Task unloading method and device, equipment and storage medium

Abstract

The embodiment of the invention provides a task unloading method and device, equipment and a storage medium, a resource optimization mode can be determined through the number of high-performance tasks and the number of low-energy-consumption tasks, and more accurate resource configuration and task scheduling can be achieved; according to the method, the training device can keep good performance under different task scenes, the overall task processing efficiency and the reasonability of resource utilization are improved, the task processing speed and quality are guaranteed, complex tasks can be completed smoothly, the resource burden of the training device is relieved by unloading the tasks, and the training efficiency is improved. According to the invention, the system can concentrate on the execution of a local core function and the processing of other low-energy-consumption tasks, optimizes the resource cooperation between the mobile device and the cloud device in the whole system, improves the task processing capability and operation efficiency of the whole system, finally provides better and more efficient mobile application service experience for users, and improves the user experience. And reasonable allocation of computing resources is realized.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a task offloading method, device, equipment and storage medium. Background Art

[0002] In the context of mobile edge computing, there are many computing-intensive and delay-sensitive applications. Such applications usually need to process large amounts of data and complex calculations, and have high requirements for response time and computing performance. In the existing technology, the computing power and network bandwidth of mobile devices are limited, and server computing resources cannot be reasonably allocated. Summary of the invention

[0003] In view of the above problems, embodiments of the present invention are proposed to provide a task offloading method, apparatus, device and storage medium that overcome the above problems or at least partially solve the above problems.

[0004] In order to solve the above problem, an embodiment of the present invention discloses a task offloading method, which includes:

[0005] selecting at least one mobile device from a plurality of mobile devices as a training device;

[0006] Determining an offload task in the training device, wherein the offload task includes a high-performance task and a low-energy consumption task;

[0007] Determining the number of the high-performance tasks and the number of the low-energy consumption tasks;

[0008] Determining a resource optimization method for the training device according to the number of the high-performance tasks and the number of the low-energy consumption tasks;

[0009] Get the initial parameters of the actor network and the critic network;

[0010] When the resource optimization method is to optimize the latency, the first cloud device to be used to execute the offloading task is determined according to the initial parameters of the actor network;

[0011] determining a first target cloud device in the first cloud device according to the critic network initial parameters;

[0012] The unloading task of the training device is performed through the first target cloud device.

[0013] Optionally, determining the resource optimization mode of the training device according to the number of the high-performance tasks and the number of the low-energy consumption tasks includes:

[0014] If the number of the high-performance tasks is greater than the number of the low-energy consumption tasks, it is determined that the resource optimization method of the training device is to optimize the latency.

[0015] Optionally, the determining the resource optimization mode of the training device according to the number of the high-performance tasks and the number of the low-energy consumption tasks further includes:

[0016] If the number of the high-performance tasks is less than the number of the low-energy consumption tasks, it is determined that the resource optimization method of the training device is to optimize energy consumption.

[0017] Optionally, it also includes:

[0018] When the resource optimization method is to optimize energy consumption, the second cloud device to be used for executing the offloading task is determined according to the initial parameters of the actor network;

[0019] determining a second target cloud device in the second cloud device according to the critic network initial parameters;

[0020] The unloading task of the training device is performed through the second target cloud device.

[0021] Optionally, the first cloud device includes a local cloud device, a central cloud device, and an edge cloud device;

[0022] The step of determining a first target cloud device in the first cloud device according to the critic network initial parameters includes:

[0023] Determining a first delay in executing the offload task on the local cloud device;

[0024] Determining a second delay for the central cloud device to execute the offloading task;

[0025] Determining a third delay for the edge cloud device to perform the offloading task;

[0026] The minimum delay among the first delay, the second delay, and the third delay is determined as the target delay, and the cloud device corresponding to the target delay is used as the first target cloud device.

[0027] Optionally, determining a first delay for executing the offloading task on the local cloud device includes:

[0028] Acquire a first decision variable for executing the offloading task on the local cloud device;

[0029] Acquire a first rate at which the offload task is executed on the local cloud device;

[0030] A first delay for executing the offloading task on the local cloud device is determined according to the first decision variable and the first rate.

[0031] Optionally, the second cloud device includes a local cloud device, a central cloud device, and an edge cloud device;

[0032] The determining of the second target cloud device in the second cloud device according to the critic network initial parameters includes:

[0033] Determine a first energy consumption of executing the offload task on the local cloud device;

[0034] Determine a second energy consumption of the central cloud device performing the offloading task;

[0035] Determining a third energy consumption of the edge cloud device performing the offloading task;

[0036] The minimum energy consumption among the first energy consumption, the second energy consumption, and the third energy consumption is determined as the target energy consumption, and the cloud device corresponding to the target energy consumption is used as the second target cloud device.

