Task resource scheduling method and device, communication equipment, readable storage medium and program product

Abstract

The invention relates to a task resource scheduling method and device, communication equipment, a computer readable storage medium and a computer program product. The method comprises the following steps: receiving a task request concurrent by each terminal, wherein the task request carries task attribute information; based on the task attribute information, performing resource scheduling conflict detection on each task request to obtain a conflict detection result; if the conflict detection result is that the resource scheduling conflict exists, calculating a task priority corresponding to the target task request with the resource scheduling conflict; and sorting the target task requests based on the task priorities, storing the sorted target task requests in a task queue, and performing resource scheduling on tasks of the target task requests in the task queue in sequence. By adopting the method, task interruption caused by model resource conflicts can be reduced, and the stability and execution efficiency of task resource scheduling are improved.

Description

Technical Field

[0001] This application relates to the field of wireless communication technologies, and particularly to a task resource scheduling method, apparatus, communication device, computer-readable storage medium, and computer program product. Background Art

[0002] In current wireless communication networks, the management and task scheduling of AI (Artificial Intelligence) models mainly rely on centralized control and preset priority rules.

[0003] In traditional technologies, in O-RAN (Open Radio Access Network Architecture), xAPP (Near-Real-Time RIC Application) and rAPP (Non-Real-Time RIC Application) are scheduled and managed through a service management and orchestration unit. Resources are allocated mainly based on time-slot scheduling or preset task priorities, and fixed task rules are relied on to avoid conflicts.

[0004] However, in traditional technologies, the scheduling logic of task resource scheduling is relatively simple and cannot handle the problems of resource preemption and task conflicts in complex multi-task concurrent environments. Summary of the Invention

[0005] Based on this, in view of the above technical problems, it is necessary to provide a task resource scheduling method, apparatus, communication device, computer-readable storage medium, and computer program product that can reduce task interruptions caused by model resource conflicts and improve the stability and execution efficiency of task resource scheduling.

[0006] In a first aspect, this application provides a task resource scheduling method, including:

[0007] Receiving concurrent task requests from each terminal, where the task requests carry task attribute information;

[0008] Based on the task attribute information, performing resource scheduling conflict detection on each of the task requests to obtain a conflict detection result;

[0009] If the conflict detection result indicates that there is a resource scheduling conflict, calculating the task priority corresponding to the target task request with a resource scheduling conflict;

[0010] Sort each of the target task requests based on the task priority, and store the sorted target task requests in a task queue, and then perform resource scheduling for the tasks of each of the target task requests in the task queue in sequence.

[0011] In one embodiment, the detecting of resource scheduling conflicts for each of the task requests based on the task attribute information to obtain a conflict detection result includes:

[0012] Parse the task attribute information carried by each of the task requests to obtain the task objectives and model call requirements of each task;

[0013] If there are at least two tasks with the same model scheduling requirements and / or the same task objectives, determine that the conflict detection result of each of the task requests is that there are resource scheduling conflicts for at least two tasks.

[0014] In one embodiment, the task attribute information further includes priority parameter information. When the conflict detection result is that there are resource scheduling conflicts, calculating the task priority corresponding to the target task request with resource scheduling conflicts includes:

[0015] When the conflict detection result is that there are resource scheduling conflicts for at least two tasks, calculate the task priority corresponding to the tasks with resource scheduling conflicts based on the priority parameter information.

[0016] In one embodiment, the sorting each of the target task requests based on the task priority and storing the sorted target task requests in a task queue includes:

[0017] Sort each of the target task requests in descending order of the task priority to obtain the sorting result of each of the target task requests;

[0018] Based on the sorting result, store each of the target task requests in a task queue.

[0019] In one embodiment, the task resources include resources of the artificial intelligence model type. The performing of resource scheduling for the tasks of each of the target task requests in the task queue in sequence includes:

[0020] In response to the target task request in the task queue, detect the model status of the target artificial intelligence model associated with the task of the target task request;

[0021] If the model status is the idle state, determine that the target task request with a higher current priority obtains the resource scheduling permission, and leave the other target task requests with a lower priority in the task queue.

[0022] In one embodiment, the method further includes:

[0023] Obtain the target artificial intelligence model based on the model call permission obtained from the target task request.

[0024] Perform task locking on the target artificial intelligence model through the status identifier of the occupancy status, allocate computing resources, and start executing the task of the target task request with the highest current priority.

[0025] In one embodiment, the method further includes:

[0026] After the task execution of each target task request is completed, update the status identifier of the target artificial intelligence model associated with the target task request to the idle state, and release the task lock on the target artificial intelligence model.

[0027] In one embodiment, the method further includes:

[0028] After the task execution of each target task request is completed, record the execution information of the task corresponding to the target task request based on the system log.

[0029] In a second aspect, the present application further provides a task resource scheduling device, and the device includes:

[0030] A receiving module, configured to receive concurrent task requests from each terminal, where the task requests carry task attribute information;

[0031] A detection module, configured to perform resource scheduling conflict detection on each of the task requests based on the task attribute information to obtain a conflict detection result;

[0032] A calculation module, configured to calculate the task priority corresponding to the target task request with a resource scheduling conflict if the conflict detection result indicates the existence of a resource scheduling conflict;

[0033] A processing module, configured to sort each of the target task requests based on the task priority, store the sorted target task requests in a task queue, and perform resource scheduling on the tasks of each of the target task requests in the task queue in sequence.

