A method and system for accelerating task unloading
By constructing a task & computing power mapping table and updating it at low cost when necessary, the shortcomings of existing task offloading schemes in terms of adaptability and efficiency are solved, and rapid decision-making and low-complexity processing of task offloading are realized.
Patent Information
- Application Number
- CN202611348920.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-02
- Publication Date
- 2026-11-17
AI Technical Summary
Existing task offloading solutions are insufficient in adapting to dynamic changes in task characteristics and node computing resources. Static offloading cannot adapt to changes and is ineffective, while dynamic offloading increases latency and complexity due to frequent execution of optimization algorithms, making it difficult to meet the needs of latency-sensitive tasks.
By constructing a task & computing power mapping table, the optimal unloading strategy under different combinations of task characteristics and computing power resources can be solved in advance, and low-cost updates can be performed when task characteristics and computing power resources change, thereby reducing decision latency and system complexity.
It achieves a balance between unloading effectiveness and adaptability, while reducing the latency of task unloading decisions and the complexity of system implementation. It overcomes the high latency and high complexity problems caused by frequent optimization calculations in existing technologies, and adapts to the dynamic changes in task characteristics and node computing resources.
Smart Images

Figure CN123114579A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of information and communication technology, specifically a method and system for accelerating task unloading. Background Technology
[0002] The statements in this section only refer to the background art related to this invention and do not necessarily constitute prior art.
[0003] With the rapid development of mobile internet and the Internet of Things, the complexity of tasks running on terminal devices (such as video processing and AI inference) is increasing, while the computing power and storage resources of edge devices are relatively limited, making it difficult to meet the latency and energy consumption requirements of task processing. To address this, task offloading technology is widely adopted: offloading all or part of the tasks to edge nodes and / or cloud nodes with stronger computing power for processing, in order to reduce task processing latency and extend terminal battery life.
[0004] Existing task offloading solutions mainly fall into two categories. One is static offloading, which offloads tasks to edge nodes or cloud nodes according to preset fixed rules. This method is simple to implement, but it cannot adapt to dynamic changes in task characteristics and node computing resources, resulting in poor offloading performance. The other is dynamic offloading, which uses intelligent optimization algorithms to solve for the optimal offloading ratio in real time based on the current task characteristics and the communication and computing resources of each node. This method has good offloading performance and strong adaptability, but the intelligent optimization algorithm must be executed before each task offloading. On the one hand, this increases the execution latency of the algorithm, making the offloading decision itself part of the task processing latency. On the other hand, frequent optimization calculations also increase the complexity of system implementation, making it difficult to meet the needs of latency-sensitive tasks. Summary of the Invention
[0005] This invention provides a method and system for accelerating task unloading. By constructing and reusing a task & computing power mapping table, the decision-making latency and implementation complexity of task unloading are reduced while taking into account both unloading effectiveness and adaptability.
[0006] To achieve the above objectives, the present invention adopts the following technical solution: The first aspect of this invention discloses a task unloading acceleration method, comprising the following steps: Respond to the user's task unloading request, obtain the task characteristics of the current task and the real-time computing resources of the terminal, edge, and cloud nodes; The task & computing power mapping table is queried to determine whether the difference between the current task's task characteristics and real-time computing power resources and the corresponding nearest entry in the task & computing power mapping table exceeds a preset threshold. The task & computing power mapping table is pre-constructed in the following way: For different combinations of task characteristics and computing power resources, the task unloading processing optimization model is solved to obtain the optimal task segmentation and unloading strategy. The task unloading processing optimization model is used to describe an optimization problem with task segmentation and unloading strategy as decision variables, minimizing task processing latency as the objective, and subject to latency and energy consumption constraints. If the preset threshold is not exceeded, the task segmentation and unloading ratio of the current task are calculated by fitting the task & computing power mapping table based on the key parameters with the greatest difference, and the task is unloaded according to the calculated unloading ratio. If the preset threshold is exceeded, the optimal task splitting and unloading strategy is solved in real time for the current task, the task is unloaded according to the solution results, and the obtained unloading strategy is added as a new entry to the task & computing power mapping table.
[0007] Furthermore, the construction process of the task offloading processing optimization model is as follows: under the conditions of limited task processing latency and energy consumption, the optimization objective is to minimize the task offloading processing latency, and it is constrained by the maximum tolerable task processing time, the maximum energy consumption of task processing, and the total number of tasks; as shown in the following formula: ; ; in, The maximum tolerable time for task processing. This represents the maximum energy consumption for task processing. For the total number of tasks, To account for the delay in task unloading processing, Energy consumption is consumed during task unloading. These represent the workload of unloading tasks to edge nodes and cloud nodes, respectively. The amount of tasks stored for edge nodes.