[0037] The present invention also discloses a task unloading device, which comprises:

[0038] A selection module, configured to select at least one mobile device from a plurality of mobile devices as a training device;

[0039] A first determination module is used to determine an offload task in the training device, wherein the offload task includes a high-performance task and a low-energy consumption task;

[0040] A quantity determination module, used to determine the number of the high-performance tasks and the number of the low-energy consumption tasks;

[0041] A mode determination module, used for determining a resource optimization mode of the training device according to the number of the high-performance tasks and the number of the low-energy consumption tasks;

[0042] Parameter acquisition module, used to obtain the initial parameters of the actor network and the critic network;

[0043] A first device determination module, configured to determine a first cloud device to be used for executing the offloading task according to the initial parameters of the actor network when the resource optimization method is to optimize the latency;

[0044] A second device determination module, configured to determine a first target cloud device in the first cloud device according to the critic network initial parameters;

[0045] An execution module is used to execute the uninstallation task of the training device through the first target cloud device.

[0046] The present invention also discloses an electronic device, comprising: a processor, a memory, and a computer program stored in the memory and capable of running on the processor, wherein the computer program implements the steps of the task unloading method as described above when executed by the processor.

[0047] The present invention also discloses a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned task offloading method are implemented.

[0048] The embodiments of the present invention include the following advantages:

[0049] The present invention discloses a task offloading method. The present invention can determine the resource optimization mode according to the number of high-performance tasks and low-energy consumption tasks, and can realize more accurate resource allocation and task scheduling; it can enable the training equipment to maintain good performance in different task scenarios, improve the overall task processing efficiency and the rationality of resource utilization, ensure the speed and quality of task processing, and enable complex tasks to be completed smoothly; and by offloading tasks, it reduces the resource burden of the training equipment, allowing it to focus more on the execution of local core functions and the processing of other low-energy consumption tasks, optimizes the resource coordination between mobile devices and cloud devices in the entire system, improves the overall task processing capability and operation efficiency of the system, and ultimately provides users with a better quality and more efficient mobile application service experience, and realizes reasonable allocation of computing resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 is a flowchart of a task offloading method provided by an embodiment of the present invention;

[0051] Figure 2 is a schematic diagram of a communication framework provided by an embodiment of the present invention;

[0052] Figure 3 It is a structural block diagram of a task offloading device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0053] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and understandable, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0054] One of the core concepts of the embodiments of the present invention is that the present invention can determine the resource optimization method according to the number of high-performance tasks and low-energy tasks, so as to achieve more accurate resource allocation and task scheduling; it can enable the training equipment to maintain good performance in different task scenarios, improve the overall task processing efficiency and the rationality of resource utilization, ensure the speed and quality of task processing, and enable complex tasks to be completed smoothly. Moreover, by offloading tasks, the resource burden of the training equipment is reduced, so that it can focus more on the execution of local core functions and the processing of other low-energy tasks, optimize the resource coordination between mobile devices and cloud devices in the entire system, improve the overall task processing capability and operating efficiency of the system, and ultimately provide users with a better and more efficient mobile application service experience and realize reasonable allocation of computing resources.

[0055] Reference Figure 1 , shows a flowchart of a task offloading method provided by an embodiment of the present invention, the method may include the following steps:

[0056] Step 101, selecting at least one mobile device from a plurality of mobile devices as a training device;

[0057] In an embodiment of the present invention, in many application scenarios involving mobile edge computing, distributed reinforcement learning, etc., there are numerous mobile devices, each of which has different resource conditions. At least one of the mobile devices is selected as a training device, and subsequent reinforcement learning training related to task offloading and resource optimization is carried out based on them. This can avoid indiscriminate complex operations on all mobile devices, improve efficiency and be more targeted. At the same time, the experience and training results accumulated by these selected devices may be subsequently promoted and applied to other similar devices, thereby enhancing the overall system's ability to process tasks.

[0058] In one example, taking a mobile phone terminal as an example, mobile phones with a remaining power of more than 50% and a memory usage rate of less than 70% can be selected as training devices.

[0059] Step 102, determining the offload tasks in the training device, where the offload tasks include high-performance tasks and low-energy consumption tasks.

[0060] In the embodiment of the present invention, due to the limitations of the mobile device's own battery power supply, processor performance, etc., the tasks generated can be classified according to the demand for computing resources and energy consumption. High-performance tasks are generally those that have high computing power requirements and require a large amount of processor computing resources, memory resources, etc. to be completed quickly and with high quality, such as complex 3D game rendering, high-definition video real-time editing, etc.; while low-energy consumption tasks are relatively less demanding on computing resources and consume less power during execution, such as simple text information processing, sensor data collection and preliminary analysis, etc. Distinguishing between these two types of tasks is very critical for subsequently reasonably selecting the offloading method and determining the resource optimization strategy according to the device resource status and optimization goals, because different types of tasks have different performance, energy consumption, and impact on the overall system resource utilization when they are offloaded to different devices (such as different cloud devices).

