Network decision-making method and device, network equipment, storage medium and computer program product

Abstract

The invention discloses a network decision method and device, network equipment, a storage medium and a computer program product, and the method comprises the steps: determining a second task and a first index of a second calculation node according to the arrangement information of a first task and the capability information of a first calculation node; wherein the arrangement information comprises a relationship between second tasks obtained by splitting the first task; the second computing node represents the first computing node meeting a first condition; the first index of each second computing node is obtained by disassembling the QoS index of the first task; sending first information to each second computing node, wherein the first information is used for inquiring whether the second computing node can complete a second task of the second computing node and meets a first index of the second computing node; and determining to accept or reject the first task according to the feedback of each second computing node about the first information.

Description

Technical Field

[0001] This application relates to the field of communication technology, and in particular to a network decision-making method, apparatus, network equipment, storage medium, and computer program product. Background Technology

[0002] The integration and coordination of communication and computing in the control plane includes two forms: external computing resources and internal computing resources. Existing technical solutions all belong to the external computing resource solution, which has problems such as difficulty in interaction and inaccurate evaluation of Quality of Service (QoS), making it impossible for the network side to accurately determine whether it can execute the computing tasks initiated by the terminal. Summary of the Invention

[0003] To address the related technical issues, embodiments of this application provide a network decision-making method, apparatus, network device, storage medium, and computer program product.

[0004] The technical solution of this application embodiment is implemented as follows:

[0005] This application provides a network decision-making method applied to network entities, the method comprising:

[0006] Based on the orchestration information of the first task and the capability information of the first computing node, the second task and the first indicator of the second computing node are determined; wherein, the orchestration information includes the relationship between the second tasks obtained by decomposing the first task; the second computing node represents the first computing node that satisfies the first condition; the first indicator of each second computing node is obtained by decomposing the quality of service (QoS) indicator of the first task;

[0007] Send a first message to each second computing node, the first message being used to inquire whether the second computing node can complete the second task of the second computing node and meet the first indicator of the second computing node;

[0008] Based on the feedback from each second computing node regarding the first information, a decision is made to accept or reject the first task.

[0009] In the above scheme, determining the second task and the first indicator of the second computing node based on the orchestration information of the first task and the capability information of the first computing node includes:

[0010] Based on the capability information of the first computing node, a task-level computing topology graph is generated; the task-level computing topology graph represents the topological relationships of the second computing node.

[0011] Based on the task-level computing topology and the orchestration information of the first task, a computing partitioning scheme is determined; the computing partitioning scheme includes the second task and data flow of each second computing node;

[0012] According to the computational segmentation scheme, the QoS indicators of the first task are decomposed to obtain the first indicator of each second computing node.

[0013] In the above scheme, the step of decomposing the QoS indicators of the first task according to the computational segmentation scheme to obtain the first indicator of each second computing node includes:

[0014] If the total transmission delay of the computation segmentation scheme is less than the delay of the first task, the QoS indicators of the first task are decomposed according to the computation segmentation scheme to obtain the first indicator of each second computing node.

[0015] The method in the above scheme further includes:

[0016] If the total transmission delay of the calculation segmentation scheme is greater than or equal to the delay of the first task, and the number of times the calculation segmentation scheme is determined is less than the number of times the first calculation is performed, the calculation segmentation scheme is re-determined based on the task-level calculation topology diagram and the arrangement information of the first task.

[0017] In the above scheme, the first indicator includes one or more of the following:

[0018] Calculate latency;

[0019] Calculation accuracy;

[0020] Computing resources, which represent the minimum computing capacity that can be allocated to the second task;

[0021] Storage capacity.

[0022] In the above scheme,

[0023] Computing resources and / or computational precision are used to generate new task-level computational topology graphs; and / or

[0024] Calculate latency and / or storage capacity to determine whether the computation partitioning scheme is feasible.

[0025] In the above scheme, the QoS metric includes latency;

[0026] The step of decomposing the QoS metrics of the first task according to the computational segmentation scheme to obtain the first metric of each second computing node includes:

[0027] Based on the total transmission delay of the computation segmentation scheme, the delay of the first task is decomposed according to the second condition to obtain the computation delay of each second computation node.

[0028] In the above scheme, the second condition includes: the computation latency of the second computing node is proportional to the computation capacity of the second computing node, and / or the actual total latency for completing the first task is minimized.

[0029] In the above scheme, the sum of the total transmission delay of the calculation segmentation scheme and the total calculation delay of the calculation segmentation scheme is less than the delay of the first task; wherein,

[0030] The total computation time for calculating the partitioning scheme is determined based on the computation time of each second computing node.

[0031] In the above scheme, the total transmission delay and the total computation delay of the segmentation scheme are obtained according to the principle of serial summation and parallel maximum value taking.

[0032] The method in the above scheme further includes:

[0033] Based on the calculation segmentation scheme, determine the transmission delay of each transmission path;

[0034] The total transmission delay of the segmentation scheme is determined based on the transmission delay of each transmission path.

[0035] In the above scheme, the computing resources are determined based on the ratio of the computing capacity of different second computing nodes.

[0036] The method in the above scheme further includes:

[0037] A second computation node is added to the task-level computation topology graph if one or more of the following conditions are met, resulting in a new task-level computation topology graph:

[0038] The total transmission delay of the segmentation scheme is greater than or equal to the delay of the first task;

[0039] The number of times the partitioning scheme is calculated is greater than or equal to the number calculated in the first iteration;

[0040] The number of times the task-level computation topology graph is generated is less than the number of times it is generated the second time.

[0041] In the above scheme, generating a task-level computing topology graph based on the capability information of the first computing node includes:

[0042] Based on the capability information of the first computing node, candidate computing nodes that meet the first condition are identified among the first computing nodes;

[0043] N second computing nodes are selected from the candidate computing nodes in descending order of computing capacity; N is a positive integer, and the sum of the computing capacities of the N second computing nodes is greater than the total computing amount of the first task.

[0044] Generate or update the task-level computing topology graph based on the topological relationships between the N second computing nodes.

[0045] In the above scheme, generating a task-level computing topology graph based on the capability information of the first computing node includes:

[0046] Based on the capability information of the first computing node, a second computing node that meets the first condition is determined from the first computing nodes;

[0047] Based on the topological relationships between the second computing nodes, a task-level computing topology graph is generated.

[0048] In the above scheme, the first condition includes being able to allocate computing resources for the first task and meeting the computing accuracy requirements of the first task.

[0049] The method in the above scheme further includes:

[0050] Request capability information from the first computing node;

[0051] Receive capability information fed back from the first computing node; among which...

[0052] The capability information includes computing capacity and / or computing accuracy.

[0053] In the above scheme, the network entity is deployed in a centralized unit (CU).

[0054] In the above scheme, the computing nodes include one or more of the following:

[0055] terminal;

[0056] CU;

[0057] Distributed Unit (DU).

[0058] This application provides a network decision-making device, including:

[0059] The determining unit is configured to determine the second task and the first indicator of the second computing node based on the orchestration information of the first task and the capability information of the first computing node; wherein, the orchestration information includes the relationship between the second tasks obtained by decomposing the first task; the second computing node represents the first computing node that satisfies the first condition; the first indicator of each second computing node is obtained by decomposing the QoS indicator of the first task;

[0060] The sending unit is used to send first information to each second computing node, the first information being used to query whether the second computing node can complete the second task of the second computing node and meet the first indicator of the second computing node.

[0061] The decision-making unit is used to decide whether to accept or reject the first task based on the feedback from each second computing node regarding the first information.

[0062] This application provides a network device, including: a processor and a communication interface; wherein,

[0063] The processor is configured to determine a second task and a first indicator of a second computing node based on the orchestration information of the first task and the capability information of the first computing node; wherein, the orchestration information includes the relationship between the second tasks obtained by decomposing the first task; the second computing node represents the first computing node that satisfies the first condition; and the first indicator of each second computing node is obtained by decomposing the QoS indicator of the first task.

[0064] The communication interface is used to send first information to each second computing node, the first information being used to inquire whether the second computing node can complete the second task of the second computing node and meet the first indicator of the second computing node.

[0065] The processor is also configured to decide whether to accept or reject the first task based on feedback from each second computing node regarding the first information.

[0066] This application provides a network device, characterized in that it includes a processor and a memory for storing computer programs capable of running on the processor.

[0067] When the processor runs the computer program, it executes the steps of any of the above methods.

[0068] This application provides a storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of any of the above methods.

[0069] This application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of any of the above methods.

[0070] In a network decision-making method, apparatus, network device, storage medium, and computer program product provided in this application embodiment, a network entity determines a second task and a first indicator of a second computing node based on the orchestration information of a first task and the capability information of a first computing node; wherein, the orchestration information includes the relationship between the second tasks obtained by decomposing the first task, the second computing node represents the first computing node that satisfies the first condition, and the first indicator of each second computing node is obtained by decomposing the QoS indicator of the first task; first information is sent to each second computing node, the first information being used to inquire whether the second computing node can complete the second task of the second computing node and satisfy the first indicator of the second computing node; based on the feedback of each second computing node regarding the first information, a decision is made to accept or reject the first task. As can be seen, in this embodiment, the network entity can decompose the QoS indicators of the first task to obtain the first indicators of each second computing node. Each second computing node evaluates whether it can meet the corresponding first indicators while completing the corresponding second task. This enables all second computing nodes to participate in evaluating whether the network side can complete the first task according to the QoS indicators of the first task, which can save network side decision time. Based on the evaluation results fed back by all second computing nodes, the network entity can accurately evaluate whether the QoS indicators of the first task can be met, thereby deciding whether to accept the first task. This reduces the difficulty of interaction between nodes and improves decision efficiency and accuracy. Attached Figure Description

[0071] Figure 1 This is a diagram illustrating an example architecture of a network decision-making system according to an embodiment of this application.

