Edge computing power scheduling method, device and equipment based on time delay perception optimization

Abstract

The invention belongs to the technical field of edge computing power scheduling, and relates to an edge computing power scheduling method, device and equipment based on time delay perception optimization. The method comprises the following steps: acquiring an edge computing environment, and constructing a multi-edge service provider computing power sharing framework; according to a multi-edge service provider computing power sharing framework, the service time delay is minimized as a target, and the computing task load undertaken by edge service providers, the scheduling decision of the edge service providers, the scheduling matrix of the edge service providers and the computing task load processed by the edge service providers are constrained. Establishing an optimization problem of edge computing power task scheduling; solving an optimization problem of edge computing power task scheduling to obtain a scheduling matrix of multiple edge service providers; and according to the scheduling matrix of the multi-edge service providers, performing edge computing power scheduling of the multi-edge service providers. By adopting the method and the device, efficient computing power resource sharing and efficient edge computing power task scheduling can be realized.

Description

Technical Field

[0001] The present application relates to the field of edge computing power scheduling technology, and in particular to edge computing power scheduling methods, devices and equipment based on latency-aware optimization. Background Art

[0002] Edge computing is a cutting-edge technology that has emerged in recent years and has become a hot topic of great concern in the industrial and academic circles at home and abroad. Its basic idea is to provide cloud computing and IT service capabilities to application developers and content providers at the edge of the network close to the terminal devices, thereby creating a big data processing service environment with high performance, low latency, and high bandwidth. In fact, edge computing and 5G technologies complement each other, jointly providing more localized services to a variety of terminal devices, ensuring lower network latency, and thus improving the experience of end users. To this end, telecommunications companies usually deploy edge computing servers near base stations to provide services to users nearby, becoming the de facto edge service provider (Edge Service Provider, ESP).

[0003] However, with the widespread deployment of edge computing power, how to achieve coordination and sharing of these decentralized computing power infrastructures among ESPs is the key to unleashing the potential of edge computing performance, reducing ESP costs, and maximizing social welfare. Computing power sharing means interconnecting decentralized computing resources, and its main means of implementation is computing power scheduling.

[0004] In the prior art, research on computing power scheduling at home and abroad focuses on four aspects: the characteristics of computing resources, the characteristics of computing tasks, the types of task scheduling, and the optimization angles of task scheduling. The computing power scheduling method based on edge computing mainly includes: edge layer task allocation and resource scheduling mechanism to achieve optimal edge server allocation of tasks; a two-layer task offloading collaboration framework, the inner layer faces the inside of ESP, and adopts a load balancing scheme based on the alternating multiplier method (ADMM) to achieve collaborative scheduling of internal nodes; the outer layer faces the outside of ESP, and uses a non-cooperative game model based on incomplete information for resource pricing and external task offloading.

[0005] However, although the current methods have studied computing power scheduling from different angles, they have ignored the impact of local computing and remote computing on users in terms of service latency. In addition, a single ESP usually has limited computing power, and its service scope is far from covering a wide range of end users. Its service resources are also insufficient to meet massive requests, which means that performance degradation, response delays, and even service interruptions are very likely to occur during peak service hours. Summary of the invention

[0006] Based on this, it is necessary to provide an edge computing power scheduling method, device and equipment based on latency-aware optimization to address the above-mentioned technical problems, which can achieve efficient computing power resource sharing and efficient edge computing power task scheduling.

[0007] The edge computing power scheduling method based on latency-aware optimization includes: Obtain edge computing environment and build a computing power sharing framework for multiple edge service providers; Based on the computing power sharing framework of multiple edge service providers, with the goal of minimizing service latency, the optimization problem of edge computing task scheduling is established with the amount of computing tasks undertaken by edge service providers, the scheduling decisions of edge service providers, the scheduling matrix of edge service providers, and the amount of computing tasks processed by edge service providers as constraints; Solve the optimization problem of edge computing task scheduling and obtain the scheduling matrix of multiple edge service providers; The edge computing power of multiple edge service providers is scheduled according to the scheduling matrix of multiple edge service providers.

