Elastic configuration and two-way collaborative adaptation system for application computing task

Abstract

The invention provides an elastic configuration and bidirectional collaborative adaptation system for application computing tasks, and relates to the technical field of computers. Equipment and network states are monitored in real time by applying an elastic configuration mechanism of a computing task, execution tasks and nodes in a task graph are dynamically adjusted, task response time is optimized, meanwhile, collaborative optimization is performed between an equipment end and a cloud end by applying a bidirectional collaborative adaptation algorithm of computing network resources, computing resources are reasonably distributed, and the computing efficiency is improved. The limitation of single optimization dimension in the prior art is overcome, in addition, the system adopts a self-adaptive learning mechanism, a task unloading strategy and a resource management scheme are continuously optimized by utilizing machine learning and a neural network model, the long-term performance is improved, and the technical scheme not only improves the dynamic response capability of the system, but also improves the reliability of the system. In addition, global optimization of resources is realized in a multi-user and multi-task environment, and task execution time and energy consumption are reduced.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a flexible configuration and bidirectional collaborative adaptation system for application computing tasks. Background Art

[0002] In modern mobile computing and Internet of Things (IoT) environments, computation offloading technology is widely used to offload resource-intensive tasks from resource-constrained devices (such as smartphones, tablets, and IoT devices) to more powerful cloud computing platforms or edge servers. The main purpose of this technology is to overcome the limitations of computing power and battery life of mobile devices and improve the processing speed and efficiency of complex tasks. Existing computation offloading technologies usually rely on the following implementation methods: Static offloading strategy: offloading specific types of tasks to the cloud according to predefined rules or strategies. For example, the device will decide whether to offload it to the cloud based on the complexity of the task or the amount of data. This strategy is simple to implement but lacks dynamic adaptability; Dynamic offloading strategy: using real-time monitoring information (such as network bandwidth, remaining power of the device, computing resources, etc.) to dynamically adjust the task offloading decision. Dynamic offloading strategies usually use intelligent algorithms such as decision trees, genetic algorithms, or deep learning to calculate the optimal offloading solution in real time, so as to make adaptive decisions under different environmental conditions; Partial offloading strategy: decomposing the task into multiple subtasks, some of which are executed locally and some in the cloud. This strategy attempts to find a balance between computing resources and communication overhead to optimize overall performance. Edge computing offloading strategy: By processing tasks on edge nodes closer to users, network latency and bandwidth consumption can be reduced. The goal of this strategy is to reduce the time and frequency of data transmission and improve response speed.

[0003] Due to the volatility of the computing network environment and the adjustable characteristics of application computing tasks, computing task offloading has certain difficulties. In terms of technical implementation, the existing computing offloading system usually includes the following key components: monitoring module: used to monitor the network status (such as bandwidth, delay), device status (such as power, CPU load) and real-time progress of task execution in real time; decision module: based on the data provided by the monitoring module, through preset rules or optimization algorithms, decide whether and how to offload tasks to the cloud or edge nodes; task allocation module: according to the instructions of the decision module, split the task and allocate it to the appropriate execution node (local device, edge server or cloud server); execution and feedback module: responsible for the specific task execution, whether it is executed locally or transmitted to the cloud or edge node through the network. After the task is executed, the execution performance data (such as time, resource consumption) is collected to provide a basis for future offloading decisions; these components work together to enable the computing offloading system to dynamically adjust task allocation and optimize performance. However, although the existing technology provides various offloading strategies, there are still significant defects. Existing static or quasi-dynamic strategies are slow to respond to rapidly changing environments (such as sudden network congestion, sudden drop in device power), which can easily lead to performance bottlenecks or resource waste. Due to the lack of real-time adjustment and two-way feedback mechanisms, these strategies are difficult to maintain efficiency in dynamic environments. At present, some work is limited to considering application tasks as independent and indivisible units or divisible directed acyclic graphs (DAGs) with fixed dependencies when considering offloading decisions, and performing unified scheduling. Few studies focus on the dynamic configurability of application tasks, that is, the complexity and computing power requirements of application tasks can be adjusted in real time according to application requirements or currently available computing power. Although partial offloading strategies attempt to find a balance between computing and communication overhead, existing algorithms are often too complex or highly dependent on parameter settings, making it difficult to achieve ideal results in practical applications. In addition, partial offloading strategies also have difficulties in task decomposition and dependency processing, especially in complex task scenarios, where dependencies between tasks and communication overhead will significantly affect the overall performance. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention provides a flexible configuration and bidirectional collaborative adaptation system for application computing tasks, which solves the problem that the existing computing offloading technology has obvious deficiencies and limitations when facing dynamically changing environments and complex multi-tasking processing requirements.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: a flexible configuration and bidirectional collaborative adaptation system for application computing tasks, including a data collection module, a bidirectional collaborative adaptation module for application computing tasks and computing network resources, a resource scheduling module, and a task execution feedback module, wherein the data collection module is used to monitor and collect device status, network status, and task characteristic data in real time; The bidirectional collaborative adaptation module of the application computing task and the computing network resources is used to dynamically adjust the execution tasks and execution nodes in the task graph according to the data; and combines the application task execution and the computing network resource status to achieve efficient collaborative adaptation between the local device and the remote computing node; The resource scheduling module is used to reasonably allocate computing tasks according to real-time data of device and network status; The task execution feedback module is used to collect task execution results, including task execution time and energy consumption, and optimize task offloading strategies and resource management through a closed-loop feedback mechanism.

