Data transmission method and device, equipment, storage medium and computer program product

Abstract

The invention provides a data transmission method and device, equipment, a storage medium and a computer program product, and relates to the technical field of communication, and the data transmission method comprises the steps: obtaining a first transmission time delay of first task data; the first task data is transmission data of a first subtask under a first artificial intelligence (AI) task; determining an AI service corresponding to the first task data according to a first mapping relationship and the first transmission delay; the first mapping relationship comprises a corresponding relationship between the AI service and the transmission delay of the task data of the AI task; and sending the task data to the AI service. According to the scheme, the problem that the end-to-end performance of the AI task in the dynamic environment cannot be guaranteed in a data transmission scheme for computing resource scheduling in the prior art can be well solved.

Description

Technical Field

[0001] The present application relates to the field of communication technologies, and in particular, to a data transmission method, apparatus, device, storage medium, and computer program product. Background Art

[0002] A Kubernetes (open-source system) cluster consists of a control plane and a set of worker machines for running containerized applications, and these worker machines are called nodes. Each cluster requires at least one worker node to run Pods (the smallest deployment unit). Among them, the worker nodes host the Pods that make up the application load. The control plane manages the worker nodes and Pods in the cluster. In a production environment, the control plane usually runs across multiple computers, and a cluster usually runs multiple nodes to provide fault tolerance and high availability.

[0003] In the prior art, it is generally believed that the implementation of AI (Artificial Intelligence) services for base stations needs to be achieved by configuring containers through the Kubernetes framework, so as to achieve the purpose of flexible scheduling of computing resources. This implementation method needs to follow the Kubernetes framework, pre-configure the container resource size limit, and complete the deployment of container image components such as the running environment and running model required for AI services before providing low-latency AI services to users; when the user's performance requirements such as the latency of the AI task calculation process change, a request needs to be reported again to the control node in the Kubernetes framework, and containers that meet the user's needs are reconfigured according to the user's needs. The reconfiguration of containers in this process is usually in seconds or minutes, which is likely to cause the failure of AI tasks with low-latency requirements proposed by users.

[0004] Therefore, although the method of providing AI services by configuring containers through the Kubernetes framework on the base station side can achieve flexible scheduling of computing resources, it is difficult to adapt to scenarios where the user's wireless environment and task computing requirements change rapidly and the latency requirements for the proposed AI tasks are high. At the same time, the scheduling period of radio air interface resources is short, at the level of one time slot, usually 1 ms, and it is difficult to synchronize with the time of computing resource container configuration; when the user's wireless channel condition suddenly deteriorates, and the access network MAC (Media Access Control) layer quickly adjusts the MCS (Modulation and Coding Scheme) and scheduling strategy, resulting in an increase in the air interface latency of the AI task, it is difficult to make changes in a timely manner due to the long time for computing resource container configuration, thus it is difficult to meet the end-to-end latency requirements of the user's AI tasks.

[0005] Therefore, the data transmission scheme for computing resource scheduling in the prior art has problems such as being unable to guarantee the end-to-end performance of AI tasks in a dynamic environment. Summary of the Invention

[0006] The purpose of this application is to provide a data transmission method, device, equipment, storage medium and computer program product, so as to solve the problem that the data transmission scheme for computing resource scheduling in the prior art cannot guarantee the end-to-end performance of AI tasks in a dynamic environment.

[0007] To solve the above technical problems, an embodiment of this application provides a data transmission method, which is applied to a network device and includes: Obtain the first transmission delay of the first task data; the first task data is the transmission data of the first subtask under the first artificial intelligence (AI) task. Determine the AI service corresponding to the first task data according to the first mapping relationship and the first transmission delay; the first mapping relationship includes: the corresponding relationship between the AI service and the transmission delay of the task data of the AI task. Send the task data to the AI service.

[0008] Optionally, the obtaining of the first transmission delay of the first task data includes: Determine the sending time and the end time of the first task data according to the task identifier of the first AI task; the sending time is carried in the first data packet of the first task data, and the end time is the time when the last data packet of the first task data is received. Obtain the first transmission delay of the first task data according to the sending time and the end time.

[0009] Optionally, it further includes: Receive the first quality of service (QoS) flow corresponding to the first subtask sent by the terminal; the first data packet and the last data packet are carried through the first QoS flow.

[0010] Optionally, it further includes: Cache the first QoS flow to obtain the first task data when it is determined according to the second mapping relationship that the category of the first QoS flow is the AI service QoS flow. Wherein, the second mapping relationship includes: the corresponding relationship between the QoS flow identifier and the QoS flow category.

[0011] Optionally, it further includes: When it is determined according to the second mapping relationship that the category of the received QoS flow is the communication service QoS flow, forward the received QoS flow to the core network user plane function (UPF).

[0012] Optionally, it further includes: Obtain the QoS flow identifier corresponding to the first parameter information of the AI task; the first parameter information includes at least one of priority, first communication rate, and second transmission delay; According to the QoS flow identifier, obtain the second mapping relationship, the fourth mapping relationship, and the QoS parameter information corresponding to the QoS flow corresponding to the AI task; the fourth mapping relationship includes: the mapping relationship between AI task data and the QoS flow.

[0013] Optionally, it further includes: Determine the first mapping relationship according to the communication status information of the terminal, the computing status information of the AI service, and the AI task requirement information corresponding to the first AI task; the terminal is the terminal that initiates the first AI task.

[0014] Optionally, the determining the first mapping relationship according to the communication status information of the terminal, the computing status information of the AI service, and the AI task requirement information corresponding to the first AI task includes: Determine the candidate AI service corresponding to the first AI task according to the task type of the first AI task and the AI task requirement information; Determine the first mapping relationship according to the communication status information, computing status information, AI task requirement information, and candidate AI service.

[0015] Optionally, the determining the first mapping relationship according to the communication status information, computing status information, AI task requirement information, and candidate AI service includes: Obtain the difference between the end-to-end delay and the average delay; the end-to-end delay is the end-to-end delay corresponding to the sub-AI task in the AI task requirement information, and the average delay is the average delay for the candidate AI service to complete the next sub-task of the first AI task in the computing status information; Obtain the transmission delay threshold when the sub-task is executed on the candidate AI service according to the difference; Determine the first mapping relationship according to the task identifier of the first AI task, the transmission delay threshold, and the IP information of the candidate AI service.

[0016] Optionally, it further includes: Obtain the minimum resource block (RB) number requirement information corresponding to the candidate AI service for the first AI task; Determine the RB allocation strategy corresponding to the first AI task according to the minimum RB number requirement information, the data volume information of the sub-task under the first AI task, the second communication rate corresponding to the terminal, and the remaining RB quantity in the air interface; the second communication rate is the communication rate that can be achieved by allocating one RB to the terminal; Determine a third communication rate and a third transmission delay corresponding to the first AI task according to the RB allocation policy.

[0017] Optionally, obtaining the minimum resource block (RB) number requirement information corresponding to the candidate AI service for the first AI task includes: Obtain the minimum communication rate requirement information corresponding to the candidate AI service according to the data volume information of the subtasks under the first AI task and the transmission delay threshold when the subtasks are executed on the candidate AI service; Obtain a second communication rate that can be achieved by allocating one resource block (RB) to the terminal according to the second information corresponding to the terminal; the second information includes at least one of the multiple input multiple output (MIMO) layer number, modulation order, and coding rate of data transmission; Obtain the minimum RB number requirement information corresponding to the candidate AI service according to the minimum communication rate requirement information and the second communication rate.

[0018] Optionally, it further includes: Determine the user's initial AI service corresponding to the subtask according to the fourth transmission delay and the transmission delay threshold when a subtask under the first AI task is executed on the candidate AI service; the fourth transmission delay is determined according to the RB allocation policy corresponding to the first AI task; Obtain a third mapping relationship according to the IP information of the user's initial AI service and the AI task identifier of the first AI task; the third mapping relationship includes: the mapping relationship between AI task data and the AI task identifier.

[0019] Optionally, it further includes: Send at least one of the following to the terminal: The QoS parameter information corresponding to the QoS flow corresponding to the AI task; The third mapping relationship, which includes: the mapping relationship between the user's initial AI service and the AI task identifier; The fourth mapping relationship, which includes: the mapping relationship between AI task data and the QoS flow.

[0020] An embodiment of the present application further provides a data transmission method, which is applied to a terminal and includes: Obtain first task data; the first task data is the transmission data of the first subtask under the first AI task; Send the first task data to the network device according to the third mapping relationship, the fourth mapping relationship, and the QoS parameter information corresponding to the QoS flow corresponding to the AI task. Among them, the third mapping relationship includes: the mapping relationship between AI task data and AI task identifiers; the fourth mapping relationship includes: the mapping relationship between AI task data and QoS flows.

[0021] Optionally, the sending the first task data to a network device according to the third mapping relationship, the fourth mapping relationship, and QoS parameter information corresponding to the QoS flow corresponding to the AI task includes: Obtaining a second QoS flow corresponding to the first task data according to the third mapping relationship and the fourth mapping relationship; Obtaining a first QoS flow corresponding to the first subtask according to the QoS parameter information corresponding to the QoS flow corresponding to the AI task and the second QoS flow; Sending the first QoS flow to the network device, where the first QoS flow carries the first task data.

[0022] Optionally, the obtaining a second QoS flow corresponding to the first task data according to the third mapping relationship and the fourth mapping relationship includes: Packing the task data of the first subtask under the first AI task according to the third mapping relationship and the AI task data sending time information to obtain an AI task data packet corresponding to the first subtask; Performing QoS mapping on the AI task data packet according to the fourth mapping relationship to obtain a second QoS flow corresponding to the first task data.

[0023] Optionally, it further includes: Receiving at least one of the following sent by the network device: QoS parameter information corresponding to the QoS flow corresponding to the AI task; The third mapping relationship; The fourth mapping relationship.

[0024] An embodiment of this application further provides a data transmission device, which is applied to a network device and includes: A first acquisition module, configured to acquire a first transmission delay of first task data; the first task data is transmission data of a first subtask under a first artificial intelligence (AI) task; A first determination module, configured to determine an AI service corresponding to the first task data according to a first mapping relationship and the first transmission delay; the first mapping relationship includes: the corresponding relationship between the AI service and the transmission delay of the task data of the AI task; A first sending module, configured to send the task data to the AI service.

[0025] Optionally, the acquiring the first transmission delay of the first task data includes: Determine the sending time and end time of the first task data according to the task identifier of the first AI task; the sending time is carried in the first data packet of the first task data, and the end time is the time when the last data packet of the first task data is received; Obtain the first transmission delay of the first task data according to the sending time and the end time.

[0026] Optionally, it further includes: A first receiving module, configured to receive a first Quality of Service (QoS) flow corresponding to a first subtask sent by a terminal; the first data packet and the last data packet are carried through the first QoS flow.

[0027] Optionally, it further includes: A first caching module, configured to cache the first QoS flow to obtain the first task data when it is determined according to the second mapping relationship that the category of the first QoS flow is an AI service QoS flow; Wherein, the second mapping relationship includes: the corresponding relationship between the QoS flow identifier and the QoS flow category.

[0028] Optionally, it further includes: A second sending module, configured to forward the received QoS flow to the core network user plane function (UPF) when it is determined according to the second mapping relationship that the category of the received QoS flow is a communication service QoS flow.

