uses the embedding representation output from the previous layer as input, repeatedly performing attention-based information aggregation, thereby capturing the complex dependencies of nodes more deeply layer by layer. Finally, the output of the last graph attention layer is transformed by a linear transformation layer to map the resource pressure score of the current node.

[0060] For example, firstly, a directed graph of networked platforms is constructed, containing three nodes: a provincial platform, a municipal platform, and a county platform, and directed edges are established from the province to the city and from the city to the county. Then, the latest resource utilization data of each platform is acquired, such as the municipal platform's memory utilization rate of 60% and bandwidth throughput of 80%, and the county platform's memory utilization rate of 90% and bandwidth throughput of 85%, and these are loaded into the graph as corresponding node attribute features. The directed graph of networked platforms is input into a trained network model. When calculating the resource pressure score of the municipal platform, its internal graph attention layer will focus on the transmission effect from the high load state of the downstream county platform; similarly, when calculating the resource pressure score of the provincial platform, it will comprehensively evaluate the overall pressure situation of the municipal and county platforms. After multi-layer processing by the model, a quantified resource pressure score is finally generated for the provincial, municipal, and county platforms, for example, 0.3, 0.7, and 0.85 respectively.

[0061] In another implementation of this application, before step S120: inputting the directed graph of the networking platform into the trained network model to obtain the resource pressure score of each node, the method further includes the process of training the network model, specifically including:

[0062] First, a historical directed graph sample set of network platforms is obtained. This sample set includes multiple historical training samples. Each historical training sample includes a directed graph of network platforms at a historical moment and a resource pressure score interval representing the actual operating status of each network platform at that historical moment. The actual resource bottleneck score interval is determined based on whether a significant decline in service quality or a resource alarm event occurred at that time. Then, for each historical training sample, the following steps are performed: The directed graph of network platforms in the historical training sample is input into an initialized graph neural network model. The graph neural network model generates a predicted resource pressure score for each node in the graph. Based on the predicted resource pressure score generated by the graph neural network model and the corresponding resource pressure score interval in the historical training sample, a loss function value for the resource pressure score interval is determined. The loss function value is used to measure the deviation between the predicted result and the actual state. If the loss function value does not meet the preset training stopping condition, the internal parameters of the resource pressure score interval, such as the weight parameters in the graph attention layer, are adjusted using a backpropagation algorithm to obtain an updated resource pressure score interval. The process then returns to the previous step of inputting the directed graph of network platforms into the resource pressure score interval to obtain the predicted resource pressure score, until the loss function value meets the training stopping condition. Finally, when the training stopping condition is met, the trained network model is obtained.

[0063] S130: Determine the task path risk value for each video retrieval task based on the service quality information of each network platform along each path and the resource pressure score of the corresponding node.

[0064] The task path risk value is a quantitative indicator used to characterize the health and resource utilization efficiency of the entire path traversed by a video retrieval task from its initiator to the target device. A higher task path risk value indicates greater pressure on the platform nodes the task passes through and poorer transmission quality, making the task an inefficient and high-risk resource consumer.

[0065] First, the service quality information (SQI) of each network platform along the path of the video retrieval task, such as network latency and packet loss rate, is converted into a standardized SQI score. Second, the SQI score of each network platform along the path is weighted and summed with the resource pressure score of the corresponding node to obtain an initial path risk value. Then, the dispersion of the resource pressure scores of all network platforms along the path is calculated; this dispersion reflects the balance of path pressure. Finally, the initial path risk value is adjusted using this dispersion. For example, when the path pressure distribution is extremely uneven, i.e., there are one or two bottleneck nodes with extremely high pressure, the final path risk value is increased accordingly. In this way, a path risk value that comprehensively reflects the path status is determined for each ongoing video retrieval task.

[0066] S140: Obtain the target priority information of the target video retrieval task. If the target priority information is high priority and the number of concurrent paths of all currently executing video retrieval tasks is greater than the concurrent path threshold or the total bandwidth requirement of all currently executing video retrieval tasks and the target video retrieval task is greater than the bandwidth requirement threshold, then the video retrieval task with low priority information is identified as a candidate video retrieval task.

[0067] A target video access task refers to a newly received video access task awaiting processing and scheduling. The concurrent channel threshold is a system-preset critical value used to identify the maximum total number of concurrent video channels that the entire cascaded network can support. The bandwidth requirement threshold is a system-preset critical value used to identify the maximum load percentage that the total bandwidth of the entire cascaded network's egress can support; for example, it could be 90% of the total bandwidth. Candidate video access tasks refer to a set of low-priority tasks selected from all currently executing video access tasks based on their priority information when preemption conditions are triggered.

[0068] Upon receiving a new target video retrieval task, its priority information is first parsed. If the target video retrieval task does not belong to the high-priority category, it is processed according to the normal procedure, which will not be explained in detail here. If it belongs to the high-priority category, a dual-threshold judgment mechanism is immediately initiated. On the one hand, the concurrent number of all currently executing video retrieval tasks is obtained and compared with a preset concurrent number threshold. On the other hand, the total bandwidth utilization of all currently executing video retrieval tasks is calculated and added to the estimated bandwidth requirement obtained after bitstream analysis of the newly accessed target video retrieval task. The sum of the two is compared with a preset bandwidth requirement threshold. The preemption condition is triggered if and only if the concurrent number is greater than the concurrent number threshold, or the total bandwidth requirement is greater than the bandwidth requirement threshold. Once the condition is triggered, the dataset of all currently executing video retrieval tasks obtained in step S110 is traversed, and all video retrieval tasks with low priority information are filtered out to form a candidate video retrieval task list. This list is then passed to subsequent steps for the final preemption decision.