[0061] Step 103, determining the number of high-performance tasks and the number of low-energy consumption tasks;

[0062] In the embodiments of the present invention, different proportions of high-performance tasks and low-energy consumption tasks mean different requirements for computing resources, energy consumption control, etc. For example, when the number of high-performance tasks is large, some tasks can be offloaded to cloud devices with powerful computing capabilities to ensure that the tasks are completed on time; and when the number of low-energy consumption tasks accounts for a large proportion, more emphasis can be placed on reasonably scheduling resources on local mobile devices for processing to reduce unnecessary network transmission overhead, etc.

[0063] Step 104, determining a resource optimization method for the training device based on the number of high-performance tasks and the number of low-energy consumption tasks.

[0064] In an embodiment of the present invention, resource optimization methods for training equipment may include optimization based on latency and optimization based on energy consumption. Specifically, the resource optimization method for training equipment may be selected based on the number of high-performance tasks and low-energy consumption tasks.

[0065] Step 105, obtaining initial parameters of the actor network and the critic network;

[0066] In the embodiment of the present invention, the initial parameters of the actor network are mainly responsible for generating corresponding actions according to the current state, and the initial parameters of the critic network perform value assessment on the current state and action, and assist the actor network in optimizing the strategy.

[0067] In some simple scenarios or when there is no prior experience to refer to, the parameters of the actor network and the critic network can be randomly assigned to put them in an initial state where training can begin. For example, for an actor network in the form of a simple fully connected neural network, the connection weights between neurons in each layer can be initialized by randomly generating values ​​within a given interval (such as [-0.1, 0.1]).

[0068] Step 106: When the resource optimization method is to optimize the latency, the cloud device to be used for executing the offloading task is determined according to the initial parameters of the actor network.

[0069] In the embodiment of the present invention, when it is determined that the resource optimization method is to optimize latency, a corresponding action decision can be generated based on the probability distribution of different actions, that is, to determine the specific cloud device to be used to perform the offloading task.

[0070] Assuming that there are cloud devices A, B, and C to choose from for offloading tasks, the actor network outputs a probability of 30% for selecting cloud device A, a probability of 50% for selecting cloud device B, and a probability of 20% for selecting cloud device C based on the input status information and its own initial parameters. Then, random sampling is performed according to this probability distribution, and finally it is determined to send the offloading task to a specific cloud device. It is expected that the goal of latency optimization can be achieved by selecting this cloud device. Because different cloud devices differ in computing power, network access location, current load, etc., they will have different impacts on the processing latency of the task.

[0071] Step 107 , determining a target cloud device in the cloud devices according to the initial parameters of the critic network.

[0072] In the embodiment of the present invention, the current state-action pair can be evaluated for value based on the initial parameters of the critic network, that is, to judge how good or bad the selection of a certain cloud device is for achieving overall resource optimization. The value evaluation function constructed by its own parameters comprehensively considers various factors to give a numerical value representing the value.

[0073] For example, if the critic network evaluates the value of cloud device A as 0.6, the value of cloud device B as 0.8, and the value of cloud device C as 0.5, cloud device B will be determined as the target cloud device.

[0074] Step 108, executing the unloading task of the training device through the target cloud device.

[0075] In an embodiment of the present invention, when the target cloud device is determined, the training device will send the previously determined offloading tasks to the target cloud device via network transmission. After receiving these tasks, the target cloud device will use its own powerful computing resources to process them according to the corresponding task processing logic. After the processing is completed, the results will be transmitted back to the training device via the network.

[0076] The present invention discloses a task offloading method. The present invention can determine the resource optimization mode according to the number of high-performance tasks and low-energy consumption tasks, so as to realize more accurate resource allocation and task scheduling. It can enable the training equipment to maintain good performance in different task scenarios, improve the overall task processing efficiency and the rationality of resource utilization, ensure the speed and quality of task processing, and enable complex tasks to be completed smoothly. Moreover, by offloading tasks, the resource burden of the training equipment is reduced, so that it can focus more on the execution of local core functions and the processing of other low-energy consumption tasks, optimize the resource coordination between mobile devices and cloud devices in the whole system, improve the overall task processing capability and operation efficiency of the system, and finally provide users with a better and more efficient mobile application service experience.

[0077] In one embodiment of the present invention, the resource optimization method of the training device is determined based on the number of high-performance tasks and the number of low-energy tasks, including: if the number of high-performance tasks is greater than the number of low-energy tasks, then the resource optimization method of the training device is determined to be to optimize latency.