[0006] Preferably, the bidirectional collaborative adaptation module of application computing tasks and computing network resources includes three parts: 1. Application elastic configuration part. The elastic configuration part of the application task constructs an elastic splitting model of application tasks and an elastic configuration algorithm for application tasks. The former is used to split the tasks in the task graph according to the task characteristics and the real-time status of the device resources; the latter is used to dynamically adjust the execution tasks and execution nodes in the task graph to optimize the utilization efficiency of computing resources and the task response time; 2. Task offloading decision part. The part is based on the task offloading algorithm of bidirectional collaborative adaptation, considering multiple factors such as the computing power of the device and the cloud, network conditions, and task complexity, and makes reasonable sub-task offloading decisions; 3. Computing network resource scheduling part. The part adopts a dynamic resource scheduling algorithm based on bidirectional collaborative adaptation to optimize the allocation of computing network resources in real time, so as to overcome the limitations of the single optimization dimension of the prior art.

[0007] Preferably, the resource scheduling module reasonably regulates its own computing power and network resources according to the decision of the bidirectional collaborative adaptation module of the application computing task and the computing network resources, and adjusts the required computing power according to the needs of task execution.

[0008] Preferably, the task execution feedback module can adjust the task offloading strategy and resource management scheme in real time through a closed-loop feedback mechanism to ensure long-term performance improvement and user experience optimization.

[0009] A method for elastic configuration and bidirectional collaborative adaptation of computing offload tasks in mobile computing and Internet of Things environments, specifically comprising the following steps: S1. Collect key performance indicators and network status data of equipment in real time; S2. Dynamically adjust the offloading strategy of the application computing task based on the data, including selecting different execution tasks and execution nodes; S3. Allocate computing tasks to avoid idle or overloaded resources; S4. Collect task execution results, including task execution time and energy consumption, and optimize task offloading strategies and resource management solutions through a closed-loop feedback mechanism Preferably, the dynamically adjusting the offloading strategy of the application computing task includes: Split the tasks in the task graph using the elastic splitting model of application tasks; By applying the elastic configuration algorithm of the task, different execution tasks and execution nodes are selected in the task graph to achieve efficient and elastic configuration of computing tasks.

[0010] Preferably, the allocation of computing tasks includes: Use a task offloading algorithm based on bidirectional collaborative adaptation, taking into account multiple factors such as computing power on the device and cloud, network conditions, and task complexity; Real-time optimization of the allocation of computing network resources overcomes the limitations of the single optimization dimension of existing technologies and achieves global optimization allocation of computing resources.

[0011] Preferably, the method adopts an adaptive learning mechanism, including but not limited to machine learning or deep learning models, to dynamically optimize task offloading strategies and resource management solutions to improve the stability and reliability of the system in a dynamically changing environment.

