Computing task fusion scheduling method under edge network and related equipment

Abstract

One or more embodiments of the invention provide a computing task fusion scheduling method under an edge network and related equipment. The method comprises the following steps: acquiring current information of all base stations and edge servers in the edge network; obtaining a scheduling strategy of the task according to the current information of the base station and the edge server; and scheduling the task according to the scheduling strategy. Through the application, the communication delay can be reduced, and the service quality is improved.

Description

Technical Field

[0001] One or more embodiments of the present application relate to the field of communication technology, and in particular, to a method for intelligent scheduling of computing tasks in an edge network and related equipment. Background Art

[0002] The multimedia video field has shown a booming development trend in recent years. Multimedia technology is developing rapidly and becoming an important part of the information technology field. The massive heterogeneous data computing tasks generated by new multimedia applications pose two major challenges to user experience: latency and energy consumption.

[0003] Edge networks effectively reduce data transmission delays and improve the response speed of multimedia services by placing computing resources close to end users. However, computing resources in edge networks are scarce and computational costs are high. In this case, designing a reasonable computing task offloading solution is the key to determining whether edge devices can better process locally generated multimedia data.

[0004] However, the related technologies have not proposed a task scheduling solution that integrates the global information of the edge network. Summary of the invention

[0005] In view of this, the purpose of one or more embodiments of the present application is to propose a computing task integrated scheduling method and related equipment in an edge network to solve the problems raised by the background technology.

[0006] Based on the above objectives, one or more embodiments of the present application provide a method for intelligent scheduling of computing tasks in an edge network. The method includes:

[0007] Obtaining current information of all base stations and edge servers in the edge network;

[0008] Obtaining a scheduling strategy for the task according to current information of the base station and the edge server;

[0009] The tasks are scheduled according to the scheduling strategy.

[0010] Optionally, the scheduling strategy is the optimal solution of the following objective function;

[0011]

[0012]

[0013] in, represents the kth task The communication delay, Represents the kth task The computational delay of represents the kth task The computing energy consumption, Represents the kth task The number of transfers between the base stations, represents the kth task The amount of data, Represents the kth task The transmission rate of the i-th transfer between the base stations, R BS-ES Represents the kth task The transmission rate between the base station and the edge server, Indicates the number of floating point operations that the edge server hardware can perform per second. Indicates the degree of parallelism, Represents edge server Energy factors related to CPU architecture, Indicates the number of CPU cycles required for calculation. Indicates the number of CPU cycles that the virtual machine can execute per second.

[0014] Optionally, the scheduling strategy is generated by a scheduling strategy generation model, and the scheduling strategy model includes an action neural network and a value evaluation neural network;

[0015] The scheduling strategy generation model performs the following operations until the preset conditions are met:

[0016] The action neural network generates an action based on the current information, and the action indicates a scheduling strategy;

[0017] The value evaluation neural network generates an evaluation value of the action.

[0018] Optionally, generating an action according to the current information includes:

[0019] Generate the current state of the edge network based on the current information of the base station and the edge server represents the current information of the edge server, q k =(c k ,g k ,d k ) represents the current information of the base station, Indicates the average CPU speed of this iteration, Indicates the average GPU speed of this iteration, represents the transmission rate between the edge servers, c k Indicates the number of CPU cycles required for the task calculation, g k Indicates the number of GPU cycles required for the task calculation, d k Indicates the amount of data calculated by the task;

[0020] The action is generated according to the current state.

[0021] Optionally, after generating the action, the action neural network further performs the following operations:

[0022] Update the parameters of the action neural network according to the following formula;

[0023]

[0024] in, Expressing expectation, r t (θ) represents the reward value of the tth iteration, T avg represents the average delay of all the tasks. avg represents the average energy consumption of all the tasks, and a and b represent the preset weight parameters A t represents the action of the tth iteration, ∈ represents the IO conflict factor.

[0025] Optionally, after generating the evaluation value of the action, the value evaluation neural network further performs the following operations:

[0026] Update the parameters of the value evaluation neural network according to the following formula;

[0027]

[0028] Among them, V ω (s t ) represents the value function estimation based on the current parameter ω, td_target t Represents the target value calculated based on the sampling process.

