Task unloading methods, devices and storage media

By optimizing the device-to-device offloading strategy in IoT devices and utilizing resource pool resources, the problem of limited computing resources in IoT devices is solved, achieving efficient utilization of resource pools and maximizing total task utility, while reducing total task latency.

CN117176656BActive Publication Date: 2026-11-17CHINA MOBILE COMM GRP TERMINAL +1
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202210585795.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-05-26
Publication Date
2026-11-17
Estimated Expiration
2042-05-26

AI Technical Summary

Technical Problem

Existing IoT devices, due to limited computing resources, struggle to efficiently utilize scarce communication and computing resources. This leads to a strained relationship between resource-constrained applications and devices, preventing the maximization of resource pool utilization. Traditional methods have failed to effectively address the issues of total task latency and resource utilization.

Method used

By acquiring the characteristics of data stream computing tasks, establishing computational and communication constraints, and aiming to maximize task utility, we optimize the device-to-device offloading strategy, construct a mixed-integer linear programming problem, optimize the device-to-device offloading strategy, utilize resource pool resources, optimize the device-to-device offloading strategy, and maximize the total utility of successfully executed tasks.

Benefits of technology

With limited resources, the device-to-device offloading strategy was optimized to maximize the total utility of successfully executed tasks, improve resource pool utilization and system efficiency, and reduce total task latency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117176656B_ABST
    Figure CN117176656B_ABST
Patent Text Reader

Abstract

This application proposes a task offloading method, apparatus, and storage medium. The method includes: acquiring the characteristics of a data stream computing task, including the required computing resources, required processing density, ratio of output data size to input data size, and task utility; establishing computing constraints and communication constraints based on the characteristics of the data stream computing task; determining the data stream computing task to be offloaded with the goal of maximizing task utility; optimizing the device-to-device offloading strategy; and rationally utilizing the resource pool with the goal of maximizing the total utility of successfully executed tasks when the resource pool is limited.
Need to check novelty before this filing date? Find Prior Art

Description

[Technical Field]

[0001] This application relates to the field of computer network technology, and in particular to a task offloading method, apparatus and storage medium. [Background Technology]

[0002] In recent years, the explosive growth of resource-intensive and latency-sensitive Internet of Things (IoT) applications, such as wearable devices, video surveillance, and connected cars, has posed significant challenges to underlying IoT devices with limited computing resources. Due to physical size constraints, IoT devices typically possess only limited energy capacity and computing resources. Therefore, the contradiction between ever-increasing user demands and the limited capabilities of IoT devices has become a challenging problem. To alleviate the escalating tension between resource-constrained applications and resource-constrained IoT devices, task offloading has been proposed as an effective solution.

[0003] In fog / edge computing, computing and storage resources are pushed from the network core to the network edge, closer to IoT devices. If as much mobile workload as possible is aggressively offloaded to edge nodes, the cost and operating expenses of those nodes will increase significantly, as more resources need to be configured to match peak demand. In each edge device with limited computing and communication capabilities, a natural challenge is how to efficiently utilize scarce communication and computing resources to maximize overall system efficiency. Traditional methods often focus on offloading tasks used for mobile edge computing by reducing the uplink latency, computation latency, and downlink latency of user tasks to account for their impact on total task latency. However, these methods are somewhat one-sided and may not maximize the utilization of the resource pool.

[0004] Therefore, providing a task offloading method that maximizes the overall system efficiency of the computing power network resource pool is a problem that urgently needs to be solved. [Summary of the Invention]

[0005] This application provides a task offloading method, apparatus, and storage medium that can optimize device-to-device offloading strategies and rationally utilize the resource pool with the goal of maximizing the total utility of successfully executed tasks when the resource pool is limited.

[0006] In a first aspect, embodiments of this application provide a task offloading method applied to a computing power network resource pool. The method includes: acquiring the characteristics of a data stream computing task, the characteristics of which include the required amount of computing resources, the required processing density, the ratio of output data size to input data size, and task utility; establishing computing constraints and communication constraints based on the characteristics of the data stream computing task; and determining the data stream computing task to be offloaded with the goal of maximizing task utility.

[0007] The first beneficial effect is that by acquiring the characteristics of data stream computing tasks, including the amount of computing resources required, the required processing density, the ratio of output data size to input data size, and the characteristics of task utility, constructing computing constraints and communication constraints, and determining the data stream computing tasks to be unloaded with the goal of maximizing task utility, the device-to-device unloading strategy can be optimized. Under the condition of limited resource pool, the resource pool can be rationally utilized with the goal of maximizing the total utility of successfully executed tasks.

[0008] In one possible implementation, the computing network resource pool includes resource devices and task devices. The step of establishing computational and communication constraints based on the characteristics of the data stream computing tasks, and determining the data stream computing tasks to be unloaded with the goal of maximizing task utility, includes:

[0009] Establish the calculation formula for maximizing the utility of the task: Where x p It is a binary decision variable, x, when the data stream of the supporting task device p is used to compute the task. p =1, when x does not support data stream computation tasks on task device p. p =0, y dp E is a decision variable representing the amount of computing resources allocated from resource device d to task device p, where p∈P means that task device p belongs to the set of task devices P. p The data stream of the task device p represents the task utility of the calculation task;

[0010] Establish the calculation formulas for computational constraints and communication constraints: Constraint 1: This indicates that the data stream computation task of task device p is selected when x is selected. p When = 1, the aggregate computing resources should meet its resource requirement C. p Where d represents resource equipment, Γ p Represents the set of resource devices capable of establishing device-to-device connections with task device p and offloading tasks; Constraint 2: This means that the total amount of computing resources provided by resource device d should not exceed its capacity c. d , where I i Let R represent the set of task devices capable of establishing device-to-device connections with resource device d; Constraint 3: Where y dp b represents the computing resources provided by resource device d for a data stream computation task on task device p. ip This indicates the data transfer rate between resource device i and task device p, where a connection has been established. and These represent the maximum affordable input data stream rate and output data stream rate for real-time processing, respectively, where D... pK represents the processing density of a data stream computation task. p The data stream of task device p represents the ratio of the output data size to the input data size of the task; Constraint 4: Represents the decision variable y dp Feasible space; Constraint 5: Describing decision variable x p The feasible space;

[0011] The data flow computation task to be unloaded is solved based on the calculation formula for maximizing task utility and the calculation formulas for computational constraints and communication constraints.