[0078] In the embodiments of the present invention, high-performance tasks usually have high requirements for computing resources, such as involving complex algorithm calculations, large amounts of data processing, or requiring high graphics rendering capabilities, such as running large 3D games, performing real-time editing of high-definition videos, and executing training or reasoning of deep learning models. When these tasks are executed on mobile devices, if they only rely on local limited processor performance, memory capacity and other resources, they often take a long time to complete, and may even cause tasks to become stuck or unable to be executed smoothly due to insufficient resources. In addition, since high-performance tasks are generally at a critical link in the business process, the timeliness of their processing has a great impact on the user experience of the entire application. For example, in real-time video calls, if the processing delay is too long, the picture seen by the user will be significantly delayed, seriously affecting the communication effect.

[0079] When the number of high-performance tasks is greater than the number of low-energy tasks, it means that the main task load currently faced by the training device is high-performance tasks that have high resource requirements and may take a long time. This imbalance in the number of tasks indicates that relying solely on local resources for processing is likely to be unable to meet the needs of all tasks being completed on time and with high quality, causing the device to fall into difficulties such as tight computing resources and task queuing, which in turn leads to a significant increase in the overall task processing delay. Therefore, a resource optimization method that focuses on reducing the total task processing delay is needed to ensure that these high-performance tasks and the overall task can be successfully completed within an acceptable time frame.

[0080] In one embodiment of the present invention, the resource optimization method of the training device is determined based on the number of high-performance tasks and the number of low-energy consumption tasks, and also includes: if the number of high-performance tasks is less than the number of low-energy consumption tasks, then the resource optimization method of the training device is determined to be to optimize energy consumption.

[0081] In the embodiments of the present invention, low-energy tasks are generally those tasks that have relatively low requirements for computing resources and consume less power during execution, such as simple text editing, basic sensor data collection and preliminary analysis, ordinary file viewing and other operations. Although such tasks individually do not put much pressure on device resources and energy consumption, when there are a large number of them, the overall accumulation will also occupy a certain amount of device resources. In addition, since mobile devices are powered by batteries, power is a relatively limited resource. If too many low-energy tasks are not properly allocated, it may also cause the device to consume power too quickly, affecting the device's battery life and the length of time the user normally uses the mobile device.

[0082] When the number of high-performance tasks is less than the number of low-energy tasks, it means that the main task load faced by the training device at this time is a large number of low-energy tasks. In this case, although a single low-energy task does not consume much resources and energy, the continuous operation of many low-energy tasks may cause the device resources to be dispersed and continuously consume power. Compared with the urgent need for computing power when high-performance tasks dominate, it is more necessary to focus on how to ensure the smooth completion of these low-energy tasks while minimizing the energy consumption of the device and extending the battery life of the device to meet the needs of users to use mobile devices for a long time to perform various basic operations. Therefore, it is particularly important to determine the resource optimization method to optimize energy consumption.

[0083] In one embodiment of the present invention, it also includes: when the resource optimization method is to optimize energy consumption, determining the second cloud device to be used to execute the offloading task according to the initial parameters of the actor network; determining the second target cloud device in the second cloud device according to the initial parameters of the critic network; and executing the offloading task of the training device through the second target cloud device.

[0084] In an embodiment of the present invention, when the resource optimization method is determined to be to optimize energy consumption, it means that the focus is on how to reduce energy consumption as much as possible during the execution of the offloading task to extend the battery life of the training equipment and achieve efficient and energy-saving overall resource utilization.

[0085] For example, assuming that there are cloud devices A, B, and C as candidate second cloud devices, the actor network may output a probability of 30% for selecting cloud device A, a probability of 50% for selecting cloud device B, and a probability of 20% for selecting cloud device C based on the above-mentioned input state information and its own initial parameters. Then, based on these probabilities, it can be determined that the second cloud device is cloud device B.

[0086] The critic network will analyze all the second cloud devices, such as the determined cloud device B, which is 3. For example, cloud device B includes cloud device b1 with a value of 0.6, cloud device b2 with a value of 0.7, and cloud device b3 with a value of 0.5, and then select the cloud device with the highest value as the second target cloud device. In this example, cloud device b2 will be determined as the second target cloud device.

[0087] In one embodiment of the present invention, a first cloud device includes a local cloud device, a central cloud device, and an edge cloud device; a target cloud device is determined in the cloud device according to the initial parameters of the critic network, including: determining a first delay for executing an offloading task in the local cloud device; determining a second delay for the central cloud device to execute the offloading task; determining a third delay for the edge cloud device to execute the offloading task; determining the minimum delay among the first delay, the second delay, and the third delay as the target delay, and using the cloud device corresponding to the target delay as the first target cloud device.