[0008] In one embodiment, the optimization problem of edge computing task scheduling includes:

[0009] in,

[0010] In the formula, For ESP In the period The scheduling strategy, For ESP In the period The scheduling matrix, For an ESP, It is a collection of edge computing servers belonging to ESP or ESP. For ESP In the period The service delay, For ESP The amount of local computing resources, For ESP In the period The amount of computing tasks undertaken within For ESP The unit calculation delay is For ESP The unit upload / download transmission delay, For the period ESP Dispatch to ESP The computational tasks For ESP To ESP The transmission delay between For ESP The unit upload / download transmission delay, For ESP The unit calculation delay is For ESP In the period Scheduling strategy.

[0011] In one embodiment, the amount of computing tasks undertaken by the edge service provider, the scheduling decision of the edge service provider, the scheduling matrix of the edge service provider, and the amount of computing tasks processed by the edge service provider are used as constraints, including:

[0012] ; ; ; ; ;

[0013] In the formula, For ESP In the period The total amount of computing tasks undertaken The proportion of For another ESP, For ESP In the period The amount of computing tasks that need to be scheduled, For ESP In the period The amount of computational tasks processed internally.

[0014] In one embodiment, the optimization problem of edge computing task scheduling is solved to obtain a scheduling matrix of multiple edge service providers, including: Initialize the scheduling matrix, calculate the objective function value of the scheduling matrix, and obtain the objective function change value; Construct a Lagrangian function based on the objective function and constraints of the optimization problem of edge computing task scheduling; According to the Lagrangian function and the change value of the objective function, the minimum quadratic programming problem is constructed; Adjust the Lagrange multiplier to meet the KKT condition and obtain the updated value of the Lagrange multiplier; According to the updated value of Lagrange multiplier, the least quadratic programming problem is solved to obtain the optimal solution of the change value of the objective function; Perform line search based on the optimal solution of the objective function change value to obtain the calculation step length; According to the calculation step size, the scheduling matrix is ​​updated along the search direction, and it is iterated to recalculate the objective function value of the scheduling matrix until the preset conditions are met, and the current scheduling matrix is ​​output as the scheduling matrix of the multi-edge service provider.

[0015] In one embodiment, a Lagrangian function is constructed according to the objective function and constraints of the optimization problem of edge computing task scheduling, including:

[0016] In the formula, is the Lagrangian function, For the The task scheduling matrix for the iteration is For the The Lagrange multipliers corresponding to the constraints are: For the constraints.

[0017] In one embodiment, a minimum quadratic programming problem is constructed according to the Lagrangian function and the objective function change value, including:

[0018] In the formula, is the change value of the objective function, for The Hessian matrix of for The gradient of For Constraint The Jacobian matrix of .

[0019] In one embodiment, a multi-edge service provider computing power sharing framework includes: a user layer, an edge layer, and a cloud layer connected in sequence, wherein the user layer is further connected to the cloud layer; The user layer includes multiple terminal devices that dynamically generate computing tasks and hope to obtain services; The edge layer includes multiple edge service providers that provide responsive computing power for computing tasks; The cloud layer includes multiple computing centers, providing the user layer with greater responsive computing power than the edge layer.

[0020] In one embodiment, the edge layer further comprises: a service analyzer, a central optimizer and a policy executor connected in sequence; the service analyzer is also connected to the user layer, the service analyzer is also connected to each edge service provider, and the policy executor is also connected to each edge service provider; The service analyzer continuously analyzes the computing tasks at the user level, describes the service characteristics and outputs them to the central optimizer; The central optimizer uses the service features output by the service analyzer as input, executes the optimization algorithm for resource sharing and service revenue, generates a scheduling matrix for computing tasks and service pricing for each edge service provider, and thus forms a computing power sharing strategy and outputs it to the strategy executor; The policy executor redirects each computing task to the best edge service provider based on the computing power sharing strategy output by the central optimizer.

[0021] The edge computing power scheduling device based on latency perception optimization includes: The acquisition module is used to obtain the edge computing environment and build a computing power sharing framework for multiple edge service providers; A modeling module is used to establish an optimization problem for edge computing task scheduling based on a multi-edge service provider computing power sharing framework, with the goal of minimizing service latency, and with the amount of computing tasks undertaken by edge service providers, the scheduling decisions of edge service providers, the scheduling matrix of edge service providers, and the amount of computing tasks processed by edge service providers as constraints; The solution module is used to solve the optimization problem of edge computing task scheduling and obtain the scheduling matrix of multiple edge service providers; The scheduling module is used to schedule the edge computing power of multiple edge service providers according to the scheduling matrix of multiple edge service providers.