[0012] In one possible implementation, the computing network resource pool includes resource devices and task devices. When the data stream computing task to be unloaded consists of multiple data stream computing tasks, the method further includes: obtaining the task resource occupancy coefficient of each data stream computing task; deleting the data stream computing task when the task resource occupancy coefficient of the data stream computing task is greater than 1, wherein the task resource occupancy coefficient represents the ratio between the computing task demand of the data stream computing task and the total available resources owned by the resource devices; obtaining the task utility of each data stream computing task; obtaining the urgency index of each data stream computing task, wherein the urgency index represents the product of the task resource occupancy coefficient and the task utility of the data stream computing task; and sorting the multiple data stream computing tasks from high to low task execution priority according to the task utility from large to small and the urgency index from large to small.

[0013] In one possible implementation, the computing network resource pool includes resource devices and task devices. When the data stream computing task to be unloaded consists of multiple data stream computing tasks, the method further includes: obtaining the task resource occupancy coefficient of each data stream computing task; deleting the data stream computing task when the task resource occupancy coefficient of the data stream computing task is greater than 1, wherein the task resource occupancy coefficient represents the ratio between the computing task requirement of the data stream computing task and the total available resources owned by the resource devices; obtaining the task utility of each data stream computing task; obtaining the urgency index of each data stream computing task, wherein the urgency index represents the product of the task resource occupancy coefficient and the task utility of the data stream computing task; obtaining the unit utility of each data stream computing task, wherein the unit utility represents the ratio between the task utility of the data stream computing task and the computing task requirement of the data stream computing task; and sorting the multiple data stream computing tasks from high to low priority according to the unit utility from high to low, the task utility from high to low, and the urgency index from high to low.

[0014] In one possible implementation, when multiple data stream computing tasks to be unloaded are executed simultaneously, the method further includes: allocating the selected multiple data stream computing tasks to be unloaded to resource devices; the allocation of the selected multiple data stream computing tasks to be unloaded to resource devices includes: constructing a computing power resource pool flow network, using computing resources as network flows, resource devices as flow providers, and task devices as flow demanders, calculating the maximum flow provided by the computing power resource pool flow network; checking whether the maximum flow provided by the resource is equal to the total flow required by the selected task devices, if so, executing task unloading, otherwise adjusting the task devices so that the maximum flow provided by the resource is equal to the total flow required by the selected task devices.

[0015] In one possible implementation, the computing power network resource pool includes resource devices and task devices. When the data stream computing task to be unloaded consists of multiple data stream computing tasks, the method further includes: each task device negotiating with all its adjacent resource devices to obtain the available resources owned by the resource devices; the task device sending a computing power resource request to the resource device; the resource device prioritizing the computing power resource requests; and the resource device providing computing power resources to the task device with the highest execution priority.

[0016] In one possible implementation, the task device sends a computing resource request to the resource device, and the resource device prioritizes the computing resource requests, including: the task device sending a computing resource request to the resource device comprising a computing task requirement, task utility, and urgency index, wherein the urgency index represents the product of the task resource occupancy coefficient of the task device's data stream computing task and the task utility, and the task resource occupancy coefficient represents the ratio between the computing task requirement of the task device's data stream computing task and the available resources of the resource device; and the resource device prioritizing the multiple computing resource requests from highest to lowest according to the task utility from highest to lowest and the urgency index from highest to lowest.

[0017] In one possible implementation, the resource device provides computing resources to the task device with the highest execution priority, including: for a given computing resource request, the resource device i provides the task device p with a maximum resource amount y′. it for: Where C p Indicates the computing task requirements of the task device, c i This represents the available resources owned by resource device i. This represents the amount of computing resources allocated from resource device i to task device p.

[0018] Secondly, embodiments of this application provide a task offloading device applied to a computing power network resource pool. The device includes: a task feature acquisition module, used to acquire features of a data stream computing task, the features of which include the required amount of computing resources, the required processing density, the ratio of output data size to input data size, and task utility; and a task offloading determination module, used to establish computing constraints and communication constraints based on the features of the data stream computing task, and determine the data stream computing task to be offloaded with the goal of maximizing task utility.

[0019] Thirdly, embodiments of this application provide a computer-readable storage medium storing computer instructions that cause the computer to perform the method provided in the first aspect.

[0020] It should be understood that the second and third aspects of the embodiments of this application are consistent with the technical solutions of the first aspect of the embodiments of this application, and the beneficial effects achieved by each aspect and the corresponding feasible implementation are similar, and will not be described again. [Attached Image Description]

[0021] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 A flowchart illustrating a task unloading method provided in an embodiment of this application;

[0023] Figure 2 A flowchart illustrating a task sorting method provided in an embodiment of this application;

[0024] Figure 3 A flowchart illustrating yet another task sorting method provided in an embodiment of this application;

[0025] Figure 4 A flowchart illustrating a task allocation method provided in an embodiment of this application;

[0026] Figure 5 A schematic diagram of the structure of a computing resource pool flow network provided in an embodiment of this application;

[0027] Figure 6 A flowchart illustrating a task allocation method provided in an embodiment of this application;

[0028] Figure 7 This is a schematic diagram of the structure of a task unloading device provided in an embodiment of this application.

Detailed Implementation Methods

[0029] To better understand the technical solutions in this specification, the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0030] It should be understood that the described embodiments are merely some, not all, of the embodiments in this specification. All other embodiments obtained by those skilled in the art based on the embodiments in this specification without inventive effort are within the scope of protection of this specification.

[0031] The terminology used in the embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this specification. The singular forms “a,” “the,” and “the” used in the embodiments of this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise.

[0032] To facilitate understanding of the technical solutions of this application by those skilled in the art, the technical terms involved in this application are explained below.