[0012] The present invention provides a flexible configuration and bidirectional collaborative adaptation system for application computing tasks. It has the following beneficial effects: The present invention provides a system for elastic configuration and bidirectional collaborative adaptation of application computing tasks. The technical solution of the present invention innovatively combines the elastic configuration mechanism of application computing tasks and the bidirectional collaborative adaptation unloading algorithm of application and computing network, and introduces an adaptive learning mechanism to form a bidirectional adaptive feedback system. Through the continuous updating of machine learning models and neural network models, the system continuously optimizes task scheduling strategies and resource allocation schemes in a dynamically changing environment, thereby achieving more efficient computing unloading. Compared with the prior art, the present invention responds to environmental changes in real time through the elastic configuration mechanism of application computing tasks. The system can adaptively adjust the execution structure of the application graph when the network status and device resources change, and select different execution tasks and execution nodes in the task graph, thereby selecting reasonable computing tasks and avoiding performance degradation. This patent proposes a task unloading algorithm based on bidirectional collaborative adaptation, which takes into account multiple factors such as the computing power, network conditions, and task complexity of the device side and the cloud side, makes reasonable subtask unloading decisions, and dynamically schedules computing network resources, thereby achieving global optimal allocation of computing resources and overcoming the limitations of the single optimization dimension of the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 It is a schematic diagram of heterogeneous edge device unloading of the present invention; Figure 2 A schematic diagram of an elastic configuration model for application computing tasks of the present invention; Figure 3 It is a schematic diagram of the implementation principle of the system architecture diagram of the present invention; Figure 4 It is a schematic block diagram of the module system of the present invention. DETAILED DESCRIPTION

[0014] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0015] like Figure 1-4 As shown, an embodiment of the present invention provides a flexible configuration and bidirectional collaborative adaptation system for application computing tasks, including a data collection module, a bidirectional collaborative adaptation module for application computing tasks and computing network resources, a resource scheduling module, and a task execution feedback module. The data collection module is used to monitor and collect device status, network status, and task characteristic data in real time; The resource scheduling module is used to reasonably allocate computing tasks based on real-time data of device and network status; The bidirectional collaborative adaptation module of application computing tasks and computing network resources is used to combine the application task execution and computing network resource status to achieve efficient collaborative adaptation between local devices and remote computing nodes; The task execution feedback module is used to collect task execution results, including task execution time and energy consumption, and optimize task offloading strategies and resource management through a closed-loop feedback mechanism.

[0016] The bidirectional collaborative adaptation module of application computing tasks and computing network resources includes three parts: 1. Application elastic configuration part. The elastic configuration part of application tasks constructs an elastic splitting model of application tasks and an elastic configuration algorithm for application tasks. The former is used to split the tasks in the task graph according to the task characteristics and the real-time status of device resources; the latter is used to dynamically adjust the execution tasks and execution nodes in the task graph to optimize the utilization efficiency of computing resources and the task response time; 2. Task offloading decision part. The task offloading algorithm of the part is based on bidirectional collaborative adaptation. It considers multiple factors such as the computing power of the device and the cloud, network conditions, and task complexity to make reasonable sub-task offloading decisions. 3. Computing network resource scheduling part. The part adopts a dynamic resource scheduling algorithm based on bidirectional collaborative adaptation to optimize the allocation of computing network resources in real time, overcoming the limitations of the single optimization dimension of the prior art.

[0017] The resource scheduling module rationally regulates its own computing power and network resources according to the decision of the two-way collaborative adaptation module of application computing tasks and computing network resources, and adjusts the computing power required according to the needs of task execution.

[0018] The task execution feedback module can adjust the task offloading strategy and resource management plan in real time through a closed-loop feedback mechanism to ensure long-term performance improvement and user experience optimization.

[0019] A method for elastic configuration and bidirectional collaborative adaptation of computing offload tasks in mobile computing and Internet of Things environments, specifically comprising the following steps: S1. Collect key performance indicators and network status data of equipment in real time; S2. Dynamically adjust the offloading strategy of application computing tasks based on data, including selecting different execution tasks and execution nodes; S3. Allocate computing tasks to avoid idle or overloaded resources; S4. Collect task execution results, including task execution time and energy consumption, and optimize task offloading strategies and resource management schemes through a closed-loop feedback mechanism.