[0034] In a third aspect, the present application further provides a communication device, including a memory and a processor, where the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0035] Receive concurrent task requests from each terminal, where the task requests carry task attribute information;

[0036] Based on the task attribute information, perform resource scheduling conflict detection on each of the task requests to obtain a conflict detection result;

[0037] When the conflict detection result indicates that there is a resource scheduling conflict, calculate the task priority corresponding to the target task request with the resource scheduling conflict;

[0038] Sort each of the target task requests based on the task priority, and store the sorted target task requests in a task queue, and sequentially perform resource scheduling for the tasks of each of the target task requests in the task queue.

[0039] Fourthly, the present application further provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0040] Receive concurrent task requests from each terminal, where the task requests carry task attribute information;

[0041] Based on the task attribute information, perform resource scheduling conflict detection on each of the task requests to obtain a conflict detection result;

[0042] When the conflict detection result indicates that there is a resource scheduling conflict, calculate the task priority corresponding to the target task request with the resource scheduling conflict;

[0043] Sort each of the target task requests based on the task priority, and store the sorted target task requests in a task queue, and sequentially perform resource scheduling for the tasks of each of the target task requests in the task queue.

[0044] Fifthly, the present application further provides a computer program product, including a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0045] Receive concurrent task requests from each terminal, where the task requests carry task attribute information;

[0046] Based on the task attribute information, perform resource scheduling conflict detection on each of the task requests to obtain a conflict detection result;

[0047] When the conflict detection result indicates that there is a resource scheduling conflict, calculate the task priority corresponding to the target task request with the resource scheduling conflict;

[0048] Sort each of the target task requests based on the task priority, and store the sorted target task requests in a task queue, and sequentially perform resource scheduling for the tasks of each of the target task requests in the task queue.

[0049] The above task resource scheduling method, device, communication device, computer-readable storage medium, and computer program product, the method includes: receiving task requests concurrently sent by each terminal, where the task requests carry task attribute information; based on the task attribute information, performing resource scheduling conflict detection on each of the task requests to obtain a conflict detection result; if the conflict detection result indicates that there is a resource scheduling conflict, calculating the task priority corresponding to the target task request with the resource scheduling conflict; sorting each of the target task requests based on the task priority and storing the sorted target task requests in a task queue, and sequentially performing resource scheduling on the tasks of each of the target task requests in the task queue. By using this method, based on the task attribute information of each concurrently sent task request, resource scheduling conflict detection is performed on each task request to obtain a conflict detection result, thereby avoiding resource scheduling conflicts between target task requests that call the same resource, increasing the accuracy and rationality of resource scheduling, and further, by means of the functions of the task queue and task priority, sorting each concurrently sent target task request to reduce task interruptions caused by model resource conflicts and improve the stability and execution efficiency of task resource scheduling. Brief Description of the Drawings

[0050] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following will briefly introduce the drawings required for use in the description of the embodiments of the present application or related technologies. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other related drawings can be obtained based on these drawings.

[0051] Figure 1 It is an application environment diagram of the task resource scheduling method in an embodiment;

[0052] Figure 2 It is a schematic architecture diagram of the RAN AI Layer of a base station in an embodiment;

[0053] Figure 3 It is a schematic flowchart of the task resource scheduling method in an embodiment;

[0054] Figure 4 It is a schematic flowchart of the step of determining that there is a task resource scheduling conflict in a task request in an embodiment;

[0055] Figure 5 It is a schematic flowchart of the step of calculating the task priority corresponding to a task with a resource scheduling conflict in an embodiment;

[0056] Figure 6 It is a schematic flowchart of sorting each task based on the task priority and storing it in a task queue in an embodiment;

[0057] Figure 7 It is a schematic flowchart of the steps for a task to obtain resource scheduling permission for a target task request with high priority in an embodiment;

[0058] Figure 8 It is a schematic flowchart of the steps for starting to execute a task of a target task request with high priority in an embodiment;

[0059] Figure 9 It is a schematic flowchart of the steps for releasing the task lock of a target artificial intelligence model in an embodiment;

[0060] Figure 10 It is a schematic flowchart of the steps for recording task execution information in an embodiment;

[0061] Figure 11 It is a schematic flowchart of a specific example process of a task resource scheduling method in an embodiment;

[0062] Figure 12 It is a structural block diagram of a task resource scheduling device in an embodiment;

[0063] Figure 13 It is an internal structure diagram of a communication device in an embodiment. Detailed implementation manners

[0064] In order to make the objectives, technical solutions and advantages of the present application clearer and more understandable, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0065] The task resource scheduling method provided by the embodiments of the present application can be applied to an application environment as shown in Figure 1 . Among them, the terminal 102 communicates with the base station 104 through the network. In the AI RAN environment, the base station includes the RAN AI Layer (Radio Access Network Artificial Intelligence Layer, the functional layer introducing artificial intelligence technology in the radio access network). As shown in Figure 2 . As the core management and control unit of AI RAN, the RAN AI Layer includes multiple functional modules, specifically: a model management functional module, a data management service module, a service management and orchestration module, and a communication and computing resource scheduling module. Among them,

[0066] Model management functional module: Real-time monitor the usage status of the AI model, including resource occupancy, call frequency, etc. Dynamically analyze the model resource usage situation among tasks and identify potential conflicts.