[0033] Computing Power Network: The computing power network represents a new architecture, protocol, and technology exploration for the convergence of computing and networking. It publishes current computing capabilities and network conditions as routing information to the network, which then routes data stream computing task packets to the appropriate computing nodes, achieving optimal user experience, optimal computing resource utilization, and optimal network efficiency. Through the computing power network's built-in dynamic routing capability for data stream computing tasks, it dynamically and flexibly schedules tasks based on real-time computing resource performance, network performance, cost, and other multi-dimensional factors according to business needs, thereby improving resource utilization, network efficiency, and the user experience. For edge computing scenarios, the computing power network can be used to create edge computing networks, enabling edge-to-edge collaboration. Leveraging the multi-instance and multi-replica characteristics of services, it achieves user proximity access and service load balancing, addressing issues such as complex deployment, low efficiency, and low resource reuse, thus facilitating large-scale edge computing deployment.

[0034] Edge computing refers to an open platform that integrates network, computing, storage, and application capabilities, located closer to the source of objects or data, to provide the nearest service. Applications initiate requests at the edge, resulting in faster network service responses and meeting the basic needs of industries in real-time business, application intelligence, security, and privacy protection. With the increasing number of IoT devices, data volume is growing at an astonishing rate, and the massive network load makes traditional cloud computing solutions unsuitable. Edge computing reduces network traffic and shifts some demands to the edge for resolution. Positioned between physical entities and industrial connections, edge computing is considered a crucial point of convergence for 5G, the Industrial Internet, and the Internet of Things, capable of driving leaps in related industries.

[0035] Fog computing refers to deploying network computing power at the network edge. It is a multi-layered computing architecture from the cloud to the device, with multiple functions such as distributed computing, communication, storage, control, and networking. Utilizing the computing power of fog nodes at the network edge to process, store, and apply data generated by network-connected terminals can significantly reduce the latency of task transmission and processing. Fog computing provides a new approach to the intelligence and informatization of the Internet of Things (IoT). Each device can be regarded as an independent fog node, forming a fog computing network together with other fog nodes such as mobile terminals, base stations, computers, and servers. This allows for the localized and distributed processing of user-requested tasks within the network, effectively reducing the latency of task transmission and processing.

[0036] Task offloading: Task offloading is a key technology in edge computing. It provides computing resources for resource-constrained devices to run compute-intensive applications, accelerating computing speed and saving energy. More specifically, task offloading in edge computing offloads the data stream computing tasks of mobile terminals to the edge cloud environment, solving the shortcomings of devices in terms of resource storage, computing performance, and energy efficiency.

[0037] Device-to-Device (D2D) communication is a novel technology that allows terminals to communicate directly by reusing cell resources under the control of the system, thus addressing the problem of spectrum scarcity in wireless communication systems to some extent. D2D communication shares resources with users under the control of the cell network, thereby improving spectrum utilization. Furthermore, it offers benefits such as reducing the burden on cellular networks, decreasing mobile terminal battery power consumption, increasing bit rates, improving the robustness of network infrastructure failures, and supporting new small-scale point-to-point data services.

[0038] In recent years, the explosive growth of resource-intensive and latency-sensitive Internet of Things (IoT) applications, such as wearable devices, video surveillance, and connected cars, has posed significant challenges to underlying IoT devices with limited computing resources. Due to physical size constraints, IoT devices typically possess only limited energy capacity and computing resources. Therefore, the contradiction between ever-increasing user demands and the limited capabilities of IoT devices has become a challenging problem. To alleviate the escalating tension between resource-constrained applications and resource-constrained IoT devices, task offloading has been proposed as an effective solution.

[0039] In fog / edge computing, computing and storage resources are pushed from the network core to the network edge, closer to IoT devices. This new paradigm addresses the limitations of traditional cloud computing architectures in supporting mobile computing, geographic distribution, location awareness, and low latency. Since IoT devices are served by nearby edge nodes, fog / edge computing can provide enhanced computing and communication services to IoT devices without compromising stringent low-latency requirements. Furthermore, fog / edge computing has the potential to extend battery life and reduce the cost of cellular networking for IoT devices. However, aggressively offloading as much mobile workload as possible to edge nodes would significantly increase the cost and operating expenses of those edge nodes, as more resources would need to be configured to match peak demand.

[0040] With limited computing and communication capabilities in each edge device, a natural question is how to efficiently utilize scarce communication and computing resources to maximize the efficiency of the entire system. Traditional methods often involve offloading tasks used for mobile edge computing by considering the impact of uplink transmission latency, computing latency, and downlink transmission latency on the total latency of user tasks. However, these methods are somewhat one-sided and may not maximize the utilization of the resource pool.

[0041] To facilitate understanding of the embodiments of this application by those skilled in the art, the concepts and symbols involved in this application are described below.

[0042] The edge devices in the computing power network resource pool are defined as a set D = {1, 2, 3, ..., N}. Edge devices are categorized into two types: task devices and resource devices to describe the device model. For each device d ∈ D, its computing power is defined as c. n Δ is represented by the number of CPU cycles per second. The set of devices that can successfully establish a device-to-device connection with device n is defined as Δ. n The data transfer rate between a pair of devices that have established a connection is defined as b. nm , denoted as the number of bits per unit time, where m∈Δ n .

[0043] Task devices: In a subset P ∈ D of all edge devices D, each device has its own data stream computation task, denoted as p ∈ P. Define Γ as the set of potential resource devices for task devices p ∈ P, i.e., the set of resource devices that can establish device-to-device connections with task device p and offload tasks. p .

[0044] Resource devices: Devices that have no tasks at the current moment are called resource devices, denoted by R = D\P. For a resource device d ∈ R, the set of task devices near it that can establish device-to-device connections with resource device d is defined as I. i .

[0045] Any device that has no task at the current moment can be called a resource device, while one that has its own data stream computing task can be called a task device.

[0046] For a dataflow computation task with task device p∈P, it can be defined by a set of parameters. <C p D p ,K p E p >Describe its characteristics, here C p D represents the amount of computing resources required for the task. p The parameter K represents the processing density of the task (in bits of CPU cycles), i.e., the number of CPU cycles required to process a unit of input data. p E represents the ratio of the output data size to the input data size of the data stream computation task for device p∈P. p The utility of a data flow computation task represents the value of a task device p∈P. That is, from the perspective of the system, different tasks have different importance, priority, or value.