[0072] Figure 2 This is a diagram illustrating an example architecture of a network decision-making system according to an embodiment of this application.

[0073] Figure 3 This is a schematic diagram of a network decision-making method according to an embodiment of this application;

[0074] Figure 4 This is a schematic diagram illustrating the interaction process between the network entity and the DU in an embodiment of this application;

[0075] Figure 5 This is a schematic diagram of the network decision-making method in an application embodiment of this application;

[0076] Figure 6 This is an example diagram of a task-level computing topology graph according to an embodiment of this application;

[0077] Figure 7 for Figure 6 Example diagram of the computation partitioning scheme corresponding to the task-level computation topology graph in the diagram;

[0078] Figure 8 To determine Figure 7 An example diagram illustrating the computational latency of the second computation node in the computational partitioning scheme;

[0079] Figure 9 This is a schematic diagram of the network decision-making device structure according to an embodiment of this application;

[0080] Figure 10 This is a schematic diagram of the network device structure according to an embodiment of this application. Detailed Implementation

[0081] In 6G (6th Generation Mobile Communication Technology) networks, how to provide users with computing services and guarantees that offer lower latency jitter and higher overall efficiency based on the efficient collaboration of communication and distributed computing resources is an unsolved problem. An important technical approach is the integration and collaboration of communication and computing in the control plane. That is, given that computing resources are built into 6G network equipment, how to design control plane signaling procedures oriented towards computing services so that end-users can be informed of the network's computing service capabilities anytime, anywhere, request and use computing services from the network, and obtain quality of service guarantees in dynamic and complex wireless network environments and during mobility.

[0082] Currently, the integration and collaboration of communication and computing in the control plane, i.e., network-to-computing convergence, includes two forms: external computing resources and intrinsic computing resources. External computing resources, such as edge computing, involve the joint optimization of the communication resources of network nodes and the computing resources of computing nodes through management plane functions. Intrinsic computing resources are the inherent computing resources of network nodes; network nodes possess not only control and forwarding capabilities but also computing capabilities. Existing technical solutions all fall under the category of external computing resource solutions, mainly including edge computing and network-to-computing collaboration within the bearer network.

[0083] Edge computing (EC) / Mobile edge computing (MEC): Provides computing services in a distributed manner at the network edge closer to the user, which can reduce the consumption of network resources while providing lower latency, so as to better serve some industry applications, such as low-latency, high-bandwidth scenarios such as video acceleration, autonomous driving, augmented reality (AR) / virtual reality (VR), as well as network applications including non-real-time wireless protocol processing and network optimization.

[0084] Network-Computing Collaboration: The routing layer introduces a network-computing integration mechanism. Through in-line control signaling of IP layer protocols such as Border Gateway Protocol (BGP), Interior Gateway Protocol (IGP), or Segment Routing IPv6 (SRv6) based on Internet Protocol Version 6, computing and network performance information is disseminated and synchronized. Therefore, the routing layer can combine current computing capabilities, application performance, and network conditions to route computing task packets to the corresponding computing nodes, thereby achieving global optimization of connectivity and computing resources in the network.

[0085] The collaboration of communication and distributed computing resources requires the terminal to initiate an application, and the network side and the terminal to work together to complete the computing task. There are two main ways to determine whether the network side accepts the computing task application initiated by the terminal: terminal side decision and network side decision. Network side decision is further divided into centralized QoS evaluation and distributed QoS evaluation.

[0086] Terminal-side decision-making: The terminal generates a computation orchestration diagram based on the task content and obtains a computation topology diagram from the network side. Based on this, it generates a computation partitioning scheme, determines whether the computation task can be completed in accordance with QoS requirements through network-side cooperation, and thus decides whether to send a computation service request to the network side.

[0087] Network-side decision-making: When a terminal initiates a computing service request, the network side decides whether to accept the request based on the task-level computing topology and the computing orchestration diagram sent by the terminal. This can be achieved using centralized QoS assessment (where the decision node collects resource status data from each computing node) or by distributing QoS calculation metrics across computing nodes and using a distributed QoS assessment approach.

[0088] The disadvantages of edge computing: EC / MEC is only physically deployed within the communication network, but logically, computing and communication have independent management and control systems; the deployment of EC / MEC-based applications achieves joint optimization of communication and computing resources through the management plane, which has certain defects in dynamism and makes it difficult to achieve the unification of network and computing on the control plane, so as to respond promptly to user movement and network changes.

[0089] The disadvantages of the bearer network's communication and computing collaboration: Since the bearer network is the basic network that provides connections between the Radio Access Network (RAN) and the Core Network (CN), it operates at a different protocol layer than the service network and cannot directly perceive users and sessions, or provide user access control, adapt to the dynamic wireless channel environment, or manage user mobility.

[0090] In summary, existing solutions mainly suffer from problems such as interaction difficulties and inaccurate QoS assessment, specifically manifested as follows:

[0091] 1) Neither edge computing nor bearer network-to-grid computing collaboration can achieve the unification of network and computing in the control plane, and cannot directly interact with the terminal for control signaling, thereby providing computing services;

[0092] 2) The terminal side cannot obtain the latest task-level computing topology map when deciding whether to initiate a computing service request. It can only make a rough estimate of QoS indicators and cannot make an accurate judgment.

[0093] 3) When deciding whether to accept a computing service request, the network side needs to consider the task-level computing topology. In the process of generating the task-level computing topology, the computing capacity of each node needs to be updated. The computing volume is large, the interaction overhead with each computing node is large, and since the network side does not know the internal computing execution characteristics of the computing nodes, it cannot accurately assess the QoS of the computing task based only on the resource status.

[0094] Based on this, in various embodiments of this application, the network entity determines the second task and first indicator of the second computing node according to the orchestration information of the first task and the capability information of the first computing node; wherein, the orchestration information includes the relationship between the second tasks obtained by decomposing the first task, the second computing node represents the first computing node that meets the first condition, and the first indicator of each second computing node is obtained by decomposing the QoS indicator of the first task; first information is sent to each second computing node, the first information being used to inquire whether the second computing node can complete the second task of the second computing node and meet the first indicator of the second computing node; based on the feedback of each second computing node regarding the first information, a decision is made to accept or reject the first task. It can be seen that, in the embodiments of this application, the network entity can decompose the QoS indicator of the first task to obtain the first indicator of each second computing node, and each second computing node evaluates whether it can meet the corresponding first indicator while completing the corresponding second task, realizing that all second computing nodes participate in evaluating whether the network side can complete the first task according to the QoS indicator of the first task, which can save the network side's decision-making time; based on the evaluation results fed back by all second computing nodes, the network entity can accurately evaluate whether the QoS indicator of the first task can be met, thereby deciding whether to accept the first task, reducing the interaction difficulty between nodes, and improving decision-making efficiency and accuracy.

[0095] The present application will now be described in further detail with reference to the accompanying drawings and embodiments.

[0096] This application provides a network decision-making method applied to network entities, which can be described as logical entities, network functions, or network elements. Network entities can be named Task Anchors (TAs). Network entities can be co-located with the CU (Complex Unit) of a network device; for example, the network entity is deployed within the CU. Network entities can also be deployed independently; for example, a network entity can be an independently deployed network function, network element, or network device.

[0097] Figure 1 This diagram illustrates an example architecture of a network decision system when network entities are deployed in a CU. Figure 1The DU and CU can belong to the same network device, such as a base station, or they can belong to different network devices. The terminal, such as user equipment (UE) and / or DU, can be used as a computing node, or the terminal can only be used to initiate the first task. Optionally, the MEC host can also be used as a computing node. The terminal can include at least a task scheduler module, an operation orchestrate module, and a radio resource control (RRC) entity. The operation orchestrate module is used to orchestrate or split the task to obtain the task orchestration information. The CU can include at least a network entity TA, an operation split module, and an RRC entity (not shown in the figure).

[0098] Figure 2 This diagram illustrates an example of a network decision system architecture when network entities are independently deployed network functions. Figure 2 As shown, a network decision-making system includes at least network entities, network devices, an Access and Mobility Management Function (AMF), and terminals. Terminals are used to initiate the first task and can also act as computing nodes. Network devices include one CU and one or more DUs, and the CU and / or DU can act as computing nodes.

[0099] like Figure 3 As shown, network decision-making methods include:

[0100] Step 301: Based on the orchestration information of the first task and the capability information of the first computing node, determine the second task and the first indicator of the second computing node.

[0101] The orchestration information includes the relationship between the second tasks obtained by splitting the first task; the second computing node represents the first computing node that satisfies the first condition; and the first index of each second computing node is obtained by decomposing the QoS index of the first task.

[0102] Here, the network entity obtains the orchestration information of the first task and the capability information of the first computing node. The orchestration information of the first task can be provided by the terminal that initiated the first task, and there can be multiple first computing nodes. Based on the capability information of the first computing nodes, a second computing node that meets the first condition is selected from the first computing nodes, and there can be one or more second computing nodes. Based on the orchestration information of the first task and the capability information of the second computing node, the second task of each second computing node is determined. Based on the capability information of each second computing node, the QoS indicators of the first task are decomposed to obtain the first indicator of each second computing node. Decomposition can be described as splitting, decomposing, or dividing. The capability information of the first computing node includes at least computing capability information and may also include storage capability information. The computing capability information may include one or more of the following: computing capacity, remaining or available computing capacity, available computing resources, and computing precision; computing capacity represents the size of the supported computing volume or the maximum supported computing volume; computing capacity can also be described as computing resource capacity. Storage capability information may include storage capacity and / or available storage resources.