[0067] Business Management Orchestration Module: Responsible for receiving task requests and calculating task priorities to generate a scheduling sequence.

[0068] General Calculation Resource Scheduling Module: According to the conflict detection results, adjust the task scheduling order and execute the resource allocation strategy.

[0069] In addition, it also includes other base station units (RRU and BBU): The RRU (Remote Radio Unit) is responsible for the transceiver of wireless signals, and the BBU (Baseband Unit) is deployed on the infrastructure layer based on CPU and GPU, integrating computing resources through virtualization technology, and cooperating with the RAN AI Layer to implement task execution.

[0070] In an exemplary embodiment, as Figure 3 shown, a task resource scheduling method is provided. Taking the base station in Figure 1 as an example for illustration, it includes the following steps 302 to step 308. Among them:

[0071] Step 302, receive concurrent task requests from each terminal.

[0072] Among them, the task request carries task attribute information.

[0073] In implementation, in the current wireless communication network, when multiple terminals concurrently send task requests to the base station, since multiple task requests may involve the invocation of the same resources (for example, model resources), therefore, resource preemption and task conflicts may occur. The base station needs to perform task scheduling management on each task request of the terminal. In this way, the base station receives concurrent task requests from two or more terminal devices at the same time. Each task request carries task attribute information, and the task attribute information may include, but is not limited to, parameters such as task objectives, roles of the tasks, model invocation requirements, priorities, etc. The specific content of the task attribute information is not limited in the embodiments of the present disclosure. The base station can parse the task attribute information carried in each task request to clarify the task requirements of each task request, so as to provide resources for each task request based on the task requirements.

[0074] Specifically, each terminal device sends a task request to the AI RAN base station through a wireless link, and the AI RAN base station forwards the request to the RAN AI Layer. The RAN AI Layer manages and orchestrates the concurrent task requests.

[0075] Step 304, based on the task attribute information, perform resource scheduling conflict detection on each task request to obtain a conflict detection result.

[0076] In implementation, the base station determines whether there is a resource scheduling conflict among the concurrent task requests based on the task attribute information of the concurrent task requests parsed from each terminal, from dimensions such as the task execution time period, the type of resources required for the task, etc., so as to obtain the conflict detection result among the concurrent task requests.

[0077] Optionally, the conflict detection result may include a list of conflicting tasks, the type of conflict (such as resource conflict, time conflict, dependency conflict, etc.), and the severity of the conflict, etc. The embodiments of the present disclosure do not limit the content information included in the conflict detection result.

[0078] Step 306, if the conflict detection result indicates that there is a resource scheduling conflict, calculate the task priority corresponding to the target task request with the resource scheduling conflict.

[0079] In implementation, after the base station performs resource scheduling conflict detection on the concurrent task requests of each terminal, if the conflict detection result obtained by the base station includes that the task corresponding to the target task request among the concurrent task requests has a resource scheduling conflict, the base station calculates the task priority corresponding to each target task request with the resource scheduling conflict based on the priority parameter included in the task attribute information of the target task request.

[0080] Step 308, sort the target task requests based on the task priority, and store the sorted target task requests in the task queue, and perform resource scheduling on the tasks of each target task request in the task queue in turn.

[0081] In implementation, after determining the task priorities corresponding to each target task request, the base station sorts the target task requests in descending order of the task priority, and the higher the priority, the more preferentially the task of the target task request is executed for resource scheduling. In this way, the base station stores the sorted target task requests in the task queue. When responding to each task request, when the task request to be responded is a target task request with a resource scheduling conflict, the base station can perform resource scheduling on the tasks of each target task request according to the order of the target task requests in the task queue to avoid the resource scheduling conflict existing among the target task requests.

[0082] Optionally, the number of task queues for task resource scheduling may be one or multiple. The embodiments of the present disclosure do not limit the number of task queues for managing the execution order.

[0083] Optionally, when there is only one task queue corresponding to task resource scheduling, when the base station faces concurrent task requests, it comprehensively sorts the target task requests with resource scheduling conflicts. For example, among the concurrent task requests, there are task request A, task request B, task request C, and task request D. Among them, there is a resource scheduling conflict that requires the same resource to be called between task request A and task request B, and there is a resource scheduling conflict that requires the same resource to be called between task request C and task request D. Therefore, task request A, task request B, task request C, and task request D are all target task requests and need to be subject to task resource scheduling. At this time, since there is only one task queue for task resource scheduling, the service management and orchestration module in the base station calculates the priorities of each task request (at this time, task requests A to D are all target task requests), and then comprehensively sorts each task request based on the priorities, and stores them in the task queue according to the sorting order of each task request after sorting. For example, after calculation, the priority of task request A is higher than that of task request B, and the priority of task request C is higher than that of task request D. At the same time, after comprehensive evaluation, it is determined that the resource scheduling tasks of task request A and task request B types have higher priorities than the resource scheduling of task request C and task request D types. Therefore, the final comprehensive sorting order of each task request is: task request A, task request B, task request C, task request D.