[0069] For example, assume a preset concurrent access threshold of 1000 channels and a bandwidth requirement threshold of 90%. At this time, a high-priority target video retrieval task initiated by the provincial command center is received. Through continuous monitoring of the entire multi-level network platform, it is detected that the currently executing video retrieval task has reached 990 concurrent channels, with a total bandwidth utilization of 88%, while the estimated bandwidth requirement of the newly accessed target video retrieval task is 3%.

[0070] A dual threshold judgment is performed: the concurrent channel count of 990 channels does not exceed the 1000-channel threshold; however, the total bandwidth demand will reach 91%, exceeding the 90% bandwidth demand threshold. Therefore, the preemption condition is triggered. Then, all running video access tasks in the multi-level network platform are traversed, and all running video access tasks with low priority information, such as all daily inspection tasks initiated by ordinary users and ordinary case review tasks initiated by specific business personnel, are filtered out to form a candidate video access task list containing multiple candidates.

[0071] S150: Based on the priority information, cross-level depth information, and task path risk value of the candidate video retrieval tasks, determine the video retrieval tasks to be scheduled through Bayesian inference, and release the resources occupied by the video retrieval tasks to be scheduled for execution of the target video retrieval task.

[0072] Bayesian inference is a probabilistic and statistical decision-making method that combines prior knowledge and new observational evidence to infer the probability of a hypothesis being true. In this method, this inference process is used to calculate the probability that each candidate video retrieval task is the current optimal preemption target. The video retrieval task to be scheduled refers to the specific video retrieval task identified from the candidate video retrieval task list after Bayesian inference that will be forcibly interrupted by the system and have its occupied resources released.

[0073] Bayesian inference is performed on each candidate video retrieval task in the candidate video retrieval task list. First, the priority information of each candidate video retrieval task—that is, the low-priority sub-levels of low-priority candidate video retrieval tasks—is transformed into a prior probability for task scheduling. This prior probability reflects the initial probability that the candidate video retrieval task will be preempted. Second, the cross-level depth information and task path risk value of the candidate video retrieval task are used together as observational evidence, and a task state conditional probability is calculated based on this evidence. This conditional probability measures the degree to which the current running state of the task has a negative impact on the entire cascaded network. Next, a final task scheduling score is calculated for each candidate video retrieval task by multiplying the prior probability for task scheduling with the task state conditional probability. After scoring all candidate video retrieval tasks, the task with the highest task scheduling score is selected as the video retrieval task to be scheduled. Finally, a session release command is sent to the platform and client involved in the video retrieval task to reclaim the concurrent channels and network bandwidth resources it occupies, and these reclaimed resources are allocated to the waiting target video retrieval task.

[0074] Once a video retrieval task to be scheduled is identified, the signaling scheduling service immediately initiates a resource reclamation and reallocation process. First, the signaling scheduling service obtains the user's identity attributes and client identifier from the status information of the task. Then, it assembles a notification message containing the preemption result, clearly indicating which high-priority target video retrieval task required resources, and attaching relevant information about the target task. Next, based on the client identifier of the task, the signaling scheduling service locates the corresponding network connection socket and actively pushes this notification message to its client. Simultaneously, the signaling scheduling service sends a session release command to the media forwarding platform through which the video retrieval task is being routed. This command includes the identifier of the video forwarding stream to be terminated. Upon receiving the command, the media forwarding platform immediately interrupts the corresponding media stream transmission and cleans up the relevant session resources, thus completing the reclamation of concurrent channels and network bandwidth. After the resources are successfully released, the signaling scheduling service seamlessly allocates these newly reclaimed resources to high-priority target video retrieval tasks in the waiting queue, establishes new video sessions and media streaming channels for them, and ensures that critical tasks can be executed immediately.

[0075] This embodiment first constructs a directed graph of the network platform and uses a graph attention model to calculate the resource pressure score of each network platform node. Then, it combines the actual service quality information of each video retrieval task at each level of the platform with the resource pressure score to generate a task path risk value that comprehensively reflects its global impact on the entire cascaded network. Finally, when making preemption decisions, it selects the low-priority task with the greatest impact on network bottlenecks and the lowest efficiency based not only on the static priority of the task but also on the task's cross-level depth and task path risk value, using Bayesian inference. Therefore, this application can achieve accurate identification of preemption targets and optimized reallocation of network resources, thereby improving the scheduling efficiency of high-priority retrieval tasks and effectively alleviating network congestion in complex multi-level cascaded network environments, thus improving user experience.

[0076] In one feasible implementation, before determining the video retrieval task to be scheduled based on the priority information, cross-level depth information, and task path risk value of the candidate video retrieval task in step S150 through Bayesian inference, the method further includes:

[0077] Obtain the user history behavior logs for the candidate video viewing task.

[0078] User history behavior logs refer to the collection of all historical actions associated with a user. This log can include several key behavioral metrics, such as the total number of video viewing tasks initiated by the user, the average duration of viewed videos, the distribution of viewing task types, and the frequency of viewing videos marked as key areas.

[0079] Once the list of candidate video access tasks is determined, the user identifier corresponding to each task in the list is extracted. Then, using the user identifier as an index, a query is performed in the user behavior database that stores the user's historical behavior logs to retrieve the user's historical behavior logs within a preset time window, such as all relevant operation records in the past thirty days, i.e., user historical behavior logs containing multiple of the aforementioned key behavioral indicators, and then associated with the corresponding candidate video access tasks.