[0022] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented: Obtain edge computing environment and build a computing power sharing framework for multiple edge service providers; Based on the computing power sharing framework of multiple edge service providers, with the goal of minimizing service latency, the optimization problem of edge computing task scheduling is established with the amount of computing tasks undertaken by edge service providers, the scheduling decisions of edge service providers, the scheduling matrix of edge service providers, and the amount of computing tasks processed by edge service providers as constraints; Solve the optimization problem of edge computing task scheduling and obtain the scheduling matrix of multiple edge service providers; The edge computing power of multiple edge service providers is scheduled according to the scheduling matrix of multiple edge service providers.

[0023] The above-mentioned edge computing power scheduling method, device and equipment based on delay-aware optimization, aimed at the challenge of limited resources of a single ESP in CoEdge, established and solved the optimization problem (DSO) of edge computing power task scheduling with the goal of minimizing service delay, so as to avoid performance degradation, response delay and even service interruption that are very likely to occur during peak service hours, realize efficient computing power resource sharing and efficient edge computing power task scheduling, and provide users with responsive services. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is an application scenario diagram of an edge computing power scheduling method based on latency-aware optimization in one embodiment; Figure 2 A schematic diagram of a process of an edge computing power scheduling method based on latency-aware optimization in one embodiment; Figure 3 is a schematic diagram of an edge computing environment in one embodiment; Figure 4 A schematic diagram of a computing power sharing framework for multiple edge service providers in one embodiment; Figure 5 A schematic diagram of a framework of an edge layer in a computing power sharing framework of multiple edge service providers in one embodiment; Figure 6 The minimum service delay of CoEdge and UnCoEdge in a specific embodiment is shown with ESP as the research object, where (a) is 35-CoEdge and 35-UnCoEdge, and (b) is 55-CoEdge and 55-UnCoEdge; Figure 7 The minimum service delay of CoEdge and UnCoEdge in a specific embodiment is studied with a telecom operator as the research object, where (a) is 35-CoEdge and 35-UnCoEdge, and (b) is 55-CoEdge and 55-UnCoEdge; Figure 8 It is a structural block diagram of an edge computing power scheduling device based on latency-aware optimization in one embodiment; Fig. 9 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical solutions and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present application.

[0026] It should be noted that all directional indications in the embodiments of the present application (such as up, down, left, right, front, back, etc.) are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0027] In addition, the descriptions of "first", "second", etc. in this application are only for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of this application, "multiple groups" means at least two groups, such as two groups, three groups, etc., unless otherwise clearly and specifically defined.

[0028] In this application, unless otherwise clearly specified and limited, the terms "connection", "fixation", etc. should be understood in a broad sense. For example, "fixation" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection, an electrical connection, a physical connection, or a wireless communication connection; it can be a direct connection, or an indirect connection through an intermediate medium, or it can be the internal connection of two elements or the interaction relationship between two elements, unless otherwise clearly defined. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0029] In addition, the technical solutions between the various embodiments of the present application can be combined with each other, but it must be based on the fact that ordinary technicians in the field can implement it. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such combination of technical solutions does not exist and is not within the scope of protection required by this application.

[0030] The edge computing power scheduling method based on latency-aware optimization provided in this application can be applied to Figure 1 In the application scenario diagram shown, the terminal 102 communicates with the server 104 through the network, and the terminal 102 may include but is not limited to various personal computers, laptops, smart phones, tablet computers and portable wearable devices, and the server 104 may be various portal websites, servers corresponding to the backend of the work system, etc.

[0031] This application provides an edge computing power scheduling method based on latency-aware optimization, such as Figure 2 The flowchart shown in FIG. 1 shows a method for applying the method to Figure 1 The following is an example of a terminal in the example, including: Step 202, obtain the edge computing environment and build a computing power sharing framework for multiple edge service providers.

[0032] Specifically: Obtain an edge computing environment and build a multi-edge service provider computing power sharing framework including: a user layer, an edge layer, and a cloud layer; the user layer, the edge layer, and the cloud layer are connected in sequence, and the user layer is also connected to the cloud layer.

[0033] The user layer includes multiple terminal devices, which dynamically generate computing tasks and hope to obtain services. Specifically: The user layer includes a large number of terminal devices such as computers, smart phones, and smart IoT terminals. They are the source and end point of data transmission, and they will dynamically generate computing tasks and hope to obtain high-quality services.