[0029] Optionally, scheduling the task according to the scheduling policy includes:

[0030] Any of the base stations forwards the transfer task to the edge server according to the scheduling strategy, or forwards the transfer task to the edge server through other base stations; the number of forwarding times of the transfer task is less than a preset number, and there is no transfer task for the base station scheduled according to the scheduling strategy.

[0031] Based on the same inventive concept, one or more embodiments of the present application further provide a computing task intelligent scheduling device under an edge network, including:

[0032] An acquisition module, configured to acquire current information of all base stations and edge servers in the edge network;

[0033] A calculation module, configured to obtain a scheduling strategy for the task according to current information of the base station and the edge server;

[0034] The scheduling module is configured to schedule the tasks according to the scheduling strategy.

[0035] Based on the same inventive concept, one or more embodiments of the present application also provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for intelligent scheduling of computing tasks in an edge network as described in any one of the above items is implemented.

[0036] Based on the same inventive concept, one or more embodiments of the present application also provide a non-transitory computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable the computer to execute any of the above-mentioned computing task integration and intelligent scheduling methods under the edge network.

[0037] From the above, it can be seen that the computing task integrated scheduling method under the edge network provided by one or more embodiments of the present application obtains the current information of all base stations and edge servers in the edge network; obtains the scheduling strategy of the task according to the current information of the base station and the edge server; and schedules the task according to the scheduling strategy.

[0038] One or more embodiment methods provided in the present application determine the scheduling strategy for all tasks in the edge network based on the global current information of the base station and edge server of the edge network, thereby reducing the latency and energy consumption of the edge network.

[0039] The computing task integration and intelligent scheduling device, electronic device and computer-readable storage medium provided in the present application are all capable of implementing the steps of the above-mentioned computing task integration and intelligent scheduling method under the edge network, and therefore also have the beneficial effects of the above-mentioned computing task integration and intelligent scheduling method under the edge network. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate one or more embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only one or more embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0041] Figure 1 A flowchart of a method for intelligent scheduling of computing tasks in an edge network according to one or more embodiments of the present application;

[0042] Figure 2A schematic diagram of the structure of a computing task intelligent scheduling device under an edge network according to one or more embodiments of the present application;

[0043] Figure 3 A schematic diagram of the architecture of an edge network of one or more embodiments of the present application;

[0044] Figure 4 A schematic diagram of the hardware structure of an electronic device according to one or more embodiments of the present application. DETAILED DESCRIPTION

[0045] In order to make the objectives, technical solutions and advantages of the present disclosure more clearly understood, the present disclosure is further described in detail below in combination with specific embodiments and with reference to the accompanying drawings.

[0046] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in one or more embodiments of the present application should be understood by people with ordinary skills in the field to which the present disclosure belongs. The "first", "second" and similar words used in one or more embodiments of the present application do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word cover the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0047] As mentioned in the background technology section, with the popularization of mobile Internet and the continuous advancement of 5G technology, people's demand for low-latency, low-energy consumption, and high-quality multimedia services is increasing. The transmission and processing of multimedia data such as video, audio, and images have become the core elements of various application scenarios, such as real-time video calls, VR cloud games, and live video broadcasts. The current development of the multimedia field shows a trend of diversification, intelligence, and collaboration, and edge networks, as emerging technologies, play an important role in it. In this context, it is particularly important to conduct in-depth research and exploration of the application and optimization strategies of collaborative task scheduling in edge networks.

[0048] Edge networks can effectively reduce data transmission delays and improve response speed, but they also have the disadvantages of scarce computing resources and expensive computing costs on edge servers. In addition, in edge networks, communication base stations as schedulers and edge servers as computing providers are often in random and uneven topological distributions, and related technologies have not yet proposed a global solution that integrates all information. In this case, designing a reasonable computing task scheduling scheme is the key to determining whether edge devices can better process locally generated multimedia data.