[0084] Optionally, when there are multiple task queues corresponding to task resource scheduling, when the base station faces concurrent task requests, for the target task requests with resource scheduling conflicts, they are sorted separately according to the different resource scheduling types of the task requests. For example, among the concurrent task requests, there are task request A, task request B, task request C, and task request D. Among them, there is a resource scheduling conflict that requires the same resource to be called between task request A and task request B, and there is a resource scheduling conflict that requires the same resource to be called between task request C and task request D. Therefore, task request A, task request B, task request C, and task request D are all target task requests and need to be subject to task resource scheduling. At this time, since there are multiple task queues for task resource scheduling, the service management and orchestration module in the base station calculates the priorities of each task request (at this time, task requests A to D are all target task requests), and then sorts the task requests of different types separately based on the priorities, and stores them in different task queues according to the sorting order of each task request after sorting. For example, after calculation, the priority of task request A is higher than that of task request B, and the priority of task request C is higher than that of task request D. Therefore, the final sorting of each task request is:

[0085] In task queue 1: task request A, task request B;

[0086] In task queue 2: task requests C and D.

[0087] In the above task resource scheduling method, based on the task attribute information of each concurrent task request, resource scheduling conflict detection is performed on each task request to obtain a conflict detection result, thereby avoiding resource scheduling conflicts between target task requests that call the same resource, increasing the accuracy and rationality of resource scheduling. Furthermore, by means of the functions of the task queue and task priority, each concurrent target task request is sorted to reduce task interruptions caused by model resource conflicts and improve the stability and execution efficiency of task resource scheduling.

[0088] In an exemplary embodiment, as Figure 4 shown, step 304 includes steps 402 to 404. Among them:

[0089] Step 402, parse the task attribute information carried by each task request to obtain the task objectives and model call requirements of each task.

[0090] In implementation, the RAN AI Layer of the base station receives task requests from the network and parses the task attribute information carried by each task request. The task attribute information contains the characteristic information of the task, such as task timeliness, resource requirements, priority rules, etc. Furthermore, taking the example that each current task involves the call of an AI model, the resource requirements of each task request are the call requirements of the AI model. Therefore, the RAN AI Layer of the base station can determine the task objectives and model call requirements of each task based on the resource requirements included in the task attribute information.

[0091] Step 404, if there are at least two tasks with the same model scheduling requirements and / or the same task objectives, determine that the conflict detection result of each task request is that there are at least two task resource scheduling conflicts.

[0092] In implementation, if there are at least two tasks with the same model scheduling requirements and / or the same task objectives, the base station RAN AI Layer determines that the conflict detection result of each task request is that there are at least two task resource scheduling conflicts. Specifically, task A corresponding to task request A is a path planning task, and task B corresponding to task request B is a dynamic environment perception task. Thus, for these two tasks, the service management and orchestration module in the RAN AI Layer receives task request A and task request B. The service management and orchestration module parses that the AI model resource that both task request A and task request B need to call is the "environment perception AI model", that is, the corresponding task objectives are the same (both may involve environment perception and environment detection), and running this "environment perception AI model" both requires GPU resources, that is, the model scheduling requirements are the same. Then the service management and orchestration module identifies that there is a resource scheduling conflict for the tasks corresponding to these two task requests. To avoid conflicts when tasks A and B call the model, the base station ensures the orderly use of model resources through a conflict detection and dynamic scheduling mechanism.

[0093] In this embodiment, by integrating a task scheduling conflict detection mechanism in the service management and orchestration module of the base station RAN AI Layer, conflict detection for each concurrent task request is performed to determine the conflict detection result of the task request, so as to avoid task resource scheduling conflicts when performing task scheduling. The stability of task resource scheduling is improved.

[0094] In an exemplary embodiment, as Figure 5 shown, the task attribute information further includes priority parameter information. When the conflict detection result in step 306 is that there is a resource scheduling conflict, the specific processing procedure for calculating the task priority corresponding to the target task request with a resource scheduling conflict includes:

[0095] Step 501, if the conflict detection result is that there are resource scheduling conflicts for at least two tasks, based on the priority parameter information, calculate the task priorities corresponding to the tasks with resource scheduling conflicts.

[0096] In implementation, when the conflict detection result indicates that there is a task resource scheduling conflict for at least two tasks, that is, when the task request involves sharing AI model resources, the conflict detection process is automatically started. The model management function module in the base station RAN AI Layer calculates the task priorities corresponding to the tasks with resource scheduling conflicts based on the priority parameter information in the task attribute information carried by the task request. For example, among concurrent task requests, if there is a resource scheduling conflict between the tasks corresponding to task request A and task request B, the model management function module in the base station calculates the task priority of the task corresponding to task request A based on the priority parameter information carried in task request A. Similarly, the model management function module in the base station calculates the task priority of the task corresponding to task request B based on the priority parameter information carried in task request B. The specific method for calculating the task priorities of each task request can be that the base station calculates the task priority of the task corresponding to the task request based on the priority parameter information in the task attribute information and a preset weighted summation algorithm, and obtains the task priority of the task corresponding to the task request. Among them, the priority parameter information included in the task attribute information includes task timeliness, task importance level, priority rules, etc. The embodiments of the present disclosure do not limit the parameter types included in the priority parameter information and the specific calculation method of the task priority of the task request.

[0097] In this embodiment, the model management function module of the base station calculates the task priorities corresponding to the two task requests with resource scheduling conflicts, and then uses the magnitude of the task priorities as the sorting basis between the two concurrent task requests, so as to effectively avoid the competition for model resources and task conflicts in a multi-task concurrent environment and ensure the execution continuity of high-priority tasks.

[0098] In an exemplary embodiment, as Figure 6 shown, in step 308, the target task requests are sorted based on the task priorities, and the sorted target task requests are stored in the task queue. The specific processing process includes:

[0099] Step 601, sort the target task requests in descending order of task priority to obtain the sorting result of each target task request.