[0008] Furthermore, the task & computing power mapping table is a multi-dimensional mapping table, which includes at least the task quantity, task complexity, end communication rate, end computing rate, edge communication rate, edge computing rate, cloud communication rate, and cloud computing rate. It is used to record the optimal task partitioning and unloading strategies corresponding to different task characteristics and computing power resource combinations.
[0009] Furthermore, the most recent entry is the entry in the task & computing power mapping table that is closest to the task characteristics and real-time computing power resources of the current task. Determine whether the differences between the current task's characteristics and real-time computing resources and the corresponding entries in the task & computing power mapping table exceed a preset threshold, including: When the absolute value of the difference between the task volume initiated by the user and the task volume of the corresponding most recent entry is greater than the task volume difference threshold, it is determined that an update is needed. When the absolute value of the difference between the communication resources or computing resources of the endpoint, edge, or cloud node and the corresponding resource in the most recent entry is greater than the computing resource difference threshold, it is determined that an update is needed.
[0010] Furthermore, when the difference does not exceed a preset threshold, the difference between each parameter in the task characteristics and real-time computing resources of the current task and the corresponding nearest entry in the task & computing power mapping table is obtained, and the parameter with the largest difference is taken as the key parameter. Key parameters are selected from the following: task volume, end node communication rate, end node computing rate, edge node communication rate, edge node computing rate, cloud node communication rate, and cloud node computing rate; the differences are expressed as a percentage of the differences.
[0011] Furthermore, the fitting calculation of the task segmentation and unloading ratio for the current task includes: Using the key parameter with the greatest difference as the independent variable, and the proportion of tasks offloaded to edge nodes and the proportion of tasks offloaded to cloud nodes as the dependent variables, the functional relationship between the key parameter and the offloading ratio is obtained by fitting the data of the entries adjacent to the key parameter in the task & computing power mapping table. Substitute the value of this key parameter in the current task into the function relationship to calculate the proportion of tasks unloaded to the edge node and the proportion of tasks unloaded to the cloud node. The proportion of tasks processed by the user is determined by subtracting the proportion of tasks unloaded to edge nodes and cloud nodes from the total number of tasks.
[0012] Furthermore, the unloading strategy obtained from the solution is added as a new entry to the task & computing power mapping table, including: based solely on the task characteristics and real-time resources of the computing power nodes of the current task, using an intelligent optimization algorithm to solve for the optimal task unloading target, and adding the optimal task segmentation and unloading strategy corresponding to the current task as a new entry to the task & computing power mapping table, while other entries in the task & computing power mapping table remain unchanged.
[0013] Furthermore, the task characteristics of the current task include the task volume and task complexity; the real-time computing resources of the end, edge, and cloud nodes include the communication resources, computing resources, and storage resources of the user end, edge nodes, and cloud nodes.
[0014] Furthermore, the optimal task splitting and unloading strategy includes the proportion of tasks unloaded to edge nodes, the proportion of tasks unloaded to cloud nodes, and the proportion of tasks processed by the user end; wherein, the proportion of tasks processed by the user end is the total number of tasks minus the proportion of tasks unloaded to edge nodes and the proportion of tasks unloaded to cloud nodes.
[0015] A second aspect of the present invention discloses a task offloading acceleration system, comprising: The task and computing power acquisition unit is used to collect the task characteristics of user-initiated tasks in real time, and periodically or non-periodically collect the communication, computing, and storage resources available at the user end, edge nodes, and cloud nodes. The task unloading model and algorithm unit is used to build a task unloading processing optimization model based on the collaboration of computing resources. Based on the task requirement characteristics, computing node resource status and task unloading processing optimization model collected by the task and computing power acquisition unit, it uses intelligent optimization algorithms to optimize the task unloading processing latency and output the corresponding task segmentation and unloading strategies. The mapping table construction and decision-making unit is used to construct a task & computing power mapping table based on the task segmentation and unloading strategy output by the task unloading model and algorithm unit; determine whether the task & computing power mapping table needs to be updated; if the task & computing power mapping table needs to be updated, send a task & computing power mapping table update request to the update unit, receive the update result returned by the update unit, add the update result to the task & computing power mapping table, and push the updated unloading result to the execution unit; if the task & computing power mapping table does not need to be updated, based on the user-initiated task characteristics and the computing power resources of cloud, edge, and terminal nodes, query the task & computing power mapping table, fit the unloading ratio based on the parameter with the largest difference, and push the unloading strategy to the execution unit. The update unit is used to optimize the task unloading target based on the update instructions output by the mapping table construction and decision unit, the task requirement characteristics collected by the task and computing power acquisition unit and the real-time resources of computing power nodes, and the intelligent optimization algorithm, and then send the results to the mapping table construction and decision unit. The execution unit is used to execute the mapping table construction and the unloading strategy pushed by the decision unit to complete the task unloading process.