[0047] To ensure real-time on-site data stream processing, the task device p must achieve an aggregated data processing rate ε. p (in bits per second), therefore the computing power required to process the data stream is ε. p D p In practice, some IoT devices typically have very limited computing resources, such as ε. p D p >c p c p This represents the computing resources that the device can provide. Therefore, device d must offload a portion of its data stream computing tasks to nearby devices, and the aggregate computing capacity provided by the resource devices should reach C. p =ε p D p -c p If ε p D p <c p If so, the device has sufficient resources and does not require a resource pool.

[0048] Based on the above problems, this application provides a task unloading method that optimizes the device-to-device unloading strategy and maximizes the utilization of resource pool resources.

[0049] Figure 1 This is a flowchart illustrating a task unloading method provided in an embodiment of this application, as shown below. Figure 1 As shown, the above task offloading method is applied to a computing power network resource pool, and the method may include:

[0050] Step 101: Obtain the characteristics of the data stream computing task, which include the required computing resources, the required processing density, the ratio of output data size to input data size, and the task utility;

[0051] Step 102: Establish computational constraints and communication constraints based on the characteristics of the data stream computation task, and determine the data stream computation task to be unloaded with the goal of maximizing task utility.

[0052] It should be noted that a large number of devices at the network edge in the computing power network resource pool share computing resources through device-to-device collaboration. For edge facilities and IoT devices, the computing power network resource pool can reduce resource allocation and improve network capacity and energy efficiency. The task utility mentioned here refers to the importance, priority, or value of a task from a system perspective.

[0053] This application embodiment obtains the characteristics of a data stream computing task, including the required amount of computing resources, the required processing density, the ratio of output data size to input data size, and the task utility characteristics, constructs computing constraints and communication constraints, and determines the data stream computing tasks to be unloaded with the goal of maximizing task utility. This method can optimize the device-to-device unloading strategy and, in the case of a limited resource pool, rationally utilize the resource pool with the goal of maximizing the total utility of successfully executed tasks.

[0054] In some embodiments, due to the limited computing and communication capabilities in each edge device, in order to efficiently utilize scarce communication and computing resources and maximize the efficiency of the entire system, the D2D offloading strategy is optimized with the goal of maximizing the total utility of successfully executed tasks. Under the constraints of computing and communication capacity limitations, the problem of maximizing total utility is expressed as the following mixed integer linear programming problem.

[0055] Here, computational and communication constraints are established based on the characteristics of the data stream computation task. The data stream computation task to be unloaded is determined with the goal of maximizing task utility. This may include:

[0056] Establish the calculation formula for maximizing the utility of the task: Where x p It is a binary decision variable, x, when the data stream of the supporting task device p is used to compute the task. p =1, when x does not support data stream computation tasks on task device p. p =0, y dp E is a decision variable representing the amount of computing resources allocated from resource device d to task device p, where p∈P means that task device p belongs to the set of task devices P. p The data stream of the task device p represents the task utility of the calculation task.

[0057] Establish the calculation formulas for computational constraints and communication constraints:

[0058] Constraint 1: This indicates that the data stream computation task of task device p is selected when x is selected. p When = 1, the aggregate computing resources should meet its resource requirement C. p Where d represents resource equipment, Γ p This represents the set of resource devices that can establish device-to-device connections with task device p and perform task unloading.

[0059] Constraint 2: This means that the total amount of computing resources provided by resource device d should not exceed its capacity c. d , among which, I i R represents the set of task devices that can establish device-to-device connections with resource device d;

[0060] Constraint 3: Where y dp b represents the computing resources provided by resource device d for a data stream computation task on task device p. ip This indicates the data transfer rate between resource device i and task device p, where a connection has been established. and These represent the maximum affordable input data stream rate and output data stream rate for real-time processing, respectively. Therefore, the aggregated input and output data rates should not exceed the D2D link capacity between devices d and p, where D... p K represents the processing density of a data stream computation task. p This represents the ratio of the output data size to the input data size of the data stream calculation task for the task device p;

[0061] Constraint 4: Represents the decision variable y dp The feasible space;

[0062] Constraint 5: Describing decision variable x p The feasible space.

[0063] The data stream computation task to be unloaded can be solved based on the calculation formula for maximizing task utility and the calculation formulas for computational and communication constraints.

[0064] Here, the key idea in the problem statement of the formula for maximizing task utility is to select a suitable set of data stream computation tasks to optimize the total utility of the system, subject to constraints 1-5 above.

[0065] For example, the formula for maximizing task utility represents the maximum total task utility of the selected dataflow computing task among the supported dataflow computing tasks. If tasks on task devices p1, p2, p3... are supported, the utility of the tasks running on them is E... p1 E p2 E p3 The purpose of the formula for maximizing task utility is to maximize the utility of the selected task.

[0066] For example, there are five task devices running different computing tasks. However, the devices running the tasks cannot meet the resource requirements of all tasks, so task offloading is necessary. The tasks running on these five task devices form a data stream task set, but the resource pool can only support a maximum of three tasks. Therefore, two tasks need to wait. How to select which devices to use for computing and which to leave idle requires the optimal combination of maximum utility based on the formula for maximizing task utility.

[0067] If the task utility of these five tasks is E p The vector x of the supported device is [5,4,8,9,1]. p If the value is [1,0,1,1,0], then the formula for maximizing task utility in this case means selecting three tasks with utilities of 8, 9, and 5 for initial support. The sum of these utilities is 8*1 + 9*1 + 5*1 + 1*0 + 4*0 = 22. If the vector x of the supported device... p If the value is [1,1,1,0,0], then the utility sum is 5*1 + 8*1 + 4*1 + 9*0 + 1*0 = 17. In this case, the total utility sum for the supported tasks is not the maximum. In practical applications, the combination with the maximum utility sum is found among multiple possible combinations of supported devices.

[0068] Understandably, in order to maximize the overall system efficiency of the computing power network resource pool, the above method expresses the utility maximization problem (UMP) under the constraints of device computing and communication capacity as a mixed integer linear programming (MILP) problem, thereby optimizing the device-to-device offloading strategy.