[0034] The edge layer includes multiple edge service providers that provide responsive computing power for computing tasks. Specifically: The edge layer includes multiple ESPs (or their deployed edge computing servers), which are responsible for providing computing, storage and other resources for computing tasks, and providing the technology and platform required for computing tasks to run, thereby responding to computing power service needs in real time.

[0035] The edge layer also includes: a service analyzer, a central optimizer and a policy executor; the service analyzer, the central optimizer and the policy executor are connected in sequence, the service analyzer is also connected to the user layer, the service analyzer is also connected to each edge service provider, and the policy executor is also connected to each edge service provider.

[0036] in: The service analyzer continuously analyzes the computing tasks at the user layer, describes the service characteristics and outputs them to the central optimizer. Specifically: The service analyzer is essentially a service characteristic extraction module, which is responsible for continuously analyzing the requests of terminal users. It can obtain and analyze the request information of all terminal users to different ESPs within a time and space range in real time, including geographic location, data traffic, data transmission rate, and service delay tolerance, so as to accurately describe the service characteristics and output them.

[0037] The central optimizer takes the service features output by the service analyzer as input, executes the optimization algorithm for resource sharing and the optimization algorithm for service benefits, generates a scheduling matrix for computing tasks and service pricing for each edge service provider, thereby forming a computing power sharing strategy and outputting it to the strategy executor. Specifically: The central optimizer is the central module of the scheduler, responsible for executing the optimization algorithm for resource sharing and service benefits. It takes the service features output by the service analyzer as basic input, generates a scheduling matrix for computing tasks and a service pricing plan for each ESP by executing the optimization algorithm, thereby forming a computing power scheduling strategy and outputting it.

[0038] The policy executor redirects each computing task to the best edge service provider according to the computing power sharing strategy output by the central optimizer. Specifically: The policy executor is the specific implementation module of the computing power scheduling strategy, responsible for redirecting the service to the best ESP. This redirection can be assisted by the existing Domain Name System (DNS) resolution technology in the edge network. Therefore, after receiving the service request, the local DNS server first resolves to the DNS server of the corresponding ESP, and then the ESP uses the DNS server of the scheduler to modify its CNAME record. Finally, according to the scheduling strategy and CNAME record optimized by the central optimizer for the service request, the policy executor redirects the service request to its best ESP.

[0039] The cloud layer includes multiple computing centers, which provide the user layer with greater response computing power than the edge layer. Specifically: the cloud layer includes multiple remote large computing centers such as intelligent computing centers and data centers, which can provide the user layer with various network services similar to those of the edge layer, but compared with the edge layer, its resource inventory is richer and the server scale is larger.

[0040] In this step, if Figure 3 The general edge computing environment shown in the figure abstracts the edge computing environment into an undirected graph .in, Representation diagram A point set whose elements are ESPs or edge computing servers affiliated with ESPs; Representation diagram The edge set, whose elements are the data transmission links between ESPs or edge computing servers. Specifically, express A specific ESP or edge computing server in Figure 1. There are multiple ESPs in the network, and each ESP has multiple edge computing servers to respond to user service requests. Each edge server can not only directly process computing tasks from users, but also process computing tasks from other edge servers. express Different edge computing servers of the same ESP are connected through a local area network (LAN), and edge computing servers of different ESPs are connected through the Internet.

[0041] In fact, if the user request is processed locally, the request will be distributed to the edge server of the ESP via the LAN for processing, which has a lower latency due to its proximity to the user; if the user is processed remotely, the request will be distributed to the edge server of another ESP via the Internet with the help of the edge server of the ESP, which has a higher latency due to the need for cross-ESP wide area transmission. Therefore, the latency of the former can be ignored in the process of processing the request. Then, in Figure In , all edge computing servers belonging to the same ESP are modeled as a node.

[0042] In addition, to describe the Accepted service requests, introduction time period , and use It represents the total computing tasks from users undertaken by the multi-edge service provider computing power sharing framework (CoEdge) during this period, and it is assumed that the overall computing resources of CoEdge can meet all computing requests from users.

[0043] For this edge computing environment, a multi-edge service provider computing power sharing framework is proposed. Figure 4 As shown in the figure, the framework diagram of the edge layer in the multi-edge service provider computing power sharing framework is as follows Figure 5 shown.

