Task offloading method and device based on approximate computing reuse in cloud-edge collaboration

Abstract

The present invention discloses a task offloading method and device based on approximate computing reuse in cloud-edge collaboration. Based on the cloud-edge collaboration scenario in edge computing, the method employs approximate computing reuse technology on edge nodes to reduce computational latency. The joint optimization problem of latency and energy consumption in this scenario is modeled as a MINLP (mixed integer nonlinear programming) problem. Finally, a two-stage heuristic algorithm based on the Karush-Kuhn-Tucker (KKT) condition and greedy thinking is designed. By rationally planning offloading decisions, reuse decisions, and computing resource allocation, the combined loss of latency and energy consumption is reduced.

Description

Technical Field

[0001] The present invention relates to the field of edge computing, and specifically to a task offloading method and device based on approximate computing reuse in a cloud-edge collaborative architecture. Background Art

[0002] In recent years, with the gradual development of internet technology, a wide variety of terminal products and applications have emerged, generating massive computing demands. In particular, new-generation intelligent applications, such as AR and VR, place extremely high demands on device computing power while also requiring low task processing latency to ensure a superior user experience. However, personal terminal devices are often limited by computing power, storage capacity, and battery life, making it difficult for these applications to be met solely by terminal devices. While traditional cloud computing models offer massive computing resources, the remote location of personal terminal devices leads to high latency in transmitting tasks, impacting user experience. Furthermore, the large number of task requests places significant pressure on cloud centers. To address these issues, the concept of edge computing has emerged. Its concept is to deploy computing and storage resources on devices at the edge of the network, enabling them to provide services to nearby devices and alleviate pressure on cloud centers. It should be noted that edge computing is not intended to replace cloud computing, but rather to complement it, enabling collaborative computing between the cloud and the edge.

[0003] In order to further meet the low-latency requirements of tasks, researchers proposed approximate computing reuse and applied it to edge computing. Approximate computing reuse refers to caching already executed tasks and their calculation results on edge nodes. When encountering a new task request, the cached similar task calculation results are directly returned. This technology is suitable for deployment on edge nodes and can greatly reduce task latency, but existing related research has ignored the impact of approximate computing reuse itself on offloading decisions. This is because, on the one hand, the approximate computing reuse process itself requires a certain delay, and the unlimited use of approximate computing reuse will bring additional delays; on the other hand, the controller cannot know the similarity between the tasks cached by the edge node and the tasks to be processed, and therefore cannot predict whether similar tasks can be found in the cache before making an offloading decision. Summary of the Invention

[0004] Purpose of the invention: In response to the problems and shortcomings of the existing technology, the present invention proposes a task offloading method and device based on approximate computing reuse in cloud-edge collaboration, taking into account the impact of approximate computing reuse on offloading decisions, and solving the task offloading and resource allocation problems caused by using approximate computing reuse technology in cloud-edge collaboration scenarios.

[0005] Technical solution: To achieve the above invention objectives, the present invention provides a task offloading method based on approximate computing reuse under cloud-edge collaboration, including the following steps:

[0006] (1) Obtaining the global system information of the edge computing network, including communication status information, computing task information, computing capacity information, and system accuracy requirements. The communication status information includes the wireless link status information between the terminal device and the edge node, and the core network link status information between the edge node and the cloud center. The computing task information includes the computing overhead of the task ω i , the data size of the task l i , the maximum tolerable delay of the task d i The computing capability information includes the computing capability of the terminal device. The computing power of edge nodes f E ;

[0007] (2) Based on the system global information and edge computing network architecture, a formal problem of jointly optimizing system latency and energy consumption is established: The formalized problem satisfies the constraints of total computing resources, system accuracy requirements, and maximum tolerable task delay; where α represents the balancing factor, t i represents the task execution delay, ε i represents the energy consumption of task execution, X represents the task offloading decision matrix, and offloading to the terminal device is represented as Offloading to the edge node is represented as Offloading to the cloud center is represented by in, and Z represents the task reuse decision matrix, and the decision variable is represented by z i ∈{0,1}, K represents the task computing resource allocation ratio matrix, and the matrix element is represented by k i ∈[0,1], N is the number of tasks to be processed;