[0080] Adjust the priority sub-levels in the priority information of candidate video retrieval tasks based on user historical behavior logs.

[0081] All candidate video retrieval tasks belong to the low priority category, which contains multiple sub-levels. Adjusting the priority sub-levels in the priority information of candidate video retrieval tasks refers to dynamically and temporarily raising or lowering the original sub-levels based on the user's historical behavior.

[0082] First, based on the acquired user historical behavior logs, a dynamic behavior score is calculated for each user in a candidate video viewing task. Specifically, two key indicators are extracted: the frequency of key viewing areas and the average viewing duration. According to a preset scoring standard, the frequency of key viewing areas is divided into multiple intervals; the higher the frequency interval, the higher the base score. Simultaneously, the standard also divides the average viewing duration into multiple intervals; the lower the duration interval, the higher the bonus score. The user's dynamic behavior score is obtained by adding their base score and bonus score. After obtaining the dynamic behavior score, the priority sub-level of the user's currently initiated candidate video viewing task is adjusted based on this score. For example, a user with a high dynamic behavior score can have their current task's priority sub-level temporarily increased by one or more levels. This dynamically adjusted new priority sub-level will replace the original static sub-level.

[0083] For example, the low-priority category is divided into three sub-levels—Level 1, Level 2, and Level 3—based on importance from highest to lowest. The candidate video retrieval task list contains two tasks: Task A is initiated by User A, and Task B is initiated by User B. The original priority information for both tasks is low-priority category, and the sub-level is the lowest, Level 3.

[0084] The historical behavior logs of users A and B were retrieved. Analysis revealed that user A accessed key areas 10% of the time in the past month, with an average access time of 2 minutes. User B, on the other hand, accessed key areas 80% of the time, with an average access time of only 30 seconds. Dynamic behavior scores were calculated for users A and B based on preset scoring criteria. For example, the criteria stipulate that a high base score is awarded for accessing key areas more than 70% of the time, and a high bonus score is awarded for an average access time of less than 1 minute. Accordingly, user B received a high base score and a high bonus score, resulting in a final dynamic behavior score of 95 points; while user A did not meet the high score criteria for either criterion, resulting in a dynamic behavior score of 60 points. According to preset adjustment rules, users with a dynamic behavior score higher than 90 points can temporarily upgrade the priority sub-level of their current task by one level. Therefore, the priority information for task B was adjusted, temporarily upgrading its sub-level from level 3 to level 2, while the priority sub-level of task A remained unchanged at level 3. Ultimately, this dynamically adjusted new priority sub-level will replace the original static sub-level.

[0085] In one feasible implementation, the method further includes:

[0086] Obtain the task progress information for the candidate video retrieval task.

[0087] Task progress information is a metric used to quantify the degree to which a video retrieval task has been completed. For historical video playback tasks, task progress information can be the percentage of time already played out out of the total time; for real-time preview tasks, task progress information can be the duration for which the task has been playing steadily and continuously.

[0088] When each video retrieval task is created, its related duration information, i.e., task progress information, is recorded. After a candidate video retrieval task list is generated, the real-time recorded duration information is obtained for each task in the list. If the task is a historical recording playback, its played duration is divided by the total duration of the recording file to obtain a completion percentage as the task progress information. If the task is a live preview, its total duration from the start of playback to the current time is directly obtained as the task progress information.

[0089] In step S150, based on the priority information, cross-level depth information, and task path risk value of the candidate video retrieval tasks, the video retrieval tasks to be scheduled are determined through Bayesian inference, including:

[0090] Based on the priority information, cross-level depth information, task path risk value, and task progress information of the candidate video retrieval tasks, the video retrieval tasks to be scheduled are determined through Bayesian inference.

[0091] When performing Bayesian inference on each task in the candidate video retrieval task list, firstly, a prior probability for task scheduling is determined based on the priority information of the candidate video retrieval tasks. Secondly, when calculating the conditional probability of task state, not only are the cross-level depth information and task path risk values ​​of the candidate video retrieval tasks selected, but also newly acquired task progress information is further quantified as part of the observational evidence. Specifically, the higher the progress of a task, the lower its corresponding weight or value in calculating the conditional probability of task state, meaning that the system tends to believe that a task that is about to be completed has a relatively small negative impact on the system. Finally, by multiplying the prior probability of task scheduling with this new conditional probability of task state that includes consideration of task progress, a more comprehensive task scheduling score is obtained, and the video retrieval tasks to be scheduled are determined based on this score.

[0092] For example, the candidate video retrieval task list contains two tasks: Task C is a historical video playback task that has been played for 58 minutes and has a total duration of 60 minutes; Task D is a live preview task that has just started for 2 minutes. All other parameters of the two tasks, including priority information, cross-level depth information, and task path risk value, are exactly the same.

[0093] After entering the Bayesian inference process, the task progress information for both tasks is first obtained. Task C's progress is 58 divided by 60, approximately 96.7%. Task D's progress is 2 minutes in duration. When calculating the conditional probability of the task state, because Task C's progress is extremely high, its corresponding value in the calculation will be very low; while Task D's progress is very low, its corresponding value will be higher. Therefore, although other risk indicators are the same for both tasks, the final calculated conditional probability of the task state for Task C will be much lower than that for Task D. Since the prior probabilities of both tasks are also the same, Task D's score will be significantly higher than Task C's in the final task scheduling score calculation. Ultimately, Task D, with the higher task scheduling score, is selected as the task to be scheduled for video playback, thus avoiding interrupting a task that is about to finish playing and preserving most of its consumed resource value.