[0049] Therefore, one or more embodiments of the present application propose a method for intelligent scheduling of computing tasks in an edge network. Figure 3 As shown, in the edge network architecture of this application, three types of nodes are distributed, namely, communication base stations, edge servers and users. Among them, the communication base station provides communication services such as data scheduling and transfer for users within a certain range around it. Depending on the geographical location and the service range of the communication base station, the user can be connected to multiple communication base stations through wireless links respectively. With the help of these wireless links, users can upload multimedia computing tasks such as cloud rendering and receive task results. The controller deployed on the communication base station can reasonably schedule and transfer user tasks collected by the communication base station based on local system information such as task queues and server loads. There are some large-capacity optical fiber connections between communication base stations and communication base stations, and between communication base stations and edge servers to provide a certain rate of task data transmission. As a computing node, the edge server supports parallel computing with IO conflicts. The communication base station can offload tasks to any adjacent computing node.

[0050] refer to Figure 1 The computing task intelligent scheduling method under the edge network of one or more embodiments of the present application includes the following steps:

[0051] Step S101: Acquire current information of all base stations and edge servers in the edge network.

[0052] In the process of implementing this application, the applicant took into account the characteristics that communication base stations can only perceive local environment and task information, and proposed a scheduling scheme for decision-making computing tasks from base station equipment to edge servers based on limited system information, so as to achieve global effective scheduling through the collaboration of multiple communication base stations, and realize the centralized and optimized utilization of computing resources, thereby improving the efficiency of multimedia processing.

[0053] According to the above content, the current information of the base station and the edge server in this application may include: the current average CPU speed, the current average GPU speed, the transmission rate between edge servers, the number of CPU cycles required for task calculation, the number of GPU cycles required for task calculation, and the amount of data calculated by the task, one or more of which are included in the current average CPU speed, the current average GPU speed, and the transmission rate between edge servers as base station information, and the number of CPU cycles required for task calculation, the number of GPU cycles required for task calculation, and the amount of data calculated by the task as edge server information.

[0054] Step S102: Obtaining a scheduling strategy for the task according to current information of the base station and the edge server.

[0055] In the technical solution of the present application, considering that the edge server is responsible for task calculation, the task scheduling principle of the present application can be set as: forwarding the transit task of the base station to the edge server connected to it in communication or forwarding it to other edge servers through the base station connected to it; sending other tasks of the base station to the edge server connected to it in communication.

[0056] The scheduling strategy means that at the beginning of each time slot, all base stations determine the service range according to their geographical locations, determine the communication base station to establish a wireless link connection according to the service range, and the task allocation plan for each base station.

[0057] In some embodiments, the above scheduling strategy follows the following scheduling principle: the number of transfers of the transfer task is less than 10, and the base station receiving the transfer task also has other tasks.

[0058] That is, in some embodiments, a communication base station set may be defined as (BS1, BS2...BS N ), communication base station BS n The task sequence in the t-slot queue is The edge server set is (ES1, ES2...ES M ). For each task in the task queue of the communication base station, the controller on the communication base station can decide to offload it to an adjacent edge server or transfer it to other adjacent communication base stations, and then offload it to an adjacent edge server through the communication base station. However, based on the above principles, if the task queue of the base station that receives the transfer task request has been cleared or the transfer task has reached the maximum number of loops, the base station can refuse to receive the transfer task, and the task can be offloaded to the adjacent edge server for execution. As the only node with computing capabilities, the edge server will perform parallel calculations on all tasks offloaded from adjacent base stations in each time slot, and finally return the task results. Task q can be represented by (d, c, g, p, l) n,twhere d represents the amount of data, c represents the number of CPU cycles required for calculation, g represents the number of GPU flops required for calculation, p represents the priority, and I represents the maximum number of times that can be transferred between base stations.

[0059] The present invention adopts a distributed algorithm in a multimedia edge network. Each communication base station deploys a controller as a decision-making agent. Different agents have different decision-making strategies. In the multi-agent iterative collaborative decision-making model, different agents compete and cooperate with each other. In each round of iteration, on the one hand, agents can transfer tasks between each other, and on the other hand, the unloading decision of the agent will also affect the perception of the server load by other agents. When the task queues of all base stations are cleared, the multi-agents reach a global consensus decision.