[0088] In an embodiment of the present invention, the local cloud device generally refers to a cloud computing resource deployed in a local network environment relatively close to the training device. It can be in the same local area network as the mobile device or within a close network coverage range.

[0089] Central cloud equipment is generally a large-scale cloud computing infrastructure provided by large cloud service providers, which is centrally deployed in data centers and has massive computing resources (such as a large number of high-performance processors, large-capacity memory and storage devices, etc.), powerful network bandwidth, and a highly reliable operation and maintenance guarantee system. Cloud services such as Amazon's AWS, Microsoft's Azure, and Alibaba Cloud rely on this type of central cloud equipment.

[0090] Edge cloud devices are located between local cloud devices and central cloud devices. They are cloud computing resources deployed close to data sources or user terminals. They are designed to extend cloud computing capabilities to the edge of the network and reduce the long-distance round-trip delay of data transmission to the remote central cloud. At the same time, they can provide more powerful computing resource support than local cloud devices. For example, edge computing servers deployed next to 5G base stations, or edge cloud computing nodes set up in some large campuses, factories and other places are all edge cloud devices.

[0091] For local devices, due to the short distance, only the task delay needs to be considered.

[0092] For central cloud devices, their network transmission delay is often greatly affected by factors such as long-distance data transmission and network intermediate nodes. Data needs to be transmitted from mobile devices through multiple network routers, switches and other devices across the wide area network to the central cloud devices located in a distant data center. In this process, network congestion and link switching between different network operators may occur, resulting in increased transmission time.

[0093] For edge cloud devices, since they are close to mobile devices, their network transmission latency is usually between local cloud devices and central cloud devices. On the one hand, it avoids the high latency problem of long-distance transmission like central cloud devices, and on the other hand, it is not limited by the overly limited local network environment like local cloud devices.

[0094] By comparing the first delay, second delay, and third delay calculated previously, find the smallest value as the target delay. For example, in the above example, the first delay is 15 milliseconds, the second delay is 35 milliseconds, and the third delay is 18 milliseconds, then the smallest 15 milliseconds is the target delay. The purpose of selecting the minimum delay is to meet the requirements of delay optimization in the resource optimization method, ensure that the offload task can be completed at the fastest speed, and improve the real-time performance of the entire task processing process.

[0095] In one embodiment of the present invention, determining a first delay in executing an offload task on a local cloud device includes: obtaining a first decision variable for executing the offload task on the local cloud device; obtaining a first rate for executing the offload task on the local cloud device; and determining a first delay in executing the offload task on the local cloud device based on the first decision variable and the first rate.

[0096] In the embodiment of the present invention, the first delay can be calculated by the following formula (1):

[0097] T1=δ i,l ×d i / R l ,δ i,l =1 Formula (1)

[0098] Among them, δi,l refers to the first decision variable, T1 refers to the first time delay, R l refers to the first rate, and di refers to the distance from the local cloud device to the training device.

[0099] In an embodiment of the present invention, when calculating the second time delay, it can be calculated by formula (2):

[0100] T2=δ i,e ×(d i / R e +t i,e (trans)),δ i,e =1 Formula (2)

[0101] Where, T2 in formula (2) refers to the second delay, t i,e (trans) refers to the transmission delay when the offload task is migrated to the central cloud device, δ i,e refers to the second decision variable, R e Refers to the second rate at which offload tasks are executed in the central cloud device.

[0102] In an embodiment of the present invention, when calculating the third time delay, it can be calculated by formula (3):

[0103] T3=δ i,c ×(d i / R c +t i,c (trans)),δ i,c =1 Formula (3)

[0104] Among them, T3 refers to the third delay, R c refers to the third rate of executing offload tasks on edge cloud devices, t i,c (trans) refers to the transmission delay when the offload task is migrated to the edge cloud device, δ i,c Refers to the third decision variable for performing offloading tasks at the edge cloud device.

[0105] In one embodiment of the present invention, the second cloud device includes a local cloud device, a central cloud device, and an edge cloud device; a second target cloud device is determined in the cloud device according to the initial parameters of the critic network, including: determining a first energy consumption of performing an offloading task in the local cloud device; determining a second energy consumption of performing an offloading task in the central cloud device; determining a third energy consumption of performing an offloading task in the edge cloud device; determining the minimum energy consumption among the first energy consumption, the second energy consumption, and the third energy consumption as the target energy consumption, and using the cloud device corresponding to the target energy consumption as the second target cloud device.

[0106] In the embodiment of the present invention, for the local device, since the distance is relatively short, only the task processing energy consumption needs to be considered.