[0094] Figure 2 This illustration shows a flowchart of a method for determining a scheduled video retrieval task using Bayesian inference, according to an embodiment of this application. Figure 2 As shown, the method includes steps S210 to S230.

[0095] In one feasible implementation, based on the priority information, cross-level depth information, task path risk value, and task progress information of candidate video retrieval tasks, the video retrieval tasks to be scheduled are determined through Bayesian inference, including:

[0096] S210: Based on the priority sub-levels in the priority information, determine the prior probability of task scheduling for candidate video retrieval tasks from the preset mapping relationship between levels and probabilities. The lower the priority sub-level, the higher the prior probability of task scheduling.

[0097] The prior probability of task scheduling is a numerical value used to represent the initial probability that a candidate video retrieval task will be selected as the preemption target based on its original priority level, without considering any dynamic running state.

[0098] In the priority information of candidate video retrieval tasks with low priority, specific sub-levels are extracted. Then, using this sub-level as the key, a query is performed in a pre-defined mapping table of levels and probabilities to determine the specific prior probability value for task scheduling corresponding to that sub-level. The mapping table of levels and probabilities is shown in Table 1. The smaller the number of the priority sub-level, the higher its importance within the low-priority category. The prior probability for task scheduling is a value between 0 and 1; the higher the value, the greater the initial probability that the task will be selected as the preemption target, without considering other dynamic factors.

[0099]

[0100] Table 1

[0101] S220: After quantifying the cross-level depth information, task path risk value, and task progress information of the candidate video retrieval task, calculate the product to obtain the task state conditional probability. The deeper the cross-level depth information, the greater the probability value of the task state conditional probability. The greater the risk value of the task path risk value, the greater the probability value of the task state conditional probability. The lower the progress information of the task progress, the greater the probability value of the task state conditional probability.

[0102] Task state conditional probability is used to measure the severity of the negative impact on the entire cascaded network, given the assumption that a task will be preempted, based on the combined performance of current observational evidence such as cross-level depth, path risk, and task progress.

[0103] First, the cross-level depth information, task path risk value, and task progress information of each candidate video retrieval task are quantified, transforming these raw data from different dimensions into standardized probability factors that can be multiplied. This quantification process follows three core principles: the deeper the cross-level depth, the larger the corresponding probability factor; the higher the task path risk value, the larger the corresponding probability factor; and the higher the task progress, the smaller the corresponding probability factor. After obtaining the probability factors corresponding to all indicators, the three probability factors are multiplied to obtain a task state conditional probability reflecting the current running state of the task.

[0104] For example, a preset transformation function is used to process the cross-level depth information, task path risk value and task progress information of each candidate video retrieval task, so as to transform these raw data of different dimensions into standardized probability factors.

[0105] For cross-level depth information, the corresponding probability factor Fdepth is determined by a predefined piecewise function. For example, when the cross-level depth is 1, Fdepth is 0.3; when the cross-level depth is 2, Fdepth is 0.5; when the cross-level depth is 3, Fdepth is 0.7; and when the cross-level depth is greater than or equal to 4, Fdepth is 0.9.

[0106] For the task path risk value, its probability factor Frisk is determined by a variant of the arctangent function. The higher the task path risk value, the more smoothly its corresponding probability factor approaches 1. The calculation process is shown in formula (1):

[0107] (1)

[0108] Where Frisk represents the probability factor corresponding to the risk value of the task path, R represents the specific value of the risk value of the task path, π is the mathematical constant pi, and k is a preset constant used to adjust the sensitivity of the risk value.

[0109] For task progress information, its probability factor Fprog is determined by a linearly decreasing function. The higher the task progress, the lower the corresponding probability factor. The calculation process is shown in formula (2):

[0110] (2)

[0111] Where Fprog represents the probability factor corresponding to the task progress information, and P represents the percentage value of the task progress information.

[0112] S230: By calculating the product of the prior probability of task scheduling and the conditional probability of task state, the task scheduling score of the candidate video retrieval task is obtained, and the candidate video retrieval task with the highest task scheduling score is determined as the video retrieval task to be scheduled.

[0113] For each candidate video retrieval task, its corresponding prior scheduling probability is multiplied by its corresponding conditional probability of the task state. The result of this product is the final task scheduling score for that candidate video retrieval task. After calculating the task scheduling scores for all tasks in the candidate video retrieval task list, all tasks are sorted in descending order of their scores, and the candidate video retrieval task with the highest score is selected as the final video retrieval task to be scheduled.

[0114] For example, it is necessary to determine a video retrieval task to be scheduled from a list containing two candidate video retrieval tasks, namely task C and task D. Task C has a higher priority sub-level of 1, so its prior scheduling probability is determined to be a lower value, such as 0.2, based on the preset mapping relationship. Task D has the lowest priority sub-level of 3, so its prior scheduling probability is determined to be a higher value, such as 0.8. Assume that task C has a cross-level depth of 1, a task path risk value of 2.0, and a task progress of 96.7%; while task D has a cross-level depth of 3, a task path risk value of 4.5, and a task progress of only 3.3%. Quantifying these data, since task C has a low cross-level depth and path risk value, and its task progress is extremely high, the multiplication of the probability factors of these three indicators results in a very low task state conditional probability. Conversely, task D has a high cross-level depth and path risk value, and its task progress is extremely low, so its final calculated task state conditional probability is significantly higher than that of task C. Finally, the final task scheduling score is calculated by multiplying the prior probability of each task scheduling task by the conditional probability of the task state. Since task D has a higher prior probability and a much higher conditional probability than task C, its final task scheduling score is significantly the highest. Therefore, task D is selected as the video retrieval task to be scheduled, and a resource release operation is performed on it.