[0008] (3) According to the optimization problem established in step (2) and the system global information, the initial reuse decision, the initial task offloading decision and the initial computing resource allocation strategy are determined based on the initial value selection algorithm;

[0009] (4) According to the initial reuse decision, initial task offloading decision, initial computing resource allocation strategy, optimization problem, and system global information in step (3), the optimal offloading decision is determined based on the offloading strategy optimization algorithm.

[0010] The present invention provides a task offloading device based on approximate computing reuse in cloud-edge collaboration, comprising:

[0011] The data acquisition module is used to obtain the global system information of the edge computing network, including communication status information, computing task information, computing capacity information, and system accuracy requirements. The communication status information includes the wireless link status information between the terminal device and the edge node, and the core network link status information between the edge node and the cloud center. The computing task information includes the computing cost of the task. i , the data size of the task l i , the maximum tolerable delay of the task d i The computing capability information includes the computing capability of the terminal device f i L , computing power of edge nodes f E ;

[0012] The problem building module is used to establish a formal problem for jointly optimizing system latency and energy consumption based on the system's global information and edge computing network architecture: The formalized problem satisfies the constraints of total computing resources, system accuracy requirements, and maximum tolerable task delay; where α represents the balancing factor, t i represents the task execution delay, ε i represents the energy consumption of task execution, X represents the task offloading decision matrix, and offloading to the terminal device is represented as Offloading to the edge node is represented as Offloading to the cloud center is represented by in, and Z represents the task reuse decision matrix, and the decision variable is represented by z i ∈{0,1}, K represents the task computing resource allocation ratio matrix, and the matrix element is represented by k i ∈[0,1], N is the number of tasks to be processed;

[0013] The initial value selection module is used to determine the initial reuse decision, initial task offloading decision, and initial computing resource allocation strategy based on the initial value selection algorithm according to the optimization problem established by the problem construction module and the system global information obtained by the data acquisition module;

[0014] The decision determination module is used to determine the optimal offloading decision based on the offloading strategy optimization algorithm according to the initial reuse decision of the initial value selection module, the initial task offloading decision, the initial computing resource allocation strategy, the optimization problem established by the problem construction module, and the system global information obtained by the data acquisition module.

[0015] The present invention provides a computer device, comprising:

[0016] one or more processors;

[0017] Memory; and

[0018] One or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and when the programs are executed by the processors, the steps of the task offloading method based on approximate computing reuse under cloud-edge collaboration as described above are implemented.

[0019] The present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the task offloading method based on approximate computing reuse under cloud-edge collaboration as described above are implemented.

[0020] Beneficial effects: The present invention specifically provides how to perform task offloading and resource allocation under the classic cloud-edge collaborative network architecture to optimize edge computing based on approximate computing reuse. In view of the impact of approximate computing reuse on unloading decisions, the approximate computing reuse model is characterized for the first time by the reuse success rate, approximate computing reuse delay and reuse accuracy, and the system reuse accuracy constraint in this scenario is proposed for the first time. At the same time, the problem is modeled as a mixed integer linear programming problem that jointly optimizes latency and energy consumption. By optimizing task offloading decisions, approximate computing reuse decisions, and computing resource allocation strategies, the advantages of approximate computing reuse are maximized to reduce system overhead. The method of the present invention reduces the combined loss of latency and energy consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a schematic diagram of the system architecture for cloud-edge collaboration;

[0022] Figure 2 It is a schematic diagram of approximate computing reuse applied to edge nodes;