[0107] For central cloud devices, their energy consumption and delay are often greatly affected by factors such as long-distance data transmission and intermediate network nodes. Data needs to be transmitted from mobile devices through multiple network routers, switches and other devices across the wide area network to the central cloud devices located in a distant data center. This process may encounter network congestion, link switching between different network operators and other problems, resulting in increased energy consumption.

[0108] For edge cloud devices, since they are close to mobile devices, their network transmission delay is usually between local cloud devices and central cloud devices.

[0109] After calculating the first energy consumption, the second energy consumption, and the third energy consumption, the minimum energy consumption can be determined as the target energy consumption, and the cloud device corresponding to the target energy consumption can be used as the second target cloud device to ensure that task processing can achieve effective control of energy consumption and rational use of resources while meeting functional requirements, which is in line with the requirements of terminal devices such as mobile devices for energy saving and efficient operation.

[0110] In an embodiment of the present invention, the first energy consumption can be calculated by formula (4):

[0111] E1=δ i,l ×P l / T i,l ,δ i,l =1 Formula (4)

[0112] Among them, E1 refers to the first energy consumption, P l Refers to the energy consumption of executing tasks on local cloud devices, T i,l Refers to the first latency of executing offload tasks on local cloud devices.

[0113] The second energy consumption can be calculated by formula (5):

[0114] E2=δ i,e ×(P i,e / T i,e +P e / T i,e (trans)),δ i,e =1 Formula (5)

[0115] Among them, E2 refers to the second energy consumption, P i,e Refers to the energy consumption of executing tasks in the central cloud device, T i,e It refers to the second delay of executing the offload task in the central cloud device, P e Refers to the transmission power consumption of the central cloud equipment.

[0116] The third energy consumption can be calculated by formula (6):

[0117] E3=δ i,c ×(P i,c / T i,c +P c / T i,c (trans)),δ i,c =1 Formula (6)

[0118] Among them, E3 refers to the third energy consumption, P i,c Refers to the energy consumption of executing tasks on edge cloud devices, T i,c It refers to the third latency of executing the offload task on the edge cloud device, P c Refers to the transmission power consumption of edge cloud devices.

[0119] like Figure 2 , showing a communication framework schematic diagram provided by an embodiment of the present invention. The central cloud serves as the main data and storage node, responsible for coordinating and managing the entire network resources. For applications with high real-time requirements, the central cloud cannot meet the requirements, thus generating an edge cloud. The edge cloud is a server or device cluster deployed at the edge of the network, which has certain computing and storage capabilities and can process data requests from mobile devices. By migrating some tasks to the edge cloud, the delay in data transmission in the network is reduced and the response speed of the application is improved. The core network is the network part located in the center of the mobile communication network, which is mainly responsible for user data processing, session management, mobility management, and connection with external networks. It is equivalent to the "brain" of the entire mobile communication system, coordinating and controlling various communication resources to ensure that mobile users can smoothly make voice calls, send text messages, and use mobile data services.

[0120] The present invention discloses a task offloading method. The present invention can determine the resource optimization mode according to the number of high-performance tasks and low-energy consumption tasks, so as to realize more accurate resource allocation and task scheduling. It can enable the training equipment to maintain good performance in different task scenarios, improve the overall task processing efficiency and the rationality of resource utilization, ensure the speed and quality of task processing, and enable complex tasks to be completed smoothly. Moreover, by offloading tasks, the resource burden of the training equipment is reduced, so that it can focus more on the execution of local core functions and the processing of other low-energy consumption tasks, optimize the resource coordination between mobile devices and cloud devices in the whole system, improve the overall task processing capability and operation efficiency of the system, and finally provide users with a better and more efficient mobile application service experience.

[0121] It should be noted that, for the sake of simplicity, the method embodiments are described as a series of action combinations, but those skilled in the art should be aware that the embodiments of the present invention are not limited by the order of the actions described, because according to the embodiments of the present invention, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily required by the embodiments of the present invention.

[0122] Reference Figure 3 , shows a structural block diagram of a device provided by an embodiment of the present invention, and the device may include the following modules:

[0123] A selection module 201, configured to select at least one mobile device from a plurality of mobile devices as a training device;

[0124] A first determination module 202 is used to determine an offload task in the training device, where the offload task includes a high-performance task and a low-energy consumption task;

[0125] A quantity determination module 203, used to determine the number of high-performance tasks and the number of low-energy consumption tasks;

[0126] A mode determination module 204, used to determine a resource optimization mode of the training device according to the number of high-performance tasks and the number of low-energy consumption tasks;

[0127] A parameter acquisition module 205 is used to acquire initial parameters of the actor network and initial parameters of the critic network;

[0128] A first device determination module 206, configured to determine a first cloud device to be used for performing the offloading task according to the initial parameters of the actor network when the resource optimization method is to optimize the latency;

[0129] A second device determination module 207, configured to determine a first target cloud device in the first cloud device according to the critic network initial parameters;

[0130] The execution module 208 is used to execute the unloading task of the training device through the first target cloud device.