[0103] It should be noted that the first task is initiated or requested by the terminal from the network side. The first task can be understood as a computational task, which may carry a computational service request or service request initiated by the terminal from the network side. The second task can be understood as a subtask obtained by splitting or dividing the first task. There can be one or more second tasks obtained from the first task, and there can be one or more second tasks on the same second computing node. The orchestration information of the first task can indicate the execution order of the second tasks and / or the relationship between the second tasks; multiple second tasks can be executed in parallel or serially; the relationship between the second tasks can be understood as a serial-parallel relationship between the second tasks, which includes serial and / or parallel relationships. Serial can also be described as connected in series or serially, and parallel can be described as connected in parallel or parallelly. The orchestration information can be described as an orchestration diagram, a computation orchestration diagram, or a computational task orchestration diagram. The first indicator can be understood as a computation indicator, a QoS indicator, a computation indicator requirement, etc.

[0104] In order to select a second computing node that can collaboratively execute the first task, in one embodiment, the first condition includes being able to allocate computing resources for the first task and meeting the computing accuracy of the first task.

[0105] Here, the computational precision of the first task can be carried within the first task itself, or within the related requests for the first task. Computational resources can be allocated to the first task; this can be described as allocable computational resources or the existence of available computational resources. Computational resources are used to determine the supported computational capacity and / or computational load.

[0106] To reduce the difficulty of interaction and the complexity of implementation, in one embodiment, the network entity is deployed on the CU.

[0107] Considering that the capabilities of the first computing node may change dynamically, in order for the network entity to assign a second task and a first metric that match the capabilities of the second computing node, and to accurately assess whether the QoS metrics of the first task can be met, in one embodiment, before step 301, the method further includes:

[0108] Request capability information from the first computing node;

[0109] Receive capability information fed back from the first computing node; among which...

[0110] The capability information includes computing capacity and / or computing accuracy.

[0111] Here, the network device can send a first request to the first computing node, which is used to solicit capability information from the first computing node; the first computing node sends a first response to the network entity according to the first request, which carries the capability information of the first computing node; the network entity receives the first response sent by the first computing node and obtains the capability information of the first computing node by parsing the first response. Solicitation can be described as a request, inquiry, or retrieval.

[0112] It should be noted that the computing capacity included in the capability information can be understood as the minimum computing capacity that can be allocated to the second task, or the maximum available computing capacity. Computational precision is used to measure the accuracy of floating-point operations in a computer; commonly used precisions include 16-bit and 32-bit.

[0113] To enable the network side to accurately assess whether the QoS indicators of the first task can be met, thereby improving the decision-making efficiency and accuracy of the network side, in one embodiment, determining the second task and the first indicator of the second computing node based on the orchestration information of the first task and the capability information of the first computing node includes:

[0114] Based on the capability information of the first computing node, a task-level computing topology graph is generated; the task-level computing topology graph represents the topological relationships of the second computing node.

[0115] Based on the task-level computing topology and the orchestration information of the first task, a computing partitioning scheme is determined; the computing partitioning scheme includes the second task and data flow of each second computing node;

[0116] According to the computational segmentation scheme, the QoS indicators of the first task are decomposed to obtain the first indicator of each second computing node.

[0117] Here, based on the capability information of the first computing node, a second computing node that meets the first condition can be determined from the first computing node. A task-level computing topology graph can be generated based on the determined second computing node. The task-level computing topology graph can be understood as a computing topology graph with tasks (e.g., the first task) as the granularity. The task-level computing topology graph includes the topological relationship or connection relationship between the determined second computing node and the second computing node. The orchestration information of the first task is mapped to the task-level computing topology graph to generate a computing partitioning scheme. The computing partitioning scheme includes the data flow direction between the second tasks of each second computing node and the second computing node. The data flow direction can be understood as the data transmission direction or data transmission relationship. The data flow direction can indicate the sender and receiver of the data. According to the computing partitioning scheme, the QoS indicators of the first task are split or decomposed into each second computing node included in the computing partitioning scheme to obtain the first indicator of each second computing node.

[0118] It should be noted that a task-level computation topology can be described as a computation topology; a computation partitioning scheme can be described as a computation task partitioning scheme or a task-level computation partitioning scheme. The data flow between second computation nodes can be determined based on the relationships between the second tasks contained in the orchestration information of the first task. These relationships can be understood as serial-parallel relationships between the second tasks. When the capability information of the first computation node includes computational capacity and / or computational precision, the computational capacity and / or computational precision are used to determine the second computation nodes, thereby generating the task-level computation topology. For example, the network entity can exclude first computation nodes that cannot be used for the first task based on the computational precision of the first computation node and the computational precision of the first task, and select first computation nodes whose computational precision meets the requirements of the first task. The computational precision of the first task can be provided or determined by the network entity based on relevant information of the first task. The network entity can also sort the first computation nodes according to their computational capacity, prioritizing the second computation nodes with larger computational capacities among the first computation nodes. All second computation nodes use a uniform computational precision to minimize the number of second computation nodes participating in computation while meeting the QoS indicators of the first task, thereby minimizing the computational latency and transmission latency of the computation partitioning scheme and improving network-side decision-making efficiency.

[0119] In one embodiment, the first indicator includes one or more of the following:

[0120] Calculate latency;

[0121] Calculation accuracy;

[0122] Computing resources, which represent the minimum computing capacity that can be allocated to the second task;

[0123] Storage capacity.

[0124] Here, computing resources can be described as computational resources. Storage capacity represents the maximum amount of binary data that can be stored. The network entity can estimate the amount of data that each second computing node needs to transmit to the next second computing node based on the task-level computing topology, and allocate storage capacity according to the estimated data volume. Since this application example uses a synchronization algorithm, the storage capacity of each second computing node should be able to temporarily store all data to be transmitted.

[0125] The network entity can determine the computation delay of each second computing node based on the delay of the first task; or, based on the delay of the first task and the transmission delay between each second computing node, it can determine the computation delay of each second computing node.

[0126] Considering that the time or total latency required for all second computing nodes to collaboratively complete the first task needs to take into account the computation and transmission latency of the second computing nodes, the total latency for all second computing nodes to collaboratively complete the first task, i.e., the total latency of the computation partitioning scheme, is equal to the sum of the total computation latency and the total transmission latency of the computation partitioning scheme. In other words, the total latency of the computation partitioning scheme = the total computation latency of the computation partitioning scheme + the total transmission latency of the computation partitioning scheme.

[0127] The output data transmission of the second task within the second computing node is synchronous. The computation results within the second computing node must be uniformly calculated and then packaged before being transmitted to the next second computing node. Therefore, the network entity can determine the serial-parallel relationship between the second computing nodes in the computation partitioning scheme based on the serial-parallel relationship between the second tasks. Based on this relationship, and considering the computation latency of the second tasks executed by the second computing nodes, the total computation latency of the computation partitioning scheme is determined according to the principles of serial summation and taking the maximum value in parallel processing. The network entity can also determine the transmission latency of each transmission path based on the amount of data and bandwidth required to be transmitted by each second computing node in the computation partitioning scheme. Furthermore, based on the serial-parallel relationship between the second computing nodes and the transmission latency of each transmission path, the total transmission latency of the computation partitioning scheme is determined according to the principles of serial summation and taking the maximum value in parallel processing.

[0128] In order to reduce the computation latency of each second computing node executing the second task in the computation partitioning scheme, in one embodiment, the computing resources are determined according to the ratio of the computing capacity of different second computing nodes.

[0129] Here, the larger the computing capacity of the second computing node, the larger the value of the minimum computing capacity represented by the capability resources in the first indicator of the second computing node.

[0130] In one embodiment, computing resources and / or computational precision are used to generate a new task-level computational topology graph; and / or

[0131] Calculate latency and / or storage capacity to determine whether the computation partitioning scheme is feasible.

[0132] Here, when the computational partitioning scheme is not feasible and it is necessary to regenerate or update the task-level computational topology, a second computational node can be added or removed from the task-level computational topology based on its computing power resources and / or computational precision, thereby obtaining a new task-level computational topology. It should be noted that when generating the task-level computational topology for the first time, the second computational node can be determined from the first computational nodes based on the computing capacity and / or computational precision contained in the capability information of the first computational node, and the task-level computational topology can be generated based on the determined computational node.

[0133] Each second computing node needs to estimate whether it can meet the first indicator while completing the corresponding second task. If the first indicator includes storage capacity, and any second computing node in the computing partitioning scheme cannot meet the storage capacity in the first indicator, then the computing partitioning scheme is not feasible. If the first indicator includes storage capacity, and all second computing nodes in the computing partitioning scheme can meet the storage capacity in the first indicator, then the computing partitioning scheme is feasible.

[0134] The computation latency of the second computing node is used to determine the total latency of the computation partitioning scheme, to judge whether the total latency of the computation partitioning scheme is less than or equal to the latency of the first task. If the first metric includes computation latency and the total latency of the computation partitioning scheme is less than or equal to the latency of the first task, the computation partitioning scheme is feasible; if the first metric includes computation latency and the total latency of the computation partitioning scheme is greater than the latency of the first task, the computation partitioning scheme is infeasible. Alternatively, if the total latency of the computation partitioning scheme is less than the latency of the first task, the computation partitioning scheme is feasible; if the total latency of the computation partitioning scheme is greater than or equal to the latency of the first task, the computation partitioning scheme is infeasible, requiring modification, updating, or redetering of the computation partitioning scheme, or rejection of the first task.