[0029] Optionally, it further includes: A second obtaining module, configured to obtain the QoS flow identifier corresponding to the first parameter information of the AI task; the first parameter information includes at least one of priority, first communication rate, and second transmission delay; A first processing module, configured to obtain the second mapping relationship, the fourth mapping relationship, and the QoS parameter information corresponding to the QoS flow corresponding to the AI task according to the QoS flow identifier; the fourth mapping relationship includes: the mapping relationship between the AI task data and the QoS flow.

[0030] Optionally, it further includes: A second determining module, configured to determine the first mapping relationship according to the communication status information of the terminal, the computing status information of the AI service, and the AI task requirement information corresponding to the first AI task; the terminal is the terminal that initiates the first AI task.

[0031] Optionally, the determining the first mapping relationship according to the communication status information of the terminal, the computing status information of the AI service, and the AI task requirement information corresponding to the first AI task includes: Determine the candidate AI service corresponding to the first AI task according to the task type of the first AI task and the AI task requirement information; Determine the first mapping relationship according to the communication status information, the computing status information, the AI task requirement information, and the candidate AI service.

[0032] Optionally, the determining the first mapping relationship according to the communication status information, the computing status information, the AI task requirement information, and the candidate AI service includes: Obtain the difference between the end-to-end delay and the average delay; the end-to-end delay is the end-to-end delay corresponding to the sub-AI task in the AI task requirement information, and the average delay is the average delay for the candidate AI service to complete the next sub-task of the first AI task in the computing status information; Obtain the transmission delay threshold when the sub-task is executed on the candidate AI service according to the difference; Determine the first mapping relationship according to the task identifier of the first AI task, the transmission delay threshold, and the IP information of the candidate AI service.

[0033] Optionally, it further includes: A third obtaining module, configured to obtain the minimum resource block (RB) number requirement information corresponding to the candidate AI service for the first AI task; A third determining module, configured to determine the RB allocation strategy corresponding to the first AI task according to the minimum RB number requirement information, the data volume information of the sub-task under the first AI task, the second communication rate corresponding to the terminal, and the remaining RB number in the air interface; the second communication rate is the communication rate that can be achieved by allocating one RB to the terminal; A fourth determining module, configured to determine the third communication rate and the third transmission delay corresponding to the first AI task according to the RB allocation strategy.

[0034] Optionally, the obtaining the minimum resource block (RB) number requirement information corresponding to the candidate AI service for the first AI task includes: Obtain the minimum communication rate requirement information corresponding to the candidate AI service according to the data volume information of the sub-task under the first AI task and the transmission delay threshold when the sub-task is executed on the candidate AI service; Obtain the second communication rate that can be achieved by allocating one resource block (RB) to the terminal according to the second information corresponding to the terminal; the second information includes at least one of the multiple input multiple output (MIMO) layer number, modulation order, and coding rate of data transmission; Obtain the minimum RB number requirement information corresponding to the candidate AI service according to the minimum communication rate requirement information and the second communication rate.

[0035] Optionally, it further includes: A fifth determination module, configured to determine a user's initial AI service corresponding to the subtask according to a fourth transmission delay and a transmission delay threshold when a subtask under the first AI task is executed on a candidate AI service; the fourth transmission delay is determined according to an RB allocation policy corresponding to the first AI task; A second processing module, configured to obtain a third mapping relationship according to the IP information of the user's initial AI service and the AI task identifier of the first AI task; the third mapping relationship includes: a mapping relationship between AI task data and an AI task identifier.

[0036] Optionally, it further includes: A third sending module, configured to send at least one of the following to a terminal: QoS parameter information corresponding to a QoS flow corresponding to an AI task; The third mapping relationship, where the third mapping relationship includes: a mapping relationship between a user's initial AI service and an AI task identifier; A fourth mapping relationship, where the fourth mapping relationship includes: a mapping relationship between AI task data and a QoS flow.

[0037] An embodiment of the present application further provides a data transmission device, which is applied to a terminal and includes: A fourth acquisition module, configured to acquire first task data; the first task data is transmission data of a first subtask under a first AI task; A fourth sending module, configured to send the first task data to a network device according to a third mapping relationship, a fourth mapping relationship, and QoS parameter information corresponding to a QoS flow corresponding to an AI task; Wherein, the third mapping relationship includes: a mapping relationship between AI task data and an AI task identifier; the fourth mapping relationship includes: a mapping relationship between AI task data and a QoS flow.

[0038] Optionally, the sending the first task data to the network device according to a third mapping relationship, a fourth mapping relationship, and QoS parameter information corresponding to a QoS flow corresponding to an AI task includes: Obtaining a second QoS flow corresponding to the first task data according to the third mapping relationship and the fourth mapping relationship; Obtaining a first QoS flow corresponding to the first subtask according to QoS parameter information corresponding to a QoS flow corresponding to an AI task and the second QoS flow; Sending the first QoS flow to the network device, where the first QoS flow carries the first task data.

[0039] Optionally, obtaining the second QoS flow corresponding to the first task data according to the third mapping relationship and the fourth mapping relationship includes: Packaging the task data of the first subtask under the first AI task according to the third mapping relationship and the AI task data sending time information to obtain the AI task data packet corresponding to the first subtask; Performing QoS mapping on the AI task data packet according to the fourth mapping relationship to obtain the second QoS flow corresponding to the first task data.

[0040] Optionally, it further includes: A second receiving module, configured to receive at least one of the following sent by a network device: The QoS parameter information corresponding to the QoS flow corresponding to the AI task; The third mapping relationship; The fourth mapping relationship.

[0041] An embodiment of the present application further provides a data transmission device, where the data transmission device is a network device, including: a processor and a transceiver; The processor is configured to obtain the first transmission delay of the first task data; the first task data is the transmission data of the first subtask under the first artificial intelligence (AI) task; Determine the AI service corresponding to the first task data according to the first mapping relationship and the first transmission delay; the first mapping relationship includes: the correspondence between the AI service and the transmission delay of the task data of the AI task; Send the task data to the AI service through the transceiver.

[0042] Optionally, obtaining the first transmission delay of the first task data includes: Determine the sending time and the end time of the first task data according to the task identifier of the first AI task; the sending time is carried in the first data packet of the first task data, and the end time is the time when the last data packet of the first task data is received; Obtain the first transmission delay of the first task data according to the sending time and the end time.

[0043] Optionally, the processor is further configured to: Receive, through the transceiver, the first quality of service (QoS) flow corresponding to the first subtask sent by a terminal; the first data packet and the last data packet are carried through the first QoS flow.

[0044] Optionally, the processor is further configured to: In the case that the class of the first QoS flow is determined to be an AI service QoS flow according to the second mapping relationship, cache the first QoS flow to obtain the first task data; Wherein, the second mapping relationship includes: the correspondence between the QoS flow identifier and the QoS flow class.

[0045] Optionally, the processor is further configured to: In the case that the class of the received QoS flow is determined to be a communication service QoS flow according to the second mapping relationship, forward the received QoS flow to the core network user plane function UPF through the transceiver.

[0046] Optionally, the processor is further configured to: Obtain the QoS flow identifier corresponding to the first parameter information of the AI task; the first parameter information includes at least one of: priority, first communication rate, and second transmission delay; According to the QoS flow identifier, obtain the second mapping relationship, the fourth mapping relationship, and the QoS parameter information corresponding to the QoS flow corresponding to the AI task; the fourth mapping relationship includes: the mapping relationship between the AI task data and the QoS flow.

[0047] Optionally, the processor is further configured to: Determine the first mapping relationship according to the communication state information of the terminal, the computing state information of the AI service, and the AI task requirement information corresponding to the first AI task; the terminal is the terminal that initiates the first AI task.

[0048] Optionally, the determining the first mapping relationship according to the communication state information of the terminal, the computing state information of the AI service, and the AI task requirement information corresponding to the first AI task includes: Determine the candidate AI service corresponding to the first AI task according to the task type of the first AI task and the AI task requirement information; Determine the first mapping relationship according to the communication state information, the computing state information, the AI task requirement information, and the candidate AI service.

[0049] Optionally, the determining the first mapping relationship according to the communication state information, the computing state information, the AI task requirement information, and the candidate AI service includes: Obtain the difference between the end-to-end delay and the average delay; the end-to-end delay is the end-to-end delay corresponding to the sub-AI task in the AI task requirement information, and the average delay is the average delay of the candidate AI service completing the next sub-task of the first AI task in the computing state information; Based on the difference, obtain the transmission delay threshold when the subtask is executed on the candidate AI service; Based on the task identifier of the first AI task, the transmission delay threshold, and the IP information of the candidate AI service, determine the first mapping relationship.

[0050] Optionally, the processor is further configured to: Obtain the minimum resource block (RB) number requirement information corresponding to the candidate AI service for the first AI task; Based on the minimum RB number requirement information, the data volume information of the subtask under the first AI task, the second communication rate corresponding to the terminal, and the remaining RB number in the air interface, determine the RB allocation strategy for the first AI task; the second communication rate is the communication rate that can be achieved by allocating one RB to the terminal; Based on the RB allocation strategy, determine the third communication rate and the third transmission delay corresponding to the first AI task.

[0051] Optionally, the obtaining the minimum resource block (RB) number requirement information corresponding to the candidate AI service for the first AI task includes: Based on the data volume information of the subtask under the first AI task and the transmission delay threshold when the subtask is executed on the candidate AI service, obtain the minimum communication rate requirement information corresponding to the candidate AI service; Based on the second information corresponding to the terminal, obtain the second communication rate that can be achieved by allocating one resource block (RB) to the terminal; the second information includes at least one of the number of multiple-input multiple-output (MIMO) layers for data transmission, the modulation order, and the coding rate; Based on the minimum communication rate requirement information and the second communication rate, obtain the minimum RB number requirement information corresponding to the candidate AI service.

[0052] Optionally, the processor is further configured to: Based on the fourth transmission delay and the transmission delay threshold when a subtask under the first AI task is executed on the candidate AI service, determine the user's initial AI service corresponding to the subtask; the fourth transmission delay is determined according to the RB allocation strategy corresponding to the first AI task; Based on the IP information of the user's initial AI service and the AI task identifier of the first AI task, obtain the third mapping relationship; the third mapping relationship includes: the mapping relationship between AI task data and the AI task identifier.

[0053] Optionally, the processor is further configured to: Through the transceiver, send at least one of the following to the terminal: QoS parameter information corresponding to the QoS flow corresponding to the AI task; A third mapping relationship, where the third mapping relationship includes: the mapping relationship between the user's initial AI service and the AI task identifier; A fourth mapping relationship, where the fourth mapping relationship includes: the mapping relationship between AI task data and the QoS flow.

[0054] An embodiment of the present application further provides a data transmission device, where the data transmission device is a terminal, including: a processor and a transceiver; The processor is configured to obtain first task data; the first task data is the transmission data of the first subtask under the first AI task; According to the third mapping relationship, the fourth mapping relationship, and the QoS parameter information corresponding to the QoS flow corresponding to the AI task, send the first task data to the network device through the transceiver.

[0055] Optionally, the step of sending the first task data to the network device through the transceiver according to the third mapping relationship, the fourth mapping relationship, and the QoS parameter information corresponding to the QoS flow corresponding to the AI task includes: According to the third mapping relationship and the fourth mapping relationship, obtain the second QoS flow corresponding to the first task data; According to the QoS parameter information corresponding to the QoS flow corresponding to the AI task and the second QoS flow, obtain the first QoS flow corresponding to the first subtask; Through the transceiver, send the first QoS flow to the network device, where the first QoS flow carries the first task data.