[0016] A third aspect of the present invention discloses a computer program product including computer-readable instructions, which, when executed on an electronic device, enable the electronic device to implement the aforementioned task offloading acceleration method.
[0017] A fourth aspect of the present invention discloses an electronic device including at least one processor and a memory connected to the processor, the memory being used to store a computer program; the processor being used to execute the computer program, enabling the electronic device to implement the aforementioned task offloading acceleration method.
[0018] Compared with existing technologies, one or more of the above technical solutions have the following beneficial effects: 1. By pre-constructing a task & computing power mapping table, the optimization results are solidified and reused in the form of a mapping table, eliminating the need to re-execute optimization every time a user initiates a task unloading request. When the task & computing power mapping table is queried and the difference is determined not to exceed a preset threshold, based on the existing data in the mapping table, a functional relationship between the key parameter with the largest difference and the unloading ratio is fitted. The unloading ratio is then calculated by substituting the current parameters, and task unloading is executed. This method effectively reduces the decision latency and system implementation complexity of task unloading, overcoming the shortcomings of existing dynamic unloading schemes, which suffer from high decision latency and high complexity due to the need to execute intelligent optimization algorithms before each unloading.
[0019] 2. When the difference exceeds the preset threshold, the optimal unloading strategy is solved only for the current task and added as a new entry to the task & computing power mapping table. This achieves low-cost updating of the mapping table, enabling the mapping table to be dynamically refreshed as task characteristics and computing power resources change. This makes up for the shortcomings of the existing static unloading scheme that cannot adapt to the dynamic changes in task characteristics and node computing power resources. Thus, while reducing task unloading latency, it also takes into account the unloading effect and adaptability. Attached Figure Description
[0020] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.
[0021] Figure 1 This is a schematic flowchart of a task unloading acceleration method provided in one or more embodiments of the present invention; Figure 2 This is a schematic diagram of a task unloading acceleration system architecture provided for one or more embodiments of the present invention. Detailed Implementation
[0022] The present invention will be further described below with reference to the accompanying drawings and embodiments.
[0023] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.
[0024] The reason why existing dynamic unloading schemes need to re-execute the intelligent optimization algorithm every time they are unloaded is that they do not effectively reuse the results of historical unloading decisions, resulting in repeated execution of optimization calculations in the same or similar scenarios. While static unloading does not require algorithm overhead, it is difficult to guarantee the unloading effect because it is unaware of changes in task characteristics and node resources.
[0025] Therefore, this solution provides a method and system for accelerating task unloading, achieving acceleration while sacrificing some unloading accuracy. Since task unloading processing latency consists of transmission latency and processing latency, the unloading ratio is theoretically not strictly linearly related to changes in parameters such as task volume and computing resources. However, considering the trade-off between unloading accuracy and computational complexity, a fitting calculation method based on a task & computing power mapping table is adopted. This reduces task unloading decision latency and system implementation complexity while maintaining acceptable unloading performance, thus balancing processing latency, unloading effectiveness, and adaptability. By pre-solving the optimal unloading strategy under different task characteristics and computing resource combinations and constructing a mapping table, the table is consulted first during unloading. Simultaneously, through difference threshold judgment, fitting calculation, and incremental updates, the mapping table can be maintained at low cost and dynamically refreshed, avoiding repeated execution of intelligent algorithms and ensuring that the unloading strategy matches the real-time environment. In scenarios where task characteristics and computing resources are relatively stable and the task & computing power mapping table does not need to be updated, decisions can be made by querying the mapping table without executing intelligent optimization algorithms, thereby reducing the processing latency of task unloading. When the mapping table needs to be updated, the unloading effect is guaranteed by solving in real time and the results are backfilled into the mapping table, taking into account the processing latency of subsequent tasks.
[0026] A method for accelerating task unloading includes the following steps: Respond to the user's task unloading request, obtain the task characteristics of the current task and the real-time computing resources of the terminal, edge, and cloud nodes; The task & computing power mapping table is queried to determine whether the differences between the current task's task characteristics and real-time computing power resources and the corresponding entries in the task & computing power mapping table exceed a preset threshold. The task & computing power mapping table is pre-constructed in the following way: For different combinations of task characteristics and computing power resources, the task unloading processing optimization model is solved to obtain the optimal task segmentation and unloading strategy. The task unloading processing optimization model is used to describe an optimization problem with task segmentation and unloading strategy as decision variables, minimizing task processing latency as the objective, and subject to latency and energy consumption constraints. If the preset threshold is not exceeded, the task segmentation and unloading ratio of the current task will be calculated based on the key parameters with the greatest difference, and the task will be unloaded according to the calculated unloading ratio. If the preset threshold is exceeded, the optimal task splitting and unloading strategy is solved in real time for the current task, the task is unloaded according to the solution results, and the obtained unloading strategy is added as a new entry to the task & computing power mapping table.