[0131] The present invention discloses a task offloading device. The present invention can determine the resource optimization method according to the number of high-performance tasks and low-energy consumption tasks, and can realize more accurate resource allocation and task scheduling. It can enable the training equipment to maintain good performance in different task scenarios, improve the overall task processing efficiency and the rationality of resource utilization, ensure the speed and quality of task processing, and enable complex tasks to be completed smoothly. Moreover, by offloading tasks, the resource burden of the training equipment is reduced, so that it can focus more on the execution of local core functions and the processing of other low-energy consumption tasks, optimize the resource coordination between mobile devices and cloud devices in the entire system, improve the overall task processing capability and operation efficiency of the system, and finally provide users with a better and more efficient mobile application service experience.

[0132] In one embodiment of the present invention, the mode determination module includes:

[0133] The first method determines a submodule, which is used to determine that the resource optimization method of the training device is to optimize the latency if the number of high-performance tasks is greater than the number of low-energy consumption tasks.

[0134] In one embodiment of the present invention, the mode determination module includes:

[0135] The second method determines a submodule, which is used to determine that the resource optimization method of the training device is to optimize energy consumption if the number of high-performance tasks is less than the number of low-energy consumption tasks.

[0136] In one embodiment of the present invention, it also includes:

[0137] A second determination module is used to determine the second cloud device to be used for executing the offloading task according to the initial parameters of the actor network when the resource optimization method is to optimize energy consumption;

[0138] A third determination module is used to determine a second target cloud device in the second cloud device according to the critic network initial parameters;

[0139] The task execution module is used to execute the unloading task of the training device through the second target cloud device.

[0140] In an embodiment of the present invention, the first cloud device includes a local cloud device, a central cloud device, and an edge cloud device; the second device determination module includes:

[0141] A first delay determination submodule, used to determine a first delay for executing an offload task on a local cloud device;

[0142] A second time delay determination submodule, used to determine a second time delay for the central cloud device to perform the offloading task;

[0143] A third delay determination submodule, used to determine a third delay for the edge cloud device to perform the offloading task;

[0144] The target delay determination submodule is used to determine the minimum delay among the first delay, the second delay, and the third delay as the target delay, and to use the cloud device corresponding to the target delay as the first target cloud device.

[0145] In one embodiment of the present invention, the first delay determination submodule includes:

[0146] A first acquisition unit, used to acquire a first decision variable for executing an offloading task on a local cloud device;

[0147] A second acquisition unit, configured to acquire a first rate at which the offload task is executed on the local cloud device;

[0148] The delay determination unit is used to determine a first delay for executing the offload task on the local cloud device according to the first decision variable and the first rate.

[0149] In an embodiment of the present invention, the second cloud device includes a local cloud device, a central cloud device, and an edge cloud device; the third determination module includes:

[0150] A first energy consumption determination submodule, used to determine a first energy consumption of executing an offload task on a local cloud device;

[0151] A second energy consumption determination submodule, used to determine a second energy consumption of the central cloud device to perform the offloading task;

[0152] A third energy consumption determination submodule, used to determine the third energy consumption of the edge cloud device performing the offloading task;

[0153] The target energy consumption determination submodule is used to determine the minimum energy consumption among the first energy consumption, the second energy consumption, and the third energy consumption as the target energy consumption, and to use the cloud device corresponding to the target energy consumption as the second target cloud device.

[0154] The present invention discloses a task offloading device. The present invention can determine the resource optimization method according to the number of high-performance tasks and low-energy consumption tasks, and can realize more accurate resource allocation and task scheduling. It can enable the training equipment to maintain good performance in different task scenarios, improve the overall task processing efficiency and the rationality of resource utilization, ensure the speed and quality of task processing, and enable complex tasks to be completed smoothly. Moreover, by offloading tasks, the resource burden of the training equipment is reduced, so that it can focus more on the execution of local core functions and the processing of other low-energy consumption tasks, optimize the resource coordination between mobile devices and cloud devices in the entire system, improve the overall task processing capability and operation efficiency of the system, and finally provide users with a better and more efficient mobile application service experience.

[0155] As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0156] An embodiment of the present invention further provides an electronic device, including:

[0157] It includes a processor, a memory, and a computer program stored in the memory and capable of running on the processor. When the computer program is executed by the processor, the various processes of the above-mentioned task offloading method embodiment are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0158] An embodiment of the present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the various processes of the above-mentioned task offloading method embodiment are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0159] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0160] It will be appreciated by those skilled in the art that the embodiments of the present invention may be provided as methods, devices, or computer program products. Therefore, the embodiments of the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the embodiments of the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0161] The embodiments of the present invention are described with reference to the flowcharts and / or block diagrams of the methods, terminal devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0162] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0163] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device so that a series of operating steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable terminal device to implement the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0164] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the embodiments of the present invention.