[0100] In implementation, the model management function module in the base station RAN AI Layer sorts the detected target task requests with resource scheduling conflicts in descending order of task priority according to a preset priority sorting rule, and obtains the sorting result of each target task request.

[0101] Step 602, store each target task request in the task queue based on the sorting result.

[0102] In implementation, based on the sorting result, the model management function module stores each target task request into the task queue. Specifically, if the processing progress of the task request at the current moment has reached the target task request, after determining the sorting result of each target task request, the model management function module can directly execute the target task request that needs to be executed first, and store the other target task requests into the task queue, so as to avoid storing all the target task requests into the task queue and then immediately reading the target task request that needs to be executed first, resulting in redundant operations on the target task request that needs to be executed first.

[0103] In this embodiment, based on the sorting result of each sorted target task request, the target task request is stored into the task queue, thereby introducing a task queue adjustment strategy, reducing task interruption caused by task resource scheduling conflicts, and improving task stability and task execution efficiency.

[0104] In an exemplary embodiment, as Figure 7 shown, the task resources include resources of the artificial intelligence model type. The specific processing process of sequentially performing resource scheduling on the tasks of each target task request in step 308 includes:

[0105] Step 701, in response to the target task request in the task queue, detect the model status of the target artificial intelligence model associated with the task of the target task request.

[0106] In implementation, the base station, in response to the target task request in the task queue, detects the model status of the target artificial intelligence model associated with the task of the target task request. Specifically, the model management function module in the RAN AI Layer obtains the model resources of the target artificial intelligence model through the internal communication interface, and then detects the real-time model status of the target artificial intelligence model.

[0107] Step 702, if the model status is the idle state, determine that the target task request with the current highest priority obtains the resource scheduling permission, and leave the other target task requests with lower priority in the task queue.

[0108] In implementation, when the model state is in the idle state, the base station determines that the target task request with the highest current priority obtains the resource scheduling permission, and leaves the other target task requests with lower priority in the task queue. Taking the target artificial intelligence model as the "environment perception AI model" as an example, when the target task requests (i.e., there are resource scheduling conflicts) are task request A and task request B, and the target artificial intelligence model that both task request A and task request B need to call is the "environment perception AI model", the model management function module checks the current state of the "environment perception AI model" and finds that it is in the "idle" state (i.e., not occupied state). Then, the model management function module enables the conflict detection function to analyze the task priorities of the two tasks corresponding to task request A and task request B, and finds that the priority of task A corresponding to task request A is higher than that of task B. Then, the model management module in the base station determines that task A corresponding to task request A obtains the call permission, and marks the model state corresponding to the "environment perception AI model" as "occupied" for task A to call. And task request B (with lower priority) is left in the task queue.

[0109] In this embodiment, detecting the idle state of the target artificial intelligence model can timely allocate resources to the task request with higher priority, avoid resource idleness, and thus improve the overall resource utilization rate. Moreover, resource scheduling is performed according to the task priority to ensure that high-priority tasks can obtain resources first, reduce their waiting time, and improve the task execution efficiency.

[0110] In an exemplary embodiment, as Figure 8 shown, the method further includes:

[0111] Step 801, obtain the target artificial intelligence model based on the model call permission obtained by the target task request.

[0112] In implementation, after determining that the conflict detection is completed, the base station confirms that task A obtains the model call permission. In this way, the base station obtains the target artificial intelligence model based on the model call permission obtained by the target task request. For example, taking the "environment perception AI model" in the above embodiment as an example, the model management function module calls the "environment perception AI model" based on the confirmed model call permission of task A.

[0113] Step 802, lock the task of the target artificial intelligence model through the status identifier of the occupied state, allocate computing resources, and start executing the task of the target task request with the highest current priority.

[0114] In implementation, after invoking the target artificial intelligence model, in order to prevent other tasks from interfering with the current task's application of the target artificial intelligence model, the base station locks the task of the target artificial intelligence model through the status identifier of the occupancy status, and records the task information of the task being executed in the lock information. For example, in the lock information, it is recorded that: the occupant is task A, the expected execution time is T seconds, and the resource requirement is GPU.

[0115] In this way, after updating the status identifier of the target artificial intelligence model, the model management function module of the base station allocates computing resources for the target artificial intelligence model. For example, the general computing resource scheduling module in the base station RAN AI Layer allocates the required GPU resources for task A and starts task execution. In this way, by starting to execute the task of the current target task request with a high priority through this target artificial intelligence model, task A can successfully call the model and complete the path planning task, and the model resources will not be interfered by other tasks during this period.

[0116] In this embodiment, obtaining the target artificial intelligence model based on the task priority and locking its task can effectively ensure that high-priority tasks obtain computing resources first and start execution, thereby improving the task processing efficiency and resource utilization rate of the system. Locking the model through the status identifier of the occupancy status avoids resource conflicts and task contention, ensures that high-priority tasks can exclusively occupy the required resources, reduces waiting time, and speeds up the task completion speed.

[0117] In an exemplary embodiment, as Figure 9 shown, in the mechanism of dynamically locking the model, after executing a certain task, the target artificial intelligence model can be released. Then the method further includes:

[0118] Step 901, after the execution of each target task request is completed, update the status identifier of the target artificial intelligence model associated with the target task request to the idle state, and release the task lock on the target artificial intelligence model.