[0135] To improve the efficiency of generating task-level computing topology graphs, in one embodiment, generating the task-level computing topology graph based on the capability information of the first computing node includes:

[0136] Based on the capability information of the first computing node, a second computing node that meets the first condition is determined from the first computing nodes;

[0137] Based on the topological relationships between the second computing nodes, a task-level computing topology graph is generated.

[0138] Here, the first condition includes the availability of computing resources for the first task and meeting the computational accuracy requirements of the first task. The topological relationship between the second computing nodes can be described as the connection relationship between the second computing nodes.

[0139] To minimize the number of second computing nodes involved in the computation and to quickly generate a concise task-level computing topology and computation partitioning scheme, in one embodiment, generating the task-level computing topology based on the capability information of the first computing node includes:

[0140] Based on the capability information of the first computing node, candidate computing nodes that meet the first condition are identified among the first computing nodes;

[0141] N second computing nodes are selected from the candidate computing nodes in descending order of computing capacity; N is a positive integer, and the sum of the computing capacities of the N second computing nodes is greater than the total computing amount of the first task.

[0142] Generate or update the task-level computing topology graph based on the topological relationships between the N second computing nodes.

[0143] Here, the computing capacity of the second computing node is used to indicate or determine the maximum amount of computation supported by the second computing node. It should be noted that when iteratively updating the task-level computing topology and / or the computing partitioning scheme, if the computing partitioning scheme in a certain iteration cannot meet the computing requirements or the QoS indicators of the first task, then a new task-level computing topology needs to be generated iteratively. Computing nodes from the candidate computing nodes can be added to the task-level computing topology in each iteration, with one or more nodes added per iteration. In practical applications, during the iterative update of the task-level computing topology, the principle of minimizing the number of second computing nodes in the task-level computing topology can also be followed, adding or removing second computing nodes from the task-level computing topology.

[0144] Considering the computational latency of the second computing node when providing computing services or executing computing tasks, if the total transmission latency of the computational partitioning scheme is greater than or equal to the latency of the first task, then the computational partitioning scheme cannot meet the latency requirement of the first task, and the first task can be rejected. Alternatively, the computational partitioning scheme can be modified or updated so that the total transmission latency of the updated computational partitioning scheme is less than the latency of the first task, thereby reducing unnecessary time consumption and improving the decision-making efficiency on the network side. Based on this, in one embodiment, the QoS indicators of the first task are decomposed according to the computational partitioning scheme to obtain the first indicator of each second computing node, including:

[0145] If the total transmission delay of the computation segmentation scheme is less than the delay of the first task, the QoS indicators of the first task are decomposed according to the computation segmentation scheme to obtain the first indicator of each second computing node.

[0146] Here, the total transmission delay of the computation partitioning scheme can be determined based on the serial-to-parallel relationship between the second computing nodes; it can be determined whether the total transmission delay of the computation partitioning scheme is less than the delay of the first task; if the total transmission delay of the computation partitioning scheme is less than the delay of the first task, the QoS indicators of the first task can be decomposed according to the computation partitioning scheme to obtain the first indicator of each second computing node.

[0147] To accurately determine the total transmission delay of the computational segmentation scheme and improve the decision-making efficiency and accuracy on the network side, in one embodiment, the method further includes:

[0148] Based on the calculation segmentation scheme, determine the transmission delay of each transmission path;

[0149] The total transmission delay of the segmentation scheme is determined based on the transmission delay of each transmission path.

[0150] Here, the transmission delay of each transmission path (the transmission path between every two second computing nodes) can be determined based on the data flow direction between the second computing nodes in the computation partitioning scheme; or, the transmission delay of each transmission path can be determined based on the amount of data that each second computing node needs to transmit and the transmission bandwidth in the computation partitioning scheme; and the total transmission delay of the computation partitioning scheme can be determined based on the transmission delay of each transmission path and the serial-to-parallel relationship between each second computing node.

[0151] It should be noted that, since different second computing nodes can be connected in series or in parallel in the computational partitioning scheme, the total transmission delay of the computational partitioning scheme can be determined based on the transmission delay of each transmission path and the series-parallel relationship between the various second computing nodes in the computational partitioning scheme.

[0152] Furthermore, in order to improve the accuracy of the total transmission delay of the determined segmentation scheme, in one embodiment, the total transmission delay of the segmentation scheme is calculated according to the principle of serial summation and parallel maximum value taking.

[0153] Here, in calculating the total transmission delay of the segmentation scheme, for a set of parallel transmission paths, the maximum transmission delay among the transmission delays of the set of parallel transmission paths is taken as the final transmission delay of that set of parallel transmission paths; for serial or series transmission paths, the transmission delays are summed. A set of parallel transmission paths can consist of multiple transmission paths with different starting points and different ending points.

[0154] Considering that the total transmission delay of the segmentation calculation scheme is greater than or equal to the delay of the first task, the segmentation calculation scheme can be modified or updated. To prevent wasting time by updating the segmentation calculation scheme infinitely, a maximum number of times the segmentation calculation scheme can be determined can be set. Therefore, if the number of times the segmentation calculation scheme is determined is greater than or equal to the maximum number of times, and the total transmission delay of the updated segmentation calculation scheme is greater than or equal to the delay of the first task, the first task is rejected, thereby improving decision-making efficiency. Based on this, in one embodiment, the method further includes:

[0155] If the total transmission delay of the calculation segmentation scheme is greater than or equal to the delay of the first task, and the number of times the calculation segmentation scheme is determined is less than the number of times the first calculation is performed, the calculation segmentation scheme is re-determined based on the task-level calculation topology diagram and the arrangement information of the first task.

[0156] Here, "first number" can be understood as the maximum number of times a partitioning scheme can be determined, or the maximum number of iterations for calculating a partitioning scheme. The number of times a partitioning scheme can be determined can be understood as the number of iterations for calculating a partitioning scheme.

[0157] Considering that the initially determined computational partitioning scheme may not meet the QoS requirements of the first task, it is necessary to redetermine the task-level computational topology and / or computational partitioning scheme. Based on this, in one embodiment, the method further includes:

[0158] A second computation node is added to the task-level computation topology graph if one or more of the following conditions are met, resulting in a new task-level computation topology graph:

[0159] The total transmission delay of the segmentation scheme is greater than or equal to the delay of the first task;

[0160] The number of times the partitioning scheme is calculated is greater than or equal to the number calculated in the first iteration;

[0161] The number of times the task-level computation topology graph is generated is less than the number of times it is generated the second time.

[0162] Here, during the process of updating or generating a new task-level computation topology, one or more second computation nodes can be selected from the candidate computation nodes that meet the first condition, in descending order of computational capacity. These selected second computation nodes are then added to the task-level computation topology. The task-level computation topology is updated based on the topological relationships between the second computation nodes, resulting in a new task-level computation topology. The first number can be understood as the maximum number of times a computation partitioning scheme can be determined, or the maximum number of iterations for a computation partitioning scheme; the number of times a computation partitioning scheme is determined can be understood as the number of iterations for a computation partitioning scheme. The second number can be understood as the maximum number of times a task-level computation topology can be determined, or the maximum number of iterations for a task-level computation topology; the number of times a computation partitioning scheme is determined can be understood as the number of iterations for a computation partitioning scheme.

[0163] In order to accurately assess whether the latency of the first task can be met, in one embodiment, the QoS metric includes latency;

[0164] The step of decomposing the QoS metrics of the first task according to the computational segmentation scheme to obtain the first metric of each second computing node includes:

[0165] Based on the total transmission delay of the computation segmentation scheme, the delay of the first task is decomposed according to the second condition to obtain the computation delay of each second computation node.

[0166] Here, the difference between the latency of the first task and the total transmission latency of the computation partitioning scheme is determined, resulting in the first difference value. Based on the computational capacity of each second computation node and the first difference value, the latency of the first task is decomposed according to the second condition, yielding the computational latency of each second computation node. The second condition can be understood as the constraint condition followed when decomposing the latency of the first task. The first difference value can be understood as the maximum value of the total computational latency of the computation partitioning scheme.

[0167] In order to reduce the processing time of the first task, in one embodiment, the second condition includes: the computing latency of the second computing node is proportional to the computing capacity of the second computing node, and / or the actual total latency for completing the first task is minimized.

[0168] Here, the actual total delay for completing the first task can be used interchangeably with the total delay for calculating the partitioning scheme mentioned above. The actual total delay for the first task is equal to the sum of the total transmission delay for calculating the partitioning scheme and the total computation delay for calculating the partitioning scheme.

[0169] In one embodiment, the sum of the total transmission delay of the computation segmentation scheme and the total computation delay of the computation segmentation scheme is less than the delay of the first task; wherein,

[0170] The total computation time for calculating the partitioning scheme is determined based on the computation time of each second computing node.

[0171] Here, the computational partitioning scheme is feasible if the sum of the total transmission delay and the total computation delay of the computational partitioning scheme is less than the delay of the first task.

[0172] For example, as described above, the total delay of calculating the segmentation scheme equals the total computation delay of calculating the segmentation scheme plus the total transmission delay of calculating the segmentation scheme. If the total delay of calculating the segmentation scheme is less than the delay of the first task, then step 302 is executed; if the total delay of calculating the segmentation scheme is greater than or equal to the delay of the first task, the network entity can directly reject the first task, or modify or redetermine the segmentation scheme.

[0173] It should be noted that the sum of the total transmission delay of the calculation segmentation scheme and the total calculation delay of the calculation segmentation scheme is less than the delay of the first task. This can also be understood as: the first difference is greater than or equal to the total calculation delay of the calculation segmentation scheme. The first difference is the difference between the delay of the first task and the total transmission delay of the calculation segmentation scheme.

