An end-to-end cloud coordination low-latency response guarantee method

Abstract

This invention relates to a method for ensuring low-latency response in edge-cloud collaborative processing, comprising: S1: acquiring the load level of each local processing node and the network latency between each local processing node, edge processing node, and cloud processing node, and sending this information to the cloud processing node; S2: dynamically allocating the processing position of each task according to the load level of each local processing node and the network latency between each local processing node, edge processing node, and cloud processing node. The edge-cloud collaborative low-latency response method of this invention effectively balances real-time performance and computational cost, and better adapts to dynamically changing community environments.

Description

Technical Field

[0001] This invention relates to the field of electronic information data processing, and in particular to a method for ensuring low-latency response in edge-cloud collaboration. Background Technology

[0002] A smart community refers to a community that achieves intelligent community management, refined services, and efficient operation through technologies such as the Internet of Things (IoT), big data, and artificial intelligence. This requires the use of numerous IoT devices, such as access control systems, cameras, elevator sensors, and environmental monitors. However, with the surge in the number of IoT devices within the community, the resulting data traffic is also growing exponentially, placing extremely high demands on the real-time performance and reliability of the system's data transmission. This is especially true in situations such as security alarms, fire warnings, fall detection for the elderly, or emergency calls, where millisecond-level response times are essential; any delay can have serious consequences.

[0003] Current smart communities primarily employ a centralized "edge-cloud" processing architecture, where all data is uploaded to cloud servers for processing and decision-making. However, the combined latency of data transmission networks and cloud processing in this "edge-cloud" architecture results in high latency, making it difficult to meet millisecond-level response requirements. Furthermore, the lack of dynamic awareness of network status and edge processing node load leads to rigid resource scheduling, easily causing response timeouts during peak periods. Moreover, in the event of network outages or cloud center failures, critical local services such as access control linkages or emergency alarms will be unable to operate autonomously, resulting in service paralysis and severely inadequate reliability. Additionally, the lack of differentiated protection mechanisms means that high-priority services such as fire alarms compete for resources with low-priority services like environmental data reporting, causing multi-task QoS conflicts. Summary of the Invention

[0004] Based on this, the purpose of this invention is to enable smart communities to dynamically adapt to changes in network conditions and load, and to operate autonomously in abnormal situations, providing a method for ensuring low-latency response through end-edge-cloud collaboration.

[0005] A method for ensuring low-latency response through edge-cloud collaboration includes the following steps: S1: Obtain the load level of each local processing node and the network latency between each local processing node, edge processing node and cloud processing node, and send it to the cloud node; S2: Dynamically allocate the processing location of each task based on the load level of each local processing node and the network latency between each local processing node, edge processing node, and cloud processing node; the detailed steps of the task offloading decision model in processing data include: S21: Predict the response time of each task on the cloud processing node and the corresponding local processing node and edge processing node respectively; S22: Based on the response time of each node, a reward function is used to calculate the score of each node; the calculation formula of the reward function is: , , in, This refers to the actual execution response time. For resource utilization rate; As a penalty for violations, the longer the actual execution response time of the task exceeds the deadline, the larger the value will be. For energy consumption costs, , , and These are the weighting coefficients; This is the minimum quality requirement for the task chain. This is a link quality penalty item. Link quality index

[0006] , in, For CPU utilization, For memory utilization, For bandwidth utilization, , and As weight; S23: Assign the task to the node with the highest score for processing.

[0007] Compared to existing technologies, this invention effectively balances the real-time performance and computational consumption among different processing nodes by setting different computing resources and functions for the edge and cloud. At the same time, by calculating the reward function of each task in different processing nodes, it improves the prediction and judgment of the real-time performance required by each task and the allocation effect, as well as the allocation efficiency. This enables dynamic allocation and transmission optimization of tasks among the three types of processing nodes, better adapting to the dynamically changing community environment.

[0008] Furthermore, when an edge processing node is simultaneously assigned to multiple tasks, a decision is made based on task priority to determine whether to preempt the task currently being executed on the edge processing node; wherein the step of determining whether preemption can be performed includes: a. Compare the priorities of the two tasks. If the priority of the new task is higher than that of the currently executing task, further judgment is made; otherwise, the currently executing task is skipped. , , in, The deadline for the task. This is used to identify critical tasks; the highest priority task is assigned a value of 1, and the rest are assigned a value of 0. and As weight, Prioritize tasks that require preemption. This sets the priority of the tasks being executed in the edge processing node.