[0119] In implementation, after the execution of each target task request is completed, the model management function module in the base station RAN AI Layer updates the status identifier of the target artificial intelligence model associated with the target task request to the idle state, and releases the task lock on the target artificial intelligence model so that the tasks of other target task requests can continue to be executed. For example, after task A is completed, the model management function module updates the status identifier representing the model state of the "environmental perception AI model" to the "idle" status identifier, and at the same time clears the record of task A in the model lock information. Further, based on the processing order of the models in the task queue, the model management module confirms that task B obtains the call permission for this target artificial intelligence model according to the priority of the next task in the task queue.

[0120] In this embodiment, after each target task request is executed, the status flag of the target artificial intelligence model is updated to the idle state in a timely manner and the task lock is released, which can effectively improve the resource utilization rate and the system response efficiency. After the lock is released, the target artificial intelligence model can be immediately called by other task requests, avoiding resource idleness and ensuring that the computing resources are always in an efficient operation state. At the same time, this dynamic status update mechanism can achieve rapid resource recycling and reallocation, support efficient scheduling and execution of multiple tasks, reduce task waiting time, and improve the overall throughput of the system.

[0121] In an exemplary embodiment, as Figure 10 shown, the method further includes:

[0122] Step 1001, after the task execution of each target task request is completed, record the execution information of the task corresponding to the target task request based on the system log.

[0123] In implementation, after the task execution of each target task request is completed, the log and audit module of the RAN AI Layer of the base station will record the execution information of the task corresponding to the target task request based on the system log, so as to facilitate subsequent operations such as analyzing the execution situation of this task. Specifically, the log and audit module records the target tasks with resource scheduling conflicts. For example, the execution information of the execution processes of task A and task B, and the execution information may include, but is not limited to: model call time, resource occupancy, conflict detection and resolution process, priority adjustment history, and other information. The specific content of the execution information is not limited in the embodiments of the present disclosure. In addition, the log and audit module can also give optimization suggestions for further adjusting the base station performance after resource scheduling conflict management. For example, the performance optimization suggestion can be: "Analyze the task execution efficiency and model usage, and provide optimization suggestions for subsequent task scheduling." In this way, the system log record can achieve: providing data support, optimizing the model scheduling strategy, and improving the overall performance of the system.

[0124] In this embodiment, by recording the execution information of the task in detail, a traceable task execution history can be provided for the system, which is convenient for subsequent problem troubleshooting, performance analysis, and fault diagnosis, thereby improving the maintainability and reliability of the system. And the log record can provide data support for the statistics and analysis of task execution, help optimize the task scheduling strategy and resource allocation mechanism, and further improve the efficiency and performance of the system.

[0125] In an exemplary embodiment, as Figure 11 shown, a specific instance flowchart of a task resource scheduling method is provided, and the exemplary flowchart includes:

[0126] Step 1101, each terminal device sends multiple task requests to the AI RAN node. For example, task request A corresponding to task A, and task request B corresponding to task B. Among them, task A is a path planning task that needs to call the "environmental perception AI model" and occupy GPU resources; task B is a dynamic environmental perception task that needs to call the "environmental perception AI model".

[0127] Step 1102, the AI RAN node sends the collected task requests to the service management and orchestration module in the RAN AI Layer of the base station.

[0128] Step 1103, the service management and orchestration module parses parameters such as task objectives, model call requirements, and priorities in each task request, detects potential conflicts, and activates the conflict detection mechanism.

[0129] Step 1104, when the task request involves sharing AI model resources (i.e., there may be resource scheduling conflicts), the service management and orchestration module calculates the priorities of each task, forwards the task requests to the model management function module, and automatically activates the conflict detection process.

[0130] Step 1105, the model management function module checks the current status of the "environmental perception AI model" and finds that the "environmental perception AI model" is in the "idle" state (not occupied). Then it activates the conflict detection function to analyze the priorities of task A and task B, and finds that the priority of task A (path planning task) is higher than that of task B (dynamic environmental perception task).

[0131] Step 1106, the model management function module sends the resource requirements of task A and task B and the priorities of each task to the service management and orchestration module.

[0132] Step 1107, the service management and orchestration module determines according to the priority rules that task A can call the model resources first, and task B needs to enter the waiting queue.

[0133] Step 1108, the service management and orchestration module requests model general computing resources from the general computing resource scheduling module, and the service management and orchestration module requests to call the "environmental perception AI model" from the model management function module.

[0134] Step 1109, the model management function module distributes the "environmental perception AI model" to each AI RAN node.

[0135] Step 1110, the general computing resource scheduling module notifies the AI RAN node that task A can use the "environmental perception AI model first, and task B can use it after a certain period of time".

[0136] Step 1111, the AI RAN node performs Task A based on the "environmental perception AI model" and feeds back the task processing result to the terminal.

[0137] It should be understood that although the steps in the flowcharts involved in the above embodiments are sequentially shown according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless there is a clear description in this article, the execution of these steps has no strict order limit, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily executed at the same moment, but can be executed at different moments. The execution order of these steps or stages is not necessarily sequential, but can be executed alternately or in turn with at least a part of other steps or steps or stages in other steps.

[0138] Based on the same inventive concept, an embodiment of the present application also provides a task resource scheduling device for implementing the above-mentioned task resource scheduling method. The solution provided by this device to solve the problem is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of the task resource scheduling device provided below can refer to the limitations on the task resource scheduling method in the above text, and will not be repeated here.