[0027] By pre-constructing a task & computing power mapping table, the optimization results of the intelligent optimization algorithm are solidified and reused in the form of a mapping table. This eliminates the need to re-execute the intelligent optimization algorithm every time a user initiates a task uninstallation request. Instead, the system queries the task & computing power mapping table and determines that the difference does not exceed a preset threshold. Based on the key parameter with the largest difference, the uninstallation ratio is fitted and the task uninstallation is executed. This reduces the decision latency and system implementation complexity of task uninstallation, and overcomes the shortcomings of existing dynamic uninstallation schemes, which require the execution of the intelligent optimization algorithm before each uninstallation, resulting in high decision latency and high complexity.
[0028] Meanwhile, when the difference exceeds a preset threshold, the optimal unloading strategy is solved only for the current task and added as a new entry to the task & computing power mapping table. This achieves low-cost updates to the mapping table, enabling the mapping table to be dynamically refreshed as task characteristics and computing power resources change. This makes up for the shortcomings of existing static unloading schemes that cannot adapt to dynamic changes in task characteristics and node computing power resources. Thus, while reducing task unloading latency, it also takes into account unloading effect and adaptability.
[0029] The following is combined Figure 1 This section introduces the overall process of accelerating task unloading, including the following steps: Build an optimization model for task offloading and processing based on collaborative computing resources; Statistical analysis of task demand characteristics over a period of time to obtain computing node resources; Utilize intelligent optimization algorithms to optimize task unloading targets; Construct a task & computing power mapping table; When a user initiates a task uninstallation process, the task & computing power mapping table is queried. Determine whether the mapping table needs to be updated. If it does, based on task requirements and computing resources, use intelligent optimization algorithms to optimize the unloading target, unload the task according to the optimization results, and add the optimization results to the mapping table. If no update is required, obtain the key parameter that differs most from the mapping table; based on the task & computing power mapping table, calculate the unloading ratio for the parameter with the largest difference, and unload the task based on this ratio.
[0030] The specific process is detailed below.
[0031] 1) Build an optimization model for task unloading and processing based on the collaboration of computing resources.
[0032] Taking a cloud-edge-device architecture as an example, the task offloading and processing latency It can be represented as: ; in, Unload to edge nodes respectively The workload of cloud nodes; edge node The amount of tasks stored.
[0033] The user end will handle the task volume. Uninstall to The time delay.
[0034] yes The latency of task processing.
[0035] It refers to the latency of the user's own task processing.
[0036] To deliver tasks to the user end The latency of unloading to the cloud node and the latency of the cloud node returning the processing result after processing. Among them: For the transmission rate between the user terminal and the cloud node, This is the ratio of the size of the task processing result to the amount of task.
[0037] To process tasks for cloud nodes The time delay. The rate at which cloud nodes process tasks.
[0038] Energy consumption of task unloading processing It can be represented as: = + + + ; This indicates the energy consumption for task processing on the user's end. This represents the transmission energy consumption when the user terminal offloads tasks to edge nodes and cloud nodes. Energy consumption for task processing of edge nodes. Energy consumption for task processing on cloud nodes.
[0039] Under conditions of limited task processing latency and energy consumption, the minimum task unloading processing latency can be expressed as: ; ; in: The maximum tolerable time for task processing. W represents the maximum energy consumption for task processing, and W represents the total number of tasks.
[0040] 2) Statistically analyze the characteristics of task demand over a period of time, such as task volume and complexity. Obtain available communication, computing, and storage resources for user terminals, edge nodes, and cloud nodes.
[0041] 3) Based on the task demand features obtained from the above statistical analysis, the resource status of computing power nodes, and the task offloading processing optimization model based on computing power resource collaboration, an intelligent optimization algorithm is used to optimize the task offloading processing latency, and output the corresponding task segmentation and offloading strategy. For example, 80% of the task volume is offloaded to the edge node for processing, 10% of the task volume is offloaded to the cloud node for processing, and the remaining 10% of the task volume is processed by the user terminal.
[0042] The intelligent optimization algorithm may adopt existing mature algorithms, and this embodiment takes the particle swarm algorithm as an example.
[0043] Preferably, the particle swarm algorithm is introduced to minimize the task offloading processing latency . One particle represents one task offloading amount, and 0< <W, 0< <W. The optimal position of the particle is set to , and the optimal position of the particle swarm is The fitness function FUN of the particle is set to , and the number of iterations is Nt. The MECs for offloading are set as M={M1, M2, ..., M n}. One MEC is selected from n MECs for task offloading, the fitness function of the particle is calculated, and the current fitness value FUN is compared with the corresponding particle . When , the current FUN is set as the new ; the current is compared with , if , the current is set as the new , and after each iteration, and values are updated. This iteration process continues until Nt is reached. The minimum task offloading processing latency is found by evaluating the fitness value of each particle.