[0020] Dynamically adjust the offloading strategies of application computing tasks including: Split the tasks in the task graph using the elastic splitting model of application tasks; By applying the elastic configuration algorithm of the task, different execution tasks and execution nodes are selected in the task graph to achieve efficient and elastic configuration of computing tasks.

[0021] Distributing computing tasks includes: Use a task offloading algorithm based on bidirectional collaborative adaptation, taking into account multiple factors such as computing power on the device and cloud, network conditions, and task complexity; Real-time optimization of the allocation of computing network resources overcomes the limitations of the single optimization dimension of existing technologies and achieves global optimization allocation of computing resources. The method adopts an adaptive learning mechanism, including but not limited to machine learning or deep learning models, to dynamically optimize task offloading strategies and resource management solutions to improve the stability and reliability of the system in a dynamically changing environment.

[0022] Elastic configuration model and algorithm for application computing tasks like Figure 1 As shown in FIG, the present invention conducts in-depth research on the application characteristics of complex multi-tasks and constructs a highly dynamically configurable application computing model. Specifically, for the same application service, task graph structures of different complexity can be configured according to the requirements of its service quality. Represents a set of task graphs with different qualities of service for application G, and uses Indicates time t The generated application service selects the application graph structure. Among them, the application task graph It is represented as a directed acyclic graph, which contains multiple subtask nodes. , subtask Superscript and The starting task and the ending task are marked respectively. The more complex the application graph structure is, the higher the service quality provided to users when the application task is completed.

[0023] For application task graph Any subtask in It has dependencies with other subtasks, using and Respectively The collection of predecessor and successor subtasks of The last completed predecessor task is defined as , whose index is , in order to further consider its impact.

[0024] This method is used to characterize the elastic configurability of the application. This method of task splitting enables the system to more effectively understand and handle the interrelationships between tasks, providing an important basis for subsequent scheduling and execution. At the same time, the following quantitative method SLO is given based on the completion latency and computing power requirements of the application tasks: in, Represents the maximum allowed time Complete the termination task within The possibility score is used here to represent the service quality of the task and use it as a metric to select appropriate subtask configurations and execution nodes from the application graph. Represents the estimated computing power required to execute the overall application task.

[0025] Design of task offloading and dynamic resource scheduling algorithm based on bidirectional cooperative adaptation The online adaptation and bidirectional collaborative adaptation algorithms for computing offloading tasks based on deep reinforcement learning (DRL) can effectively deal with complex decision-making problems in computing offloading scenarios. By utilizing the interaction between the agent and the environment in reinforcement learning, the algorithm can learn the optimal offloading decision strategy under dynamic network environments and resource conditions. The following is the specific implementation idea of ​​this algorithm: Overall structure of the algorithm The core of the algorithm is a deep Q network (DQN) or deep deterministic policy gradient algorithm (DDPG), which can optimize computing offloading and resource allocation decisions in a dynamic environment. The system collects the status of devices and networks, and combines the characteristics of tasks to adapt the computing offloading strategy online to achieve two-way collaborative adaptation between applications and computing networks. Among them, the following elements are included: State (S): The current state of the system, which consists of information such as device status, network status, and task characteristics.

[0026] Action (A): Compute offloading strategy, including whether to offload, to which node to offload, and resource (bandwidth and computing power) allocation strategy.

[0027] Reward (R): Evaluates the quality of the current decision, taking into account factors such as computing latency, energy consumption, and network transmission time.

[0028] Policy: The rules that determine how to move from state to action. Deep neural networks are used to approximate the optimal policy.

[0029] State Space The system state vector contains the current resource status of the device and network, as well as the characteristics of the task. It can be expressed in vector form as: in, Indicates the CPU utilization of the current device. Indicates the memory usage of the current device. Current available network bandwidth, : Current network packet loss rate; Task_info indicates the information of the current subtask, including the size of the subtask, the computational complexity of the subtask, the execution location and status of the pre- and post-task tasks, and the service quality requirements of the subtask. Devices_info indicates the available computing resources, task queue execution status, and network bandwidth of the currently connected edge node. These can be obtained through the data collection module.