[0056] Optionally, the step of obtaining the second QoS flow corresponding to the first task data according to the third mapping relationship and the fourth mapping relationship includes: According to the third mapping relationship and the AI task data sending time information, package the task data of the first subtask under the first AI task to obtain the AI task data packet corresponding to the first subtask; According to the fourth mapping relationship, perform QoS mapping on the AI task data packet to obtain the second QoS flow corresponding to the first task data.

[0057] Optionally, the processor is further configured to: Receive, through the transceiver, at least one of the following sent by the network device: The QoS parameter information corresponding to the QoS flow corresponding to the AI task; The third mapping relationship; The fourth mapping relationship.

[0058] An embodiment of the present application further provides a data transmission device, including a memory, a processor, and a program stored on the memory and executable on the processor; when the processor executes the program, it implements the data transmission method on the network device side or the terminal side as described above.

[0059] An embodiment of the present application further provides a readable storage medium, on which a program is stored, and when the program is executed by a processor, it implements the steps in the data transmission method on the network device side or the terminal side as described above.

[0060] An embodiment of the present application further provides a computer program product, including computer instructions, and when the computer instructions are executed by a processor, they implement the steps of the data transmission method on the network device side or the terminal side as described above.

[0061] The beneficial effects of the above technical solutions of the present application are as follows: In the above solution, the data transmission method obtains the first transmission delay of the first task data; the first task data is the transmission data of the first subtask under the first artificial intelligence (AI) task; according to the first mapping relationship and the first transmission delay, it determines the AI service corresponding to the first task data; the first mapping relationship includes: the corresponding relationship between the AI service and the transmission delay of the task data of the AI task; and sends the task data to the AI service; it can support the realization that the called AI service can dynamically adapt to the transmission delay of the task data (the change of air interface resource scheduling), that is, the AI task computing resource scheduling can dynamically adapt to the change of air interface resource scheduling (can be synchronized with the air interface resource scheduling in real time to a certain extent), and can support changing the called computing resources according to the real-time communication conditions (dynamic environment), meeting the end-to-end delay requirements of the user's AI task, and well solving the problem that the data transmission scheme for computing resource scheduling in the prior art cannot guarantee the end-to-end performance of the AI task in a dynamic environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 Schematic diagram of the data transmission method process for the embodiment of the present application Figure 1 ; Figure 2 Schematic diagram of the data transmission method process for the embodiment of the present application Figure 2 ; Figure 3 Schematic diagram of the functional framework of the AI computing service data adaptation protocol for the embodiment of the present application Figure 1 ; Figure 4 Schematic diagram of the functional framework of the AI computing service data adaptation protocol for the embodiment of the present application Figure 2 ; Figure 5 Schematic diagram of the specific implementation framework of the data transmission method for the embodiment of the present application; Figure 6 Structural diagram of the data transmission device according to the embodiment of the present application Figure 1 ; Figure 7 Structural diagram of the data transmission device according to the embodiment of the present application Figure 2 ; Figure 8 Structural diagram of the data transmission equipment according to the embodiment of the present application Figure 1 ; Figure 9 Structural diagram of the data transmission equipment according to the embodiment of the present application Figure 2 . Detailed implementation manners

[0063] To make the technical problems, technical solutions and advantages to be solved by the present application clearer, the following will be described in detail with reference to the accompanying drawings and specific embodiments.

[0064] In view of the problem that the existing data transmission scheme for computing resource scheduling in the prior art cannot guarantee the end-to-end performance of AI tasks in a dynamic environment, the present application provides a data transmission method, which is applied to a network device, such as Figure 1 shown, and includes: Step 11: (Receiving first task data) Obtaining the first transmission delay of the first task data; the first task data is the transmission data of the first subtask under the first artificial intelligence (AI) task; Step 12: Determining the AI service corresponding to the first task data according to the first mapping relationship and the first transmission delay; the first mapping relationship includes: the corresponding relationship between the AI service and the transmission delay of the task data of the AI task; Step 13: Sending the task data to the AI service.

[0065] Among them, determining the AI service may include: determining information such as the IP address of the AI service (such as determining the destination IP address of the data packet), which is not limited herein.

[0066] The data transmission method provided by the embodiment of the present application obtains the first transmission delay of the first task data; the first task data is the transmission data of the first subtask under the first artificial intelligence (AI) task; determines the AI service corresponding to the first task data according to the first mapping relationship and the first transmission delay; the first mapping relationship includes: the corresponding relationship between the AI service and the transmission delay of the task data of the AI task; and sends the task data to the AI service. It can support the realization that the called AI service can dynamically adapt to the transmission delay of the task data (the change of radio resource scheduling), that is, the computing resource scheduling of the AI task can dynamically adapt to the change of radio resource scheduling (can achieve real-time synchronization with the radio resource scheduling to a certain extent), and can support changing the called computing resources according to the real-time communication conditions (dynamic environment) to meet the end-to-end delay requirements of the user's AI task, and can well solve the problem that the data transmission scheme for computing resource scheduling in the prior art cannot guarantee the end-to-end performance of the AI task in a dynamic environment.

[0067] Among them, the obtaining of the first transmission delay of the first task data includes: determining the sending time and the end time of the first task data according to the task identifier of the first AI task; the sending time is carried in the first data packet of the first task data, and the end time is the time when the last data packet of the first task data is received; and obtaining the first transmission delay of the first task data according to the sending time and the end time. In this way, the first transmission delay can be accurately obtained.

[0068] Further, the data transmission method further includes: receiving the first quality of service (QoS) flow corresponding to the first subtask sent by the terminal; the first data packet and the last data packet are carried by the first QoS flow. In this way, the transmission of the first task data can be specifically realized.

[0069] In the embodiment of the present application, the data transmission method further includes: caching the first QoS flow to obtain the first task data when it is determined according to the second mapping relationship that the category of the first QoS flow is an AI service QoS flow; where the second mapping relationship includes: the corresponding relationship between the QoS flow identifier and the QoS flow category. In this way, the shunting of QoS can be supported and realized.

[0070] Further, the data transmission method further includes: forwarding the received QoS flow to the core network user plane function (UPF) when it is determined according to the second mapping relationship that the category of the received QoS flow is a communication service QoS flow. In this way, the shunting of QoS can be further supported and realized.

[0071] In an embodiment of the present application, the data transmission method further includes: obtaining a QoS flow identifier corresponding to first parameter information of an AI task; the first parameter information includes at least one of priority, a first communication rate, and a second transmission delay; according to the QoS flow identifier, obtaining the second mapping relationship, the fourth mapping relationship, and QoS parameter information corresponding to the QoS flow corresponding to the AI task; the fourth mapping relationship includes: the mapping relationship between AI task data and the QoS flow. In this way, the second mapping relationship, the fourth mapping relationship, and the QoS parameter information corresponding to the QoS flow corresponding to the AI task can be accurately obtained.

[0072] Further, the data transmission method further includes: determining the first mapping relationship according to the communication state information of the terminal, the computing state information of the AI service, and the AI task requirement information corresponding to the first AI task; the terminal is the terminal that initiates the first AI task. In this way, the first mapping relationship can be accurately obtained.

[0073] Among them, the determining the first mapping relationship according to the communication state information of the terminal, the computing state information of the AI service, and the AI task requirement information corresponding to the first AI task includes: determining a candidate AI service corresponding to the first AI task according to the task type of the first AI task and the AI task requirement information; determining the first mapping relationship according to the communication state information, the computing state information, the AI task requirement information, and the candidate AI service. In this way, the first mapping relationship can be specifically obtained.

[0074] In an embodiment of the present application, the determining the first mapping relationship according to the communication state information, the computing state information, the AI task requirement information, and the candidate AI service includes: obtaining the difference between the end-to-end delay and the average delay; the end-to-end delay is the end-to-end delay corresponding to a sub-AI task (i.e., a subtask under the AI task) in the AI task requirement information, and the average delay is the average delay for the candidate AI service to complete (or execute) a subtask corresponding to the first AI task in the computing state information; according to the difference, obtaining a transmission delay threshold when the subtask is executed on the candidate AI service; determining the first mapping relationship according to the task identifier of the first AI task, the transmission delay threshold, and the IP information of the candidate AI service. In this way, the first mapping relationship can be more specifically obtained.

[0075] Further, the data transmission method further includes: obtaining the minimum resource block (RB) number requirement information corresponding to the candidate AI service for the first AI task; determining the RB allocation strategy for the first AI task according to the minimum RB number requirement information, the data volume information of the subtasks under the first AI task, the second communication rate corresponding to the terminal, and the remaining RB number in the air interface; the second communication rate is the communication rate that can be achieved by allocating one RB to the terminal (or understood as the communication rate that the terminal can achieve after allocating one RB to the terminal); determining the third communication rate and the third transmission delay corresponding to the first AI task according to the RB allocation strategy. In this way, the third communication rate and the third transmission delay can be accurately obtained, thereby supporting the acquisition of the first parameter information of the AI task. For example, the third communication rate and the third transmission delay are respectively used as the first communication rate and the second transmission delay, but it is not limited thereto.

[0076] Among them, the obtaining of the minimum resource block (RB) number requirement information corresponding to the candidate AI service for the first AI task includes: obtaining the minimum communication rate requirement information corresponding to the candidate AI service according to the data volume information of the subtasks under the first AI task and the transmission delay threshold when the subtasks are executed on the candidate AI service; obtaining the second communication rate that can be achieved by allocating one resource block (RB) to the terminal according to the second information corresponding to the terminal; the second information includes at least one of the multiple input multiple output (MIMO) layer number, modulation order, and coding rate of data transmission; obtaining the minimum RB number requirement information corresponding to the candidate AI service according to the minimum communication rate requirement information and the second communication rate. In this way, the minimum resource block (RB) number requirement information can be accurately obtained.

[0077] Further, the data transmission method further includes: determining the user's initial AI service corresponding to the subtask according to the fourth transmission delay and the transmission delay threshold when a subtask under the first AI task is executed on the candidate AI service; the fourth transmission delay is determined according to the RB allocation strategy corresponding to the first AI task; obtaining the third mapping relationship according to the IP information of the user's initial AI service and the AI task identifier of the first AI task; the third mapping relationship includes: the mapping relationship between the AI task data and the AI task identifier. In this way, the third mapping relationship can be accurately obtained.

[0078] In the embodiment of the present application, the data transmission method further includes: sending at least one of the following to the terminal: QoS parameter information corresponding to the QoS flow corresponding to the AI task; a third mapping relationship, where the third mapping relationship includes: a mapping relationship between the user's initial AI service and the AI task identifier, and a fourth mapping relationship; the fourth mapping relationship includes: a mapping relationship between the AI task data and the QoS flow. This can support the terminal to accurately obtain at least one of the QoS parameter information corresponding to the QoS flow corresponding to the AI task, the third mapping relationship, and the fourth mapping relationship.

[0079] The embodiment of the present application also provides a data transmission method, which is applied to a terminal, as Figure 2 shown, and includes: Step 21: Obtain first task data; the first task data is the transmission data of the first subtask under the first AI task; Step 22: Send the first task data to the network device according to the third mapping relationship, the fourth mapping relationship, and the QoS parameter information corresponding to the QoS flow corresponding to the AI task; where the third mapping relationship includes: a mapping relationship between the AI task data and the AI task identifier; the fourth mapping relationship includes: a mapping relationship between the AI task data and the QoS flow.