[0044] Preferably, when the particle swarm algorithm is used to solve for the minimum value of the task offloading processing latency , the solution of the objective function and the selected offloading are obtained. The task segmentation and offloading strategy is: the task is divided into 3 parallel subtasks according to , and the segmented subtasks are offloaded to , and the subtasks are offloaded to the cloud node.
[0045] 4) Construct a task & computing power mapping table. The task & computing power mapping table is multi-dimensional and includes at least the following dimensions: task volume, task complexity, end communication rate, end computing rate, edge communication rate, edge computing rate, cloud communication rate, and cloud computing rate.
[0046] Taking a task complexity of 50 cycles / bit as an example, the task & computing power mapping table is shown in Table 1.
[0047] It should be noted that the task & computing power mapping table in this scheme is a multi-dimensional mapping table. Table 1 is only used as an example with a task complexity of 50 cycles / bit.
[0048] For the dimensional expansion of the task & computing power mapping table, the following approach is adopted: each task & computing power mapping table has fixed values for all dimensions except task volume and edge-cloud computing power; when a certain dimension has N values, N task & computing power mapping tables need to be maintained.
[0049] For example, when the task complexity includes three cases: 30 cycles / bit, 50 cycles / bit, and 80 cycles / bit, in addition to Table 1 (task complexity of 50 cycles / bit), two task & computing power mapping tables corresponding to task complexities of 30 cycles / bit and 80 cycles / bit need to be maintained.
[0050] Similarly, if two more dimensions, A and B, are added, with M and N possible values for A and B respectively, then M×N task & computing power mapping tables need to be maintained. The dimensions and sampling method of a single task & computing power mapping table are shown in Table 1. Adding new dimensions only increases the number of tables maintained, and the size of a single table is controllable.
[0051] Table 1 Task & Computing Power Mapping Table
[0052] For example, bold text indicates a task size of 1 Mbit with a complexity of 50 cycles / bit. The user-side communication rate is 10 Mbps, the computing power is 20 cycles / s, and the edge nodes... When the communication rate is 10Mbps and the computing power is 100 cycels / s, and the cloud node's communication rate is 5Mbps and its computing power is 300 cycels / s, 1 Mbit * 80% = 0.8 Mbit of workload will be offloaded to the edge node. The processing involves offloading 1Mbit * 10% = 0.1Mbit of the task to the cloud node for processing, while the user terminal processes 1Mbit * (1 - 80% - 10%) = 0.1Mbit of the task.
[0053] While the combinations of task characteristics and computing resources in task unloading scenarios are diverse, they also exhibit repetition. Storing the optimal strategies by combination allows for the reuse of historical optimization results, replacing repetitive calculations with queries. Therefore, this solution pre-constructs a task & computing power mapping table to solidify and reuse the solutions from the intelligent optimization algorithm. This eliminates the need to repeatedly execute the intelligent optimization algorithm during subsequent task unloading, reducing task unloading decision latency and system implementation complexity. It also solves the problems of high decision latency and excessive complexity caused by existing dynamic unloading schemes that require executing the intelligent optimization algorithm before each unloading operation.
[0054] 5) User-initiated task uninstallation processing.
[0055] 6) Based on the user-initiated task characteristics and the computing power resources of cloud, edge, and terminal nodes, query the entries in the task & computing power mapping table that correspond to the current task characteristics and computing power resources, and compare the differences between the current task characteristics and computing power resources and the corresponding entries.
[0056] It should be noted that in this scheme, the differences between the current task's characteristics and real-time computing resources and the task & computing power mapping table refer to the differences with the most recent entry in the task & computing power mapping table. For example, when the current task's workload is 10.8 Mbit, the most recent entry in the task & computing power mapping table with a workload of 10.8 Mbit is 10 Mbit, and the difference between the two is 8%, which does not exceed the preset threshold, so there is no need to update the mapping table.
[0057] 7) Determine if the task & computing power mapping table needs updating. If updating is required, proceed to 8); otherwise, proceed to 9). Preferably, the method for determining if the task & computing power mapping table needs updating is as follows: If the absolute value of the difference between the amount of tasks initiated by the user and the amount of tasks in the task & computing power mapping table is greater than the task amount difference threshold (e.g., 10%), the task & computing power mapping table needs to be updated.
[0058] If the absolute value of the difference between the communication or computing resources of the endpoint, edge, or cloud node and the corresponding resources in the task & computing power mapping table is greater than the computing power resource difference threshold (e.g., 10%), the task & computing power mapping table needs to be updated.