[0060] In some embodiments, before performing task scheduling or after receiving other transfer tasks, the tasks on the base station may be sorted according to priority.

[0061] In the process of implementing this application, the applicant found that due to the difficulty in accurately modeling edge energy consumption and the high complexity of joint optimization, existing methods rarely consider edge energy consumption performance. In edge networks, the energy consumption of communication base stations and edge computing nodes will bring huge costs, and excessive energy consumption may also cause problems such as node crashes and downtime.

[0062] Therefore, this application proposes an overhead model, which consists of three parts: transmission delay, computation delay and computation energy consumption.

[0063] The transmission delay model can be set as

[0064] Specifically, the applicant considers that the edge network architecture of this application uses a wireless link connection between users and base stations, and this part of the delay can be ignored. This application defines that each communication base station has multiple transceiver antennas, each edge server has a single transceiver antenna, and models the multi-channel to multi-channel data transmission process between base stations into a multiple-input multiple-output system. It can be understood that the signal-to-noise ratio of the MIMO system is as follows:

[0065]

[0066] Where E s represents the signal energy, N0 represents the noise power spectral density, and h represents the channel matrix. The element h in the matrix ij represents the channel gain from the transmitting antenna i of one base station to the receiving antenna j of another base station, h I and h J Represent the channel numbers of the two base stations respectively.

[0067] The communication process between the base station and the edge server is modeled as a single-input multiple-output system. The SIMO system signal-to-noise ratio is as follows:

[0068]

[0069] W represents the channel bandwidth, and the transmission rate between communication base stations can be calculated as:

[0070]

[0071] The transmission rate between the communication base station and the edge server can be calculated as:

[0072]

[0073] t time slot communication base station BS n The kth task in the task queue The communication delay can be calculated as:

[0074]

[0075] in Representation Task The number of transfers between base stations, represents the kth task It can be understood that when the task is directly offloaded to the adjacent edge computing node,

[0076] The model for calculating delay can be set as

[0077] Specifically, the edge server will create a virtual machine for each task to achieve computing isolation. Here, it can be assumed that the edge server will evenly allocate resources to each virtual machine. For multimedia computing tasks, the key factors affecting computing efficiency are the CPU execution speed, GPU execution speed, and IO conflicts. IO conflicts are caused by the degradation of computing performance due to multiple virtual machines sharing the same physical device. The calculation delay is as follows:

[0078]

[0079] in Indicates the number of CPU cycles required for calculation. Indicates the number of GPU cycles required for calculation, Indicates the number of CPU cycles that the virtual machine can execute per second. Indicates the number of gpu flops that the virtual machine can execute per second, ∈ indicates the IO conflict factor, Representation Task The edge server finally selected, Indicates the degree of parallelism, Indicates the number of CPU cycles that the edge server hardware can execute per second. Indicates the number of gpu flops that the edge server hardware can execute per second.

[0080] The overall power consumption of the multimedia edge network architecture is another optimization goal of this application. The computing power consumption of the edge server is related to multiple factors such as CPU, GPU, disk, network card, etc. Here, this application only considers the CPU as the dominant factor, and the power consumption model based on DVFS technology can be used for modeling as follows:

[0081]

[0082] in, Indicates the task The final calculation power consumption is Edge Server Energy factors related to CPU architecture.

[0083] In some embodiments, the above scheduling strategy can be generated by a scheduling strategy generation model.

[0084] In some embodiments, the above-mentioned scheduling strategy generation model may include an action neural network and a value evaluation neural network, and generate a scheduling strategy based on a proximal policy optimization (PPO) algorithm of deep reinforcement learning.

[0085] Specifically, the action neural network generates an action according to the current information, and the action indicates a scheduling strategy; the value evaluation neural network generates an evaluation value of the action.