[0044] Specifically, the data streams and task streams generated by the user layer are uploaded to the edge layer. Since ESP and its edge computing servers are interconnected and can reasonably decide on the local / remote processing of computing tasks at the edge layer, ESP can collaboratively process user service requests, thereby ensuring that the edge layer processes service requests in a timely and efficient manner. When the computing resources of the edge layer are fully utilized and most tasks are preprocessed, if there are still unfinished computing tasks, these tasks will be further submitted to the cloud layer. The cloud layer has a cloud data center with more powerful computing power and massive data storage space, which will take over the remaining tasks and process them.

[0045] Therefore, with the support of the scheduler (the scheduler includes: service analyzer, central optimizer and policy executor), CoEdge adopts a hierarchical mechanism to process computing tasks, which not only optimizes the allocation of computing resources, but also reduces network latency and improves service response speed, especially by reasonably allocating computing tasks at the edge.

[0046] Step 204, based on the multi-edge service provider computing power sharing framework, with the goal of minimizing service delay, and with the amount of computing tasks undertaken by the edge service provider, the scheduling decision of the edge service provider, the scheduling matrix of the edge service provider, and the amount of computing tasks processed by the edge service provider as constraints, establish an optimization problem for edge computing power task scheduling.

[0047] Specifically:

[0048] ; ; ; ; ; ;

[0049] in,

[0050] In the formula, For ESP In the period The scheduling strategy (scheduling strategy is a 0-1 variable indicating whether ESP is scheduled in the time period), For ESP In the period The scheduling matrix, For an ESP, It is a collection of edge computing servers belonging to ESP or ESP. For ESP In the period The service delay, For ESP In the period The total amount of computing tasks undertaken The proportion of For ESP In the period The amount of computing tasks undertaken within For ESP The amount of local computing resources, For another ESP, For ESP In the period The scheduling strategy, For the period ESP Dispatch to ESP The computational tasks For ESP In the period The amount of computing tasks that need to be scheduled, For ESP In the period The amount of computational tasks processed internally, For ESP The unit calculation delay is For ESP The unit calculation delay is For ESP The unit upload / download transmission delay, For ESP To ESP The transmission delay between For ESP The unit upload / download transmission delay, for a continuous period of time.

[0051] More specifically: First, ESP undertakes the computing tasks of users. Considering the regional differences of service requests, that is, due to the different service request volumes in different regions, the computing tasks undertaken by ESPs deployed in different regions are different. Indicates time period Internal ESP Total computing tasks undertaken Then, for the ratio, there are constraints that the ratio value is non-negative, cannot exceed 1, and the sum of the ratios is 1:

[0052] ;

[0053] Therefore, ESP In the period The amount of computing tasks undertaken for:

[0054] Then, ESP decides whether to schedule computing tasks. According to the connection mode of data transmission links in CoEdge, compared with remote processing, that is, scheduling tasks to other ESPs for processing, local processing of computing tasks does not rely on external networks and can respond to service requests faster. Therefore, in order to reduce service delays, ESP prioritizes local processing. If local resources are insufficient, it considers processing some computing tasks remotely.

[0055] Based on the above considerations, each ESP has only two options: scheduling and not scheduling. A 0-1 variable is introduced. Then, if , then ESP In the period The computing tasks need to be scheduled; the reverse is not true. ESP The amount of local computing resources. Therefore, the mathematical expression of this 0-1 variable is as follows:

[0056] Next, we discuss the different The computing task processing status of ESP.

[0057] like ESP No need to schedule computing tasks, then the time period The computing tasks of users can be completely undertaken locally. At this time, if , then ESP The computing resources available are just enough to handle the computing tasks from users, and there is no need to share resources with other ESPs, and there are no surplus resources to share with other ESPs; if , then ESP Complete the user's computing tasks locally After that, there are still remaining computing resources , which can be used to assist ESPs with insufficient resources in processing computing tasks and reduce local resource waste.