[0059] Because task characteristics and the computing resources of edge, cloud, and endpoint nodes change dynamically over time, pre-built mapping tables are at risk of becoming outdated. This solution uses a difference threshold instead of a fixed periodicity to determine the timeline. Considering the uncertainty of changes in task characteristics and computing resources, updating based on actual deviations is more in line with the pace of environmental changes than updating at a fixed period. Updates are only triggered when the difference between the task characteristics or computing resources and the corresponding entries in the mapping table exceeds a preset threshold. This ensures the synchronization of the mapping table with the real-time environment while avoiding the additional overhead of frequent updates, balancing the unloading effect and adaptability. It addresses the problem of poor adaptability caused by existing static unloading solutions being unaware of environmental changes.
[0060] It should be noted that determining whether the task & computing power mapping table needs updating includes not only judging the differences between the dimensions shown in Table 1 (task volume, end communication rate, end computing rate, edge communication rate, edge computing rate, cloud communication rate, cloud computing rate) and the most recent entries, but also judging the differences between other dimensions (such as task complexity, edge node storage task volume, etc.) and the dimensions shown in Table 1. When the difference between any of the other dimensions and the most recent entries in the existing task & computing power mapping table exceeds a preset threshold, a new mapping table for the current task needs to be added.
[0061] For example, if the existing mapping table corresponds to three task complexity values of 30 cycles / bit, 50 cycles / bit, and 80 cycles / bit, and the current task complexity is 43 cycles / bit, the difference from the most recent value of 50 cycles / bit is 14%, which exceeds a preset threshold, then a new mapping table needs to be added for this task. When there are deviations in both the task complexity and the amount of task data stored in the edge nodes, as long as the difference between either dimension and the most recent entry in the existing mapping table exceeds a preset threshold, a new mapping table needs to be added for this task (both the task complexity and the amount of task data stored in the edge nodes are determined according to the current task's value).
[0062] 8) Based on the user-initiated task requirements and real-time computing node resources, an intelligent optimization algorithm is used to optimize the task unloading target. Tasks are unloaded according to the optimization results, and the optimization results are added to the task & computing power mapping table.
[0063] When the mapping table needs to be updated, this scheme takes into account that each entry in the mapping table is independent of the others. A full reconstruction would result in a large amount of computational waste that is irrelevant to the current changes. On-demand incremental addition can maintain the validity of the mapping table at the lowest cost. Therefore, based on the task requirements characteristics of the current task and the real-time resources of the computing nodes, the intelligent optimization algorithm is used to solve the optimal unloading target, and the unloading result of the current task is added to the mapping table as a new entry, instead of a full reconstruction of the mapping table. Other entries remain unchanged, which significantly reduces the update complexity of the mapping table and reduces the computational overhead required for the update.
[0064] 9) Obtain the key parameters that differ most from the task & computing power mapping table. Preferably, the key parameters include, but are not limited to: task volume, end node communication rate, end node computing rate, edge node communication rate, edge node computing rate, cloud node communication rate, and cloud node computing rate. The difference ratio is preferred.
[0065] 10) Based on the task & computing power mapping table, for the parameters with the greatest differences, fit and calculate the unloading ratio, and unload tasks based on this result.
[0066] The fitting calculation is as follows: Based on the data that has been accurately calculated in the task & computing power mapping table, fit the functional relationship between each parameter and the unloading ratio, and substitute the parameters of the current task into the functional relationship to calculate the corresponding unloading ratio.
[0067] Preferably, linear interpolation can be used, which is faster and simpler to calculate; alternatively, other function fitting methods such as polynomial fitting can be used to achieve the above fitting calculation.
[0068] Taking Table 1 as an example, assuming that the key parameter with the biggest difference from the task & computing power mapping table in step 9) is the task quantity, which is 10.8 Mbit, then the offloading ratio of the edge node is 80% + (80% - 80%) / (15 - 10) * (10.8 - 10) = 80%, the offloading ratio of the cloud node is 10% + (12% - 10%) / (15 - 10) * (10.8 - 10) = 10.32%, and the offloading ratio of the end node is 100% - 80% - 10.32% = 9.68%.
[0069] Because task characteristics and computing resources are continuously changing, the entries recorded in the mapping table cannot cover all values. This solution, when the difference does not exceed a preset threshold, obtains the key parameter with the largest difference from the mapping table and calculates the unloading ratio based on this parameter. This achieves an unloading ratio with controllable error with minimal computation, ensuring decision-making speed and allowing for quick approximate unloading strategies even for values not covered by the mapping table. It also expands the applicability of the mapping table, avoiding decision gaps caused by direct table lookups, thus maintaining the unloading effect without needing to execute intelligent optimization algorithms.
[0070] Based on the above process, this solution presents a task offloading acceleration system based on a task & computing power mapping table, such as... Figure 2 As shown.
[0071] for Figure 2 The functions of each unit are as follows: Task and computing power acquisition unit: Real-time acquisition of task volume, task complexity, etc., of user-initiated tasks. Periodically or non-periodically acquisition of available communication, computing, and storage resources of user terminals, edge nodes, and cloud nodes.