[0086] In some embodiments, the above state can be expressed as in represents the current information of the edge server, q k =(c k ,g k ,d k ) represents the current information of the base station, Indicates the current average CPU speed. Indicates the current average GPU speed, represents the transmission rate between the edge servers, c k Indicates the number of CPU cycles required for the task calculation, g k Indicates the number of GPU cycles required for the task calculation, d k Indicates the amount of data calculated by the task.

[0087] The above actions can be regarded as scheduling decisions, that is, the vector formed by splicing the discrete decision results of all communication base stations is the action result of the entire system in a time slot, including the communication base station BS n Choose to offload the queue head task q1 to an adjacent edge server or transfer it to an adjacent communication base station.

[0088] The action neural network in the PPO algorithm is a neural network model used to generate agent actions. Its core function is to output the probability distribution of the corresponding action according to the input state. In terms of specific structure, the action neural network consists of an input layer, a hidden layer, and an output layer. The input layer accepts the state of the current environment, which can be continuous or discrete data, representing the specific information of the environment. The hidden layer is usually composed of several fully connected layers or convolutional layers. The main task of these layers is to extract the features of the input state, thereby providing a high-level abstract representation for decision-making. The structure of the output layer depends on the type of action space. In the discrete action space, the output layer uses the softmax activation function to generate the probability distribution of each possible action, so that the network can assign a probability to each action. In the continuous action space, the output layer generates the mean and standard deviation of the action, which are used to construct a Gaussian distribution from which specific actions are sampled. The action neural network updates the policy parameters by optimizing a clipped objective function to ensure that the amplitude of the policy update is not too large, thereby maintaining the stability and efficiency of the training. This objective function combines the policy gradient method with the clipped probability ratio, so that the change between the new and old policies is limited to avoid excessive policy updates.

[0089] The value evaluation neural network in the PPO algorithm is a neural network model used to evaluate the value of a given state s. Its main function is to approximate the state value function V(s) and provide a basis for strategy optimization. In terms of specific structure, the value evaluation neural network consists of an input layer, a hidden layer, and an output layer. The input layer accepts the state of the current environment, which can be continuous or discrete data, representing specific information about the environment. The hidden layer is usually composed of several fully connected layers or convolutional layers. The main task of these layers is to extract the features of the input state, thereby providing a high-level abstract representation for the estimation of the state value. The output layer generates the state value V(s), which is the expected long-term return of the current state. The value evaluation neural network updates its parameters by minimizing the mean square error (MSE) between the actual return and the estimated return, thereby improving the accuracy of its estimation. The PPO algorithm uses the value estimate of the value evaluation neural network to calculate the advantage function where δ t =r t +γV(s t+1 )-V(s t) to measure the pros and cons of the current strategy, thereby guiding the optimization of the action neural network. The value evaluation neural network and the action neural network work together, the former provides value evaluation, and the latter adjusts the strategy based on these evaluations, enabling the intelligent agent to continuously improve its decision-making process in a complex environment and ultimately learn the optimal strategy.

[0090] The parameters of the action neural network and the value evaluation neural network can be updated as follows. n At the beginning of the i-th iteration of a time slot t, the adjacent edge server information and task queue information are first standardized. Then they are spliced ​​and input into the actor's PolicyNet, and the decision result is obtained based on probability sampling. When all iterations of time slot t are completed, the communication base station BS n We can get a set of multiple quadruple (s i ,a i ,r i ,s i+1 ) is composed of a trajectory. According to this trajectory, the communication base station BS n The gradients of PolicyNet and ValueNet can be updated multiple times in a cycle.

[0091] We use TD error to update ValueNet, and the objective function is as follows:

[0092]

[0093] The objective function we use to update PolicyNet is as follows:

[0094]

[0095] in, Expressing expectation, r t (θ) represents the reward value of the tth iteration, T avg represents the average delay of all the tasks, E avg represents the average energy consumption of all the tasks, and a and b represent the preset weight parameters A t represents the action of the tth iteration, ∈ represents the IO conflict factor. Among them, V ω (s t ) represents the value function estimation based on the current parameter ω, td_target t Represents the target value calculated based on the sampling process.