[0080] The data transmission method provided by the embodiment of the present application obtains first task data; the first task data is the transmission data of the first subtask under the first AI task; and sends the first task data to the network device according to the third mapping relationship, the fourth mapping relationship, and the QoS parameter information corresponding to the QoS flow corresponding to the AI task; where the third mapping relationship includes: a mapping relationship between the AI task data and the AI task identifier; the fourth mapping relationship includes: a mapping relationship between the AI task data and the QoS flow; and can support the network device to: obtain the first transmission delay of the first task data; the first task data is the transmission data of the first subtask under the first artificial intelligence AI task; determine the AI service corresponding to the first task data according to the first mapping relationship and the first transmission delay; the first mapping relationship includes: a corresponding relationship between the AI service and the transmission delay of the task data of the AI task; and send the task data to the AI service; thereby being able to support the realization that the called AI service can dynamically adapt to the transmission delay of the task data (changes in air interface resource scheduling), that is, the AI task computing resource scheduling can dynamically adapt to the changes in air interface resource scheduling (can be synchronized with the air interface resource scheduling in real time to a certain extent), and can support changing the called computing resources according to the real-time communication conditions (dynamic environment) to meet the end-to-end delay requirements of the user's AI task, and well solve the problem that the data transmission scheme for computing resource scheduling in the prior art cannot guarantee the end-to-end performance of the AI task in a dynamic environment.

[0081] Among them, sending the first task data to a network device according to the third mapping relationship, the fourth mapping relationship, and the QoS parameter information corresponding to the QoS flow corresponding to the AI task includes: obtaining a second QoS flow corresponding to the first task data according to the third mapping relationship and the fourth mapping relationship; obtaining a first QoS flow corresponding to the first subtask according to the QoS parameter information corresponding to the QoS flow corresponding to the AI task and the second QoS flow; and sending the first QoS flow to the network device, where the first QoS flow carries the first task data. In this way, the transmission of the first task data can be specifically implemented.

[0082] In an embodiment of the present application, obtaining a second QoS flow corresponding to the first task data according to the third mapping relationship and the fourth mapping relationship includes: packing the task data of the first subtask under the first AI task according to the third mapping relationship and the AI task data sending time information to obtain an AI task data packet corresponding to the first subtask; and performing QoS mapping on the AI task data packet according to the fourth mapping relationship to obtain a second QoS flow corresponding to the first task data. In this way, the second QoS flow can be accurately obtained.

[0083] Further, the data transmission method further includes: receiving at least one of the following sent by the network device: the QoS parameter information corresponding to the QoS flow corresponding to the AI task; the third mapping relationship; the fourth mapping relationship. In this way, at least one of the QoS parameter information corresponding to the QoS flow corresponding to the AI task, the third mapping relationship, and the fourth mapping relationship can be accurately obtained.

[0084] It should be noted here that the relevant content on the network device side and the terminal side can be referred to each other, and the repeated parts will not be elaborated.

[0085] Next, an example of the data transmission method provided in the embodiment of the present application is given. The network device is taken as a base station, and the AI service is taken as a service provided by the Kubernetes framework.

[0086] In view of the above technical problems, an embodiment of the present application provides a data transmission method, which can be specifically implemented as a method for the AI computing service data adaptation protocol function and configuration on the base station side, mainly involving: designing a CSDAP (Compute Service Data Adaptation Protocol) function between the communication user plane and the container providing the AI service (this function can be deployed on the base station), and this function can support the real-time mapping of user plane data to the AI service provided by the container according to the user's AI task requirements and real-time communication conditions, so that the computing resource scheduling can dynamically adapt to the changes in the air interface resource scheduling (to a certain extent, realizing the real-time synchronization of the AI task computing resource scheduling and the air interface resource scheduling); at the same time, this solution also provides a control plane AI task analysis function for configuring the mapping table of the user plane AI task data to the AI service recognizable by the Kubernetes framework (which can correspond to the above first mapping relationship); the content involved in this solution will be specifically introduced by examples below.

[0087] 1. The CSDAP function framework of the AI computing service data adaptation protocol; Considering that the base station side configures containers through the Kubernetes framework to provide AI services, combining the access network protocol stack on the base station side and the Kubernetes framework, a control plane AI task analysis function and a user plane CSDAP are newly added, specifically as follows Figure 3 shown.

[0088] When a user makes an AI task request, the framework can analyze through the AI task analysis function according to the user's current communication status information, current computing status information, and AI task requirement information, and configure the service (corresponding to the POD or container with different resource configurations for task execution) mapping rules with different communication transmission delays for this AI task of this user on the base station side, and send the configuration information to the CSDAP (which can correspond to determining the first mapping relationship according to the communication status information of the terminal, the computing status information of the AI service, and the AI task requirement information corresponding to the first AI task; the terminal is the terminal that initiates the first AI task); thus, during the transmission process of the user plane AI task data, according to the transmission situation (such as transmission delay) of the AI task under different communication conditions, the real-time dynamic mapping of the AI task data to the executed POD or container can be realized (which can correspond to obtaining the first transmission delay of the first task data; the first task data is the transmission data of the first subtask under the first artificial intelligence AI task; determining the AI service corresponding to the first task data according to the first mapping relationship and the first transmission delay; the first mapping relationship includes: the corresponding relationship between the AI service and the transmission delay of the task data of the AI task; sending the task data to the AI service). The functions in the framework are specifically illustrated as follows: 1. User Plane - CSDAP Function; 1) Mapping Layer; The main functions of the mapping layer include: based on the transmission delay of a subtask of the task recorded by the timer (i.e., the duration required from the terminal side to the completion of reception by the base station's cache layer), and the mapping table configured by the control plane AI task analysis function (which can correspond to the above - mentioned first mapping relationship), modify the destination IP address (i.e., the Cluster IP of the Service) of all TCP / UDP (i.e., transmission control plane or user data plane) data packets of the subtask stored in the cache layer, so as to be able to change in real - time the AI service corresponding to the data packets of the subtask of the AI task; that is, determine the AI service corresponding to the task data of the subtask and transmit it to the AI service.

[0089] Among them, the services provided by PODs with different resource configurations for the same AI task can be regarded as different AI services; due to the differences in resource configurations, the latencies of these PODs to complete the same AI task are different; when the user's communication conditions change, resulting in a higher transmission delay of AI task data and it is difficult to meet the end - to - end latency requirements of the task, the mapping layer function can map the corresponding AI task data to the services provided by PODs with more resource configurations (i.e., transmit the task data to the AI service with more available resources), in order to meet the end - to - end latency requirements of the AI task, so that the computing resource scheduling can dynamically adapt to the changes in the air interface resource scheduling.

[0090] Specifically, for example Figure 4 As shown, the AI task analysis function transmits the mapping table (which can correspond to the above - mentioned first mapping relationship) to the mapping layer; the cache layer transmits the AI task data (which can carry the task ID) to the mapping layer, and the cache layer transmits information such as the sending time of the AI task data to the timer, so that the counter determines the transmission delay of the AI task data and transmits it to the mapping layer; the mapping layer maps the AI task data to the corresponding AI service according to the mapping table and the transmission delay, such as at least one of AI service 1 (corresponding to Cluster (cluster) IP1), AI service 2 (corresponding to Cluster IP2), or AI service n (corresponding to Cluster IPn), etc.

[0091] In addition, the mapping table format received by the mapping layer function from the AI task analysis function can be as shown in the following table. The mapping layer function can store the mapping tables of multiple task IDs. The number of task IDs can depend on the AI services and the corresponding Cluster IP address information obtained by the AI task analysis function from Kube-proxy (proxy). Among them, the transmission delay thresholds T1 to Tn corresponding to the AI tasks (specifically, the transmission delay threshold corresponding to a subtask under the AI task) can be analyzed by the control plane AI task analysis function according to the requirements of the AI task and the current computing resource status information. When the recorded transmission delay (i.e., the duration obtained by the timer in CSDAP) exceeds the range (i.e., exceeds the maximum threshold), it represents the failure of the AI task.

[0092]

[0093] 2) Cache layer; The main functions of the cache layer include: receiving the AI task data packets separated by the shunt layer and classifying and packing them into sets according to task IDs; when the data packets of a certain task ID are complete (specifically, the data packets of a subtask are complete), then sending the set of data packets corresponding to the AI task ID to the mapping layer together.

[0094] In this solution, considering that the mapping layer may map the data packets belonging to a unit AI task (i.e., a subtask) to the services provided by different PODs. Therefore, in order to ensure that the data packets of the unit AI task are mapped to the services provided by the same POD, the cache layer is added to cache the data packets of the unit AI task, and a timer can be assisted to monitor the transmission delay of the data of the unit AI task in real time. Among them, the unit AI task refers to the smallest AI task that can be processed in the POD process (for example, one of the multiple pictures in an image recognition task); the AI task corresponding to a task ID can be composed of multiple unit AI tasks.

[0095] 3) Timer; The main functions of the timer include: monitoring the data transmission delay of unit AI tasks. Specifically, when the user (i.e., the terminal) sends an AI task data packet, it can attach the task ID (which can correspond to the above-mentioned task identifier) and the sending time information (which can correspond to the sending time of the above-mentioned first task data); when the cache layer receives the first data packet of a unit AI task (which can correspond to the first data packet of the above-mentioned first task data), it can send the task ID and the sending time information carried in the data packet to the timer, and the timer can establish a sending timestamp for this unit AI task; when the cache layer completely receives all the data packets of this unit AI task, it can feedback the information that the reception of the data packets of this unit task is completed to the timer; at this time, the timer establishes a reception completion timestamp for this unit AI task, and obtains the transmission delay of this unit AI task according to the difference between the sending time and the reception completion time (which can correspond to determining the sending time and the end time of the first task data according to the task identifier of the above-mentioned first AI task; the sending time is carried in the first data packet of the first task data, and the end time is the time when the last data packet of the first task data is received; according to the sending time and the end time, obtain the first transmission delay of the first task data), and send the task ID corresponding to this unit AI task and the transmission delay information to the mapping layer function.

[0096] 4) Shunting layer; The main functions of the shunting layer include: receiving the QoS (Quality of Service) flow sent by the SDAP (Service Data Adaptation Protocol) layer of the communication user plane protocol stack, distinguishing traditional communication services and AI services processed at the base station through the QoS flow information, shunting the QoS flow of traditional communication services to the UPF (User Plane Function), and unpacking the QoS flow of AI services to be processed on the base station side into service data streams and sending them to the cache layer; which can correspond to receiving the first Quality of Service QoS flow corresponding to the first subtask sent by the terminal; the first data packet and the last data packet are carried through the first QoS flow; in the case of determining that the category of the first QoS flow is an AI service QoS flow according to the second mapping relationship, caching the first QoS flow to obtain the first task data; where the second mapping relationship includes: the corresponding relationship between the QoS flow identifier and the QoS flow category; in the case of determining that the category of the received QoS flow is a communication service QoS flow according to the second mapping relationship, forwarding the received QoS flow to the core network user plane function UPF.

[0097] In this solution, it is considered to use new QoS flows to carry and distinguish AI service data. The control-plane AI task analysis function will configure and send down the mapping rules for AI task data to new QoS flows (which can correspond to the above fourth mapping relationship) to the user-side QoS rule (rule function or module), and will also inform the traffic splitting layer which new QoS flows need to be processed on the base station side (which can correspond to configuring the above second mapping relationship); at the same time, the control-plane AI task analysis function will inform the communication protocol stacks on both the base station side and the user side of the QoS parameters corresponding to the new QoS flows (which can correspond to the QoS parameter information corresponding to the QoS flows corresponding to the above AI tasks). Among them, data packets with different AI task IDs can be mapped to the same QoS flow.