[0174] To improve the accuracy of the latency (total transmission latency, total computation latency) of the obtained computational partitioning scheme, in one embodiment, the total transmission latency and the total computation latency of the computational partitioning scheme are obtained according to the principle of serial summation and parallel maximum value taking.

[0175] Here, as mentioned above, the network entity can determine the serial-to-parallel relationship of each second computing node in the computation partitioning scheme based on the serial-to-parallel relationship between the second tasks; based on the serial-to-parallel relationship of the second computing nodes, and considering the computation latency of the second computing nodes executing the second tasks, the total computation latency of the computation partitioning scheme is determined according to the principles of serial summation and taking the maximum value in parallel processing. The network entity can determine the transmission latency of each transmission path based on the amount of data and transmission bandwidth that each second computing node needs to transmit in the computation partitioning scheme; based on the serial-to-parallel relationship of each second computing node in the computation partitioning scheme, and based on the transmission latency of each transmission path, the total transmission latency of the computation partitioning scheme is determined according to the principles of serial summation and taking the maximum value in parallel processing.

[0176] Step 302: Send the first message to each second computing node.

[0177] The first information is used to inquire whether the second computing node can complete the second task of the second computing node and meet the first indicator of the second computing node.

[0178] Here, the first information sent to the second computing node may include information related to the second task of the second node and a first indicator of the second task. Upon receiving the first information, the second computing node can determine, confirm, or evaluate whether it can complete the second task based on its capability information, and if it can complete the second task, determine, confirm, or evaluate whether the first indicator of the second computing node is met. The second computing node also sends feedback about the first information to the network entity. The feedback about the first information is used to indicate whether the second task can be completed and / or whether the first indicator can be met; meeting the first indicator can also be described as achieving the first indicator.

[0179] Inquiring whether the second computing node can complete its second task and meet its first metric can also be understood as inquiring whether the second computing node can complete its corresponding second task according to the corresponding first metric.

[0180] Step 303: Based on the feedback from each second computing node regarding the first information, decide whether to accept or reject the first task.

[0181] Here, feedback regarding the first information sent by each second computing node is received. This feedback indicates whether the second computing node can complete its second task and meet its first metric. If each second computing node can complete its corresponding second task and meet its corresponding first metric, the first task is accepted or agreed upon. If any second computing node cannot complete its corresponding second task and / or cannot meet its corresponding first metric, the first task is rejected, or step 301 is executed to redetermine the second computing node, and / or redetermine its second task, and / or redetermine its first metric. For example, if any second computing node cannot complete its corresponding second task and / or cannot meet its corresponding first metric, the first task is rejected, or the task-level computing topology and / or computing partitioning scheme is redetermined.

[0182] To enable the solution to be applied to various scenarios, in one embodiment, the computing node includes one or more of the following:

[0183] terminal;

[0184] CU;

[0185] DU.

[0186] Here, the computing nodes include a first computing node and / or a second computing node. That is, the first computing node includes a terminal and / or a CU and / or a DU, and the second computing node includes a terminal and / or a CU and / or a DU. Wherein, when the first computing node and / or the second computing node includes a terminal, the terminal may be different from the terminal that initiated the first task.

[0187] It should be noted that, when the computing node includes a terminal, if the network entity is deployed within the CU, or if the network entity is co-located with the CU, then the network entity interacts with the terminal through the RRC entity; if the network entity is an independently deployed network function, network element, or network device, then the network entity interacts with the terminal through a first interface and a second interface. The first interface represents the interface between the network entity and the CU, for example... Figure 2 The Tr interface in the diagram represents the interface between the network device and the terminal, for example... Figure 2 The Uu interface in the network. Interactions between network entities and terminals include one or more of the following: the network entity sends first information to the terminal; the terminal sends feedback to the network entity regarding the first information; the network entity requests capability information from the terminal; and the terminal sends capability information back to the network entity.

[0188] When a compute node includes a CU, if the network entity is deployed within the CU, or if the network entity and the CU are co-located, then the interaction between the network entity and the CU does not involve external interfaces; if the network entity is an independently deployed network function, network element, or network device, then the network entity interacts with the CU through a first interface, for example... Figure 2 The Tr interface in the code. Interactions between network entities and CUs include one or more of the following: the network entity sends first information to the CU; the CU sends feedback to the network entity regarding the first information; the network entity requests capability information from the CU; and the CU provides feedback capability information to the network entity.

[0189] When a compute node includes a DU (Distributed Utility), if the network entity is deployed within the CU, or if the network entity is co-located with the CU, then the network entity and the DU can interact directly without involving external interfaces. If the network entity is an independently deployed network function, network element, or network device—that is, the network entity is deployed outside the CU—then the network entity interacts indirectly with the DU through the CU; in other words, the CU forwards information between the network entity and the DU. For example... Figure 4 As shown, the content of the interaction between the network entity and the DU may include one or more of the following: first information, feedback on the first information, information used to solicit capability information from the DU, and capability information fed back by the DU.

[0190] The present application will be further described in detail below with reference to application examples.

[0191] like Figure 5 As shown, network decision-making methods include:

[0192] Step 501: Set the first number and the second number.

[0193] Here, the first number is set to m, and the second number is set to n. The first number can be understood as the maximum number of iterations to calculate the partitioning scheme; the maximum number of iterations to calculate the partitioning scheme can be understood as the maximum number of times a partitioning scheme can be determined. The second number can be understood as the maximum number of iterations to calculate the topology graph at the task level; the maximum number of iterations to calculate the topology graph at the task level can be understood as the maximum number of times a partitioning scheme can be generated.

[0194] Step 502: Solicit capability information from each first computing node, including computing capacity and / or computing accuracy.

[0195] Here, computational capacity and / or computational accuracy are used to generate or update the task-level computational topology.

[0196] The first computing node includes one or more of the following: terminal, CU, DU.

[0197] Step 503: Receive capability information fed back from each first computing node.

[0198] Step 504: Generate a task-level computing topology diagram and computing partitioning scheme based on the scheduling information of the first task and the capability information of the first computing node.

[0199] Here, a task-level computation topology graph can be generated based on the capability information of the first computing node. This graph represents the topological relationships of the second computing nodes. A computation partitioning scheme can be determined based on the task-level computation topology graph and the orchestration information of the first task. The partitioning scheme includes the second task and data flow for each second computing node. For details on the specific implementation of generating the task-level computation topology graph and the computation partitioning scheme, please refer to the relevant descriptions above; they will not be repeated here.

[0200] Among them, such as Figure 6 An example of a task-level computation topology graph is given. Figure 7 The following is given by Figure 6 An example of a computation partitioning scheme corresponding to the task-level computation topology graph.

[0201] It should be noted that in the process of generating the task-level computing topology graph, the principle of minimizing the number of second computing nodes in the task-level computing topology graph can be followed. The minimum number of second computing nodes that can meet the computing requirements of the first task can be selected, and the task-level computing topology graph can be generated based on the topological or connection relationships between the selected second computing nodes.

[0202] Step 505: Determine whether the total transmission delay of the calculation segmentation scheme is less than the delay of the first task.

[0203] Here, if the number of iterations of the task-level computation topology graph is less than or equal to the second iteration, it can be determined whether the total transmission delay of the computation partitioning scheme is less than the delay of the first task. If the total transmission delay of the computation partitioning scheme is less than the delay of the first task, steps 506 to 509 are executed; if the total transmission delay of the computation partitioning scheme is greater than or equal to the delay of the first task, steps 510 to 511 are executed.

[0204] Step 506: According to the computational segmentation scheme, the QoS indicators of the first task are decomposed to obtain the first indicator of each second computation node.

[0205] Here, the first indicator includes one or more of the following:

[0206] Calculate latency;

[0207] Calculation accuracy;

[0208] Computing resources, which represent the minimum computing capacity that can be allocated to the second task;

[0209] Storage capacity.

[0210] The computing resources are determined based on the ratio of the computing capacity of different second computing nodes.

[0211] In one embodiment, the QoS metric for the first task includes latency, and step 506 may include:

[0212] Based on the total transmission delay of the computational segmentation scheme, the delay of the first task is decomposed according to the second condition to obtain the computational delay of each second computation node; where,

[0213] The second condition includes: the computation latency of the second computing node is proportional to the computation capacity of the second computing node, and / or the actual total latency for completing the first task is minimized;

[0214] The sum of the total transmission delay of the computation partitioning scheme and the total computation delay of the computation partitioning scheme is less than the delay of the first task; the total computation delay of the computation partitioning scheme is determined based on the computation delay of each second computation node.

[0215] The total transmission delay and total computation delay of the partitioning scheme are calculated according to the principles of serial summation and parallel maximum value taking.

[0216] Here, as Figure 8 As shown, the computational partitioning scheme includes six second computational nodes: CU-1, DU-1, DU-2, DU-3, DU-5, and a terminal; the known parameters are as follows:

[0217] The latency (maximum allowed latency) of the first task is t_max;

[0218] The transmission delay from DU-1 to CU-1 is t_trans1, the transmission delay from DU-2 to CU-1 is t_trans2, the transmission delay from DU-5 to CU-1 is t_trans3, the transmission delay from CU-1 to DU-3 is t_trans4, and the transmission delay from the terminal to DU-3 is t_trans5; the transmission delay between any two second computing nodes can be determined based on the transmission distance and transmission rate between the two second computing nodes.

[0219] The computing capacity of DU-1 is c1, the computing capacity of DU-2 is c2, the computing capacity of DU-5 is c3, the computing capacity of CU-1 is c4, the computing capacity of DU-3 is c5, and the computing capacity of the terminal is c6.