[0115] This embodiment introduces task progress information for candidate video retrieval tasks. In a multi-level networked platform cascading retrieval scenario, it not only assesses the global impact of a task on the entire cascading network based on its static priority, cross-level depth, and task path risk value, but also incorporates task completion progress as a new key dimension. When making preemption decisions, Bayesian inference integrates task progress information into the judgment criteria, which tends to retain low-priority tasks that are about to complete, while prioritizing tasks with lower progress. This avoids resource waste and impact on user experience caused by preempting tasks that are about to complete. Therefore, this application can further optimize user service experience and improve scheduling rationality in resource-constrained scenarios while ensuring high-priority tasks.

[0116] Figure 3 This illustration shows a flowchart of a method for determining the risk value of a video retrieval task according to an embodiment of this application. Figure 3 As shown, the method includes steps S310 to S330.

[0117] In one feasible implementation, step S130: Based on the service quality information of each video retrieval task on each path's network platform and the resource pressure score of the corresponding node, determine the task path risk value for each video retrieval task, including:

[0118] S310: Convert the service quality information of the video retrieval task on each network platform into a service quality score for the video retrieval task on each network platform.

[0119] Quality of Service (QoS) score is a numerical value used to comprehensively characterize the overall network transmission quality of a video retrieval task within a single network platform transmission segment. This score is derived from multiple raw QoS indicators for that transmission segment, such as network latency and packet loss rate.

[0120] For each video retrieval task, the service quality information corresponding to each network platform along its path is processed. First, a base score of 0.5 points is assigned to each of the two raw metrics: network latency and packet loss rate. Second, a pre-defined deduction standard is applied to the base score, dividing the numerical range of each raw metric into multiple intervals; the worse the interval in which the metric falls, the higher the deduction value. Finally, the scores after deductions for the two raw metrics are added together to obtain a final service quality score.

[0121] For example, a transmission segment has a network latency of 80 milliseconds and a packet loss rate of 0.3%. The preset deduction criteria stipulate that a latency of 51 to 100 milliseconds deducts 0.1 points, and a packet loss rate of 0.11 to 0.5% deducts 0.15 points. Therefore, the network latency score for this transmission segment is 0.5 minus 0.1 equals 0.4 points, and the packet loss rate score is 0.5 minus 0.15 equals 0.35 points. Ultimately, the service quality score for this transmission segment is 0.4 plus 0.35 equals 0.75 points. In this way, service quality information of different dimensions is transformed into a standardized service quality score between 0 and 1.

[0122] S320: The service quality score and resource pressure score corresponding to each network platform of the video retrieval task are weighted and summed to obtain the initial path risk value of each video retrieval task.

[0123] The initial path risk value is a preliminary quantification of the total risk of a video retrieval task along the entire path from the initiator to the target device, which is composed of the transmission quality of each link segment and the stress level of each platform node.

[0124] First, obtain all network platforms along the video retrieval task path. Then, for each network platform on the path, calculate a weighted sum of its service quality score and resource stress score. The weighting coefficients can be set according to actual conditions; for example, the service quality score weighting coefficient can be set to a negative value, while the resource stress score weighting coefficient can be set to a positive value. Finally, sum the weighted sums calculated for all network platforms on the path to obtain an initial path risk value that can preliminarily reflect the risk status of the entire path.

[0125] S330: Based on the distribution dispersion of resource pressure scores of all networked platforms along the video retrieval task path, the initial path risk value is adjusted to obtain the task path risk value for each video retrieval task.

[0126] Distribution dispersion is a statistical metric used to measure the degree of dispersion or concentration of a set of values. In this application, it specifically refers to the dispersion of the resource pressure scores of all network platforms traversed by a video retrieval task. A high distribution dispersion means that the pressure distribution along the path is uneven, with some platforms experiencing significantly higher pressure than others.

[0127] First, the resource stress scores of all network platforms traversed by a video retrieval task are extracted, forming a numerical sequence. Then, the dispersion of this numerical sequence is calculated, for example, by calculating its standard deviation or variance. Next, the calculated dispersion is converted into an adjustment coefficient, which can be 1 plus the dispersion. Finally, the initial path risk value is multiplied by this adjustment coefficient to obtain the final task path risk value. This ensures that when a bottleneck node with extremely high stress exists on a task's path, causing increased dispersion, its final task path risk value will be amplified accordingly, thus more accurately identifying such high-risk paths.