[0098] Before the AI task data is mapped through the QoS rule on the user side, the user side can package the AI task data sending time information, task ID, and data packet label for the base station side cache layer to unpack the information for corresponding processing.

[0099] 2. Control plane - AI task analysis function; 1) Communication status information acquisition function; Communication status information acquisition: Whenever the user's communication conditions change, resulting in changes in the MCS (Modulation and Coding Scheme) and the number of MIMO layers transmitted by the user, the user's MAC layer can be passively triggered to report communication status information. The communication status information can include the number of MIMO layers transmitted by the user, modulation order, code rate, and the remaining (or configurable) RB (Resource Block) number in the air interface. The reported information (i.e., communication status information) can be stored in the communication status information acquisition function and is ready to be called by the AI task delay analysis function at any time. Figure 3 In the above, PHY represents the physical layer, RLC represents the radio link control layer, and PDCP represents the packet data convergence protocol layer.

[0100] 2) Computing status information acquisition function; Computing status information acquisition: Based on the Kubernetes framework, where Kube-proxy runs on each node and listens for changes in services and endpoints in the API server. Once the services and endpoints on the base station node change, the information reporting of the service and the corresponding endpoints can be passively triggered; while Kubelet monitors the resource configuration of each POD and the delay size for completing a unit AI task, and then reports the average delay for the POD to complete a unit AI task, the model used, and the corresponding computing accuracy to the computing information status acquisition function. The computing status information acquisition function can establish an information table (such as the following table) based on the acquired information for the AI task delay analysis function to call.

[0101]

[0102] 3) AI task latency analysis function; Based on the information in the communication status information acquisition function and the computing status information acquisition function, when a user puts forward an AI task requirement, the AI task latency analysis function can first determine the candidate PODs that can provide services for the AI task according to the AI task type and the requirements for the computing accuracy in the requirement (which can correspond to determining the candidate AI services corresponding to the first AI task according to the task type of the first AI task and the AI task requirement information), and then configure the user plane information according to the obtained communication status information and computing status information. The user plane information can include the AI service mapping table of the base station side CSDAP mapping layer, the QoS flow identification table of the shunt layer, the QoS parameter information corresponding to the new QoS flow (the flow for carrying AI data) of the radio protocol stack, as well as the new QoS flow mapping table in the QoS rule of the user side and the task ID mapping table required for packing AI task data packets, as Figure 5 shown (which can include determining the first mapping relationship according to the communication status information, computing status information, AI task requirement information, and candidate AI services).

[0103] Among them, the AI service mapping table (which can correspond to the above first mapping relationship) has been listed above and can be used by the base station side CSDAP mapping layer to indicate the AI services mapped by AI tasks under different transmission latencies. These AI services can be provided by PODs with different resource sizes.

[0104] The QoS flow identification table (which can correspond to the above second mapping relationship) can be used by the base station side CSDAP shunt layer to distinguish the AI service QoS flows that need to be specially processed on the base station side from the communication service QoS flows that need to be transmitted to the core network UPF traditionally.

[0105] The new QoS flow mapping table (which can correspond to the above fourth mapping relationship) can indicate the new QoS flows that need to be mapped for processing different AI task data on the base station side, echoing the QoS flow identification table.

[0106] The QoS parameter information corresponding to the new QoS flow (such as the user plane configuration corresponding to the QoS flow, for example, how much resources are needed, etc.) can be used to configure the parameters of the new QoS flow in the radio protocol stack, which can support the alignment of the configurations between the user side and the base station side. The QoS parameter information corresponding to the new QoS flow can correspond to the QoS parameter information corresponding to the QoS flow corresponding to the AI task.

[0107] The task ID mapping table (which can correspond to the above-mentioned third mapping relationship) can be used to distinguish task IDs for task data packets (indicating the mapping relationship between AI task data and AI task identifiers), so as to facilitate identification by the cache layer. The task ID in this mapping table can indicate the task ID information corresponding to the Cluster IP address of the user's initial target AI service, and multiple Cluster IPs of AI services can correspond to one task ID.

[0108] Specifically, it can be as Figure 5 shown. This solution includes the following data transmission content: The user side obtains an AI task data packet (which can correspond to the above-mentioned obtaining of the first task data; the first task data is the transmission data of the first subtask under the first AI task); the user side performs a packaging operation on the AI task data packet according to the task ID mapping table and the sending time information (which can correspond to the above-mentioned packaging the task data of the first subtask under the first AI task according to the third mapping relationship and the AI task data sending time information to obtain the AI task data packet corresponding to the first subtask; the third mapping relationship includes: the mapping relationship between AI task data and AI task identifiers), and transmits it to the QoS rule, and the QoS rule maps it to the corresponding QoS flow according to the new QoS flow mapping table (which can correspond to the above-mentioned performing QoS mapping on the AI task data packet according to the fourth mapping relationship to obtain the second QoS flow corresponding to the first task data; the fourth mapping relationship includes: the mapping relationship between AI task data and QoS flows) and transmits it to the user side protocol stack; After that, the user side protocol stack processes the QoS flow according to the QoS parameter information corresponding to the new QoS flow and sends it to the base station side protocol stack (which can correspond to the above-mentioned obtaining the first QoS flow corresponding to the first subtask according to the QoS parameter information corresponding to the QoS flow corresponding to the AI task and the second QoS flow; sending the first QoS flow to the network device, and the first QoS flow carries the first task data), and then transmits it to the shunt layer; then the shunt layer identifies and shunts the QoS flow according to the QoS identification table, and further supports the mapping layer to transmit the corresponding AI data to the corresponding AI service according to the AI service mapping table.

[0109] Among them, regarding identification and shunting, specifically, it can include: the base station side protocol stack processes the received QoS flow according to the QoS parameter information corresponding to the new QoS flow, and then the shunt layer identifies and shunts the QoS flow according to the QoS identification table, but it is not limited to this.

[0110] II. The process of configuring user plane information for the AI task delay analysis function; When a user submits an AI task requirement, the AI task latency analysis function can first determine the candidate PODs that can provide services for the AI task according to the AI task type and the requirements for the computing accuracy in the requirement. Then, the AI task latency analysis function can complete the configuration of the user plane information table according to the following operations.

[0111] 1. Configuration of the AI service mapping table; Step 1-1: Invoke the end-to-end latency requirement information of the AI task (which can correspond to the end-to-end latency of the sub-AI task described in the above AI task requirement information) and the candidate POD in the computing status information acquisition function to obtain the average latency information of a unit AI task (i.e., a sub-task) (which can correspond to the average latency of a sub-task of the first AI task completed by the candidate AI service described in the above computing status information).

[0112] Step 1-2: Calculate the difference between the end-to-end latency of the AI task and the average latency of a unit AI task of different candidate PODs (corresponding to the difference between the obtained end-to-end latency and average latency above), and obtain the transmission latency threshold of this unit AI task when executed on different PODs (which can correspond to the transmission latency threshold of the sub-task when executed on the candidate AI service according to the above difference), that is, when the AI task transmission latency is within this threshold range, the AI task can be mapped to this POD for execution, which can meet the end-to-end latency requirement of the AI task.

[0113] Step 1-3: Construct an AI service mapping table with the AI task ID, transmission latency threshold, and AI service IP (which can correspond to determining the first mapping relationship according to the task identifier of the first AI task, the transmission latency threshold, and the IP information of the candidate AI service above), for configuring the mapping layer in the user plane CSDAP.

[0114] 2. Configuration of the new QoS flow mapping table, QoS flow identification table, and corresponding QoS parameter information; Step 2-1: After knowing the transmission latency thresholds of the above different candidate PODs, according to the data volume size of a unit AI task (i.e., a sub-task) in the AI task requirement, the ratio (data volume size divided by this threshold) can be calculated to obtain the minimum communication rate requirements corresponding to different candidate PODs; it can correspond to obtaining the minimum communication rate requirement information corresponding to the candidate AI service according to the data volume information of the sub-task under the first AI task and the transmission latency threshold of the sub-task when executed on the candidate AI service above.

[0115] Step 2-2: Since the total number of REs (Resource Element) within the bandwidth of a single RB (Resource Block) is 168, generally within the bandwidth of a single RB, a user (i.e., a terminal) will not be allocated more than 156 REs. Therefore, the communication rate (i.e., the achievable rate) of allocating one RB to the user can be estimated by multiplication calculation based on the number of MIMO (Multiple Input Multiple Output) layers, modulation order, and code rate of the user transmission in the communication status information acquisition function; the second communication rate that can be achieved by allocating one resource block RB to the terminal can be obtained corresponding to the second information corresponding to the terminal; the second information includes at least one of the number of multiple-input multiple-output MIMO layers, modulation order, and code rate of data transmission.

[0116] Step 2-3: The minimum RB number requirements corresponding to different candidate PODs can be calculated by the ratio of the minimum communication rate requirement corresponding to the above-mentioned POD and the communication rate of allocating one RB to the user; the minimum RB number requirement information corresponding to the candidate AI service can be obtained according to the above-mentioned minimum communication rate requirement information and the second communication rate.

[0117] Step 2-4: According to the (currently) remaining (or configurable) number of RBs in the air interface and the corresponding ones of all currently requested AI tasks (that is, all AI requests respectively): the minimum RB number requirement, the unit AI task data volume size, and the communication rate corresponding to one RB (that is, the communication rate of allocating one RB to the user obtained in step 2-2), an efficient RB allocation strategy (not actual RB allocation, that is, the node here has not yet implemented the allocation) is determined through a resource allocation algorithm to meet the minimum RB number requirement of the AI ​​task as much as possible, while giving priority to the AI ​​tasks of users with high priority and better channel conditions; corresponding to the above-mentioned minimum RB number requirement information, the data volume information of the subtask under the first AI task, the second communication rate corresponding to the terminal, and the number of RBs remaining in the air interface, the RB allocation strategy corresponding to the first AI task is determined; the second communication rate is the communication rate that can be achieved by allocating one RB to the terminal.

[0118] Among them, the parameters corresponding to an AI task (such as the minimum RB number requirement, the size of the unit AI task data volume, or the communication rate corresponding to one RB) can be obtained based on the parameters corresponding to each subtask under this AI task (such as the minimum RB number requirement, the size of the unit AI task data volume, or the communication rate corresponding to one RB); for example, the minimum RB number requirement corresponding to an AI task can be the minimum value among the minimum RB number requirements corresponding to all subtasks under this AI task, or the minimum RB number requirement corresponding to an AI task can include the minimum RB number requirements corresponding to all subtasks under this AI task, which is not limited herein.

[0119] Step 2-5: Calculate the corresponding communication rate and transmission delay according to the obtained RB allocation policy for each AI task above (specifically, it can be the RB allocation policy for each subtask) (which can correspond to determining the third communication rate and the third transmission delay corresponding to the first AI task according to the above RB allocation policy; specifically, it can be the communication rate and transmission delay corresponding to the subtask), and use them as a reference for the mapping of the initial QoS flow parameters. Furthermore, stipulate that AI tasks with priorities, communication rates, and transmission delays within a certain preset QoS parameter range are mapped to the same QoS flow, and use a new QFI (QoS flow ID) to identify this new QoS flow. Thus, construct an AI task data to new QoS flow mapping table (corresponding to the above fourth mapping relationship), send it to the user-side QoS rule (which can correspond to sending the fourth mapping relationship to the terminal), and send the parameter configuration of the new QoS flow to the radio protocol stacks on the base station side and the user side (that is, the QoS parameter information corresponding to the new QoS flow, which can correspond to sending the QoS parameter information corresponding to the QoS flow corresponding to the AI task to the terminal); it can correspond to obtaining the QoS flow identifier corresponding to the first parameter information of the AI task; the first parameter information includes at least one of the priority, the first communication rate, and the second transmission delay; according to the QoS flow identifier, obtain the fourth mapping relationship and the QoS parameter information corresponding to the QoS flow corresponding to the AI task; the fourth mapping relationship includes: the mapping relationship between the AI task data and the QoS flow. Among them, the end-to-end requirements of the periodically continuous AI task can be the requirements for each periodic sub-AI task, but it is not limited thereto.