[0069] In some embodiments, due to limited computing and communication resources, it may be insufficient to support the real-time processing of all data stream computing tasks. In order to maximize the overall utility of offloading tasks, it is necessary to consider which offloading tasks to execute. Therefore, a ranking metric is needed to prioritize data stream computing tasks. Some ranking metrics can be defined to determine the priority of tasks. The following method is used to sort the set of data stream computing tasks and then offload the tasks in sequence.

[0070] Figure 2 This is a flowchart illustrating a task sorting method provided in an embodiment of this application, as shown below. Figure 2 As shown, when the data stream computing task to be unloaded consists of multiple data stream computing tasks, the method includes:

[0071] Step 201: Obtain the task resource occupancy coefficient of each data stream computing task. When the task resource occupancy coefficient of the data stream computing task is greater than 1, delete the data stream computing task. The task resource occupancy coefficient represents the ratio between the computing task requirement of the data stream computing task and the total available resources owned by the resource device.

[0072] Step 202: Obtain the task utility of each data stream computation task;

[0073] Step 203: Obtain the urgency index of each data stream computing task, wherein the urgency index represents the product of the task resource occupancy coefficient and the task utility of the data stream computing task;

[0074] Step 204: Sort the multiple data stream computing tasks from high to low priority according to the task utility from high to low and the urgency index from high to low.

[0075] It should be noted that some ranking metrics were defined first: Task resource utilization coefficient: Among them, the resource occupancy coefficient O p The meaning is the ratio between the computational resource requirements of task p and the total available resources of the resource equipment. A smaller O p This means that task p requires fewer resources from nearby resource devices, while a larger O p It requires a higher proportion of resources, and therefore may be more difficult to sustain. When O p When the value is greater than 1, we should remove the corresponding task p from the selection set.

[0076] Utility: Tasks that bring greater benefits to the entire system rank higher. A simple but typical ranking metric is the utility E of task p itself. p For this metric, tasks that are more efficient are more likely to be selected.

[0077] Urgency Indicators: In addition to the task's utility and computational resource requirements, the available resources of nearby resource devices must also be considered. The resource occupancy of the task's devices should also be taken into account. Devices with higher efficiency but difficult to support are given higher priority for selection. Tasks with high utility but difficult to support are urgent and should be offloaded as soon as possible. Urgency Indicators

[0078] Based on the ranking criteria, devices with higher efficiency and greater difficulty in being supported should be given higher priority for selection. High-efficiency, "difficult-to-support" tasks are urgent and should be offloaded as soon as possible; otherwise, they cannot be executed. Therefore, multiple data stream computation tasks can be ranked from highest to lowest execution priority according to their task efficiency and urgency criteria.

[0079] The data stream computing task set is sorted using the above-mentioned computing resource measurement method. After sorting, as many task devices as possible are selected so that the needs of the selected task devices can be fully supported. Then, the tasks are unloaded in sequence.

[0080] The goal of sorting tasks and selecting as many task devices as possible allows for simultaneous computation of more tasks, saving time and minimizing the waiting time for later-ordered tasks. Since sorting by criteria, maximizing the number of supported devices with limited resources is a permutation and combination problem with multiple solutions, a problem can be solved. For example, with 15 total resources and six task devices, the resource requirements are 7, 8, 1, 6, 6, 3. The possible combinations are 7+8, 1+6+6+3, and 1+8+6, each corresponding to a different number of supported devices. "Selecting as many task devices as possible" is a permutation and combination problem with multiple solutions. Given a fixed total number of selected elements, the different values ​​of these elements indicate different selectable elements. A typical solution involves constructing a graph model and using methods for solving the maximum flow problem.

[0081] Figure 3 A flowchart illustrating another task sorting method provided in this application embodiment is shown below. Figure 3 As shown, when the data stream computing task to be unloaded consists of multiple data stream computing tasks, the method includes:

[0082] Step 301: Obtain the task resource occupancy coefficient of each data stream computing task. When the task resource occupancy coefficient of the data stream computing task is greater than 1, delete the data stream computing task. The task resource occupancy coefficient represents the ratio between the computing task requirements of the data stream computing task and the total available resources of the resource device.

[0083] Step 302: Obtain the task utility of each data stream computation task;

[0084] Step 303: Obtain the urgency index of each data stream computing task, wherein the urgency index represents the product of the task resource occupancy coefficient and the task utility of the data stream computing task;

[0085] Step 304: Obtain the unit utility of each data stream computing task, wherein the unit utility represents the ratio of the task utility of the data stream computing task to the computing task requirement of the data stream computing task;

[0086] Step 305: Sort the multiple data stream computing tasks from high to low priority according to the unit utility from high to low, the task utility from high to low, and the urgency index from high to low.

[0087] It should be noted that this embodiment is in Figure 2 Based on the aforementioned embodiment, another ranking metric—unit utility—is defined. Considering the consumption of task resources, this ranking metric, unit utility, is defined as the utility of a task divided by the amount of resources it requires. Compared to other tasks, devices that generate more utility while consuming the same resources are more likely to be selected; such tasks can be called cost-effective tasks. Therefore, multiple data stream computing tasks can be prioritized from high to low according to the following criteria: unit utility from high to low, task utility from high to low, and urgency index from high to low.

[0088] The above embodiments utilize a greedy heuristic to find near-optimal solutions. To ensure near-optimal global efficiency, the greedy heuristic prioritizes unloading tasks with higher cost-efficiency, i.e., tasks with a higher ratio of task utility to resource requirement. By designing evaluation indicators such as resource occupancy coefficient, utility, unit utility, and urgency to rank task devices, the device-to-device unloading strategy is optimized. With a limited resource pool, the goal is to maximize the utility of successfully executed tasks. By using the designed indicators to rank task devices, the rational utilization of the resource pool is achieved.

[0089] In some embodiments, for a centralized computing network resource pool, a network operator has multiple base stations at the network edge, and devices can establish connections with nearby base stations. The network operator can collect information from the edge devices (the requirements of the computing tasks and the computing capabilities of the devices), so it can instruct the task devices to offload tasks to maximize global utility.

[0090] In centralized computing network resources, the centralized scheduler can sort the set of task devices T according to designed metrics. It then selects as many task devices as possible to fully support their needs. Finally, it assigns the data stream computing tasks to the resource devices under the scheduler's control.