[0128] For example, to calculate the task path risk value of a video retrieval task that spans both a city-level platform and a district / county-level platform, firstly, the link service quality information on the city-level platform—namely, network latency of 30 milliseconds and packet loss rate of 0.01 percentage points—is converted into a service quality score, for example, 0.95, according to a preset deduction standard. Simultaneously, the link service quality information on the district / county-level platform—namely, network latency of 40 milliseconds and packet loss rate of 0.1 percentage points—is converted into a relatively lower service quality score, for example, 0.85. Next, the resource pressure scores for the city-level and district / county-level platforms are obtained, which are 0.7 and 0.85, respectively. Assuming the weighting coefficient for the service quality score is set to -2 and the weighting coefficient for the resource pressure score is set to +3, then the risk component for the city-level platform is 0.95 multiplied by -2 plus 0.7 multiplied by 3, equaling 0.2. The risk component for the district / county-level platform is 0.85 multiplied by -2 plus 0.85 multiplied by 3, equaling 0.85. Adding the two components yields an initial path risk value of 1.05 for the task. Finally, the dispersion of the resource stress scores of 0.7 and 0.85 is calculated, for example, its standard deviation is 0.106. This dispersion is then converted into an adjustment factor, i.e., 1 plus 0.106 equals 1.106. Multiplying the initial path risk value of 1.05 by the adjustment factor of 1.106, the final task path risk value for this video retrieval task is approximately 1.16.

[0129] In one feasible implementation, the number of concurrent paths for all video retrieval tasks being executed includes the total number of concurrent channels for all video retrieval tasks being executed in the multi-level network platform.

[0130] The concurrent stream count of all ongoing video retrieval tasks is a key indicator of the overall load on the multi-level network platform system. Specifically, the concurrent stream count includes the total number of concurrent channels currently occupied across all network platforms for transmitting real-time preview and historical playback video streams. By continuously monitoring the signaling scheduling services of each network platform, the number of video sessions being processed by each platform is counted in real time, and the session counts of all platforms are summed to obtain a total value reflecting the concurrent load of the entire cascaded network. This total value is compared with a preset concurrent stream count threshold and is an important basis for triggering the preemptive scheduling mechanism.

[0131] In one feasible implementation, the total bandwidth requirement of all ongoing video retrieval tasks and the target video retrieval task includes the sum of the total bandwidth utilization of all ongoing video retrieval tasks and the estimated bandwidth requirement of the target video retrieval task.

[0132] The total bandwidth requirement of all ongoing video retrieval tasks and the target video retrieval task is a key indicator for assessing whether the network egress bandwidth of the entire multi-level networking platform is overloaded. Specifically, the total bandwidth requirement includes the sum of two parts. The first part is the total bandwidth utilization of all currently executing video retrieval tasks. This value is obtained by collecting and accumulating the actual bandwidth occupied by all transmitted video streams in real time through streaming media probes or monitoring tools deployed at the network egress. The second part is the estimated bandwidth requirement of a newly accessed target video retrieval task. When the task request is received, the requested bitstream type is parsed, such as whether it is a high-definition main stream or a standard-definition sub-stream, and the bandwidth it will occupy is estimated based on the standard bitrate of that stream type. By adding these two parts, it is possible to determine in advance whether allowing the new task to access the network will exceed the preset bandwidth requirement threshold, thus providing a basis for decision-making on whether to trigger preemption scheduling.

[0133] Based on the same concept, this application provides a video access preemption scheduling system triggered by both concurrency and bandwidth dual thresholds. The following is a detailed description... Figure 4 This application provides a detailed description of the video access preemption scheduling system triggered by both concurrency and bandwidth thresholds, as described in the embodiments of this application.

[0134] Figure 4 This is a structural block diagram of a video access preemption scheduling system with concurrent and bandwidth dual-threshold triggering, as shown in an embodiment of this application.

[0135] like Figure 4 As shown, this system is used for cascading access on multi-level networked platforms. The system may include:

[0136] The acquisition module 410 is used to acquire service transmission data and priority information of each video retrieval task being executed in the multi-level network platform, as well as resource utilization data of each network platform. The service transmission data includes cross-level depth information and service quality information of each video retrieval task on each path of the network platform.

[0137] Module 420 is used to construct a directed graph of network platforms with each network platform as a node, the cascading relationship between network platforms as edges, and the resource utilization data of network platforms as node attribute features. The directed graph of network platforms is then input into the trained network model to obtain the resource pressure score of each node. The network model is trained based on the historical directed graph of network platforms through a graph attention mechanism.

[0138] The determination module 430 is used to determine the task path risk value of each video retrieval task based on the service quality information of the network platform in each path and the resource pressure score of the corresponding node for each video retrieval task.

[0139] The determination module 430 is also used to obtain the target priority information of the target video retrieval task. When the target priority information is high priority and the number of concurrent paths of all currently executing video retrieval tasks is greater than the number of concurrent paths threshold or the total bandwidth requirement of all currently executing video retrieval tasks and the target video retrieval task is greater than the bandwidth requirement threshold, the video retrieval task with low priority information is determined as a candidate video retrieval task.

[0140] The scheduling module 440 is used to determine the video retrieval task to be scheduled based on the priority information, cross-level depth information and task path risk value of the candidate video retrieval task through Bayesian inference, and release the resources occupied by the video retrieval task to be scheduled so that the target video retrieval task can be executed.

[0141] In one embodiment, before determining the video access task to be scheduled by Bayesian inference based on the priority information, cross-level depth information and task path risk value of the candidate video access task, the determination module 430 is also used to obtain the user's historical behavior log of the candidate video access task; and adjust the priority sub-level in the priority information of the candidate video access task based on the user's historical behavior log.

[0142] In one embodiment, the determining module 430 is further configured to obtain the task progress information of the candidate video retrieval task; and determine the video retrieval task to be scheduled by Bayesian inference based on the priority information, cross-level depth information, task path risk value and task progress information of the candidate video retrieval task.