[0120] Step 2-6: Record the QFI of the new QoS flow generated in the previous step, establish a QoS flow identification table (which can correspond to obtaining the second mapping relationship according to the above QoS flow identifier), mark these QFIs as AI service QoS flows that need to be processed on the base station side, and send them to the shunt layer of the base station side CSDAP function, but it is not limited thereto.

[0121] 3. Task ID mapping table configuration; Step 3-1: Configure the initial service IP address for the AI task data packet for the user according to the transmission delay corresponding to the RB allocation policy of each of the above AI tasks (i.e., the transmission delay calculated in Step 2-5, which can correspond to the above fourth transmission delay) and the transmission delay threshold of the candidate POD (i.e., the transmission delay threshold obtained in Step 1-2); it can correspond to determining the user's initial AI service corresponding to the sub-task according to the fourth transmission delay and the transmission delay threshold when a sub-task under the first AI task is executed on the candidate AI service; the fourth transmission delay is determined according to the RB allocation policy corresponding to the first AI task; this step can support determining in which threshold range the transmission delay estimated by the initial radio resource allocation policy (RB allocation policy) is, so as to configure the initial service IP address of the AI task data packet for the user.

[0122] Step 3-2: Configure a task ID mapping table according to the initial service IP address and the corresponding task ID (which can correspond to obtaining the third mapping relationship according to the IP information of the user's initial AI service and the AI task identifier of the first AI task; the third mapping relationship includes: the mapping relationship between the AI task data and the AI task identifier), and send it to the packaging function on the user side (which can correspond to sending the third mapping relationship to the terminal), so that the user side packaging function can determine the task ID of the AI task data packet according to the destination IP address of the AI task data, and then package the sending time information into the data packet together to support the unpacking process of the cache layer in the CSDAP layer on the base station side.

[0123] Thus, the solution provided by the embodiment of the present application has the following advantages: 1. Compared with the general AI task offloading scheme, the user plane real-time forwarding mapping scheme takes into account the real-time communication transmission process delay while considering the AI task requirements, and supports changing the mapping strategy of AI task data in real time, enabling the computing resource scheduling to dynamically adapt to the changes in the radio resource scheduling (to a certain extent, realizing the synchronization of the computing resource scheduling and the radio resource scheduling), being more adaptable to the dynamic environment and ensuring the end-to-end performance requirements of the AI task.

[0124] 2. Compared with the scheme of reconfiguring the container resource size according to the communication conditions, the user plane real-time forwarding mapping scheme reduces the process of re-analyzing the computing requirements and reconfiguring the resource size, and has more latency advantages.

[0125] The embodiment of the present application also provides a data transmission device, which is applied to a network device, as Figure 6 shown, including: A first acquisition module 61, configured to acquire a first transmission delay of first task data; the first task data is transmission data of a first sub-task under a first artificial intelligence (AI) task; The first determination module 62 is configured to determine the AI service corresponding to the first task data according to the first mapping relationship and the first transmission delay; the first mapping relationship includes: the correspondence between the AI service and the transmission delay of the task data of the AI task. The first sending module 63 is configured to send the task data to the AI service.

[0126] The data transmission device provided by the embodiment of the present application obtains the first transmission delay of the first task data; the first task data is the transmission data of the first subtask under the first artificial intelligence (AI) task; determines the AI service corresponding to the first task data according to the first mapping relationship and the first transmission delay; the first mapping relationship includes: the correspondence between the AI service and the transmission delay of the task data of the AI task; and sends the task data to the AI service; it can support the realization that the called AI service can dynamically adapt to the transmission delay of the task data (changes in air interface resource scheduling), that is, the computing resource scheduling of the AI task can dynamically adapt to the changes in air interface resource scheduling (can be synchronized with the air interface resource scheduling in real time to a certain extent), and can support changing the called computing resources according to the real-time communication conditions (dynamic environment) to meet the end-to-end delay requirements of the user's AI task, and can well solve the problem that the data transmission scheme for computing resource scheduling in the prior art cannot guarantee the end-to-end performance of the AI task in a dynamic environment.

[0127] Among them, obtaining the first transmission delay of the first task data includes: determining the sending time and the end time of the first task data according to the task identifier of the first AI task; the sending time is carried in the first data packet of the first task data, and the end time is the time when the last data packet of the first task data is received; and obtaining the first transmission delay of the first task data according to the sending time and the end time.

[0128] In the embodiment of the present application, the data transmission device further includes: a first receiving module, configured to receive the first quality of service (QoS) flow corresponding to the first subtask sent by the terminal; the first data packet and the last data packet are carried by the first QoS flow.

[0129] Further, the data transmission device further includes: a first caching module, configured to cache the first QoS flow to obtain the first task data when it is determined according to the second mapping relationship that the category of the first QoS flow is an AI service QoS flow; where the second mapping relationship includes: the correspondence between the QoS flow identifier and the QoS flow category.

[0130] In the embodiment of the present application, the data transmission device further includes: a second sending module, configured to forward the received QoS flow to the core network user plane function (UPF) when it is determined according to the second mapping relationship that the category of the received QoS flow is a communication service QoS flow.

[0131] Further, the data transmission device further includes: a second obtaining module, configured to obtain a QoS flow identifier corresponding to first parameter information of an AI task; the first parameter information includes at least one of a priority, a first communication rate, and a second transmission delay; a first processing module, configured to obtain the second mapping relationship, a fourth mapping relationship, and QoS parameter information corresponding to the QoS flow corresponding to the AI task according to the QoS flow identifier; the fourth mapping relationship includes a mapping relationship between AI task data and a QoS flow.

[0132] In the embodiment of the present application, the data transmission device further includes: a second determining module, configured to determine the first mapping relationship according to communication state information of a terminal, computing state information of an AI service, and AI task requirement information corresponding to the first AI task; the terminal is a terminal that initiates the first AI task.

[0133] Among them, determining the first mapping relationship according to the communication state information of the terminal, the computing state information of the AI service, and the AI task requirement information corresponding to the first AI task includes: determining a candidate AI service corresponding to the first AI task according to the task type of the first AI task and the AI task requirement information; determining the first mapping relationship according to the communication state information, the computing state information, the AI task requirement information, and the candidate AI service.

[0134] In the embodiment of the present application, determining the first mapping relationship according to the communication state information, the computing state information, the AI task requirement information, and the candidate AI service includes: obtaining a difference between an end-to-end delay and an average delay; the end-to-end delay is the end-to-end delay corresponding to a sub-AI task in the AI task requirement information, and the average delay is the average delay of the candidate AI service completing a next sub-task of the first AI task in the computing state information; obtaining a transmission delay threshold when the sub-task is executed on the candidate AI service according to the difference; determining the first mapping relationship according to the task identifier of the first AI task, the transmission delay threshold, and the IP information of the candidate AI service.

[0135] Further, the data transmission device further includes: a third obtaining module, configured to obtain the minimum resource block (RB) number requirement information corresponding to the candidate AI service for the first AI task; a third determining module, configured to determine an RB allocation policy for the first AI task according to the minimum RB number requirement information, the data volume information of the subtasks under the first AI task, the second communication rate corresponding to the terminal, and the remaining RB number in the air interface; the second communication rate is the communication rate that can be achieved by allocating one RB to the terminal; a fourth determining module, configured to determine a third communication rate and a third transmission delay corresponding to the first AI task according to the RB allocation policy.

[0136] Among them, obtaining the minimum resource block (RB) number requirement information corresponding to the candidate AI service for the first AI task includes: obtaining the minimum communication rate requirement information corresponding to the candidate AI service according to the data volume information of the subtasks under the first AI task and the transmission delay threshold when the subtasks are executed on the candidate AI service; obtaining the second communication rate that can be achieved by allocating one resource block (RB) to the terminal according to the second information corresponding to the terminal; the second information includes at least one of the multiple input multiple output (MIMO) layer number, modulation order, and coding rate of data transmission; obtaining the minimum RB number requirement information corresponding to the candidate AI service according to the minimum communication rate requirement information and the second communication rate.

[0137] Further, the data transmission device further includes: a fifth determining module, configured to determine the user's initial AI service corresponding to the subtask according to the fourth transmission delay and the transmission delay threshold when a subtask under the first AI task is executed on the candidate AI service; the fourth transmission delay is determined according to the RB allocation policy corresponding to the first AI task; a second processing module, configured to obtain a third mapping relationship according to the IP information of the user's initial AI service and the AI task identifier of the first AI task; the third mapping relationship includes: the mapping relationship between the AI task data and the AI task identifier.

[0138] In an embodiment of the present application, the data transmission device further includes: a third sending module, configured to send at least one of the following to the terminal: the QoS parameter information corresponding to the QoS flow corresponding to the AI task; the third mapping relationship, the third mapping relationship includes: the mapping relationship between the user's initial AI service and the AI task identifier; the fourth mapping relationship, the fourth mapping relationship includes: the mapping relationship between the AI task data and the QoS flow.

[0139] Among them, the implementation embodiments of the above data transmission method on the network device side are all applicable to the embodiment of this data transmission device and can achieve the same technical effects.

[0140] The embodiment of the present application further provides a data transmission device, which is applied to a terminal, such as Figure 7 shown, including: A fourth acquisition module 71, configured to acquire first task data; the first task data is transmission data of a first subtask under a first AI task; A fourth sending module 72, configured to send the first task data to a network device according to a third mapping relationship, a fourth mapping relationship, and QoS parameter information corresponding to a QoS flow corresponding to an AI task; Wherein, the third mapping relationship includes: a mapping relationship between AI task data and an AI task identifier; the fourth mapping relationship includes: a mapping relationship between AI task data and a QoS flow.

[0141] The data transmission device provided by the embodiment of the present application acquires first task data; the first task data is transmission data of a first subtask under a first AI task; according to a third mapping relationship, a fourth mapping relationship, and QoS parameter information corresponding to a QoS flow corresponding to an AI task, sends the first task data to a network device; wherein, the third mapping relationship includes: a mapping relationship between AI task data and an AI task identifier; the fourth mapping relationship includes: a mapping relationship between AI task data and a QoS flow; can support the network device to: acquire a first transmission delay of the first task data; the first task data is transmission data of a first subtask under a first artificial intelligence (AI) task; determine an AI service corresponding to the first task data according to a first mapping relationship and the first transmission delay; the first mapping relationship includes: a corresponding relationship between an AI service and a transmission delay of task data of an AI task; send the task data to the AI service; thereby being able to support the realization that the called AI service can dynamically adapt to the transmission delay of task data (changes in radio resource scheduling), that is, the AI task computing resource scheduling can dynamically adapt to the changes in radio resource scheduling (can be synchronized with radio resource scheduling in real time to a certain extent), and can support changing the called computing resources according to real-time communication conditions (dynamic environment), meeting the end-to-end delay requirements of the user's AI task, and well solving the problem that the data transmission scheme for computing resource scheduling in the prior art cannot guarantee the end-to-end performance of AI tasks in a dynamic environment.