[0220] if Figure 8 The computation delay of DU-1 is represented as t1, the computation delay of DU-2 as t2, the computation delay of DU-5 as t3, the computation delay of CU-1 as t4, the computation delay of DU-3 as t5, and the computation delay of the terminal as t6; according to Figure 8 The serial-to-parallel relationship between the second computing nodes in the computational partitioning scheme can be obtained by following the principles of serial summation and parallel maximum value taking:

[0221] The computation latency for the second computing node (DU-1, DU-2, DU-5, terminal) in parallel computing is: t1 = t2 = t3 = max{t1, t2, t3}, t6 = max{t1 + t4, t6}; max{t1, t2, t3} means taking the maximum value among t1, t2, and t3, and max{t1 + t4, t6} means taking the maximum value among (t1 + t4) and t6;

[0222] The computation delay for the second computing node in the serial process is: t1+t4+t5, t6+t5; that is, max{t1+t4,t6}+t5;

[0223] For the transmission delay of the second computing node in parallel computing: t_trans1=t_trans2=t_trans3=max{t_trans1,t_trans2,t_trans3},t_trans5=max{(t_trans 1+t_trans 4),t_trans5};

[0224] Transmission delay for the second computing node in serial mode: t_trans 1 + t_trans 4;

[0225] Based on the above content, it can be concluded that Figure 8 The total computation delay of the computation segmentation scheme shown is max{t1+t4,t6}+t5, and the total transmission delay t_trans is max{(t_trans 1+t_trans4),t_trans 5}. Since the sum of the total transmission delay and the total computation delay of the computation segmentation scheme is less than the delay of the first task, we can obtain: max{t1+t4,t6}+t5+max{(t_trans 1+t_trans 4),t_trans 5}≤t_max, that is, max{t1+t4,t6}+t5≤t_max-max{(t_trans 1+t_trans 4),t_trans 5}.

[0226] Assuming (t1+t4)>t6, then max{t1+t4,t6} is t1+t4. The network entities can determine t1, t4, and t5 based on max{t1+t4,t6}+t5≤t_max-max{(t_trans 1+t_trans 4),t_trans 5}, and then determine t2 and t3 based on t1. In practical applications, t1, t4, and t5 can be determined based on the computing capacities of the three second computing nodes DU-1, CU-1, and DU-3; c1:c4:c5=t1:t4:t5.

[0227] In one embodiment, computing power resources and / or computing precision are used to generate a new task-level computing topology; and / or computing latency and / or storage capacity are used to determine whether the computing partitioning scheme is feasible.

[0228] Here, when the computational partitioning scheme is not feasible and it is necessary to regenerate or update the task-level computational topology, the second computational nodes in the task-level computational topology can be added or removed based on their computing power resources and / or computational precision to obtain a new task-level computational topology. It should be noted that during the iterative update of the task-level computational topology, the principle of minimizing the number of second computational nodes in the task-level computational topology can be followed when adding or removing them.

[0229] The computation delay of each second computing node in the partitioning scheme is calculated to determine the total computation delay of the partitioning scheme. The partitioning scheme is feasible if the sum of the total transmission delay and the total computation delay of the partitioning scheme is less than the delay of the first task. For example, as mentioned above, the total delay of the partitioning scheme = the total computation delay of the partitioning scheme + the total transmission delay of the partitioning scheme. If the total delay of the partitioning scheme is less than the delay of the first task, then the partitioning scheme is feasible, and step 507 is executed; if the total delay of the partitioning scheme is greater than or equal to the delay of the first task, then the partitioning scheme is infeasible, and the first task can be directly rejected, or step 510 can be executed.

[0230] Step 507: Send first information to each second computing node in the computation partitioning scheme. The first information is used to inquire whether the second computing node can complete the second task of the second computing node and meet the first indicator of the second computing node, or to inquire whether the second computing node can complete the second task of the second computing node according to the first indicator of the second computing node.

[0231] Here, the first information can carry the second task and the first indicator. The method for any second computing node to determine whether it can achieve the first indicator is as follows:

[0232] 1) Regarding the calculation delay in the first indicator

[0233] The second computing node estimates the time required to complete the second task based on the second task carried in the first information and the computational delay of the second task, combined with the computing resources available for the second computing node. If the second computing node estimates that it can complete the second task within the computational delay of the second task carried in the first information, then the second computing node can meet the computational delay of the second task. If the second computing node estimates that it cannot complete the second task within the computational delay of the second task carried in the first information, then the second computing node does not meet the computational delay of the second task.

[0234] 2) Regarding the computing resources in the first indicator, computing resources represent the minimum computing capacity that can be allocated to the second task.

[0235] The second computing node allocates computing resources to the second task based on the second task and computing resources carried by the first information, combined with the computing resources, computing capacity, or computing power of the second computing node. If the second computing node has sufficient computing power or computing resources to complete the second task, then the second computing node meets the computing resources in the first indicator. If the second computing node does not have sufficient computing power or computing resources to complete the second task, then the second computing node does not meet the computing resources in the first indicator.

[0236] 3) For the storage capacity in the first metric, the storage capacity can be understood as the amount of data of the calculation result

[0237] The second computing node determines whether it has sufficient storage space to store the calculation result of the second task based on the second task and the storage capacity carried in the first information, combined with the available storage space size of the second computing node itself; if the second computing node has sufficient storage space to store the calculation result of the second task, then the second computing node meets the storage capacity of the first metric; if the second computing node does not have sufficient storage space to store the calculation result of the second task, then the second computing node does not meet the storage capacity of the first metric.

[0238] 4) For the calculation accuracy in the first metric

[0239] The second computing node determines whether its own calculation accuracy is higher than or equal to the calculation accuracy in the first metric

[0240] If the calculation accuracy of the second computing node is higher than or equal to the calculation accuracy in the first metric, then the second computing node meets the calculation accuracy in the first metric; if the calculation accuracy of the second computing node is lower than the calculation accuracy in the first metric, then the second computing node does not meet the calculation accuracy in the first metric.

[0241] Step 508: Receive the feedback on the first information sent by each second computing node, and decide to accept or reject the first task according to the feedback on the first information from each second computing node.

[0242] Step 509: Determine whether the stop condition is satisfied.

[0243] Here, the stop condition may include one or more of the following:

[0244] Each second computing node can meet the first metric of its respective second task;

[0245] There is a second computing node that cannot meet the first metric of the corresponding second task, the iteration count i of the calculation partitioning scheme < m, and all second computing nodes have participated in the iteration;

[0246] There is a second computing node that cannot meet the first metric of the corresponding second task, and the iteration count i of the calculation partitioning scheme ≥ m.

[0247] In the case where the stop condition is not satisfied, execute steps 510 to 511; in the case where the stop condition is satisfied, end.

[0248] Step 510: Update the iteration count i of the calculation partitioning scheme to i = i + 1, and update the iteration count j of the task-level calculation topology graph to j = j + 1.

[0249] Here, if the total transmission delay of the computational partitioning scheme is greater than or equal to the delay of the first task, it is determined whether the number of iterations of the computational partitioning scheme is less than the first number, and / or whether the number of iterations of the task-level computational topology graph is less than the second number. If the number of iterations of the computational partitioning scheme is less than the first number, and / or the number of iterations of the task-level computational topology graph is less than the second number, it indicates that the task-level computational topology graph needs to be updated or regenerated. Based on the new task-level computational topology graph and the arrangement information of the first task, the computational partitioning scheme is re-determined. At this time, the number of iterations of the computational partitioning scheme, i, is incremented and updated to i+1, and the number of iterations of the task-level computational topology graph, j, is incremented and updated to j+1.

[0250] Step 511: Add one or more new second computing nodes to the task-level computing topology graph in descending order of computing capacity.

[0251] Here, based on the capability information of the first computing node, candidate computing nodes that meet the first condition are identified from among the first computing nodes. One or more new second computing nodes are selected from the candidate computing nodes in descending order of computing capacity, and these new second computing nodes are added to the task-level computing topology graph. A new task-level computing topology graph is then generated based on the topological relationships between the second computing nodes. After step 511 is executed, step 504 is executed to determine a new computing partitioning scheme based on the new task-level computing topology graph and the arrangement information of the first task.

[0252] To implement the network decision-making method of this application embodiment, this application embodiment also provides a network decision-making device, which is installed on a network entity, such as... Figure 9 As shown, the device includes:

[0253] The determining unit 901 is configured to determine the second task and the first indicator of the second computing node based on the orchestration information of the first task and the capability information of the first computing node; wherein, the orchestration information includes the relationship between the second tasks obtained by decomposing the first task; the second computing node represents the first computing node that satisfies the first condition; and the first indicator of each second computing node is obtained by decomposing the QoS indicator of the first task.

[0254] The sending unit 902 is used to send first information to each second computing node, the first information being used to query whether the second computing node can complete the second task of the second computing node and meet the first indicator of the second computing node.

[0255] Decision unit 903 is used to decide whether to accept or reject the first task based on the feedback from each second computing node regarding the first information.

[0256] In one embodiment, the determining unit 901 is specifically used for:

[0257] Based on the capability information of the first computing node, a task-level computing topology graph is generated; the task-level computing topology graph represents the topological relationships of the second computing node.

[0258] Based on the task-level computing topology and the orchestration information of the first task, a computing partitioning scheme is determined; the computing partitioning scheme includes the second task and data flow of each second computing node;

[0259] According to the computational segmentation scheme, the QoS indicators of the first task are decomposed to obtain the first indicator of each second computing node.

[0260] In one embodiment, the determining unit 901 is specifically used to: when the total transmission delay of the computation segmentation scheme is less than the delay of the first task, decompose the QoS indicators of the first task according to the computation segmentation scheme to obtain the first indicator of each second computing node.