[0143] In one embodiment, the determining module 430 is specifically used to determine the prior probability of task scheduling for candidate video retrieval tasks based on the priority sub-levels in the priority information and from a preset mapping relationship between levels and probabilities. The lower the priority sub-level, the higher the prior probability of task scheduling. The module quantifies the cross-level depth information, task path risk value, and task progress information of the candidate video retrieval tasks and calculates their product to obtain the conditional probability of task state. The deeper the cross-level depth information, the higher the probability value of the conditional probability of task state. The higher the risk value of the task path risk value, the higher the probability value of the conditional probability of task state. The lower the progress information, the higher the probability value of the conditional probability of task state. By calculating the product of the prior probability of task scheduling and the conditional probability of task state, the task scheduling score of the candidate video retrieval task is obtained, and the candidate video retrieval task with the highest task scheduling score is determined as the video retrieval task to be scheduled.

[0144] In one embodiment, the determining module 430 is specifically used to convert the service quality information of the network platforms of the video retrieval task in each path into the service quality score of the video retrieval task corresponding to the network platforms of each path; to perform a weighted summation of the service quality score and resource pressure score of the video retrieval task corresponding to the network platforms of each path to obtain the initial path risk value of each video retrieval task; and to adjust the initial path risk value according to the distribution dispersion of the resource pressure scores of all network platforms of the video retrieval task path to obtain the task path risk value of each video retrieval task.

[0145] In one embodiment, the number of concurrent paths for all video retrieval tasks being executed includes the total number of concurrent channels for all video retrieval tasks being executed in the multi-level networking platform.

[0146] In one embodiment, the total bandwidth requirement of all ongoing video retrieval tasks and the target video retrieval task includes the sum of the total bandwidth utilization of all ongoing video retrieval tasks and the estimated bandwidth requirement of the target video retrieval task.

[0147] Figure 4 Each module in the system shown has an implementation Figures 1 to 3 The functions of each step in the process and their corresponding technical effects are described in detail here for the sake of brevity.

[0148] Figure 5 A schematic diagram of the hardware structure of an electronic device provided in one embodiment of this application is shown.

[0149] The electronic device may include a processor 510 and a memory 520 storing computer program instructions.

[0150] Specifically, the processor 510 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0151] Memory 520 may include mass storage for data or instructions. For example, and not limitingly, memory 520 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 520 may include removable or non-removable (or fixed) media. Where appropriate, memory 520 may be internal or external to the integrated gateway disaster recovery device. In a particular embodiment, memory 520 is non-volatile solid-state memory.

[0152] Memory may include read-only memory (ROM), random access memory (RAM), disk storage media devices, optical storage media devices, flash memory devices, and electrical, optical, or other physical / tangible memory storage devices. Therefore, typically, memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the method according to the first aspect of this disclosure.

[0153] The processor 510 reads and executes computer program instructions stored in the memory 520 to implement any of the concurrent and bandwidth dual-threshold triggered video access preemption scheduling methods in the above embodiments.

[0154] In one example, the electronic device may also include a communication interface 530 and a bus 540. Wherein, such as Figure 5 As shown, the processor 510, memory 520, and communication interface 530 are connected through bus 540 and complete communication with each other.

[0155] The communication interface 530 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application.

[0156] Bus 540 includes hardware, software, or both, that couples components of an online data traffic metering device together. For example, and not limitingly, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 540 may include one or more buses. Although specific buses are described and illustrated in embodiments of this application, any suitable bus or interconnect is contemplated herein.

[0157] The electronic device can execute the video access preemption scheduling method with concurrent and bandwidth dual-threshold triggering in the embodiments of this application, thereby achieving a combination of Figures 1 to 3 The video access preemption scheduling method described is triggered by both concurrency and bandwidth thresholds.

[0158] Furthermore, in conjunction with the video access preemption scheduling method triggered by both concurrency and bandwidth in the above embodiments, this application embodiment can provide a computer-readable storage medium for implementation. This computer-readable storage medium stores computer program instructions; when executed by a processor, these computer program instructions implement any one of the video access preemption scheduling methods triggered by both concurrency and bandwidth in the above embodiments.

[0159] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.

[0160] The functional blocks shown in the above-described structural diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. Programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0161] It should also be noted that the exemplary embodiments mentioned in this application describe methods or systems based on a series of steps or apparatus. However, this application is not limited to the order of the above steps; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0162] The aspects of this application have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It should be understood that each block in the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by dedicated hardware performing the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

[0163] The above description is merely a specific implementation of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application.

Claims

1. A video access preemption scheduling method triggered by both concurrency and bandwidth thresholds, applied to cascaded access on multi-level network platforms, characterized in that, include: The service transmission data and priority information of each video retrieval task being executed in the multi-level network platform are obtained, as well as the resource utilization data of each network platform. The service transmission data includes cross-level depth information and the service quality information of each video retrieval task on each path of the network platform. The cross-level depth information is the number of network platforms that the video retrieval task passes through from the initiator to the target device. Using each of the network platforms as nodes, the cascading relationships between the network platforms as edges, and the resource utilization data of the network platforms as node attribute features, a directed graph of the network platforms is constructed. The directed graph of the network platforms is then input into a trained network model to obtain a resource pressure score for each node. The network model is trained based on the historical directed graph of the network platforms using a graph attention mechanism. Based on the service quality information of each video retrieval task on each network platform along each path and the resource pressure score of the corresponding node, the task path risk value of each video retrieval task is determined; Obtain the target priority information of the target video retrieval task. If the target priority information is high priority and the number of concurrent paths of all the currently executing video retrieval tasks is greater than the number of concurrent paths threshold or the total bandwidth requirement of all the currently executing video retrieval tasks and the target video retrieval task is greater than the bandwidth requirement threshold, then the video retrieval task with the low priority information is determined as a candidate video retrieval task. Based on the priority information, cross-level depth information, and task path risk value of the candidate video retrieval tasks, a video retrieval task to be scheduled is determined through Bayesian inference, and the resources occupied by the video retrieval task to be scheduled are released to execute the target video retrieval task.