[0142] Among them, sending the first task data to a network device according to the third mapping relationship, the fourth mapping relationship, and the QoS parameter information corresponding to the QoS flow corresponding to the AI task includes: obtaining a second QoS flow corresponding to the first task data according to the third mapping relationship and the fourth mapping relationship; obtaining a first QoS flow corresponding to the first subtask according to the QoS parameter information corresponding to the QoS flow corresponding to the AI task and the second QoS flow; and sending the first QoS flow to the network device, where the first QoS flow carries the first task data.

[0143] In an embodiment of the present application, obtaining a second QoS flow corresponding to the first task data according to the third mapping relationship and the fourth mapping relationship includes: packaging the task data of the first subtask under the first AI task according to the third mapping relationship and the AI task data sending time information to obtain an AI task data packet corresponding to the first subtask; and performing QoS mapping on the AI task data packet according to the fourth mapping relationship to obtain a second QoS flow corresponding to the first task data.

[0144] Further, the data transmission device further includes: a second receiving module, configured to receive at least one of the following sent by the network device: the QoS parameter information corresponding to the QoS flow corresponding to the AI task; the third mapping relationship; and the fourth mapping relationship.

[0145] Among them, the implementation embodiments of the data transmission method on the terminal side are all applicable to the embodiments of this data transmission device and can achieve the same technical effects.

[0146] An embodiment of the present application further provides a data transmission device, where the data transmission device is a network device, as Figure 8 shown, including: a processor 81 and a transceiver 82; The processor 81 is configured to obtain a first transmission delay of the first task data; the first task data is the transmission data of the first subtask under the first artificial intelligence (AI) task; determine an AI service corresponding to the first task data according to the first mapping relationship and the first transmission delay; the first mapping relationship includes: the corresponding relationship between the AI service and the transmission delay of the task data of the AI task; send the task data to the AI service through the transceiver 82.

[0147] The data transmission device provided by the embodiment of the present application obtains the first transmission delay of the first task data; the first task data is the transmission data of the first subtask under the first artificial intelligence (AI) task; determines the AI service corresponding to the first task data according to the first mapping relationship and the first transmission delay; the first mapping relationship includes: the corresponding relationship between the AI service and the transmission delay of the task data of the AI task; sends the task data to the AI service; can support the realized called AI service to dynamically adapt to the transmission delay of the task data (the change of radio resource scheduling), that is, the computing resource scheduling of the AI task can dynamically adapt to the change of radio resource scheduling (can be synchronized with the radio resource scheduling in real time to a certain extent), can support changing the called computing resources according to the real-time communication conditions (dynamic environment), meet the end-to-end delay requirement of the user's AI task, and well solve the problem that the data transmission scheme for computing resource scheduling in the prior art cannot guarantee the end-to-end performance of the AI task in the dynamic environment.

[0148] Among them, the obtaining of the first transmission delay of the first task data includes: determining the sending time and the end time of the first task data according to the task identifier of the first AI task; the sending time is carried in the first data packet of the first task data, and the end time is the time when the last data packet of the first task data is received; obtaining the first transmission delay of the first task data according to the sending time and the end time.

[0149] Further, the processor is further configured to: receive, through the transceiver, the first quality of service (QoS) flow corresponding to the first subtask sent by the terminal; the first data packet and the last data packet are carried through the first QoS flow.

[0150] In the embodiment of the present application, the processor is further configured to: cache the first QoS flow to obtain the first task data when determining that the category of the first QoS flow is the AI service QoS flow according to the second mapping relationship; where the second mapping relationship includes: the corresponding relationship between the QoS flow identifier and the QoS flow category.

[0151] Further, the processor is further configured to: forward the received QoS flow to the core network user plane function (UPF) through the transceiver when determining that the category of the received QoS flow is the communication service QoS flow according to the second mapping relationship.

[0152] In an embodiment of the present application, the processor is further configured to: obtain a QoS flow identifier corresponding to first parameter information of an AI task; the first parameter information includes at least one of a priority, a first communication rate, and a second transmission delay; according to the QoS flow identifier, obtain the second mapping relationship, the fourth mapping relationship, and QoS parameter information corresponding to the QoS flow corresponding to the AI task; the fourth mapping relationship includes: a mapping relationship between AI task data and a QoS flow.

[0153] Further, the processor is further configured to: determine the first mapping relationship according to the communication status information of the terminal, the computing status information of the AI service, and the AI task requirement information corresponding to the first AI task; the terminal is the terminal that initiates the first AI task.

[0154] Wherein, determining the first mapping relationship according to the communication status information of the terminal, the computing status information of the AI service, and the AI task requirement information corresponding to the first AI task includes: determining a candidate AI service corresponding to the first AI task according to the task type of the first AI task and the AI task requirement information; determining the first mapping relationship according to the communication status information, the computing status information, the AI task requirement information, and the candidate AI service.

[0155] In an embodiment of the present application, determining the first mapping relationship according to the communication status information, the computing status information, the AI task requirement information, and the candidate AI service includes: obtaining a difference between an end-to-end delay and an average delay; the end-to-end delay is the end-to-end delay corresponding to a sub-AI task in the AI task requirement information, and the average delay is the average delay for the candidate AI service to complete the next sub-task of the first AI task in the computing status information; according to the difference, obtain a transmission delay threshold when the sub-task is executed on the candidate AI service; determine the first mapping relationship according to the task identifier of the first AI task, the transmission delay threshold, and the IP information of the candidate AI service.

[0156] Further, the processor is further configured to: obtain the minimum resource block (RB) number requirement information corresponding to the candidate AI service corresponding to the first AI task; determine an RB allocation strategy corresponding to the first AI task according to the minimum RB number requirement information, the data volume information of the sub-tasks under the first AI task, the second communication rate corresponding to the terminal, and the remaining RB number in the air interface; the second communication rate is the communication rate that can be achieved by allocating one RB to the terminal; determine a third communication rate and a third transmission delay corresponding to the first AI task according to the RB allocation strategy.

[0157] Among them, obtaining the minimum resource block (RB) number requirement information corresponding to the candidate AI service for the first AI task includes: obtaining the minimum communication rate requirement information corresponding to the candidate AI service according to the data volume information of the subtasks under the first AI task and the transmission delay threshold when the subtasks are executed on the candidate AI service; obtaining the second communication rate that can be achieved by allocating one resource block (RB) to the terminal according to the second information corresponding to the terminal; the second information includes at least one of the multiple input multiple output (MIMO) layer number, modulation order, and code rate of data transmission; and obtaining the minimum RB number requirement information corresponding to the candidate AI service according to the minimum communication rate requirement information and the second communication rate.

[0158] Further, the processor is further configured to: determine the user's initial AI service corresponding to the subtask according to the fourth transmission delay and the transmission delay threshold when a subtask under the first AI task is executed on the candidate AI service; the fourth transmission delay is determined according to the RB allocation policy corresponding to the first AI task; obtain a third mapping relationship according to the IP information of the user's initial AI service and the AI task identifier of the first AI task; the third mapping relationship includes: the mapping relationship between AI task data and the AI task identifier.

[0159] In an embodiment of the present application, the processor is further configured to: send, through the transceiver, at least one of the following to the terminal: the QoS parameter information corresponding to the QoS flow corresponding to the AI task; the third mapping relationship, the third mapping relationship includes: the mapping relationship between the user's initial AI service and the AI task identifier; the fourth mapping relationship, the fourth mapping relationship includes: the mapping relationship between AI task data and the QoS flow.

[0160] Among them, the implementation embodiments of the above data transmission method on the network device side are all applicable to the embodiment of this data transmission device and can achieve the same technical effect.

[0161] An embodiment of the present application further provides a data transmission device, and the data transmission device is a terminal, as Figure 9 shown, including: a processor 91 and a transceiver 92; The processor 91 is configured to obtain first task data; the first task data is the transmission data of the first subtask under the first AI task; Send the first task data to the network device through the transceiver 92 according to the third mapping relationship, the fourth mapping relationship, and the QoS parameter information corresponding to the QoS flow corresponding to the AI task.

[0162] The data transmission device provided by the embodiment of the present application obtains first task data; the first task data is the transmission data of the first subtask under the first AI task; according to the third mapping relationship, the fourth mapping relationship, and the QoS parameter information corresponding to the QoS flow corresponding to the AI task, the first task data is sent to the network device; wherein, the third mapping relationship includes: the mapping relationship between AI task data and AI task identifiers; the fourth mapping relationship includes: the mapping relationship between AI task data and QoS flows; it can support the network device to obtain the first transmission delay of the first task data; the first task data is the transmission data of the first subtask under the first artificial intelligence (AI) task; according to the first mapping relationship and the first transmission delay, determine the AI service corresponding to the first task data; the first mapping relationship includes: the corresponding relationship between the AI service and the transmission delay of the task data of the AI task; send the task data to the AI service; thereby enabling the called AI service to dynamically adapt to the transmission delay of the task data (changes in radio resource scheduling), that is, the computing resource scheduling of the AI task can dynamically adapt to the changes in radio resource scheduling (can be synchronized with radio resource scheduling in real time to a certain extent), and can support changing the called computing resources according to real-time communication conditions (dynamic environment) to meet the end-to-end delay requirements of the user's AI task, and well solve the problem that the data transmission scheme for computing resource scheduling in the prior art cannot guarantee the end-to-end performance of AI tasks in a dynamic environment.

[0163] Among them, the step of sending the first task data to the network device through the transceiver according to the third mapping relationship, the fourth mapping relationship, and the QoS parameter information corresponding to the QoS flow corresponding to the AI task includes: obtaining the second QoS flow corresponding to the first task data according to the third mapping relationship and the fourth mapping relationship; obtaining the first QoS flow corresponding to the first subtask according to the QoS parameter information corresponding to the QoS flow corresponding to the AI task and the second QoS flow; sending the first QoS flow, which carries the first task data, to the network device through the transceiver.

[0164] In the embodiment of the present application, the step of obtaining the second QoS flow corresponding to the first task data according to the third mapping relationship and the fourth mapping relationship includes: packing the task data of the first subtask under the first AI task according to the third mapping relationship and the AI task data sending time information to obtain the AI task data packet corresponding to the first subtask; performing QoS mapping on the AI task data packet according to the fourth mapping relationship to obtain the second QoS flow corresponding to the first task data.

[0165] Further, the processor is further configured to: receive, via the transceiver, at least one of the following sent by a network device: QoS parameter information corresponding to a QoS flow corresponding to the AI task; the third mapping relationship; the fourth mapping relationship.

[0166] Among them, the implementation embodiments of the above-mentioned data transmission method on the terminal side are all applicable to the embodiments of this data transmission device, and the same technical effects can also be achieved.

[0167] An embodiment of this application further provides a data transmission device, including a memory, a processor, and a program stored on the memory and executable on the processor; when the processor executes the program, it implements the data transmission method on the above-mentioned network device side or terminal side.

[0168] Among them, the implementation embodiments of the above-mentioned data transmission method on the network device side or terminal side are all applicable to the embodiments of this data transmission device, and the same technical effects can also be achieved.

[0169] An embodiment of this application further provides a readable storage medium, on which a program is stored, and when the program is executed by a processor, it implements the steps in the data transmission method on the above-mentioned network device side or terminal side.