[0030] Action Space The action space defines the decisions that the system can make at each moment, mainly including: Offload target node selection: whether to offload. If offloading is selected, decide which computing node (such as local, edge server, or cloud) the task will be offloaded to.

[0031] Resource allocation strategy: Dynamically adjust the bandwidth and computing power allocation of offloaded tasks to meet the latency and energy consumption requirements of the tasks.

[0032] Actions can be represented by vectors as: in, represents the target node (local, edge node or cloud), represents the network bandwidth allocated for offloading tasks, Indicates the allocated computing resources (such as CPU, GPU, etc.).

[0033] Reward function design The design of the reward function needs to comprehensively consider performance indicators such as computing delay, energy consumption, and network transmission time. The goal is to maximize the user experience while minimizing system resource consumption. The reward function can be designed in the following form: in, is the service quality score provided by the task at time t, Represents the energy consumed by task execution, taking into account device energy consumption. The total delay required for the task to complete, including computation time and network transmission time. , : Weight coefficient, used to adjust the balance between service quality and energy consumption delay.

[0034] In this design, if the offloading strategy improves the service quality of the task, the system will receive a positive reward; if the delay or energy consumption is too high, the system will receive a negative reward.

[0035] Training and decision-making process of reinforcement learning DQN (Deep Q-Network) Experience replay: During the interaction, the system stores each state transition (S, A, R, S') into the experience pool so that samples can be randomly drawn from the pool for training later. This method can break the correlation of state transitions and improve the training effect of the model.

[0036] Q value update: Use the target network and the main network (policy network) to estimate the action value in the current state. The Q value update formula is as follows: Among them, η is the learning rate and γ is the discount factor.

[0037] DDPG (Deep Deterministic Policy Gradient) DDPG is a good choice for optimizing continuous action spaces. Unlike DQN, DDPG uses a policy network to directly output the optimal action and optimizes the offloading decision through policy gradient updates: Policy evaluation: Use a Q network to evaluate the actions given by the policy network and continuously optimize the two networks so that the policy network can output the optimal offloading decision.

[0038] Training and decision-making process Initialization: Initialize the network parameters, experience replay pool, and environment state of DQN or DDPG.

[0039] State observation: At each time step t, the system observes the current state .

[0040] Action selection: based on the current state and policy network, select the best action , including whether to uninstall, where to uninstall, and resource allocation.

[0041] Execute action: According to the action Perform task offloading and resource allocation.

[0042] Environmental feedback: Get system feedback after executing an action, including the new state and instant rewards .

[0043] Experience storage and training: Store (S, A, R, S') into the experience pool, randomly extract samples from the experience pool, and update the network parameters.

[0044] Iterative update: Repeat the above steps and continuously update the strategy until the model converges.

[0045] Implementation principle of system architecture diagram like Figure 2 As shown, in order to better demonstrate the various modules of the present invention and their implementation principles, Figure 2 It is a description of the system architecture diagram, which helps to understand how to achieve flexible configuration and two-way collaborative adaptation of computing offloading tasks through the collaboration of various modules.

[0046] The system architecture is mainly composed of four core modules: data collection module, bidirectional collaborative adaptation module for application computing tasks and computing network resources, resource scheduling module and task execution feedback module. Each module works closely together to ensure that when unloading, application tasks can be flexibly configured, reasonable task unloading decisions can be made, and computing network resources can be reasonably scheduled.

[0047] Data collection module This module is responsible for monitoring and collecting real-time data related to device status, network status, and task characteristics, providing accurate input for subsequent resource scheduling and task offloading decisions. First, the module collects key performance indicators of the device in real time, such as CPU and GPU usage, memory usage, device temperature, battery power, etc. Device status information can reflect the computing resources currently available to the system and is the basis for decision-making on task offloading and scheduling. For example, when the device battery is low, the system may give priority to offloading computing tasks to remote devices to reduce the energy consumption of local devices. In addition to device and network status, the data collection module also collects relevant information about the task, including task size, computational complexity, data dependencies, execution latency requirements, etc. These characteristics of the task determine its resource requirements and suitable execution environment. For example, for computationally intensive tasks, the system tends to allocate them to high-performance computing nodes, while for latency-sensitive tasks, local or low-latency nodes are given priority.