2. The method according to claim 1, characterized in that, Before determining the video retrieval task to be scheduled based on the priority information, cross-level depth information, and task path risk value of the candidate video retrieval tasks using Bayesian inference, the method further includes: Obtain the user's historical behavior logs for the candidate video retrieval task; The priority sub-levels in the priority information of the candidate video retrieval task are adjusted based on the user's historical behavior logs.

3. The method according to claim 1, characterized in that, The method further includes: Obtain the task progress information of the candidate video retrieval task; The step of determining the video retrieval task to be scheduled based on the priority information, cross-level depth information, and task path risk value of the candidate video retrieval tasks through Bayesian inference includes: Based on the priority information, cross-level depth information, task path risk value, and task progress information of the candidate video retrieval tasks, the video retrieval tasks to be scheduled are determined through Bayesian inference.

4. The method according to claim 3, characterized in that, The step of determining the video retrieval task to be scheduled based on the priority information, cross-level depth information, task path risk value, and task progress information of the candidate video retrieval tasks through Bayesian inference includes: Based on the priority sub-levels in the priority information, the prior probability of task scheduling for the candidate video retrieval task is determined from the preset mapping relationship between level and probability. The lower the priority sub-level, the higher the prior probability of task scheduling. The cross-level depth information, the task path risk value, and the task progress information of the candidate video retrieval task are quantified and multiplied to obtain the task state conditional probability. The deeper the cross-level depth information, the greater the probability value of the task state conditional probability. The greater the risk value of the task path risk value, the greater the probability value of the task state conditional probability. The lower the progress of the task progress information, the greater the probability value of the task state conditional probability. The task scheduling score of the candidate video retrieval task is obtained by calculating the product of the prior probability of task scheduling and the conditional probability of task state. The candidate video retrieval task with the highest task scheduling score is determined as the video retrieval task to be scheduled.

5. The method according to claim 1, characterized in that, The step of determining the task path risk value for each video retrieval task based on the service quality information of the network platform along each path and the resource pressure score of the corresponding node includes: The service quality information of the video retrieval task on each of the network platforms is converted into a service quality score corresponding to the video retrieval task on each of the network platforms. The service quality score and resource pressure score corresponding to each network platform in each path of the video retrieval task are weighted and summed to obtain the initial path risk value of each video retrieval task; Based on the distribution dispersion of the resource pressure scores of all networked platforms along the video retrieval task path, the initial path risk value is adjusted to obtain the task path risk value for each video retrieval task.

6. The method according to claim 1, characterized in that, The concurrent number of all video retrieval tasks currently being executed includes the total number of concurrent channels for all video retrieval tasks currently being executed in the multi-level network platform.

7. The method according to claim 1, characterized in that, The total bandwidth requirement of all currently executing video retrieval tasks and the target video retrieval task includes the sum of the total bandwidth utilization of all currently executing video retrieval tasks and the estimated bandwidth requirement of the target video retrieval task.

8. A video access preemption scheduling system triggered by both concurrency and bandwidth thresholds, applied to cascaded access on multi-level networked platforms, characterized in that... The system includes: The acquisition module is used to acquire service transmission data and priority information of each video retrieval task being executed in the multi-level network platform, as well as resource utilization data of each network platform. The service transmission data includes cross-level depth information and service quality information of each video retrieval task on each path of the network platform. The cross-level depth information is the number of network platforms that the video retrieval task passes through from the initiator to the target device. The construction module is used to construct a directed graph of the network platforms with each of the network platforms as a node, the cascading relationship between the network platforms as an edge, and the resource utilization data of the network platforms as node attribute features. The directed graph of the network platforms is then input into a trained network model to obtain the resource pressure score of each node. The network model is trained based on the historical directed graph of the network platforms through a graph attention mechanism. The determination module is used to determine the task path risk value of each video retrieval task based on the service quality information of the network platform in each path and the resource pressure score of the corresponding node for each video retrieval task; The determination module is also used to obtain the target priority information of the target video retrieval task. When the target priority information is high priority and the number of concurrent paths of all the video retrieval tasks being executed is greater than the number of concurrent paths threshold or the total bandwidth requirement of all the video retrieval tasks being executed and the target video retrieval task is greater than the bandwidth requirement threshold, the video retrieval task whose priority information is low priority is determined as a candidate video retrieval task. The scheduling module is used to determine the video retrieval task to be scheduled based on the priority information, cross-level depth information and task path risk value of the candidate video retrieval task through Bayesian inference, and release the resources occupied by the video retrieval task to be scheduled for execution of the target video retrieval task.

9. An electronic device, characterized in that, The device includes: a processor and a memory storing computer program instructions; When the processor executes the computer program instructions, it implements the video access preemption scheduling method as described in any one of claims 1-7, which is triggered by both concurrency and bandwidth limitations.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer program instructions, which, when executed by a processor, implement the video access preemption scheduling method as described in any one of claims 1-7, which is triggered by both concurrency and bandwidth limitations.

Images