[0091] In some embodiments, since it is unknown how to allocate limited computing resources to selected task devices, it is not feasible to simply subtract the computing power required by the set of task devices from the remaining computing power of the resource devices to check the feasibility of the set. Due to limited computing resources, this application proposes a network flow-based method. This method treats computing resources as network flows, resource devices as flow providers, and task devices as flow consumers, constructing a computing power network resource pool capacity network. The maximum flow of the constructed network is calculated, and it is checked whether the maximum flow equals the total flow required by the selected task devices. By using a network flow-based method to determine whether the maximum capacity of the resource devices meets the total capacity of the task devices, the purpose of reasonably offloading tasks can be achieved.

[0092] When multiple data stream computing tasks to be unloaded are selected and executed simultaneously, the method further includes: allocating the selected multiple data stream computing tasks to be unloaded to resource devices.

[0093] Figure 4 This is a flowchart illustrating a task allocation method provided in an embodiment of this application, as shown below. Figure 4 As shown, allocating the selected plurality of data stream computation tasks to resource devices may include:

[0094] Step 401: Construct a computing resource pool flow network, with computing resources as network flows, resource devices as flow providers, and task devices as flow demanders, and calculate the maximum flow provided by the computing resource pool flow network;

[0095] Step 402: Check whether the maximum flow provided by the resource is equal to the total flow required by the selected task device. If yes, perform task unloading; otherwise, adjust the task device so that the maximum flow provided by the resource is equal to the total flow required by the selected task device.

[0096] It's important to note that the approach to constructing a computing resource pool network is as follows: Define a manually designed total resource device node `s` as the overall resource provider. The capacity of each resource device can be considered as allocated by `s`. The maximum capacity that `s` can allocate (i.e., the maximum resource each resource device can provide) is the maximum capacity of the constructed network. This is calculated by summing the capacities of all resource devices. However, in practical use, the maximum resource each resource device can provide must be considered. Define another manually designed total resource device node `r`. Each task node is connected to this manually defined total task node `r`. The resource requirements of all task devices converge at node `r`. Therefore, the total resource requirement of the task devices at any given moment can be considered the requirement of the manually defined node `r`. Resource devices output resources, while task nodes require resources. During the verification process, for all selected task devices, determine whether the output of each node equals the input of the resource node. If all conditions are met, the maximum capacity equals the requirement of the selected task devices.

[0097] Figure 5 This application provides a schematic diagram of the structure of a computing resource pool flow network, as shown in the embodiment of the present application. Figure 5 As shown, the task aggregation node r is connected to all task nodes, and each resource node is connected to the demand source node s. If the task node and resource node can establish a device-to-device link, an edge from the task node to the resource node will be created. The goal of the computing power resource pool flow network is to maximize the total flow from the sending node s to the receiving node r while satisfying the edge constraints. The steps for constructing the computing power resource pool flow network may include: (1) initializing a set of task devices, defined as A, which is currently an empty set; (2) creating two artificial nodes as resource device nodes (defined as s) and task device nodes (defined as r); (3) for each task device (if the task device is not in the set at this time, add it to the set first) p∈A, create a task node p and an edge from p to r with a capacity of m. p (4) For resource device i∈R, create a resource node i and an edge from s to i with capacity c. i (5) For resource devices that can establish device-to-device connections with task device p Where p∈A, add an edge from i to p with capacity O(i). The capacity formula here comes from constraint 5 in the previous embodiment, which ensures that the input and output data stream rates for real-time processing do not exceed the bandwidth, where b ip K represents the data transfer rate between resource device i and task device p, where a connection has been established. p D represents the ratio of the output data size to the input data size of the data stream computation task of the task device p. pThis indicates the processing density of the data stream computation task. (6) Calculate the maximum capacity of the network and check whether the maximum capacity is equal to the total capacity required by the selected task device. If the condition is met, unload the task and return to step (3) to continue adding device tasks. Otherwise, remove the task device from the node set. Here, edge resource pooling refers to pooling and sharing computing resources among edge devices to maximize benefits from a global perspective.

[0098] For example, if a sending node allocates resources of 5, 1, 7, 8, and 3 to five resource devices, this represents the maximum available resources for these five devices. Three task devices require resources of 7, 11, and 6 respectively. If we directly subtract resources, then 5 + 1 + 7 + 8 + 3 = 7 + 11 + 6, which satisfies the condition. However, if we allocate 5 + 3 to 7 and 7 + 8 to 11, then the remaining available resources are only 1 < 6, which is insufficient to support the task devices. There are surplus resources, but their utilization is not maximized. However, if the allocation method is to allocate 7 to 7, 3 + 8 to 11, and 5 + 1 to 6, then the resource devices can be utilized to the maximum extent, maximizing the flow provided by the resource devices to the task devices.

[0099] Traditional edge computing methods prioritize resource allocation to selected task devices, subtract the required computing resources from the computing power of the resource devices, and check the feasibility based on the remaining computing power. Although this method can offload a certain amount of tasks, it cannot offload as many tasks as possible, and may fail to offload highly urgent tasks in a timely manner.

[0100] This application treats resource devices as a capability-optimal matching problem. First, a computing power network resource pool is constructed, enabling pooling and sharing of computing resources among edge devices. Then, the maximum capacity of the constructed network is calculated to match the capacity required by the selected task devices for task offloading. Compared to simply subtracting the required computing power of the task device set from the remaining computing power of the resource devices, this approach achieves more informed and efficient decision-making regarding task offloading, allowing even a single IoT device to act as a scheduler. In centralized scenarios (where the network edge has multiple base stations), a novel computing power network resource pool is constructed, and the maximum flow of the constructed network is calculated, maximizing task offloading for task devices and improving global utility.

[0101] In some embodiments, when the computing power network resource pool is distributed, there is no centralized scheduler; that is, there are no base stations at the network edge, only local devices. All devices make their own computation offloading decisions, fully utilizing the distributed task offloading scheme. In this distributed task offloading scheme, IoT devices communicate and determine their offloading strategies locally.