[0170] Among them, the implementation embodiments of the above-mentioned data transmission method on the network device side or terminal side are all applicable to the embodiments of this readable storage medium, and the same technical effects can also be achieved.

[0171] An embodiment of this application further provides a computer program product, including computer instructions, and when the computer instructions are executed by a processor, they implement each process of the method embodiment of the data transmission method on the above-mentioned network device side or terminal side, and the same technical effects can be achieved. To avoid repetition, details are not described here again.

[0172] It should be noted that many functional components described in this specification are referred to as modules to more particularly emphasize the independence of their implementation manners.

[0173] In the embodiments of this application, a module can be implemented by software so as to be executed by various types of processors. For example, an identifiable executable code module can include one or more physical or logical blocks of computer instructions. For example, it can be constructed as an object, a process, or a function. Nevertheless, the executable code of the identified module does not need to be physically located together, but can include different instructions stored in different locations. When these instructions are logically combined together, they constitute the module and achieve the specified purpose of the module.

[0174] In fact, an executable code module can be a single instruction or many instructions, and can even be distributed over multiple different code segments, distributed in different programs, and distributed across multiple memory devices. Similarly, the operation data can be identified within the module, and can be implemented in any appropriate form and organized within any appropriate type of data structure. The operation data can be collected as a single data set, or can be distributed at different locations (including on different storage devices), and can at least partially exist only as electronic signals in a system or network.

[0175] When a module can be implemented by software, considering the level of existing hardware technology, for a module that can be implemented by software, without considering cost, those skilled in the art can build corresponding hardware circuits to implement corresponding functions. The hardware circuits include conventional very large scale integration (VLSI) circuits or gate arrays, as well as existing semiconductors such as logic chips and transistors, or other discrete components. The module can also be implemented using programmable hardware devices, such as field programmable gate arrays, programmable array logic, programmable logic devices, etc.

[0176] The above are the preferred embodiments of the present application. It should be noted that for those of ordinary skill in the art, without departing from the principles described in the present application, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present application.

Claims

1. A data transmission method, applied to a network device, characterized in that: include: Acquire a first transmission delay of first task data; The first task data is the transmission data of the first subtask under the first artificial intelligence AI task; Determine, according to the first mapping relationship and the first transmission delay, the AI ​​service corresponding to the first task data; the first mapping relationship includes: a correspondence between the AI ​​service and the transmission delay of the task data of the AI ​​task; Send the task data to the AI ​​service.

2. The data transmission method according to claim 1, characterized in that: The obtaining of a first transmission delay of the first task data comprises: Determine, according to the task identifier of the first AI task, the sending time and the end time of the first task data; the sending time is carried in the first data packet of the first task data, and the end time is the time when the last data packet of the first task data is received; A first transmission delay of the first task data is obtained according to the sending time and the end time.

3. The data transmission method according to claim 2, characterized in that: Also includes: A first quality of service QoS flow corresponding to a first subtask sent by a receiving terminal; The first data packet and the last data packet are carried by the first QoS flow.

4. The data transmission method according to claim 3, characterized in that: Also includes: When it is determined according to the second mapping relationship that the category of the first QoS flow is an AI service QoS flow, the first QoS flow is cached to obtain the first task data; The second mapping relationship includes: a corresponding relationship between a QoS flow identifier and a QoS flow category.

5. The data transmission method according to claim 4, characterized in that: Also includes: When it is determined according to the second mapping relationship that the category of the received QoS flow is a communication service QoS flow, the received QoS flow is forwarded to the core network user plane function UPF.

6. The data transmission method according to claim 4 or 5, characterized in that: Also includes: Obtain the QoS flow identifier corresponding to the first parameter information of the AI ​​task; The first parameter information includes: at least one of a priority, a first communication rate, and a second transmission delay; According to the QoS flow identifier, the second mapping relationship, the fourth mapping relationship and the QoS parameter information corresponding to the QoS flow corresponding to the AI ​​task are obtained; the fourth mapping relationship includes: a mapping relationship between AI task data and QoS flow.

7. The data transmission method according to claim 1, characterized in that: Also includes: Determining the first mapping relationship according to the communication state information of the terminal, the computing state information of the AI ​​service, and the AI ​​task requirement information corresponding to the first AI task; The terminal is a terminal that initiates the first AI task.

8. The data transmission method according to claim 7, characterized in that: The determining the first mapping relationship according to the communication state information of the terminal, the computing state information of the AI ​​service, and the AI ​​task requirement information corresponding to the first AI task includes: Determining a candidate AI service corresponding to the first AI task according to a task type of the first AI task and the AI ​​task requirement information; The first mapping relationship is determined according to the communication status information, the computing status information, the AI ​​task requirement information and the candidate AI services.

9. The data transmission method according to claim 8, characterized in that: The determining the first mapping relationship according to the communication state information, the computing state information, the AI ​​task requirement information, and the candidate AI service includes: Obtaining the difference between the end-to-end delay and the average delay; the end-to-end delay is the end-to-end delay corresponding to the sub-AI task in the AI ​​task requirement information, and the average delay is the average delay of the candidate AI service in the computing status information to complete a sub-task under the first AI task; According to the difference, a transmission delay threshold value when the subtask is executed on the candidate AI service is obtained; The first mapping relationship is determined according to the task identifier of the first AI task, the transmission delay threshold, and the IP information of the candidate AI service.

10. The data transmission method according to claim 1 or 9, characterized in that: Also includes: Obtaining minimum resource block (RB) requirement information corresponding to the candidate AI service corresponding to the first AI task; Determine, according to the minimum RB number requirement information, the data volume information of the subtask under the first AI task, the second communication rate corresponding to the terminal, and the number of RBs remaining in the air interface, the RB allocation strategy corresponding to the first AI task; The second communication rate is a communication rate that can be achieved by allocating one RB to the terminal; According to the RB allocation strategy, a third communication rate and a third transmission delay corresponding to the first AI task are determined.

11. The data transmission method according to claim 10, characterized in that: The obtaining of the minimum resource block (RB) number requirement information corresponding to the candidate AI service corresponding to the first AI task includes: Obtaining minimum communication rate requirement information corresponding to the candidate AI service according to data volume information of the subtask under the first AI task and a transmission delay threshold when the subtask is executed on the candidate AI service; According to the second information corresponding to the terminal, a second communication rate that can be achieved by allocating a resource block RB to the terminal is obtained; the second information includes at least one of the number of multiple-input multiple-output MIMO layers, the modulation order and the code rate of data transmission; According to the minimum communication rate requirement information and the second communication rate, the minimum RB number requirement information corresponding to the candidate AI service is obtained.

12. The data transmission method according to claim 1, characterized in that: Also includes: Determine, according to the fourth transmission delay and a transmission delay threshold of a subtask under the first AI task when executing on the candidate AI service, a user initial AI service corresponding to the subtask; the fourth transmission delay is determined according to the RB allocation policy corresponding to the first AI task; A third mapping relationship is obtained according to the IP information of the user's initial AI service and the AI ​​task identifier of the first AI task; the third mapping relationship includes: a mapping relationship between AI task data and the AI ​​task identifier.

13. The data transmission method according to claim 1, characterized in that: Also includes: Send at least one of the following to the endpoint: QoS parameter information corresponding to the QoS flow corresponding to the AI ​​task; A third mapping relationship, the third mapping relationship comprising: a mapping relationship between a user initial AI service and an AI task identifier; A fourth mapping relationship, the fourth mapping relationship comprising: a mapping relationship between AI task data and QoS flow.

14. A data transmission method, applied to a terminal, characterized in that: include: Obtaining first task data; The first task data is the transmission data of the first subtask under the first AI task; Sending the first task data to the network device according to the third mapping relationship, the fourth mapping relationship, and the QoS parameter information corresponding to the QoS flow corresponding to the AI ​​task; The third mapping relationship includes: a mapping relationship between AI task data and an AI task identifier; and the fourth mapping relationship includes: a mapping relationship between AI task data and a QoS flow.

15. The data transmission method according to claim 14, characterized in that: The sending the first task data to the network device according to the third mapping relationship, the fourth mapping relationship, and the QoS parameter information corresponding to the QoS flow corresponding to the AI ​​task includes: According to the third mapping relationship and the fourth mapping relationship, obtaining a second QoS flow corresponding to the first task data; Obtaining a first QoS flow corresponding to the first subtask according to the QoS parameter information corresponding to the QoS flow corresponding to the AI ​​task and the second QoS flow; The first QoS flow is sent to a network device, where the first QoS flow carries the first task data.

16. The data transmission method according to claim 15, characterized in that: The acquiring, according to the third mapping relationship and the fourth mapping relationship, the second QoS flow corresponding to the first task data includes: Packing the task data of the first subtask under the first AI task according to the third mapping relationship and the AI ​​task data sending time information to obtain an AI task data packet corresponding to the first subtask; According to the fourth mapping relationship, QoS mapping is performed on the AI ​​task data packet to obtain a second QoS flow corresponding to the first task data.

17. The data transmission method according to claim 14, characterized in that: Also includes: Receive at least one of the following from a network device: QoS parameter information corresponding to the QoS flow corresponding to the AI ​​task; The third mapping relationship; The fourth mapping relationship.

18. A data transmission device, applied to a network device, characterized in that: include: A first acquisition module, used to acquire a first transmission delay of first task data; The first task data is the transmission data of the first subtask under the first artificial intelligence AI task; A first determination module, configured to determine the AI ​​service corresponding to the first task data according to a first mapping relationship and the first transmission delay; the first mapping relationship includes: a correspondence between the AI ​​service and the transmission delay of the task data of the AI ​​task; The first sending module is used to send the task data to the AI ​​service.

19. A data transmission device, applied to a terminal, characterized in that: include: A fourth acquisition module, used to acquire first task data; The first task data is the transmission data of the first subtask under the first AI task; a fourth sending module, configured to send the first task data to the network device according to the third mapping relationship, the fourth mapping relationship, and the QoS parameter information corresponding to the QoS flow corresponding to the AI ​​task; The third mapping relationship includes: a mapping relationship between AI task data and an AI task identifier; and the fourth mapping relationship includes: a mapping relationship between AI task data and a QoS flow.

20. A data transmission device, the data transmission device being a network device, characterized in that: include: processor and transceiver; The processor is used to obtain a first transmission delay of first task data; The first task data is the transmission data of the first subtask under the first artificial intelligence AI task; Determine, according to the first mapping relationship and the first transmission delay, the AI ​​service corresponding to the first task data; the first mapping relationship includes: a correspondence between the AI ​​service and the transmission delay of the task data of the AI ​​task; The task data is sent to the AI ​​service via the transceiver.

21. A data transmission device, the data transmission device being a terminal, characterized in that: include: processor and transceiver; The processor is used to obtain first task data; The first task data is the transmission data of the first subtask under the first AI task; According to the third mapping relationship, the fourth mapping relationship and the QoS parameter information corresponding to the QoS flow corresponding to the AI ​​task, the first task data is sent to the network device through the transceiver.

22. A data transmission device, comprising a memory, a processor, and a program stored in the memory and executable on the processor; characterized in that: When the processor executes the program, the data transmission method according to any one of claims 1 to 17 is implemented.

23. A readable storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, the steps in the data transmission method according to any one of claims 1 to 17 are implemented.

24. A computer program product, characterized in that The method comprises computer instructions, which, when executed by a processor, implement the steps of the data transmission method according to any one of claims 1 to 17.

Images