[0261] In one embodiment, the determining unit 901 is further configured to:

[0262] If the total transmission delay of the calculation segmentation scheme is greater than or equal to the delay of the first task, and the number of times the calculation segmentation scheme is determined is less than the number of times the first calculation is performed, the calculation segmentation scheme is re-determined based on the task-level calculation topology diagram and the arrangement information of the first task.

[0263] In one embodiment, the first indicator includes one or more of the following:

[0264] Calculate latency;

[0265] Calculation accuracy;

[0266] Computing resources, which represent the minimum computing capacity that can be allocated to the second task;

[0267] Storage capacity.

[0268] In one embodiment,

[0269] Computing resources and / or computational precision are used to generate new task-level computational topology graphs; and / or

[0270] Calculate latency and / or storage capacity to determine whether the computation partitioning scheme is feasible.

[0271] In one embodiment, the QoS metric includes latency;

[0272] The determining unit 901 is specifically used to decompose the delay of the first task according to the second condition based on the total transmission delay of the calculation segmentation scheme, so as to obtain the calculation delay of each second calculation node.

[0273] In one embodiment, the second condition includes: the computation latency of the second computing node is proportional to the computation capacity of the second computing node, and / or the actual total latency for completing the first task is minimized.

[0274] In one embodiment, the sum of the total transmission delay of the computation segmentation scheme and the total computation delay of the computation segmentation scheme is less than the delay of the first task; wherein,

[0275] The total computation time for calculating the partitioning scheme is determined based on the computation time of each second computing node.

[0276] In one embodiment, the total transmission delay and the total computation delay of the partitioning scheme are calculated according to the principle of serial summation and parallel maximum value taking.

[0277] In one embodiment, the determining unit 901 is further configured to determine the transmission delay of each transmission path according to the calculation segmentation scheme; and to determine the total transmission delay of the calculation segmentation scheme according to the transmission delay of each transmission path.

[0278] In one embodiment, the computing resources are determined based on the ratio of the computing capacities of different second computing nodes.

[0279] In one embodiment, the determining unit 901 is further configured to add a second computing node to the task-level computing topology graph to obtain a new task-level computing topology graph if one or more of the following conditions are met:

[0280] The total transmission delay of the segmentation scheme is greater than or equal to the delay of the first task;

[0281] The number of times the partitioning scheme is calculated is greater than or equal to the number calculated in the first iteration;

[0282] The number of times the task-level computation topology graph is generated is less than the number of times it is generated the second time.

[0283] In one embodiment, the determining unit 901 is specifically used for:

[0284] Based on the capability information of the first computing node, candidate computing nodes that meet the first condition are identified among the first computing nodes;

[0285] N second computing nodes are selected from the candidate computing nodes in descending order of computing capacity; N is a positive integer, and the sum of the computing capacities of the N second computing nodes is greater than the total computing amount of the first task.

[0286] Generate or update the task-level computing topology graph based on the topological relationships between the N second computing nodes.

[0287] In one embodiment, the determining unit 901 is specifically used for:

[0288] Based on the capability information of the first computing node, a second computing node that meets the first condition is determined from the first computing nodes;

[0289] Based on the topological relationships between the second computing nodes, a task-level computing topology graph is generated.

[0290] In one embodiment, the first condition includes the availability of computing resources for the first task and meeting the computing accuracy requirements of the first task.

[0291] In one embodiment, the method further includes:

[0292] Request capability information from the first computing node;

[0293] Receive capability information fed back from the first computing node; among which...

[0294] The capability information includes computing capacity and / or computing accuracy.

[0295] In one embodiment, the network entity is deployed on a CU.

[0296] In one embodiment, the computing node includes one or more of the following:

[0297] Terminal, CU, DU.

[0298] In practical applications, the determining unit 901 and the decision-making unit 903 can be implemented by the processor in the network decision-making device, and the sending unit 902 can be implemented by the processor in the network decision-making device in combination with the communication interface.

[0299] It should be noted that the network decision-making device provided in the above embodiments is only illustrated by the division of the above-described program modules when making network decisions. In practical applications, the above processing can be assigned to different program modules as needed, that is, the internal structure of the device can be divided into different program modules to complete all or part of the processing described above. In addition, the network decision-making device and the network decision-making method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0300] Based on the hardware implementation of the above program modules, and in order to implement the method of the embodiments of this application, the embodiments of this application also provide a network device, such as... Figure 10 As shown, network device 1000 includes:

[0301] Communication interface 1001 enables information exchange with other network nodes;

[0302] The processor 1002 is connected to the communication interface 1001 to enable information interaction with other network nodes and, when running a computer program, executes the methods provided by one or more technical solutions on the network entity side. The computer program is stored in the memory 1003.

[0303] Specifically, the processor 1002 is configured to determine a second task and a first indicator of a second computing node based on the orchestration information of the first task and the capability information of the first computing node; wherein, the orchestration information includes the relationship between the second tasks obtained by decomposing the first task; the second computing node represents the first computing node that satisfies the first condition; and the first indicator of each second computing node is obtained by decomposing the QoS indicator of the first task.

[0304] The communication interface 1001 is used to send first information to each second computing node. The first information is used to query whether the second computing node can complete the second task of the second computing node and meet the first indicator of the second computing node.

[0305] The processor 1002 is further configured to decide whether to accept or reject the first task based on feedback from each second computing node regarding the first information.

[0306] In one embodiment, the processor 1002 is specifically used for:

[0307] Based on the capability information of the first computing node, a task-level computing topology graph is generated; the task-level computing topology graph represents the topological relationships of the second computing node.

[0308] Based on the task-level computing topology and the orchestration information of the first task, a computing partitioning scheme is determined; the computing partitioning scheme includes the second task and data flow of each second computing node;

[0309] According to the computational segmentation scheme, the QoS indicators of the first task are decomposed to obtain the first indicator of each second computing node.

[0310] In one embodiment, the processor 1002 is specifically configured to: when the total transmission delay of the computation segmentation scheme is less than the delay of the first task, decompose the QoS indicators of the first task according to the computation segmentation scheme to obtain the first indicator of each second computing node.

[0311] In one embodiment, the processor 1002 is further configured to, when the total transmission delay of the computational segmentation scheme is greater than or equal to the delay of the first task, and the number of times the computational segmentation scheme is determined is less than the number of times the first calculation is performed, redetermine the computational segmentation scheme based on the task-level computational topology and the arrangement information of the first task.

[0312] In one embodiment, the first indicator includes one or more of the following:

[0313] Calculate latency;

[0314] Calculation accuracy;

[0315] Computing resources, which represent the minimum computing capacity that can be allocated to the second task;

[0316] Storage capacity.

[0317] In one embodiment,

[0318] Computing resources and / or computational precision are used to generate new task-level computational topology graphs; and / or

[0319] Calculate latency and / or storage capacity to determine whether the computation partitioning scheme is feasible.

[0320] In one embodiment, the QoS metric includes latency;

[0321] The processor 1002 is specifically used to decompose the latency of the first task according to the second condition based on the total transmission latency of the computation segmentation scheme, so as to obtain the computation latency of each second computing node.

[0322] In one embodiment, the second condition includes: the computation latency of the second computing node is proportional to the computation capacity of the second computing node, and / or the actual total latency for completing the first task is minimized.

[0323] In one embodiment, the sum of the total transmission delay of the computation segmentation scheme and the total computation delay of the computation segmentation scheme is less than the delay of the first task; wherein,

[0324] The total computation time for calculating the partitioning scheme is determined based on the computation time of each second computing node.

[0325] In one embodiment, the total transmission delay and the total computation delay of the partitioning scheme are calculated according to the principle of serial summation and parallel maximum value taking.

[0326] In one embodiment, the processor 1002 is further configured to determine the transmission delay of each transmission path according to the calculation segmentation scheme; and to determine the total transmission delay of the calculation segmentation scheme according to the transmission delay of each transmission path.

[0327] In one embodiment, the computing resources are determined based on the ratio of the computing capacities of different second computing nodes.

[0328] In one embodiment, the processor 1002 is further configured to add a second computing node to the task-level computing topology graph to obtain a new task-level computing topology graph if one or more of the following conditions are met:

[0329] The total transmission delay of the segmentation scheme is greater than or equal to the delay of the first task;

[0330] The number of times the partitioning scheme is calculated is greater than or equal to the number calculated in the first iteration;

[0331] The number of times the task-level computation topology graph is generated is less than the number of times it is generated the second time.

[0332] In one embodiment, the processor 1002 is specifically used for:

[0333] Based on the capability information of the first computing node, candidate computing nodes that meet the first condition are identified among the first computing nodes;

[0334] N second computing nodes are selected from the candidate computing nodes in descending order of computing capacity; N is a positive integer, and the sum of the computing capacities of the N second computing nodes is greater than the total computing amount of the first task.

[0335] Generate or update the task-level computing topology graph based on the topological relationships between the N second computing nodes.

[0336] In one embodiment, the processor 1002 is specifically used for:

[0337] Based on the capability information of the first computing node, a second computing node that meets the first condition is determined from the first computing nodes;

[0338] Based on the topological relationships between the second computing nodes, a task-level computing topology graph is generated.

[0339] In one embodiment, the first condition includes the availability of computing resources for the first task and meeting the computing accuracy requirements of the first task.

[0340] In one embodiment, the method further includes:

[0341] Request capability information from the first computing node;

[0342] Receive capability information fed back from the first computing node; among which...

[0343] The capability information includes computing capacity and / or computing accuracy.