[0009] b. If the remaining time of the new task is less than the remaining time of the currently executing task, preempt the currently executing task. The expression is: , in, The deadline for the new task. For the current moment, This represents the remaining time for the current task.

[0010] Furthermore, the task unloading decision model is trained using the DQN algorithm and employs a Q-network to predict the expected value of the cumulative future reward; the expression for the Q-network is: , Its objective function is: , in, This is the current state. For the selected action, To provide timely rewards, For the next state, For the next action, For learning rate, This is the discount factor.

[0011] Furthermore, during training, an experience replay mechanism is used to construct an experience replay pool, and small batches of experience data are randomly sampled from the experience replay pool for the target network. Stable training; the target network The expression is: , Among them, among them, This is the current state. For the selected action, To provide timely rewards, For the next state, For the next action, This is the discount factor.

[0012] Furthermore, the formula for calculating the local processing response time is as follows: , in, Local processing time; The formula for calculating the response time of edge processing is as follows: , in, This refers to the transmission time of data uploaded to the edge processing node. Queuing time for tasks on edge processing nodes. For edge processing node processing time; The data upload time to the edge processing node The calculation formula is: , in, For task data size, For uplink bandwidth, Delay in network transmission; The task queuing waiting time of the edge processing node Based on the M / M / 1 queuing model, the calculation formula is as follows: , in, For the task arrival rate of edge processing nodes, For edge processing node processing time; The formula for calculating the response time of cloud processing is: , in, This refers to the transmission time of data uploaded to the edge processing node. Queuing time for tasks in the cloud. For cloud processing time, This refers to the data transfer time from the cloud. The expression for the link quality index is: , Among them, For data round-trip time, For packet loss rate, Available broadband, , and All are weights and their sum is 1.

[0013] Furthermore, the network status is continuously monitored, data is cached when the network is interrupted, and data is synchronized after the network is restored; during the data caching process during the network interruption, if the data buffer space is full before the network is restored, some data is selectively deleted. Among them, the network connection health index is used to judge the network status, and its expression is: , in, This represents the number of consecutive failed heartbeats. This represents the maximum number of allowed failures.

[0014] Furthermore, the criteria for selecting data to be deleted are as follows: , , in, Prioritize the data for each task. For the data caching market, d represents various tasks.

[0015] Furthermore, during network recovery, to ensure data synchronization consistency, the specific synchronization strategy is as follows: a. Sort local cached data by the timestamp of the event.

[0016] b. Check if there are data records for the corresponding time in the cloud.

[0017] c. If the cloud data is empty, upload the local data to the cloud; if the cloud data is critical data and the local data is non-critical data, download the cloud data to the local machine; if both the cloud data and the local data are critical data, merge the local and cloud data.

[0018] Furthermore, the quality of the transmission link is determined by the link quality index, and the sampling frequency is dynamically adjusted according to the link quality index to optimize the transmission quality.

[0019] Based on the same inventive concept, the present invention also includes an electronic device comprising a processor; a memory for storing a computer program executed by the processor; wherein, when the processor executes the computer program, it implements the edge-cloud collaborative low-latency response guarantee method described above.

[0020] To better understand and implement this invention, the following detailed description is provided in conjunction with the accompanying drawings. Attached Figure Description

[0021] Figure 1 This is an overall flowchart of the response protection method of the present invention. Detailed Implementation

[0022] This invention addresses the problems of high latency in centralized processing, rigid resource scheduling, insufficient reliability guarantees, and QoS conflicts among multiple tasks in current smart communities. It designs a low-latency response guarantee architecture with three-layer collaboration between "end-edge-cloud". The end layer performs data collection and basic preprocessing, the edge layer performs real-time data processing, local linkage, and intelligent decision-making, and the cloud layer performs big data analysis and global policy distribution. Furthermore, it dynamically allocates tasks and optimizes transmission among the three layers, effectively balancing real-time performance and computational consumption, and better adapting to the dynamically changing community environment.

[0023] Specifically, please refer to Figure 1 , Figure 1 This is an overall flowchart of the response guarantee method of the present invention. The present invention provides a low-latency response guarantee method for edge-cloud collaboration, which specifically includes: S1: Obtain the load level of each local processing node and the network latency between each local processing node, edge processing node, and cloud processing node. The "end-edge-cloud" three-layer node system includes local processing nodes as the end layer, edge processing nodes as the edge layer, and cloud processing nodes as the cloud layer. The local processing nodes include IoT devices such as smoke pipe sensors and access control systems. The edge processing nodes include devices deployed within the community, such as data buffers, decision modules, or scheduling modules. Each edge processing node only transmits data with the local processing nodes in its local area. The cloud processing nodes include a central cloud platform such as a big data analysis engine and a global policy distribution center. The "end-edge-cloud" three-layer nodes transmit and exchange data via the network.