[0096] Step S103: Schedule the above tasks according to the above scheduling strategy.

[0097] It can be understood that the method can be executed by any device, equipment, platform, or device cluster having computing and processing capabilities.

[0098] It should be noted that the method of one or more embodiments of the present application can be performed by a single device, such as a computer or server. The method of this embodiment can also be applied to a distributed scenario and completed by multiple devices cooperating with each other. In the case of such a distributed scenario, one of the multiple devices can only perform one or more steps in the method of one or more embodiments of the present application, and the multiple devices will interact with each other to complete the described method.

[0099] It should be noted that the above describes specific embodiments of the present application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0100] Based on the same inventive concept, corresponding to any of the above-mentioned embodiments and methods, the present application also provides a computing task intelligent scheduling device under an edge network. Figure 2 As shown, the device comprises:

[0101] An acquisition module 11 is configured to acquire current information of all base stations and edge servers in the edge network;

[0102] The calculation module 12 is configured to obtain a scheduling strategy for the task according to current information of the base station and the edge server;

[0103] The scheduling module 13 is configured to schedule the tasks according to the scheduling strategy.

[0104] For the convenience of description, the above devices are described in terms of functions and modules. Of course, when implementing one or more embodiments of the present application, the functions of each module can be implemented in the same or multiple software and / or hardware.

[0105] The apparatus of the above-mentioned embodiment is used to implement the corresponding method in the above-mentioned embodiment, and has the beneficial effects of the corresponding method embodiment, which will not be described in detail here.

[0106] Figure 4A more specific schematic diagram of the hardware structure of an electronic device provided in this embodiment is shown, and the device may include: a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. The processor 1010, the memory 1020, the input / output interface 1030, and the communication interface 1040 are connected to each other through the bus 1050 in the device.

[0107] The processor 1010 can be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application.

[0108] The memory 1020 may be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 1020 may store an operating system and other application programs. When the technical solution provided in the embodiment of the present application is implemented by software or firmware, the relevant program code is stored in the memory 1020 and is called and executed by the processor 1010.

[0109] The input / output interface 1030 is used to connect the input / output module to realize information input and output. The input / output module can be configured in the device as a component (not shown in the figure), or it can be externally connected to the device to provide corresponding functions. The input device may include a keyboard, a mouse, a touch screen, a microphone, various sensors, etc., and the output device may include a display, a speaker, a vibrator, an indicator light, etc.

[0110] The communication interface 1040 is used to connect a communication module (not shown) to realize communication interaction between the device and other devices. The communication module can realize communication through a wired mode (such as USB, network cable, etc.) or a wireless mode (such as mobile network, WIFI, Bluetooth, etc.).

[0111] The bus 1050 includes a path that transmits information between the various components of the device (eg, the processor 1010, the memory 1020, the input / output interface 1030, and the communication interface 1040).

[0112] It should be noted that, although the above device only shows the processor 1010, the memory 1020, the input / output interface 1030, the communication interface 1040 and the bus 1050, in the specific implementation process, the device may also include other components necessary for normal operation. In addition, it can be understood by those skilled in the art that the above device may also only include the components necessary for implementing the embodiment of the present application, and does not necessarily include all the components shown in the figure.

[0113] The electronic device of the above embodiment is used to implement the corresponding method in the above embodiment, and has the beneficial effects of the corresponding method embodiment, which will not be described in detail here.

[0114] The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, read-only compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device.

[0115] Those skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of the present disclosure (including the claims) is limited to these examples. Based on the concept of the present disclosure, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of different aspects of one or more embodiments of the present application as described above, which are not provided in detail for the sake of simplicity.

[0116] In addition, to simplify the description and discussion, and in order not to make one or more embodiments of the present application difficult to understand, the known power / ground connections to the integrated circuit (IC) chip and other components may or may not be shown in the provided drawings. In addition, the device can be shown in the form of a block diagram to avoid making one or more embodiments of the present application difficult to understand, and this also takes into account the fact that the details of the implementation of these block diagram devices are highly dependent on the platform on which one or more embodiments of the present application will be implemented (that is, these details should be fully within the scope of understanding of those skilled in the art). In the case of elaborating specific details (e.g., circuits) to describe exemplary embodiments of the present disclosure, it is obvious to those skilled in the art that one or more embodiments of the present application can be implemented without these specific details or when these specific details are changed. Therefore, these descriptions should be considered illustrative rather than restrictive.