[0102] Figure 6 This is a flowchart illustrating a task allocation method provided in an embodiment of this application, as shown below. Figure 6 As shown, the computing power network resource pool includes resource devices and task devices. When the data stream computing task to be unloaded consists of multiple data stream computing tasks, the method may include:

[0103] Step 501: Each task device negotiates with all its adjacent resource devices to obtain the available resources owned by the resource devices;

[0104] Step 502: The task device sends a computing power resource request to the resource device, and the resource device prioritizes the computing power resource requests.

[0105] Step 503: The resource device provides computing resources to the task device with the highest execution priority.

[0106] It should be noted that in scenarios without a centralized architecture, by leveraging decentralization, all devices make their own computational offloading decisions. This decision-making characteristic can increase autonomy in the computing power network resource pool system, helping to alleviate the heavy control and communication overhead resources brought about by complex centralized management (such as collecting data and offloading computations from a large number of IoT devices).

[0107] In some embodiments, the task device sends a computing resource request to the resource device, and the resource device sorts the computing resource requests by execution priority, including: (1) the task device sends a computing resource request to the resource device including computing task requirements, task utility and urgency index, wherein the urgency index represents the product of the task resource occupancy coefficient of the data flow computing task of the task device and the task utility, and the task resource occupancy coefficient represents the ratio between the computing task requirements of the data flow computing task of the task device and the available resources owned by the resource device; (2) the resource device sorts the multiple computing resource requests by task execution priority from high to low according to the task utility from large to small and the urgency index from large to small.

[0108] It should be noted that the task device sends a request to the resource device regarding its computing resource requirements and utility, where utility includes parameter E. p C p and urgency index η t When a task device p requests resources from a resource device, the task device p should first gather information about its environment. Each task device p must negotiate with all its neighboring resource devices d, where d∈Γ. p Resource devices prioritize tasks based on negotiation information sent by task devices, using the prioritization metric of task utility E. p and urgency η tThe system can dynamically update the ranking metric for tasks, allowing high-utility and urgent tasks to be unloaded as quickly as possible. When a task device sends a request to its neighboring resource devices, the resource devices, upon receiving these requests, rank them according to the ranking metric and select the top-ranked task device to provide assistance, determining how much resource to allocate to it. The resource devices will send positive feedback to the highest-priority task device and negative feedback to other task devices. Upon receiving feedback, task devices that meet the computational resource requirements send an acknowledgment message to the resource device. If a task device receives sufficient assistance and successfully unloads its task, or if its resource requirements are still not met after receiving feedback, the task device is removed from the set. In other words, tasks that cannot be unloaded will not send further requests if their resource requirements are not met.

[0109] In some embodiments, the resource device d provides computing resources to the task device with the highest execution priority. For a given request, the resource device d can provide the task device p with a maximum amount of resources y′. it Determined by the following formula: Where C p Indicates the computing task requirements of the task device, c i This represents the available resources owned by resource device i. This represents the amount of computing resources allocated from resource device i to task device p.

[0110] Traditional edge computing methods actively offload as much mobile workload as possible to edge nodes. While this can alleviate the computing resource demands of resource-constrained applications and IoT devices, it also leads to a significant increase in the capital and operating expenses of edge nodes.

[0111] This application embodiment solves the above problems by leveraging edge resource pools to enhance the computing power of IoT devices in edge computing. It utilizes a large number of nearby devices and facilities to perform collaborative tasks based on device-to-device (D2D) and performs task offloading locally. The task device links with the resource device through device-to-device and sends a request to the resource device. After processing the request, the resource device sends positive feedback to devices that can be offloaded to provide assistance and sends negative feedback to other devices, thereby performing task offloading.

[0112] By enabling all devices to make their own computation offloading decisions in the absence of a centralized scheduler, and utilizing local device-to-device links, the network costs and latency associated with edge computing are reduced. Furthermore, this self-decision-making capability increases autonomy within the computing power network resource pool system, helping to alleviate the heavy control and communication overhead associated with complex centralized management (such as collecting data and offloading computations from a large number of IoT devices).

[0113] Figure 7 This is a schematic diagram of a task offloading device provided in an embodiment of this application. The task offloading device is installed in a computing power network resource pool, such as... Figure 7 As shown, the above-mentioned task unloading device may include: a task feature acquisition module 61 and a task unloading determination module 62.

[0114] The task feature acquisition module 61 is used to acquire the features of the data stream computing task, including the required amount of computing resources, the required processing density, the ratio of output data size to input data size, and task utility. The task unloading determination module 62 is used to establish computing constraints and communication constraints based on the features of the data stream computing task, and determine the data stream computing task to be unloaded with the goal of maximizing task utility.

[0115] Figure 7 The task unloading device provided in the illustrated embodiment can be used to execute this specification. Figure 1 The implementation principle and technical effects of the method embodiment shown can be further referred to the relevant description in the method embodiment.

[0116] This application provides a computer-readable storage medium that stores computer instructions that cause the computer to execute this specification. Figures 1-6 The illustrated embodiment provides a task unloading method.

[0117] The aforementioned computer-readable storage medium may be any combination of one or more computer-readable media. A computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium may be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM) or flash memory, optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this document, a computer-readable storage medium may be any tangible medium that contains or stores a program that may be used by or in connection with an instruction execution system, apparatus, or device.

[0118] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including—but not limited to—electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of transmitting, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0119] Program code contained on a computer-readable medium may be transmitted using any suitable medium, including—but not limited to—wireless, wire, optical fiber, radio frequency (RF), etc., or any suitable combination thereof.

[0120] Computer program code for performing the operations described herein can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as "C" or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0121] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0122] In the description of the embodiments of the present invention, the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this specification. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0123] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this specification, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0124] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this specification includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as will be understood by those skilled in the art to which the embodiments of this specification pertain.

[0125] Depending on the context, the word "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0126] In the several embodiments provided in this specification, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0127] Furthermore, the functional units in the various embodiments of this specification can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in a combination of hardware and software functional units.

[0128] The integrated units implemented as software functional units described above can be stored in a computer-readable storage medium. These software functional units, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods described in the various embodiments of this specification. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0129] The above description is merely a preferred embodiment of this specification and is not intended to limit this specification. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this specification should be included within the scope of protection of this specification.