[0023] Figure 3 It is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0024] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0025] This embodiment provides a task offloading method based on approximate computing reuse under cloud-edge collaboration, which is described in detail using a single-cell base station cloud-edge collaboration system as an example. Figure 1 As shown in the figure, there are N terminal devices connected to the edge node in the current cell, and each terminal device generates a task request. The terminal device, edge node, and cloud center can all process task requests. The edge node is deployed as follows. Figure 2 The approximate computing reuse system shown in the figure can not only directly compute tasks, but also query the approximate cache. If a similar task that meets the requirements is found, the cached task result can be directly returned without computing. If no similar task is found, the computation will be performed again. For example, Figure 2Task G in the example chooses to query the approximate cache. If a similar task G′ is found in the cache, the result of task G′ can be directly returned. Since no such similar task actually exists in the cache, task G is transferred to the CPU for calculation after the query fails. Task F, on the other hand, chooses not to query the approximate cache, but directly uses the CPU for calculation and returns the task calculation result. After the task request is processed, the result is returned to the corresponding terminal device. To find the optimal offloading decision, reuse decision, and computing resource allocation decision, the edge node collects global system information and then makes reasonable decisions based on this system information.

[0026] Reference Figure 3 The task offloading method based on approximate computing reuse under cloud-edge collaboration described in the present invention has the following specific implementation steps:

[0027] (S1) Obtain the system global information of the edge computing network of cloud-edge collaboration.

[0028] Reference Figure 1 An edge computing network architecture with cloud-edge collaboration is established, including N terminal devices, an edge node, and a remote cloud center, where each terminal device currently has only one task request to be processed; the edge node includes a base station and several edge servers, which are regarded as a whole; the terminal devices, edge nodes, and cloud center have computing capabilities and can process computing tasks; the terminal devices and edge nodes are connected through wireless communication, and some tasks can be offloaded from the terminal devices to the edge nodes; the edge nodes and cloud center are connected through the core network, and some tasks can be offloaded from the terminal devices to the cloud center through the edge nodes as intermediaries.

[0029] After receiving the task request, the edge node collects task information and system status information. Task information includes the computational cost of the task ω i , the data size of the task l i , the maximum tolerable delay of the task d i The system status information includes communication status information, including wireless link status information between terminal devices and edge nodes, core network link status information between edge nodes and cloud centers, computing power information, including computing power of terminal devices. i L , computing power of edge nodes f E .

[0030] (S2) Based on the information collected in step (S1), a system model is established and the optimization objectives are formalized as follows:

[0031]

[0032] st:

[0033]

[0034]

[0035]

[0036] (C4)t i ≤d i

[0037]

[0038] (C6)z i ∈{0,1}

[0039] (C7)k i ∈[0,1]

[0040] Where α represents the equalization factor, t i represents the task execution delay, ε i represents the energy consumption of task execution, ρ i represents the success rate of task reuse on edge nodes, μ i represents the reuse error rate of tasks on edge nodes, μ s represents the system error rate requirement, E represents the set of tasks offloaded to the edge node, X represents the task offloading decision matrix, and offloading to the terminal device is represented as Offloading to the edge node is represented as Offloading to the cloud center is represented by Z represents the task reuse decision matrix, and K represents the task computing resource allocation ratio matrix. Constraint (C1) indicates that the allocated computing resources must not exceed the total computing resources of the edge nodes. Constraint (C2) indicates that tasks can only be processed in one place. Constraint (C3) indicates that the system accuracy must meet the system accuracy requirements. Constraint (C4) indicates that the task processing time cannot exceed the maximum tolerable task delay. Constraints (C5) to (C7) represent the value ranges of the decision variables.

[0041] According to an embodiment of the present invention, the method for calculating the task execution delay is as follows:

[0042]

[0043]

[0044]

[0045]

[0046]

[0047]

[0048]

[0049] The energy consumption of task execution is calculated according to the following method:

[0050]

[0051]

[0052]

[0053] Among them, τ i represents the time required for the task to perform approximate computation reuse on the edge node, B represents the channel bandwidth, and p i Indicates the transmission power of the terminal equipment, h i represents the channel gain from the terminal device to the edge node, σ 2 represents Gaussian background noise, r TC Indicates the transmission rate of the edge node uploading tasks to the cloud center, Indicates the transmission rate of tasks uploaded from terminal devices to edge nodes, Indicates the energy efficiency of the terminal equipment.