[0344] In one embodiment, the network entity is deployed on a CU.

[0345] In one embodiment, the computing node includes one or more of the following:

[0346] terminal;

[0347] CU;

[0348] DU.

[0349] It should be noted that the specific processing procedures of the processor 1002 and the communication interface 1001 can be understood by referring to the above method.

[0350] Of course, in practical applications, the various components in network device 1000 are coupled together through bus system 1004. It can be understood that bus system 1004 is used to implement communication between these components. In addition to a data bus, bus system 1004 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in... Figure 10 The general labeled all buses as Bus System 1004.

[0351] The memory 1003 in this embodiment is used to store various types of data to support the operation of the network device 1000. Examples of such data include any computer program used to operate on the network device 1000.

[0352] The methods disclosed in the embodiments of this application can be applied to the processor 1002, or implemented by the processor 1002. The processor 1002 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the integrated logic circuit of the hardware in the processor 1002 or by instructions in the form of software. The processor 1002 may be a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The processor 1002 can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor may be a microprocessor or any conventional processor, etc. The steps of the methods disclosed in the embodiments of this application can be directly reflected as being executed by a hardware decoding processor, or being executed by a combination of hardware and software modules in the decoding processor. The software modules may be located in a storage medium, which is located in the memory 1003. The processor 1002 reads the information in the memory 1003 and completes the steps of the aforementioned method in combination with its hardware.

[0353] In an exemplary embodiment, the network device 1000 may be implemented by one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontrollers (MCUs), microprocessors, or other electronic components to perform the aforementioned method.

[0354] It is understood that the memory (memory 1003) in the embodiments of this application can be volatile memory or non-volatile memory, or both. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), ferromagnetic random access memory (FRAM), flash memory, magnetic surface memory, optical disc, or compact disc read-only memory (CD-ROM); the magnetic surface memory can be disk storage or magnetic tape storage. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Synchronous Static Random Access Memory (SSRAM), Dynamic Random Access Memory (DRAM), Synchronous Dynamic Random Access Memory (SDRAM), Double Data Rate Synchronous Dynamic Random Access Memory (DDRSDRAM), Enhanced Synchronous Dynamic Random Access Memory (ESDRAM), SyncLink Dynamic Random Access Memory (SLDRAM), and Direct Rambus Random Access Memory (DRRAM).The memories described in the embodiments of this application are intended to include, but are not limited to, these and any other suitable types of memories.

[0355] In an exemplary embodiment, this application also provides a storage medium, namely a computer storage medium, specifically a computer-readable storage medium, such as a memory 1003 storing a computer program, which can be executed by the processor 1002 of the network device 1000 to complete the steps described in the aforementioned network entity-side method. The computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, Flash Memory, magnetic surface memory, optical disc, or CD-ROM.

[0356] For example, embodiments of this application also provide a computer program product, including a computer program that can be executed by the processor 1002 of a network device 1000 to perform the steps described in any of the foregoing methods.

[0357] It should be noted that terms such as "first" and "second" are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0358] Furthermore, the technical solutions described in the embodiments of this application can be combined arbitrarily without conflict.

[0359] The above description is merely a preferred embodiment of this application and is not intended to limit the scope of protection of this application.

Claims

1. A network decision method, characterized by, The method is applied to a network entity, and comprises the following steps: According to the arrangement information of the first task and the capability information of the first computing node, a second task of a second computing node and a first index are determined; wherein the arrangement information contains the relationship between the second tasks obtained by splitting the first task; the second computing node represents the first computing node satisfying a first condition; the first index of each second computing node is obtained by splitting the quality of service (QoS) index of the first task; First information is sent to each second computing node, and the first information is used to inquire whether the second computing node can complete the second task of the second computing node and satisfy the first index of the second computing node; According to the feedback of each second computing node on the first information, it is decided to accept or reject the first task.

2. The method of claim 1, wherein, The method further comprises the following steps: According to the capability information of the first computing node, a task-level computing topology graph is generated; the task-level computing topology graph represents the topology relationship of the second computing node; According to the task-level computing topology graph and the arrangement information of the first task, a computing segmentation scheme is determined; the computing segmentation scheme contains the second task of each second computing node and the data flow direction; According to the computing segmentation scheme, the QoS index of the first task is split to obtain the first index of each second computing node.

3. The method of claim 2, wherein, The method further comprises the following steps: In the case that the total transmission time delay of the computing segmentation scheme is less than the time delay of the first task, the QoS index of the first task is split according to the computing segmentation scheme to obtain the first index of each second computing node.

4. The method of claim 3, wherein, The method further comprises the following steps: In the case that the total transmission time delay of the computing segmentation scheme is greater than or equal to the time delay of the first task, and the number of times of determining the computing segmentation scheme is less than a first number of times, the computing segmentation scheme is re-determined according to the task-level computing topology graph and the arrangement information of the first task.

5. The method according to any one of claims 1 to 3, characterized in that, The first index comprises one or more of the following: Computing time delay; Computing accuracy; Computing resource, which represents the minimum computing capacity that can be allocated to the second task; Storage capacity.

6. The method of claim 5, wherein: The computing resource and / or the computing accuracy are used to generate a new task-level computing topology graph; and / or The computing time delay and / or the storage capacity are used to determine whether the computing segmentation scheme is feasible.

7. The method according to any one of claims 2 to 4, characterized in that, The QoS index comprises a time delay; The method further comprises the following steps: According to the total transmission time delay of the computing segmentation scheme, the time delay of the first task is split according to a second condition to obtain the computing time delay of each second computing node.

8. The method of claim 7, wherein, The second condition comprises that the computing time delay of the second computing node is proportional to the computing capacity of the second computing node, and / or the actual total time delay for completing the first task is minimum.

9. The method of claim 7, wherein: The sum of the total transmission latency of the splitting solution and the total computation latency of the splitting solution is less than the latency of the first task; wherein The total computation latency of the splitting solution is determined according to the computation latency of each second computation node.

10. The method of claim 9, wherein, The total transmission latency of the splitting solution and the total computation latency of the splitting solution are obtained according to the principle of serial summation and parallel maximum.

11. The method according to any one of claims 4, 8 to 10, characterized in that, The method further comprises: According to the splitting solution, determining the transmission latency of each transmission path; According to the transmission latency of each transmission path, determining the total transmission latency of the splitting solution.

12. The method of claim 5, wherein, The computation capacity of the algorithm resource is determined according to the ratio of the computation capacity of different second computation nodes.

13. The method of claim 4, wherein, The method further comprises: In the case of one or more of the following conditions, adding a second computation node to the task-level computation topology graph to obtain a new task-level computation topology graph: The total transmission latency of the splitting solution is greater than or equal to the latency of the first task; The number of times of determining the splitting solution is greater than or equal to the first number of times; The number of times of generating the task-level computation topology graph is less than the second number of times.

14. The method of claim 2 or 13, wherein, The method further comprises: According to the capability information of the first computation node, determining a candidate computation node that satisfies a first condition in the first computation node; According to the topology relationship between the N second computation nodes, generating or updating the task-level computation topology graph. The method further comprises:

15. The method of claim 2, wherein, According to the capability information of the first computation node, determining a second computation node that satisfies a first condition in the first computation node; According to the topology relationship between the second computation nodes, generating the task-level computation topology graph. The first condition includes that the second computation node can allocate computation resources for the first task and satisfy the computation accuracy of the first task.

16. The method according to any one of claims 1 to 4, 6, 8 to 10, 12, 14 to 15, characterized in that, The method further comprises:

17. The method of any one of claims 1 to 4, 6, 8 to 10, 12, 14 to 15, wherein, Inquiring the capability information from the first computation node; Receiving the capability information fed back by the first computation node; wherein The capability information includes computation capacity and / or computation accuracy. The network entity is deployed in a centralized unit (CU).

18. The method of claim 1, wherein, The computation node includes one or more of the following:

19. The method of claim 1, wherein, A terminal; A centralized unit (CU); A distributed unit (DU). The method further comprises:

20. A network decision device, characterized by A determining unit configured to determine a second task of a second computation node and a first index according to the arrangement information of a first task and the capability information of a first computation node; wherein the arrangement information contains the relationship between the second tasks obtained by splitting the first task; the second computation node represents the first computation node that satisfies a first condition; and the first index of each second computation node is obtained by decomposing the QoS index of the first task; A sending unit configured to send first information to each second computation node, the first information being used to inquire whether the second computation node can complete the second task of the second computation node and satisfy the first index of the second computation node; A decision unit configured to decide to accept or reject the first task according to the feedback of each second computation node on the first information. The method further comprises:

21. A network device, comprising: ​ A processor and a communication interface; wherein The processor is configured to determine, according to orchestration information of the first task and capability information of the first computing node, a second task of a second computing node and a first index; wherein the orchestration information contains relationships between the second tasks split from the first task; the second computing node represents the first computing node satisfying a first condition; and the first index of each second computing node is obtained by splitting a QoS index of the first task; The communication interface is configured to send, to each second computing node, first information for inquiring whether the second computing node can complete the second task of the second computing node and satisfy the first index of the second computing node; The processor is further configured to determine, according to feedback of each second computing node on the first information, whether to accept or reject the first task.

22. A network device, comprising: A computer program product comprising a processor and a memory for storing a computer program executable on the processor, wherein the processor is configured to execute the computer program to perform the steps of the method of any one of claims 1 to 19.

23. A storage medium having stored thereon a computer program, characterized in that The computer program, when executed by the processor, implements the steps of the method of any one of claims 1 to 19.

24. A computer program product comprising a computer program, characterized in that, The computer program, when executed by the processor, implements the method according to any one of claims 1 to 19.

Images