[0139] In an exemplary embodiment, as Figure 12 shown, a task resource scheduling device 1200 is provided, including: a receiving module 1201, a detection module 1202, a calculation module 1203, and a processing module 1204, where:

[0140] The receiving module 1201 is configured to receive concurrent task requests from each terminal, and the task requests carry task attribute information.

[0141] The detection module 1202 is configured to perform resource scheduling conflict detection on each task request based on the task attribute information to obtain a conflict detection result.

[0142] The calculation module 1203 is configured to calculate the task priority corresponding to the target task request with a resource scheduling conflict if the conflict detection result indicates the existence of a resource scheduling conflict.

[0143] The processing module 1204 is configured to sort each target task request based on the task priority and store the sorted target task requests in a task queue, and sequentially perform resource scheduling on the tasks of each target task request in the task queue.

[0144] In one embodiment, the detection module 1202 is specifically configured to parse the task attribute information carried in each task request to obtain the task objectives and model call requirements of each task;

[0145] If there are at least two tasks with the same model scheduling requirements and / or the same task objectives, it is determined that the conflict detection result of each task request is that there are at least two task resource scheduling conflicts.

[0146] In one embodiment, the task attribute information further includes priority parameter information. The calculation module 1203 is specifically configured to, when the conflict detection result is that there are at least two task resource scheduling conflicts, calculate the task priorities corresponding to the tasks with resource scheduling conflicts based on the priority parameter information.

[0147] In one embodiment, the processing module 1204 is specifically configured to sort the target task requests in descending order of task priority to obtain the sorting result of each target task request;

[0148] Based on the sorting result, each target task request is stored in the task queue.

[0149] In one embodiment, the task resources include resources of the artificial intelligence model type. The processing module 1204 is specifically configured to, in response to the target task request in the task queue, detect the model status of the target artificial intelligence model associated with the task of the target task request;

[0150] If the model status is the idle state, it is determined that the target task request with the current higher priority obtains the resource scheduling permission, and the other target task requests with lower priority are left in the task queue.

[0151] In one embodiment, the device further includes:

[0152] Based on the model call permission obtained by the target task request, obtain the target artificial intelligence model;

[0153] Perform task locking on the target artificial intelligence model through the status identifier of the occupied state, allocate computing resources, and start executing the task of the target task request with the current higher priority.

[0154] In one embodiment, the device further includes:

[0155] After the task execution of each target task request is completed, update the status identifier of the target artificial intelligence model associated with the target task request to the idle state, and release the task locking on the target artificial intelligence model.

[0156] In one embodiment, the device further includes:

[0157] After the execution of each target task request is completed, the execution information of the task corresponding to the target task request is recorded based on the system log.

[0158] Each module in the above task resource scheduling device can be implemented in whole or in part by software, hardware, and their combination. Each of the above modules can be embedded in the processor of the communication device in hardware form or independent of it, or stored in the memory of the communication device in software form, so that the processor can call and execute the operations corresponding to each of the above modules.

[0159] The network device involved in the embodiments of the present application can be a base station, which can include multiple cells that provide services for terminals. According to different specific application scenarios, the base station can also be called an access point, or can be a device in the access network that communicates with wireless terminal devices through one or more sectors on the air interface, or other names. The network device can be used to mutually replace the received airframe and Internet Protocol (IP) packets, and act as a router between the wireless terminal device and the rest of the access network, where the rest of the access network can include an Internet Protocol (IP) communication network. The network device can also coordinate the management of the attributes of the air interface. For example, the network device involved in the embodiments of the present application can be an evolved network device (eNB or e-NodeB) in a Long Term Evolution (LTE) system, a 5G base station (gNB) in a 5G network architecture (next generation system), etc., or can also be a Home evolved Node B (HeNB), a relay node, a femto, a pico, a network test device, etc. The embodiments of the present application do not limit this. In some network structures, the network device can include a centralized unit (CU) node and a distributed unit (DU) node, and the centralized unit and the distributed unit can also be geographically separated.

[0160] Those skilled in the art can understand that Figure 13 the structure shown in

[0161] In an exemplary embodiment, a communication device is provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0162] Receive task requests concurrently sent by each terminal, where the task requests carry task attribute information;

[0163] Based on the task attribute information, perform resource scheduling conflict detection on each task request to obtain a conflict detection result;

[0164] If the conflict detection result indicates that there is a resource scheduling conflict, calculate the task priority corresponding to the target task request with the resource scheduling conflict;

[0165] Sort each target task request based on the task priority, and store the sorted target task requests in a task queue, and sequentially perform resource scheduling for the tasks of each target task request in the task queue.

[0166] In one embodiment, when the processor executes the computer program, the following steps are further implemented:

[0167] Parse the task attribute information carried by each task request to obtain the task objectives and model call requirements of each task;

[0168] If there are at least two tasks with the same model scheduling requirements and / or the same task objectives, determine that the conflict detection result of each task request is that there is a resource scheduling conflict for at least two tasks.

[0169] In one embodiment, the task attribute information further includes priority parameter information. When the processor executes the computer program, the following steps are further implemented:

[0170] If the conflict detection result indicates that there is a resource scheduling conflict for at least two tasks, calculate the task priority corresponding to the tasks with the resource scheduling conflict based on the priority parameter information.