[0054] (S3) The above problem is a MINLP (mixed integer nonlinear programming) problem, which is easily known to be NP-hard. The initial values ​​of the decision variables X and Z are first determined using the initial value selection algorithm. The specific algorithm is as follows:

[0055] (3a) Obtaining system global information;

[0056] (3b) For all mission orders

[0057] (3c)

[0058] (3d) When Seasonal i =0, when Seasonal i =1;

[0059] (3e) Output the task offloading decision X and the initial value of the approximate computing reuse decision Z.

[0060] (S4) After obtaining the initial values ​​of the decision variables in step (S3), they are substituted into the unloading optimization algorithm to obtain the final solution. The unloading optimization algorithm is as follows:

[0061] (4a) Obtaining the initial values ​​of the task offloading decision X and the approximate computing reuse decision Z output in step (S3) as input values ​​of this algorithm;

[0062] (4b) For all That is, the task of offloading to the edge node calculation is updated k i The values ​​are as follows:

[0063]

[0064] where j∈E represents mission;

[0065] (4c) Determine at this time for all Does it satisfy the following expression:

[0066]

[0067] If not, proceed to step (4d); if satisfied, proceed to step (4f);

[0068] (4d) For Make it And when season Otherwise, Then if its z i =0, go to step (4b), otherwise go to step (4e);

[0069] (4e) For If i exists and if z i =1 after satisfaction Then let z i =1, and execute this step in a loop; otherwise, exit the loop and go to step (4b);

[0070] (4f) Determine whether the current state satisfies the following expression:

[0071]

[0072] If it is satisfied, go to step (4h); if not, loop through step (4g) until the expression is satisfied, then exit the loop and go to step (4b);

[0073] (4g) For Let z i =0

[0074] (4h) For all Determine whether the following expression is satisfied:

[0075]

[0076] If all conditions are met, proceed to step (4j), otherwise proceed to step (4i);

[0077] (4i) For Make it And when season Otherwise, Then if its z i =0, go to step (4b), otherwise go to step (4e);

[0078] (4j) End the algorithm and output the task offloading decision X, computing resource allocation decision K, and approximate computing reuse decision Z.

[0079] Based on the offloading decision variables calculated above, the edge node decides whether to process the task locally or offload it to the edge node or cloud center. It also allocates computing resources proportionally to the task offloaded to the edge node and decides whether to use approximate computing reuse technology. This concludes the task offloading and resource allocation decision process.

[0080] The present invention also provides a task offloading device based on approximate computing reuse under cloud-edge collaboration, comprising:

[0081] The data acquisition module is used to obtain the global system information of the edge computing network, including communication status information, computing task information, computing capacity information, and system accuracy requirements. The communication status information includes the wireless link status information between the terminal device and the edge node, and the core network link status information between the edge node and the cloud center. The computing task information includes the computing cost of the task. i , the data size of the task l i , the maximum tolerable delay of the task d i The computing capability information includes the computing capability of the terminal device. The computing power of edge nodes f E ;

[0082] The problem building module is used to establish a formal problem for jointly optimizing system latency and energy consumption based on the system's global information and edge computing network architecture:

[0083]

[0084] The formalized problem satisfies the constraints of total computing resources, system accuracy requirements, and maximum tolerable task delay; where α represents the balancing factor, t i represents the task execution delay, ε i represents the energy consumption of task execution, X represents the task offloading decision matrix, and offloading to the terminal device is represented as Offloading to the edge node is represented as Offloading to the cloud center is represented by in, and Z represents the task reuse decision matrix, and the decision variable is represented by z i ∈{0,1}, K represents the task computing resource allocation ratio matrix, and the matrix element is represented by k i ∈[0,1], N is the number of tasks to be processed;

[0085] The initial value selection module is used to determine the initial reuse decision, initial task offloading decision, and initial computing resource allocation strategy based on the initial value selection algorithm according to the optimization problem established by the problem construction module and the system global information obtained by the data acquisition module;

[0086] The decision determination module is used to determine the optimal offloading decision based on the offloading strategy optimization algorithm according to the initial reuse decision of the initial value selection module, the initial task offloading decision, the initial computing resource allocation strategy, the optimization problem established by the problem construction module, and the system global information obtained by the data acquisition module.