[0048] In addition, the data collection module will regularly monitor the device's current network bandwidth, latency, packet loss rate and other indicators to provide real-time network status feedback to the task offloading module. If the network condition is poor, the system may choose to execute the task locally first to avoid task failures caused by high latency or packet loss in data transmission. After that, the module will continuously transmit the collected data to the two-way collaborative adaptation module of the application / computing network resources to ensure that the system can make decisions based on the latest device and network status. As the basic data layer of the entire system, the data collection module supports the adaptive adjustment and optimization of the system by regularly collecting and sharing data.

[0049] Bidirectional collaborative adaptation module for application / computing network resources The bidirectional collaborative adaptation module of application / computing network resources is responsible for formulating flexible task configuration and dynamic task offloading strategies, and efficiently allocating computing resources and network bandwidth according to offloading targets, ensuring that different tasks can reasonably utilize existing resources and maximize system performance and task execution efficiency. It mainly includes the following three parts: Applying elastic configuration This part splits and flexibly configures the application tasks according to the task characteristics and the current resource status of the system to optimize the system's resource utilization and task execution efficiency. First, based on the task characteristic information (such as computational complexity, data dependencies, etc.) provided by the data collection module, the application elastic configuration module will decompose the complex task into multiple subtasks. These subtasks can be divided into different levels according to their computational intensity, data transmission volume, and service quality requirements. For example, tasks with lower service quality may only require a small amount of computation and resources, while tasks with higher service quality require more resources and more complex computations. After each task is disassembled, the application elastic configuration module organizes it into a directed acyclic graph (DAG), where each node represents a subtask and the edge represents the dependency between tasks. The task graph can be dynamically adjusted according to the quality of service (QoS) requirements, and the system selects the optimal task graph to execute based on the currently available resources and network conditions. For example, a high-complexity task graph can be selected in low-latency situations to provide higher quality services.

[0050] After completing the task splitting, this part will realize the elastic configuration of the application, that is, allowing the system to adaptively adjust the task execution mode according to the dynamic changes of resources. Specifically, the system can automatically reduce the service quality requirements of the task according to the current status of the device (such as low power or unstable network), thereby reducing resource consumption. Conversely, if the system resources are sufficient, the module can increase the execution priority and service quality of the task to improve the user experience. Finally, the application elastic configuration module passes the task graph structure and its priority information to the resource scheduling module. The resource scheduling module uses this information to further make resource allocation and scheduling decisions to ensure that subtasks are executed on the most appropriate device.

[0051] Task offloading decision part This part receives input data from the data collection module and the application elastic configuration module, and uses intelligent algorithms to make decisions. This part combines the device status, network status, and task characteristics, and uses intelligent offloading algorithms to determine which tasks need to be executed locally and which tasks should be offloaded to remote computing nodes or edge devices, and configures the bandwidth and computing power of edge devices. First, the task offloading decision part uses intelligent algorithms (such as machine learning or neural network algorithms) to analyze the real-time device and network status provided by the data collection module, and combines the task graph structure output by the application elastic configuration module to dynamically determine the execution location of the task. For example, if the CPU resources of the device are relatively tight and the network conditions are good, the module will give priority to offloading tasks to remote computing nodes for processing.

[0052] Computing network resource scheduling This part uses intelligent algorithms to output configuration information for offloading devices, allocates appropriate network resources to different nodes, and ensures that network resources will not become a system bottleneck. For tasks with high network dependence, the system will prioritize allocating more bandwidth to avoid transmission delays affecting task execution. In addition, based on the CPU, GPU, and memory utilization of the device, the resource scheduling module regulates the computing resources that can be allocated to each task in real time. High-priority or computationally intensive tasks will be given priority to obtain more computing power, while low-priority or less resource-demanding tasks will be allocated fewer resources. By dynamically adjusting the allocation of computing resources, the module ensures load balancing of tasks between different nodes and avoids node overload or resource waste.