[0024] S2: Dynamically allocate the processing location of each task based on the load level of each local processing node and the network latency between each local processing node, edge processing node, and cloud processing node. Specifically, a task offloading decision model is used to dynamically allocate the processing location of each task. This task offloading decision model is deployed in the cloud processing node. The cloud processing node obtains the network latency between different nodes and inputs it into the model for processing calculations. The task offloading decision model includes a state space. and action space , respectively represented as: , , in, Categorize tasks by priority. To meet the real-time requirements of the task, This refers to task response time, including local processing response time, edge processing response time, and cloud processing response time. CPU load rate of edge processing nodes This refers to the load rate of cloud computing. This refers to the link quality index. To process locally on the terminal device, To offload the process to the edge processing node, The process is to offload the data to the cloud. The detailed steps of the data processing by the task's offloading decision model include: S21: Predict the response time of each task on the cloud processing node, the corresponding local processing node, and the edge processing node. The formula for calculating the local processing response time is as follows: , in, This is the local processing time.

[0025] The formula for calculating the response time of edge processing is as follows: , in, This refers to the transmission time of data uploaded to the edge processing node. Queuing time for tasks on edge processing nodes This refers to the processing time for edge processing nodes.

[0026] The data upload time to the edge processing node The calculation formula is: , in, For task data size, For uplink bandwidth, This is due to delays in network transmission.

[0027] The task queuing waiting time of the edge processing node Based on the M / M / 1 queuing model, the calculation formula is as follows: , in, For the task arrival rate of edge processing nodes, This refers to the processing time for edge processing nodes.

[0028] The formula for calculating the response time of cloud processing , in, This refers to the data upload and transmission time to the cloud. Queuing time for tasks in the cloud. For cloud processing time, This refers to the transmission time of the data downloaded from the cloud.

[0029] The expression for the link quality index is: , in, For data round-trip time, For packet loss rate, Available broadband, , and All are weights and their sum is 1.

[0030] S22: Based on the response time of each node, a reward function is used to calculate the score of each node. The formula for calculating the reward function is as follows: , , in, This refers to the actual execution response time. For resource utilization rate; As a penalty for violations, the longer the actual execution response time of the task exceeds the deadline, the larger the value will be. For energy consumption costs, , , and These are the weighting coefficients; This is the minimum quality requirement for the task chain. This is a link quality penalty item. This represents the link quality index. The violation penalty is used to encourage the system to prioritize nodes that can guarantee timely task completion and avoid response timeouts; its expression is: .

[0031] The formula for calculating the resource utilization rate is as follows: , in, For CPU utilization, For memory utilization, For bandwidth utilization, , and For weights.

[0032] S23: Assign the task to the node with the highest score for processing.

[0033] The task unloading decision model is trained using the DQN algorithm. The DQN algorithm constructs a Q-network to handle numerically inexhaustible... The numerical values ​​are obtained to predict the expected value of the accumulated future reward based on the current state and the selected action. The expression of the Q network is: , Its goal is set as follows: , The objective function is obtained as follows: , in, This is the current state. For the selected action, To provide timely rewards, For the next state, For the next action, For learning rate, As a discount factor, For each parameter, This refers to parameters that are updated a certain number of times later.

[0034] During training, an experience replay mechanism is used to construct an experience replay pool, and small batches of experience data are randomly sampled from the experience replay pool to train the target network. Perform stable training. During training, the Q-function and the target network are used. The loss function is calculated, and the gradients of the parameters in the Q-function are continuously updated based on the calculation results using the backpropagation algorithm, thereby continuously minimizing the loss function. The expression for the loss function is: , Among them, among them, This is the current state. For the selected action, To provide timely rewards, For the next state, For the next action, This is the discount factor.

[0035] Furthermore, when an edge processing node is simultaneously assigned to multiple tasks, a decision is made based on task priority to determine whether to preempt the tasks currently being executed on the edge processing node. Specifically, each currently executing task on the edge processing node is sequentially assessed for preemption until it is determined that one of the currently executing tasks can be preempted. The step of determining whether preemption can be performed includes: a. Compare the priorities of the two tasks. If the priority of the new task is higher than that of the currently executing task, further judgment is made; otherwise, the currently executing task is skipped. That is: , , in, The deadline for the task. This is used to identify critical tasks; the highest priority task is assigned a value of 1, and the rest are assigned a value of 0. and As weight, Prioritize tasks that require preemption. This sets the priority of the tasks being executed in the edge processing node.