[0171] In one embodiment, when the processor executes the computer program, the following steps are further implemented:

[0172] Sort each target task request in descending order of task priority to obtain a sorting result of each target task request;

[0173] Based on the sorting result, store each target task request in a task queue.

[0174] In one embodiment, the task resources include resources of the artificial intelligence model type. When the processor executes the computer program, the following steps are further implemented:

[0175] In response to the target task request in the task queue, detect the model status of the target artificial intelligence model associated with the task of the target task request;

[0176] When the model status is the idle status, it is determined that the target task request with the highest current priority obtains the resource scheduling permission, and the other target task requests with lower priorities are stored in the task queue.

[0177] In one embodiment, the task resources include resources of the artificial intelligence model type. When the processor executes the computer program, the following steps are further implemented:

[0178] Based on the model call permission obtained by the target task request, the target artificial intelligence model is obtained.

[0179] The target artificial intelligence model is locked for tasks through the status identifier of the occupied status, computing resources are allocated, and the task of the target task request with the highest current priority is started and executed.

[0180] In one embodiment, when the processor executes the computer program, the following steps are further implemented:

[0181] After the task execution of each target task request is completed, the status identifier of the target artificial intelligence model associated with the target task request is updated to the idle status, and the task lock on the target artificial intelligence model is released.

[0182] In one embodiment, when the processor executes the computer program, the following steps are further implemented:

[0183] After the task execution of each target task request is completed, the execution information of the task corresponding to the target task request is recorded based on the system log.

[0184] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0185] In one embodiment, a computer program product is provided, including a computer program. When the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0186] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided in this application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in this application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logic devices, data processing logics based on quantum computing, artificial intelligence (AI) processors, etc., without limitation.

[0187] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered to be within the scope recorded in this application.

[0188] The above embodiments merely represent several implementation manners of the present application. The description thereof is relatively specific and detailed, but it should not be construed as a limitation to the patent scope of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all fall within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the appended claims.

Claims

1. A task resource scheduling method, characterized in that, The method includes: Receiving task requests concurrently sent by each terminal, where the task requests carry task attribute information; Based on the task attribute information, performing resource scheduling conflict detection on each of the task requests to obtain a conflict detection result; If the conflict detection result indicates that there is a resource scheduling conflict, calculating the task priority corresponding to the target task request with a resource scheduling conflict; Sorting each of the target task requests based on the task priority and storing the sorted target task requests in a task queue, and sequentially performing resource scheduling on the tasks of each of the target task requests in the task queue.

2. The method according to claim 1, wherein The performing resource scheduling conflict detection on each of the task requests based on the task attribute information to obtain a conflict detection result includes: Parsing the task attribute information carried by each of the task requests to obtain the task objectives and model call requirements of each task; If there are at least two tasks with the same model scheduling requirements and / or the same task objectives, determining that the conflict detection result of each of the task requests is that there are at least two task resource scheduling conflicts.

3. The method according to claim 2, characterized in that, The task attribute information further includes priority parameter information. The calculating the task priority corresponding to the target task request with a resource scheduling conflict if the conflict detection result indicates that there is a resource scheduling conflict includes: If the conflict detection result indicates that there are resource scheduling conflicts for at least two tasks, calculating the task priority corresponding to the tasks with resource scheduling conflicts based on the priority parameter information.

4. The method according to claim 1, characterized in that, The sorting each of the target task requests based on the task priority and storing the sorted target task requests in a task queue includes: Sorting each of the target task requests in descending order of the task priority to obtain a sorting result of each of the target task requests; Based on the sorting result, storing each of the target task requests in a task queue.

5. The method according to any one of claims 1 to 4, characterized in that, The task resources include resources of the artificial intelligence model type. The sequentially performing resource scheduling on the tasks of each of the target task requests in the task queue includes: In response to the target task request in the task queue, detecting the model state of the target artificial intelligence model associated with the task of the target task request; If the model state is an idle state, determining that the target task request with a higher current priority obtains the resource scheduling permission and leaving the other target task requests with a lower priority in the task queue.

6. The method according to claim 5, wherein The method further includes: Based on the model call permission obtained for the target task request, obtaining the target artificial intelligence model; Locking the task of the target artificial intelligence model through the status identifier of the occupied state, allocating computing resources, and starting to execute the task of the target task request with a higher current priority.

7. The method according to claim 1 or 6, characterized in that The method further includes: After the task of each of the target task requests is completed, updating the status identifier of the target artificial intelligence model associated with the target task request to an idle state and releasing the task lock on the target artificial intelligence model.

8. The method according to claim 1, wherein The method further includes: After the task of each of the target task requests is completed, recording the execution information of the task corresponding to the target task request based on the system log.

9. A task resource scheduling device, characterized in that, The device includes: a receiving module, configured to receive task requests concurrently sent by each terminal, where the task requests carry task attribute information; a detection module, configured to perform resource scheduling conflict detection on each of the task requests based on the task attribute information to obtain a conflict detection result; a calculation module, configured to calculate the task priorities corresponding to the target task requests with resource scheduling conflicts if the conflict detection result indicates the existence of resource scheduling conflicts; a processing module, configured to sort each of the target task requests based on the task priorities and store the sorted target task requests in a task queue, and sequentially perform resource scheduling on the tasks of each of the target task requests in the task queue.

10. A communication device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 8.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 8.

12. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 8.

Images