[0087] The specific execution process of the initial value selection algorithm and the unloading strategy optimization algorithm can refer to the description of the above method embodiment, which will not be repeated here.

[0088] The present invention also provides a computer device comprising: one or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and when the programs are executed by the processors, the steps of the task offloading method based on approximate computing reuse under cloud-edge collaboration as described above are implemented.

[0089] The present invention also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the task offloading method based on approximate computing reuse under cloud-edge collaboration as described above are implemented.

[0090] The above is an exemplary embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A task offloading method based on approximate computing reuse in cloud-edge collaboration, characterized in that: The following steps are involved: (1) Obtaining the global system information of the edge computing network, including communication status information, computing task information, computing capacity information, and system accuracy requirements. The communication status information includes the wireless link status information between the terminal device and the edge node, and the core network link status information between the edge node and the cloud center. The computing task information includes the computing overhead of the task ω i , the data size of the task l i , the maximum tolerable delay of the task d i The computing capability information includes the computing capability of the terminal device f i L , computing power of edge nodes f E ; (2) Based on the system global information and edge computing network architecture, a formal problem of jointly optimizing system latency and energy consumption is established: The formalized problem satisfies the constraints of total computing resources, system accuracy requirements, and maximum tolerable task delay. Where α represents the equalization factor, t i represents the task execution delay, ε i represents the energy consumption of task execution, X represents the task offloading decision matrix, and offloading to the terminal device is represented as Offloading to the edge node is represented as Offloading to the cloud center is represented by in, and Z represents the task reuse decision matrix, and the decision variable is represented by z i ∈{0,1}, K represents the task computing resource allocation ratio matrix, and the matrix element is represented by k i ∈[0,1], N is the number of tasks to be processed; (3) According to the formal problem established in step (2) and the system global information, the initial reuse decision, initial task offloading decision and initial computing resource allocation strategy are determined based on the initial value selection algorithm; (4) According to the initial reuse decision, initial task offloading decision, initial computing resource allocation strategy, optimization problem, and system global information in step (3), the optimal offloading decision is determined based on the offloading strategy optimization algorithm.

2. The method according to claim 1, characterized in that In step (2), the total computing resource constraint is expressed as: The system accuracy requirement constraint is expressed as: The maximum tolerable delay constraint of the task is expressed as: t i ≤d i Among them, ρ i represents the success rate of task reuse on edge nodes, μ i represents the reuse error rate of tasks on edge nodes, μ s represents the system error rate requirement, and E represents the set of tasks offloaded to the edge nodes.

3. The method according to claim 1, characterized in that The calculation method of task execution delay is as follows: Among them, τ i represents the time required for the task to perform approximate computation reuse on the edge node, B represents the channel bandwidth, and p i Indicates the transmission power of the terminal equipment, h i represents the channel gain from the terminal device to the edge node, σ 2 represents Gaussian background noise, r TC represents the transmission rate of the edge node uploading tasks to the cloud center, ρ i Indicates the success rate of task reuse on edge nodes.

4. The method according to claim 1, wherein The calculation method of task execution energy consumption is as follows: in, represents the energy efficiency of the terminal equipment, ρ i Indicates the transmission power of the terminal equipment, Indicates the transmission rate of tasks uploaded from terminal devices to edge nodes.