[0072] Task offloading model and algorithm unit: Build a task offloading processing optimization model based on computing power resource collaboration; based on the task requirement characteristics, computing power node resource status and task offloading processing optimization model based on computing power resource collaboration collected by the task and computing power acquisition unit, use intelligent optimization algorithms to optimize task offloading processing latency and output corresponding task segmentation and offloading strategies.
[0073] Mapping Table Construction and Decision Unit: Based on the task segmentation and unloading strategies output by the task unloading model and algorithm unit, constructs a task & computing power mapping table; determines whether the task & computing power mapping table needs updating; if it does, sends an update request to the update unit, receives the update result returned by the update unit, adds the update result to the task & computing power mapping table, and pushes the updated unloading result to the execution unit; if the task & computing power mapping table does not need updating, based on the user-initiated task characteristics and the computing power resources of cloud, edge, and terminal nodes, queries the task & computing power mapping table, calculates the unloading ratio based on the parameter with the largest difference, and pushes the unloading strategy to the execution unit.
[0074] Update Unit: Based on the update instructions output by the mapping table construction and decision-making unit, and using intelligent optimization algorithms, the task unloading target is optimized according to the task requirement characteristics and real-time computing node resources collected by the task and computing power acquisition unit, and the results are sent to the mapping table construction and decision-making unit.
[0075] Execution Unit: Executes the mapping table construction and the uninstallation strategy pushed by the decision unit to complete the task uninstallation process.
[0076] By pre-constructing a task & computing power mapping table, the optimization results of the intelligent optimization algorithm are solidified and reused in the form of a mapping table. This eliminates the need to re-execute the intelligent optimization algorithm every time a user initiates a task uninstallation request. Instead, the system queries the task & computing power mapping table and determines that the difference does not exceed a preset threshold. Based on the key parameter with the largest difference, the system uses fitting to calculate the uninstallation ratio and executes the task uninstallation. This reduces the decision latency and system implementation complexity of task uninstallation and overcomes the shortcomings of existing dynamic uninstallation schemes, which require the execution of the intelligent optimization algorithm before each uninstallation, resulting in high decision latency and high complexity.
[0077] Meanwhile, when the difference exceeds a preset threshold, the optimal unloading strategy is solved only for the current task and added as a new entry to the task & computing power mapping table. This achieves low-cost updates to the mapping table, enabling the mapping table to be dynamically refreshed as task characteristics and computing power resources change. This makes up for the shortcomings of existing static unloading schemes that cannot adapt to dynamic changes in task characteristics and node computing power resources. Thus, while reducing task unloading latency, it also takes into account unloading effect and adaptability.
[0078] Correspondingly, a computer program product includes computer-readable instructions that, when executed on an electronic device, cause the electronic device to implement the aforementioned task offloading acceleration method.
[0079] Accordingly, an electronic device includes at least one processor and a memory connected to the processor, the memory being used to store computer programs; the processor is used to execute the computer programs, enabling the electronic device to implement the aforementioned task offloading acceleration method.
[0080] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for accelerating task unloading, characterized in that, Includes the following steps: Respond to the user's task unloading request, obtain the task characteristics of the current task and the real-time computing resources of the terminal, edge, and cloud nodes; The task & computing power mapping table is queried to determine whether the differences between the current task's task characteristics and real-time computing power resources and the corresponding entries in the task & computing power mapping table exceed a preset threshold. The task & computing power mapping table is pre-constructed in the following way: For different combinations of task characteristics and computing power resources, the task unloading processing optimization model is solved to obtain the optimal task segmentation and unloading strategy. The task unloading processing optimization model is used to describe an optimization problem with task segmentation and unloading strategy as decision variables, minimizing task processing latency as the objective, and subject to latency and energy consumption constraints. If the preset threshold is not exceeded, the task segmentation and unloading ratio of the current task are calculated by fitting the task & computing power mapping table based on the key parameters with the greatest difference, and the task is unloaded according to the calculated unloading ratio. If the preset threshold is exceeded, the optimal task splitting and unloading strategy is solved in real time for the current task, the task is unloaded according to the solution results, and the obtained unloading strategy is added as a new entry to the task & computing power mapping table.
2. The task unloading acceleration method as described in claim 1, characterized in that, The construction process of the task offloading processing optimization model is as follows: Under the conditions of limited task processing latency and energy consumption, the optimization objective is to minimize task offloading processing latency, and it is constrained by the maximum tolerable task processing time, the maximum energy consumption of task processing, and the total number of tasks; as shown in the following formula: ; ; in, The maximum tolerable time for task processing. This represents the maximum energy consumption for task processing. For the total number of tasks, To account for the delay in task unloading processing, Energy consumption is consumed during task unloading. These represent the workload of unloading tasks to edge nodes and cloud nodes, respectively. The amount of tasks stored for edge nodes.