[0036] b. If the remaining time of the new task is less than the remaining time of the currently executing task, preempt the currently executing task. The expression is: , in, The deadline for the new task. For the current moment, This represents the remaining time for the current task.

[0037] Furthermore, network conditions are continuously monitored, data is cached during network outages, and data is synchronized after network recovery. Network conditions are assessed using a network connection health index, expressed as: , in, This represents the number of consecutive failed heartbeats. This represents the maximum number of allowed failures.

[0038] During data caching after a network interruption, if the data buffer space is full before the network is restored, some data will be selectively deleted. The criteria for selecting data to be deleted are as follows: , , in, Prioritize the data for each task. For the data caching market, d represents various tasks.

[0039] To ensure data consistency during network recovery, the specific synchronization strategy is as follows: a. Sort local cached data by the timestamp of the event.

[0040] b. Check if there are data records for the corresponding time in the cloud.

[0041] c. If the cloud data is empty, upload the local data to the cloud; if the cloud data is critical and the local data is non-critical, download the cloud data to the local machine; if both the cloud and local data are critical, merge the local and cloud data. In this embodiment, a circular buffer is used for temporary data storage.

[0042] Furthermore, the quality of the transmission link is continuously monitored and optimized. The quality of the transmission link is determined by the link quality index, and one or more transmission optimization strategies are dynamically selected based on the link quality index.

[0043] When the link quality index is greater than or equal to 0.8, it indicates that the transmission quality is excellent. At this time, raw data transmission, high sampling rate, no compression or lightweight compression can be used.

[0044] When the link quality index is greater than or equal to 0.5 and less than 0.8, it indicates that the transmission quality is moderate. At this time, fast compression algorithms such as forward error correction or optimizations such as reducing the sampling frequency by 50% can be enabled.

[0045] When the link quality index is greater than or equal to 0.2 and less than 0.5, it indicates poor transmission quality. In this case, optimization can be achieved by using a high compression ratio, enabling multi-path redundant transmission, or further reducing the sampling frequency.

[0046] When the link quality index is less than 0.2, it indicates that the transmission quality is extremely poor. At this time, it can be regarded as a network connection interruption, and the reporting of non-critical data is suspended. Only the P1 service heartbeat and local linkage are guaranteed.

[0047] Compared to existing technologies, this invention effectively balances the real-time performance and computational consumption among different processing nodes by setting different computing resources and functions for the edge and cloud. At the same time, by calculating the reward function of each task in different processing nodes, it improves the prediction and judgment of the real-time performance required by each task and the allocation effect, as well as the allocation efficiency. This enables dynamic allocation and transmission optimization of tasks among the three types of processing nodes, better adapting to the dynamically changing community environment.

[0048] The terminology used in the embodiments of this application is for the purpose of describing particular embodiments only and is not intended to limit the embodiments of this application. The singular forms “a,” “the,” and “the” used in the embodiments and claims of this application are also intended to include the plural forms, unless the context clearly indicates otherwise. It should also be understood that, unless otherwise stated, “a plurality” means two or more; the terms “first,” “second,” “third,” etc., are used only to distinguish and not to describe a particular order or sequence, nor should they be construed as indicating or implying relative importance. The term “and / or” as used herein refers to and includes any or all possible combinations of one or more associated listed items. When the above description relates to drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. In the description of this application, those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0049] The embodiments described above are merely examples of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and the present invention also intends to include these modifications and variations.

Claims

1. A method for ensuring low-latency response through edge-cloud collaboration, characterized in that, Includes the following steps: S1: Obtain the load level of each local processing node and the network latency between each local processing node, edge processing node and cloud processing node; S2: Dynamically allocate the processing location of each task based on the load level of each local processing node and the network latency between each local processing node, edge processing node and cloud processing node; The detailed steps involved in unloading the decision model from data processing for this task include: S21: Predict the response time of each task on the cloud processing node and the corresponding local processing node and edge processing node respectively; S22: Based on the response time of each node, a reward function is used to calculate the score of each node; the calculation formula of the reward function is: ， ， in, This refers to the actual execution response time. For resource utilization rate; As a penalty for violations, the longer the actual execution response time of the task exceeds the deadline, the larger the value will be. For energy consumption costs, , , and These are the weighting coefficients; This is the minimum quality requirement for the task chain. This is a link quality penalty item. The link quality index is used; the formula for calculating the resource utilization rate is: ， in, For CPU utilization, For memory utilization, For bandwidth utilization, , and As weight; S23: Assign the task to the node with the highest score for processing.