5. The method according to claim 1, wherein The initial value selection algorithm in step (3) includes: For all tasks, set up when Seasonal i =0, Seasonal i =1, where τ i represents the time required for the task to perform approximate computation reuse on the edge node, ρ i represents the success rate of task reuse on edge nodes, p i Indicates the transmission power of the terminal equipment, Indicates the total latency of the task calculated on the local device. Represents the total latency of task offloading to the cloud center for processing, Indicates the total energy consumption of the task calculated on the local device, Represents the total energy consumption of tasks offloaded to the cloud center for processing; Output the task offloading decision X and the initial value of the approximate computing reuse decision Z.

6. The method according to claim 5, characterized in that The optimization algorithm in step (4) includes: (4a) Obtain the initial values ​​of the task offloading decision X and the approximate computing reuse decision Z output in step (3) as input values ​​of this algorithm; (4b) For all That is, the task of offloading to the edge node calculation is updated k i The values ​​are as follows: where j∈E represents mission; (4c) Determine at this time for all Does it satisfy the following expression: If not, proceed to step (4d); if satisfied, proceed to step (4f); (4d) For Make it And when season Otherwise, Then if its z i =0, go to step (4b), otherwise go to step (4e); (4e) For If i exists and if z i =1 after satisfaction Then let z i =1, and loop through this step, otherwise exit the loop and go to step (4b); (4f) Determine whether the current state satisfies the following expression: If it is satisfied, go to step (4h); if not, loop through step (4g) until the expression is satisfied, then exit the loop and go to step (4b); (4g) For Let z i =0; (4h) For all Determine whether the following expression is satisfied: If all conditions are met, proceed to step (4j), otherwise proceed to step (4i); (4i) For Make it And when season Otherwise, Then if its z i =0, go to step (4b), otherwise go to step (4e); (4j) End the algorithm and output the task offloading decision X, computing resource allocation decision K, and approximate computing reuse decision Z.

7. A task offloading device based on approximate computing reuse in cloud-edge collaboration, characterized in that: include: The data acquisition module is used to obtain the global system information of the edge computing network, including communication status information, computing task information, computing capacity information, and system accuracy requirements. The communication status information includes the wireless link status information between the terminal device and the edge node, and the core network link status information between the edge node and the cloud center. The computing task information includes the computing cost of the task. i , the data size of the task l i , the maximum tolerable delay of the task d i The computing capability information includes the computing capability of the terminal device f i L , computing power of edge nodes f E ; The problem building module is used to establish a formal problem for jointly optimizing system latency and energy consumption based on the system's global information and edge computing network architecture: The formalized problem satisfies the constraints of total computing resources, system accuracy requirements, and maximum tolerable task delay; where α represents the balancing factor, t i represents the task execution delay, ε i represents the energy consumption of task execution, X represents the task offloading decision matrix, and offloading to the terminal device is represented as Offloading to the edge node is represented as Offloading to the cloud center is represented by in, and Z represents the task reuse decision matrix, and the decision variable is represented by z i ∈{0,1}, K represents the task computing resource allocation ratio matrix, and the matrix element is represented by k i ∈[0,1], N is the number of tasks to be processed; The initial value selection module is used to determine the initial reuse decision, initial task offloading decision, and initial computing resource allocation strategy based on the initial value selection algorithm according to the formal problem established by the problem construction module and the system global information obtained by the data acquisition module; The decision determination module is used to determine the optimal offloading decision based on the offloading strategy optimization algorithm according to the initial reuse decision of the initial value selection module, the initial task offloading decision, the initial computing resource allocation strategy, the optimization problem established by the problem construction module, and the system global information obtained by the data acquisition module.

8. A computer device, characterized in that: include: one or more processors; Memory; as well as One or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and when the programs are executed by the processors, the steps of the task offloading method based on approximate computing reuse under cloud-edge collaboration as described in any one of claims 1-6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, the steps of the task offloading method based on approximate computing reuse under cloud-edge collaboration as described in any one of claims 1-6 are implemented.

Images