[0165] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprises" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal device. In the absence of further restrictions, the elements defined by the sentence "including one..." do not exclude the existence of other identical elements in the process, method, article or terminal device including the elements.

[0166] The above is a detailed introduction to a task offloading method, device, equipment and storage medium provided by the present invention. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for general technical personnel in this field, according to the idea of ​​the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A task offloading method, characterized in that: The method comprises: selecting at least one mobile device from a plurality of mobile devices as a training device; Determining an offload task in the training device, wherein the offload task includes a high-performance task and a low-energy consumption task; Determining the number of the high-performance tasks and the number of the low-energy consumption tasks; Determining a resource optimization method for the training device according to the number of the high-performance tasks and the number of the low-energy consumption tasks; Get the initial parameters of the actor network and the critic network; When the resource optimization method is to optimize the latency, the first cloud device to be used to execute the offloading task is determined according to the initial parameters of the actor network; determining a first target cloud device in the first cloud device according to the critic network initial parameters; The unloading task of the training device is performed through the first target cloud device.

2. The method according to claim 1, characterized in that The determining of the resource optimization mode of the training device according to the number of the high-performance tasks and the number of the low-energy consumption tasks includes: If the number of the high-performance tasks is greater than the number of the low-energy consumption tasks, it is determined that the resource optimization method of the training device is to optimize the latency.

3. The method according to claim 1, characterized in that The determining of the resource optimization mode of the training device according to the number of the high-performance tasks and the number of the low-energy consumption tasks further includes: If the number of the high-performance tasks is less than the number of the low-energy consumption tasks, it is determined that the resource optimization method of the training device is to optimize energy consumption.

4. The method according to claim 3, characterized in that: Also includes: When the resource optimization method is to optimize energy consumption, the second cloud device to be used for executing the offloading task is determined according to the initial parameters of the actor network; determining a second target cloud device in the second cloud device according to the critic network initial parameters; The unloading task of the training device is performed through the second target cloud device.

5. The method according to claim 1, characterized in that The first cloud device includes a local cloud device, a central cloud device, and an edge cloud device; The step of determining a first target cloud device in the first cloud device according to the critic network initial parameters includes: Determining a first delay in executing the offload task on the local cloud device; Determining a second delay for the central cloud device to execute the offloading task; Determining a third delay for the edge cloud device to perform the offloading task; The minimum delay among the first delay, the second delay, and the third delay is determined as the target delay, and the cloud device corresponding to the target delay is used as the first target cloud device.

6. The method according to claim 5, characterized in that The determining a first delay in executing the offloading task on the local cloud device includes: Acquire a first decision variable for executing the offloading task on the local cloud device; Acquire a first rate at which the offload task is executed on the local cloud device; A first delay for executing the offloading task on the local cloud device is determined according to the first decision variable and the first rate.

7. The method according to claim 4, characterized in that The second cloud device includes a local cloud device, a central cloud device, and an edge cloud device; The determining of the second target cloud device in the second cloud device according to the critic network initial parameters includes: Determine a first energy consumption of executing the offload task on the local cloud device; Determine a second energy consumption of the central cloud device performing the offloading task; Determining a third energy consumption of the edge cloud device performing the offloading task; The minimum energy consumption among the first energy consumption, the second energy consumption, and the third energy consumption is determined as the target energy consumption, and the cloud device corresponding to the target energy consumption is used as the second target cloud device.

8. A task offloading device, characterized in that: The device comprises: A selection module, configured to select at least one mobile device from a plurality of mobile devices as a training device; A first determination module is used to determine an offload task in the training device, wherein the offload task includes a high-performance task and a low-energy consumption task; A quantity determination module, used to determine the number of the high-performance tasks and the number of the low-energy consumption tasks; A mode determination module, used for determining a resource optimization mode of the training device according to the number of the high-performance tasks and the number of the low-energy consumption tasks; Parameter acquisition module, used to obtain the initial parameters of the actor network and the critic network; A first device determination module, configured to determine a first cloud device to be used for executing the offloading task according to the initial parameters of the actor network when the resource optimization method is to optimize the latency; A second device determination module, configured to determine a first target cloud device in the first cloud device according to the critic network initial parameters; An execution module is used to execute the uninstallation task of the training device through the first target cloud device.

9. An electronic device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor is used to execute the steps of the task offloading method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the task offloading method according to any one of claims 1 to 7 are implemented.

Images