2. The edge-cloud collaborative low-latency response guarantee method according to claim 1, characterized in that, When an edge processing node is simultaneously assigned to multiple tasks, a decision is made based on task priority to determine whether to preempt the tasks currently being executed on the edge processing node; wherein the step of determining whether preemption can be performed includes: a. Compare the priorities of the two tasks. If the priority of the new task is higher than that of the currently executing task, further judgment is made; otherwise, the currently executing task is skipped. ， ， in, The deadline for the task. This is used to identify critical tasks; the highest priority task is assigned a value of 1, and the rest are assigned a value of 0. and As weight, Prioritize tasks that require preemption. Priority of tasks being executed in the edge processing nodes; b. If the remaining time of the new task is less than the remaining time of the currently executing task, the currently executing task will be preempted; the expression is: ， in, The deadline for the new task. For the current moment, This represents the remaining time for the current task.

3. The edge-cloud collaborative low-latency response guarantee method according to claim 1, characterized in that, The task unloading decision model is trained using the DQN algorithm and employs a Q-network to predict the expected value of the cumulative future reward; the expression for the Q-network is: ， Its objective function is: ， in, This is the current state. For the selected action, To provide timely rewards, For the next state, For the next action, For learning rate, This is the discount factor.

4. The edge-cloud collaborative low-latency response guarantee method according to claim 3, characterized in that, During training, an experience replay mechanism is used to construct an experience replay pool, and small batches of experience data are randomly sampled from the experience replay pool for the target network. Stable training; the target network The expression is: ， Among them, among them, This is the current state. For the selected action, To provide timely rewards, For the next state, For the next action, This is the discount factor.

5. The edge-cloud collaborative low-latency response guarantee method according to claim 1, characterized in that, The formula for calculating the local processing response time is: ， in, Local processing time; The formula for calculating the response time of edge processing is as follows: ， in, This refers to the transmission time of data uploaded to the edge processing node. Queuing time for tasks on edge processing nodes For edge processing node processing time; The data upload time to the edge processing node The calculation formula is: ， in, For task data size, For uplink bandwidth, Delay in network transmission; The task queuing waiting time of the edge processing node Based on the M / M / 1 queuing model, the calculation formula is as follows: ， in, For the task arrival rate of edge processing nodes, For edge processing node processing time; The formula for calculating the response time of cloud processing is: ， in, This refers to the transmission time of data uploaded to the edge processing node. Queuing time for tasks in the cloud. For cloud processing time, This refers to the transmission time of data downloaded from the cloud. The expression for the link quality index is: ， in, For data round-trip time, For packet loss rate, Available broadband, , and All are weights and their sum is 1.

6. The edge-cloud collaborative low-latency response guarantee method according to claim 1, characterized in that, The system continuously monitors network conditions, caches data during network outages, and synchronizes data after network recovery. During data caching during network outages, if the data buffer space is full before network recovery, some data is selectively deleted. Among them, the network connection health index is used to judge the network status, and its expression is: ， in, This represents the number of consecutive heartbeat failures. This represents the maximum number of allowed failures.

7. The method for ensuring low-latency response in edge-cloud collaboration according to claim 1, characterized in that, The criteria for selecting data to be deleted are: ， ， in, Prioritize the data for each task. For the data caching market, d represents various tasks.

8. The method for ensuring low-latency response in edge-cloud collaboration according to claim 1, characterized in that, To ensure data consistency during network recovery, the specific synchronization strategy is as follows: a. Sort local cached data by the timestamp of the event occurrence; b. Check if there are data records for the corresponding time in the cloud; c. If the cloud data is empty, upload the local data to the cloud; if the cloud data is critical data and the local data is non-critical data, download the cloud data to the local machine; if both the cloud data and the local data are critical data, merge the local and cloud data.

9. The method for ensuring low-latency response in edge-cloud collaboration according to claim 1, characterized in that, The quality of the transmission link is determined by the link quality index, and the sampling frequency is dynamically adjusted according to the link quality index to improve transmission quality.

10. An electronic device, characterized in that, processor; Memory for storing computer programs executed by the processor; When the processor executes the computer program, it implements the edge-cloud collaborative low-latency response guarantee method according to any one of claims 1-9.

Images