[0117] Although the present disclosure has been described in conjunction with specific embodiments of the present disclosure, many replacements, modifications and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may use the embodiments discussed.

[0118] One or more embodiments of the present application are intended to cover all such substitutions, modifications and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of one or more embodiments of the present application should be included in the scope of protection of this disclosure.

Claims

1. A method for intelligent scheduling of computing tasks in edge networks, characterized in that: include: Obtaining current information of all base stations and edge servers in the edge network; Obtaining a scheduling strategy for the task according to current information of the base station and the edge server; The tasks are scheduled according to the scheduling strategy.

2. The method according to claim 1, characterized in that The scheduling strategy is the optimal solution of the following objective function: in, represents the kth task The communication delay, Represents the kth task The computational delay of represents the kth task The computing energy consumption, Represents the kth task The number of transfers between the base stations, represents the kth task The amount of data, Represents the kth task The transmission rate of the i-th transfer between the base stations, R BS-ES Represents the kth task The transmission rate between the base station and the edge server, Indicates the number of floating point operations that the edge server hardware can perform per second. Indicates the degree of parallelism, Represents edge server Energy factors related to CPU architecture, Indicates the number of CPU cycles required for calculation, Indicates the number of CPU cycles that the virtual machine can execute per second.

3. The method according to claim 2, characterized in that The scheduling strategy is generated by a scheduling strategy generation model, and the scheduling strategy model includes an action neural network and a value evaluation neural network; The scheduling strategy generation model performs the following operations until the preset conditions are met: The action neural network generates an action based on the current information, and the action indicates a scheduling strategy; The value evaluation neural network generates an evaluation value of the action.

4. The method according to claim 3, characterized in that The generating an action according to the current information comprises: Generate the current state of the edge network based on the current information of the base station and the edge server represents the current information of the edge server, q k =(c k , g k , d k ) represents the current information of the base station, Indicates the current average CPU speed. Indicates the current average GPU speed, represents the transmission rate between the edge servers, c k Indicates the number of CPU cycles required for the task calculation, g k Indicates the number of GPU cycles required for the task calculation, d k Indicates the amount of data calculated by the task; The action is generated according to the current state.

5. The method according to claim 3, characterized in that: After generating the action, the action neural network further performs the following operations: Update the parameters of the action neural network according to the following formula; in, Expressing expectation, r t (θ) represents the reward value of the tth iteration, T avg represents the average delay of all the tasks, E avg represents the average energy consumption of all the tasks, and a and b represent the preset weight parameters A t represents the action of the tth iteration, ∈ represents the IO conflict factor.

6. The method according to claim 3, characterized in that: After generating the evaluation value of the action, the value evaluation neural network further performs the following operations: Update the parameters of the value evaluation neural network according to the following formula; Among them, V ω (s t ) represents the value function estimation based on the current parameter ω, td_target t Represents the target value calculated based on the sampling process.

7. The method according to claim 1, characterized in that The step of scheduling the task according to the scheduling strategy includes: Any of the base stations forwards the transfer task to the edge server according to the scheduling strategy, or forwards the transfer task to the edge server through other base stations; the number of forwarding times of the transfer task is less than a preset number, and there is no transfer task for the base station scheduled according to the scheduling strategy.

8. A computing task intelligent scheduling device under edge network, characterized in that: include: An acquisition module, configured to acquire current information of all base stations and edge servers in the edge network; A calculation module, configured to obtain a scheduling strategy for the task according to current information of the base station and the edge server; The scheduling module is configured to schedule the tasks according to the scheduling strategy.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the method according to any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium, characterized in that: The non-transitory computer-readable storage medium stores computer instructions, and the computer instructions are used to cause the computer to execute any one of claims 1 to 7.

Images