Claims

1. A task offloading method, characterized by, Applied to a computing power network resource pool, the method includes: The characteristics of a data stream computing task are obtained, including the amount of computing resources required, the required processing density, the ratio of output data size to input data size, and the task utility. Based on the characteristics of the data stream computing task, computational constraints and communication constraints are established, and the data stream computing task to be unloaded is determined with the goal of maximizing task utility. The computing power network resource pool includes resource devices and task devices. When the data stream computing task to be unloaded consists of multiple data stream computing tasks, the method further includes: Obtain the task resource utilization coefficient for each data stream computing task. When the task resource utilization coefficient of the data stream computing task is greater than 1, delete the data stream computing task. The task resource utilization coefficient represents the ratio between the computing task requirements of the data stream computing task and the total available resources owned by the resource device. Obtain the task utility of each data stream computation task; Obtain the urgency index for each data stream computing task, wherein the urgency index represents the product of the task resource occupancy coefficient and the task utility of the data stream computing task; Obtain the unit utility for each data stream computing task, wherein the unit utility represents the ratio of the task utility of the data stream computing task to the computing task requirement of the data stream computing task; The data stream computing tasks are sorted from highest to lowest priority based on the unit utility, task utility, and urgency index.

2. The method of claim 1, wherein, The computing power network resource pool includes resource devices and task devices. The step of establishing computational and communication constraints based on the characteristics of the data stream computing tasks, and determining the data stream computing tasks to be unloaded with the goal of maximizing task utility, includes: A calculation formula of the task utility maximization is established: wherein is a binary decision variable, when the data stream computing task of the support task device is supported , is 1; when the data stream computing task of the support task device is not supported , is 0, is a decision variable, representing the amount of computing resources of the resource device allocated to the task device , represents that the task device belongs to the task device set , represents the task utility of the data stream computing task of the task device ; Establish the calculation formulas for computational constraints and communication constraints: Constraint 1: , indicating the mission equipment When a data stream computation task is selected, that is... At that time, aggregated computing resources should meet its resource requirements. ,in Indicates resource equipment, Indicates the ability to communicate with mission equipment A collection of resource devices that perform device-to-device connections and offload tasks; Constraint 2: , denotes a resource device The total amount of computing resources provided by the resource device should not exceed its capacity , wherein denotes a set of task devices capable of establishing a device-to-device connection with the resource device , denotes a set of task devices capable of establishing a device-to-device connection with the resource device , denotes a set of resource devices; Constraint 3: ,in Indicates a given resource device For mission equipment The computing resources provided by the dataflow computing task Indicates a resource device with an established connection. With mission equipment Data transfer rate between and These represent the maximum affordable input data stream rate and output data stream rate for real-time processing, respectively. This indicates the processing density of the data stream computation task. Indicates the mission equipment The ratio of the output data size to the input data size of a data stream computation task; Constraint 4: , represents the feasible space for the decision variable ; Constraint 5: , representing the feasible space of the decision variable ; The data flow computation task to be unloaded is solved based on the calculation formula for maximizing task utility and the calculation formulas for computational constraints and communication constraints.

3. The method of claim 1, wherein, When multiple data stream computing tasks to be unloaded are selected and executed simultaneously, the method further includes: allocating the selected multiple data stream computing tasks to be unloaded to resource devices; Assigning the selected plurality of data stream computation tasks to be unloaded to resource devices includes: Construct a computing resource pool flow network, where computing resources are treated as network flows, resource devices are treated as flow providers, and task devices are treated as flow demanders, and calculate the maximum flow provided by the computing resource pool flow network. Check whether the maximum flow provided by the resource is equal to the total flow required by the selected task device. If so, perform task unloading; otherwise, adjust the task device so that the maximum flow provided by the resource is equal to the total flow required by the selected task device.

4. The task offloading method of claim 1, wherein, The computing power network resource pool includes resource devices and task devices. When the data stream computing task to be unloaded consists of multiple data stream computing tasks, the method further includes: Each task device negotiates with all its adjacent resource devices to obtain the available resources owned by those resource devices; The task device sends a computing power resource request to the resource device, and the resource device performs priority sorting on the computing power resource request; The resource device provides computing resources to the device executing the highest priority task.

5. The method of claim 4, wherein, The resource device provides computing resources to the task device with the highest execution priority, including: For a given computing resource request, resource devices Provide mission equipment Maximum resource quantity for: ,in This indicates the computing task requirements of the task device. Indicates resource equipment The available resources available, Indicates resource equipment Assigned to task device The amount of computing resources.

6. A task offloading apparatus characterized by comprising: The device, applied to a computing power network resource pool, includes: The task feature acquisition module is used to acquire the features of the data stream computing task. The features of the data stream computing task include the amount of computing resources required, the required processing density, the ratio of the output data size to the input data size, and the task utility. The task unloading determination module is used to establish computational constraints and communication constraints based on the characteristics of the data stream computing task, and to determine the data stream computing task to be unloaded with the goal of maximizing task utility. The computing power network resource pool includes resource devices and task devices. When the data stream computing task to be unloaded consists of multiple data stream computing tasks, the resource devices are further used for: Obtain the task resource utilization coefficient for each data stream computing task. When the task resource utilization coefficient of the data stream computing task is greater than 1, delete the data stream computing task. The task resource utilization coefficient represents the ratio between the computing task requirements of the data stream computing task and the total available resources owned by the resource device. Obtain the task utility of each data stream computation task; Obtain the urgency index for each data stream computing task, wherein the urgency index represents the product of the task resource occupancy coefficient and the task utility of the data stream computing task; Obtain the unit utility for each data stream computing task, wherein the unit utility represents the ratio of the task utility of the data stream computing task to the computing task requirement of the data stream computing task; The data stream computing tasks are sorted from highest to lowest priority based on the unit utility, task utility, and urgency index.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause the computer to perform the task offloading method as described in any one of claims 1 to 5.

Citation Information

Patent Citations

  • Joint task scheduling and resource allocation method supporting uninstallation of D2D-edge server

    CN107995660A

  • Task unloading method and device for non-uniform mobile edge computing network

    CN112104693A