[0053] In addition, based on the characteristics of the intelligent algorithm, the module does not make a one-time decision, but continuously monitors the effect of task execution through interaction with the task execution feedback module. If some tasks encounter high network latency or execution failure after unloading, the module will adjust the unloading strategy in time and re-evaluate the execution location of the task to reduce the possibility of errors. Through this dynamic adjustment mechanism, the system can maintain the stability and efficiency of task execution in a constantly changing environment. After determining the task unloading, the module will also optimize the execution order of the task. The module will design the best execution order and resource allocation plan based on factors such as task priority, resource consumption, and network transmission time to ensure that the task can be processed in the best way.

[0054] Task execution feedback module The task execution feedback module is used to monitor and record the actual effect of task execution, and pass feedback information to other modules for further optimization. It monitors and manages the energy consumption of the device, and adjusts the allocation of computing tasks in combination with the strategy of the online adaptation module to optimize the device's battery life and performance.

[0055] The data collection module is responsible for real-time monitoring and collection of data such as device status, network status, and application characteristics. Specifically, the information collected by this module includes: device status (CPU and GPU utilization, memory usage, battery power), network status (network bandwidth, latency, packet loss rate), and application characteristics (application type, computing power requirements under different service qualities of the current application, and maximum runtime latency). For tasks executed locally, monitor the usage of device resources; for tasks offloaded to the edge, monitor network transmission time and edge-side execution efficiency. If an exception or failure occurs during task execution, the task execution feedback module will record the error information and notify the task offloading decision module so that the strategy can be adjusted in time to avoid further errors.

[0056] The core innovation of the present invention lies in the flexible configuration and bidirectional collaborative adaptation mechanism of computing offloading tasks, as well as the multi-dimensional computing network resource management and task scheduling strategy based on this mechanism. Specifically, the following points should be protected in this case: 1. Elastic configuration mechanism for application computing tasks: The elastic configuration mechanism for applications can monitor and evaluate the real-time status of the computing network in real time, and dynamically adjust the execution structure of the application task graph based on this information, and select different execution tasks and execution nodes in the task graph to achieve efficient and elastic configuration of computing tasks. This part needs to focus on the following key technologies: the design of the elastic splitting model of application tasks, the elastic configuration algorithm of application computing tasks, and the computing network data collection module.

[0057] 2. Bidirectional collaborative adaptation of application tasks and computing network resources: Based on 1), combined with application task execution and computing network resource status, the application / network collaborative adaptation optimization is performed between local devices and remote computing nodes (such as cloud edge nodes): while executing the decision to offload application tasks, the computing network resources are dynamically scheduled to achieve dynamic collaborative adaptation of applications and networks. This part needs to focus on protecting key technologies such as bidirectional optimization of application configuration and computing network resource scheduling, application / network collaborative adaptation algorithm based on reinforcement learning, and model training methods.

[0058] The method proposed in the present invention has broad application prospects, covering multiple technical fields and industries, which are specifically listed as follows: 1. Gaming and Entertainment For online games and multimedia entertainment applications, the technology of the present invention can optimize the computational offloading of graphics rendering, physical calculations, and artificial intelligence tasks, improve the fluency and image quality of the game, reduce latency, and provide a better user experience. This is especially important for cloud computing-based games (cloud games) and augmented reality (AR) and virtual reality (VR) applications, as these applications require real-time computing and fast response.

[0059] 2. Telemedicine and Health Monitoring In the field of telemedicine and health monitoring, portable devices and wearable devices often need to perform complex computing tasks, such as real-time monitoring of patients' physiological data or medical image analysis. By offloading these computing tasks to the edge or cloud, the technology of the present invention can reduce the computing burden of the device, extend the battery life of the device, and improve the accuracy and speed of computing tasks, providing patients with faster and more accurate diagnosis and care.

[0060] 3. Autonomous Driving and Intelligent Transportation Systems In autonomous driving and intelligent transportation systems, vehicles need to process a large amount of data from sensors in real time to respond quickly. The technology of this invention can help vehicles offload complex computing tasks to edge nodes or the cloud to optimize the use of computing resources, reduce data transmission delays and computing loads, and improve the real-time decision-making capabilities and safety of vehicles.

[0061] 4. Industrial Automation and Robotics Industrial automation and robotics systems usually involve complex real-time computing tasks, such as machine vision, path planning, and machine learning reasoning. By applying the technology of this invention, these tasks can be optimized and offloaded and calculated on edge nodes or cloud servers, allowing robots and automation systems to perform tasks more efficiently, reduce latency and energy consumption, and improve production efficiency and system reliability.