3. The task unloading acceleration method as described in claim 1, characterized in that, The task & computing power mapping table is a multi-dimensional mapping table that includes at least the task quantity, task complexity, end communication rate, end computing rate, edge communication rate, edge computing rate, cloud communication rate, and cloud computing rate. It is used to record the optimal task splitting and unloading strategies corresponding to different task characteristics and computing power resource combinations.
4. The task unloading acceleration method as described in claim 1, characterized in that, The most recent entry is the entry in the task & computing power mapping table that is closest to the task characteristics and real-time computing power resources of the current task. Determine whether the differences between the current task's characteristics and real-time computing resources and the corresponding entries in the task & computing power mapping table exceed a preset threshold, including: When the absolute value of the difference between the task volume initiated by the user and the task volume of the corresponding most recent entry is greater than the task volume difference threshold, it is determined that an update is needed; When the absolute value of the difference between the communication resources or computing resources of the endpoint, edge, or cloud node and the corresponding resource in the most recent entry is greater than the computing resource difference threshold, it is determined that an update is needed.
5. The task unloading acceleration method as described in claim 4, characterized in that, When the difference does not exceed the preset threshold, obtain the difference between each parameter and the corresponding nearest entry in the task & computing power mapping table in the task characteristics and real-time computing power resources of the current task, and take the parameter with the largest difference as the key parameter. Key parameters are selected from the following: task volume, end node communication rate, end node computing rate, edge node communication rate, edge node computing rate, cloud node communication rate, and cloud node computing rate; the differences are expressed as a percentage of the differences.
6. The task unloading acceleration method as described in claim 1, characterized in that, The fitting calculation of the task segmentation and unloading ratio for the current task includes: Using the key parameter with the greatest difference as the independent variable, and the proportion of tasks offloaded to edge nodes and the proportion of tasks offloaded to cloud nodes as the dependent variables, the functional relationship between the key parameter and the offloading ratio is obtained by fitting the data of the entries adjacent to the key parameter in the task & computing power mapping table. Substitute the value of this key parameter in the current task into the function relationship to calculate the proportion of tasks unloaded to the edge node and the proportion of tasks unloaded to the cloud node. The proportion of tasks processed by the user is determined by subtracting the proportion of tasks unloaded to edge nodes and cloud nodes from the total number of tasks.
7. The task unloading acceleration method as described in claim 1, characterized in that, The unloading strategy obtained from the solution is added as a new entry to the task & computing power mapping table. This includes: optimizing the task unloading target using an intelligent optimization algorithm based solely on the task characteristics and real-time resources of the computing power nodes of the current task, and adding the optimal task segmentation and unloading strategy corresponding to the current task as a new entry to the task & computing power mapping table, while keeping other entries in the task & computing power mapping table unchanged.
8. The task unloading acceleration method as described in claim 1, characterized in that, The task characteristics of the current task include the task volume and task complexity; the real-time computing resources of the end, edge, and cloud nodes include the communication resources, computing resources, and storage resources of the user end, edge nodes, and cloud nodes.
9. A task unloading acceleration method as described in claim 1, characterized in that, The optimal task splitting and unloading strategy includes the proportion of tasks unloaded to edge nodes, the proportion of tasks unloaded to cloud nodes, and the proportion of tasks processed by the user end; wherein, the proportion of tasks processed by the user end is the total number of tasks minus the proportion of tasks unloaded to edge nodes and the proportion of tasks unloaded to cloud nodes.
10. A task unloading acceleration system, used to implement the task unloading acceleration method as described in any one of claims 1-9, characterized in that, include: The task and computing power acquisition unit is used to collect the task characteristics of user-initiated tasks in real time, and periodically or non-periodically collect the communication, computing, and storage resources available at the user end, edge nodes, and cloud nodes. The task unloading model and algorithm unit is used to build a task unloading processing optimization model based on the collaboration of computing resources. Based on the task requirement characteristics, computing node resource status and task unloading processing optimization model collected by the task and computing power acquisition unit, it solves the optimal task unloading processing latency and outputs the corresponding task segmentation and unloading strategies. The mapping table construction and decision-making unit is used to construct a task & computing power mapping table based on the task segmentation and unloading strategy output by the task unloading model and algorithm unit; and to determine whether the task & computing power mapping table needs to be updated. The update unit is used to optimize the task unloading target based on the update instructions output by the mapping table construction and decision unit, the task requirement characteristics collected by the task and computing power acquisition unit and the real-time resources of computing power nodes, and the intelligent optimization algorithm, and then send the results to the mapping table construction and decision unit. The execution unit is used to execute the mapping table construction and the unloading strategy pushed by the decision unit to complete the task unloading process.