[0062] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A flexible configuration and two-way collaborative adaptation system for application computing tasks, comprising a data collection module, a two-way collaborative adaptation module for application computing tasks and computing network resources, a resource scheduling module, and a task execution feedback module, characterized in that: The data collection module is used to monitor and collect device status, network status and task characteristic data in real time; The bidirectional collaborative adaptation module of the application computing task and the computing network resources is used to combine the application task execution and the computing network resource status; The resource scheduling module is used to reasonably allocate computing tasks according to real-time data of device and network status; The task execution feedback module is used to collect task execution results, including task execution time and energy consumption, and optimize task offloading strategies and resource management through a closed-loop feedback mechanism.

2. The system for flexible configuration and bidirectional collaborative adaptation of application computing tasks according to claim 1, characterized in that: The bidirectional collaborative adaptation module of application computing tasks and computing network resources includes an application elastic configuration part, which includes an elastic splitting model of application tasks and an elastic configuration algorithm of application tasks. The elastic splitting model of application tasks is used to split the tasks in the task graph according to task characteristics and real-time status of device resources, and the elastic configuration algorithm of application tasks is used to dynamically adjust the execution tasks and execution nodes in the task graph.

3. The system for flexible configuration and bidirectional collaborative adaptation of application computing tasks according to claim 1, characterized in that: The bidirectional collaborative adaptation module of application computing tasks and computing network resources includes a task offloading decision part and a computing network resource scheduling part. The two parts respectively contain a task offloading algorithm based on bidirectional collaborative adaptation and a dynamic resource scheduling algorithm. The task offloading algorithm based on bidirectional collaborative adaptation is used to consider multiple factors such as the computing power of the device and the cloud, network conditions, task complexity, etc., and make reasonable subtask offloading decisions. The dynamic resource scheduling algorithm based on bidirectional collaborative adaptation is used to optimize the allocation of computing network resources in real time.

4. The system for flexible configuration and bidirectional collaborative adaptation of application computing tasks according to claim 1, characterized in that: The resource scheduling module rationally regulates its own computing power and network resources according to the decision of the bidirectional collaborative adaptation module of the application computing task and computing network resources, and adjusts the required computing power according to the needs of task execution.

5. The system for flexible configuration and bidirectional collaborative adaptation of application computing tasks according to claim 1, characterized in that: The task execution feedback module can adjust the task offloading strategy and resource management scheme in real time through a closed-loop feedback mechanism.

6. A method for elastic configuration and bidirectional collaborative adaptation of computing offload tasks in mobile computing and Internet of Things environments, characterized in that: The specific steps include: S1. Collect key performance indicators and network status data of equipment in real time; S2. Dynamically adjust the offloading strategy of the application computing task based on the data, including selecting different execution tasks and execution nodes; S3. Allocate computing tasks; S4. Collect task execution results, including task execution time and energy consumption, and optimize task offloading strategies and resource management schemes through a closed-loop feedback mechanism.

7. A method for elastic configuration and bidirectional collaborative adaptation of computing offload tasks in mobile computing and Internet of Things environments according to claim 6, characterized in that: The dynamic adjustment of the offloading strategy of the application computing task includes: Split the tasks in the task graph using the elastic splitting model of application tasks; Different execution tasks and execution nodes are selected in the task graph by applying the elastic configuration algorithm of the task.

8. The method for elastic configuration and bidirectional collaborative adaptation of computing offloading tasks in mobile computing and Internet of Things environments according to claim 6, characterized in that: The distributed computing tasks include: Use a task offloading algorithm based on two-way collaborative adaptation, taking into account multiple factors such as the computing power of the device and the cloud, network conditions, and task complexity; and optimize the allocation of computing network resources in real time.

9. A method for elastic configuration and bidirectional collaborative adaptation of computing offloading tasks in mobile computing and Internet of Things environments according to any one of claims 6 to 8, characterized in that: The method adopts an adaptive learning mechanism, including but not limited to a machine learning or deep learning model, to dynamically optimize task offloading strategies and resource management schemes.

Images