[0058] like ESP If you need to schedule computing tasks, then they should be scheduled in the time period The ESP cannot completely undertake the computing tasks from users locally, and needs to combine remote processing, that is, dispatching tasks that exceed its computing capacity to ESP with excess resources for processing. ESP In the period The amount of computing tasks that need to be scheduled within the ESP needs to share the resources of other ESPs. From the perspective of reducing service delay, ESP should process as many computing tasks as possible locally and schedule as few computing tasks as possible to other ESPs, that is, make full use of local resources. Then:

[0059] At this time, use ESP In the period The scheduling matrix in Indicates time period From ESP Dispatch to ESP (ESP There is no need to schedule computing tasks, i.e. ) computing tasks. To avoid idle resources, it is assumed that computing tasks can be quickly scheduled to the appropriate ESP through the least number of network hops (usually 1 hop) so that computing resources can be quickly allocated to the tasks. In addition, the non-negative amount of computing tasks is the basis for ensuring the effectiveness of resource allocation and the rationality of scheduling decisions. Then there is .therefore, It can be expressed as:

[0060] At the same time, to ensure that all tasks can be completed on the allocated resources, it is assumed that the computational tasks are conserved during the scheduling process, that is, during the scheduling process, the total computational tasks remain unchanged before and after the scheduling. Therefore, the scheduling matrix The sum of the elements of is equal to the initial computing task amount before scheduling:

[0061] Then, the resource utilization of ESP in sharing is analyzed. In order to ensure the normal operation of computing tasks and the continuous service of ESP to users, the computing tasks processed by ESP should be within the scope of its computing resources. For ESPs that need to schedule computing tasks, they make full use of local computing resources and process some computing tasks remotely, so the computing tasks they process are equal to the amount of local computing resources; for ESPs that do not need to schedule computing tasks, they have local resources that can be used for sharing, so the computing tasks they process may come from other ESPs in addition to users. ESP In the period The amount of computational tasks to be processed internally is:

[0062] Finally, we discuss the service latency in resource sharing. Service latency, which is the time interval from user request to response, is a key indicator for measuring service quality and user experience in high-real-time CoEdge. This application focuses on computing latency and content transmission latency, and does not involve the latency of the process from request to server reception and server response back to the user.

[0063] Computational latency refers to the time it takes for ESP to process a computing task. ESP The unit computing delay is the time consumed to process a unit computing task. When the ESP that needs to be scheduled collaborates with local and remote computing resources to complete computing tasks, in order to respond to user requests faster, it prefers to choose low latency, that is, it hopes that the computing delay scheduled to other ESPs is as small as possible from the computing delay of its own processing. This article calls it the latency-aware constraint, which is described as:

[0064] Content transmission delay refers to the time it takes to transmit a computing task in the network. Content transmission delay mainly considers the transmission delay of computing tasks between ESPs and the upload / download transmission delay of computing tasks in ESPs.

[0065] Transmission delay of computing tasks between ESPs: In CoEdge, the transmission delay between ESPs is quantified as the ratio of the computing task scheduling amount to the link transmission rate. Indicates that from ESP Dispatch to ESP The amount of computational tasks, among which, and . Unified use represents any link transmission rate. Then, ESP To ESP The transmission delay between It can be expressed as:

[0066] The upload / download transmission delay of computing tasks in ESP: The upload / download transmission delay corresponds to the process of uploading computing tasks from the ESP receiving end to the ESP computing module and downloading from the ESP computing module to the ESP sending end. Assume that for the same ESP, the upload and download only differ in the transmission direction of computing tasks, while the data transmission method, transmission protocol, etc. are the same, and the amount of data in the transmission process will not change. ESP The unit upload / download transmission delay.

[0067] For ESPs that need to schedule computing tasks, since they share resources from other ESPs, their service latency needs to consider the latency of local processing of computing tasks and the latency caused by the scheduling process. The latter includes the transmission latency between ESPs and the computing latency of other ESPs. Scheduling is a parallel process, and its latency depends on the longest subtask execution time. For ESPs that do not need to schedule computing tasks, their service latency only needs to consider the latency of local processing of computing tasks. Then ESP In the period Service delay It can be expressed as:

[0068] ;

[0069] For ESP, it is hoped that computing tasks can be completed faster by sharing resources; for CoEdge, it is hoped that the total service delay is minimized. Therefore, the DSO problem can be obtained, that is, the optimization problem of realizing efficient edge computing task scheduling with the goal of reducing service delay:

[0070] ;

[0071] In this step, the optimization problem of edge computing task scheduling is established, and the specific solution is given below.

[0072] Step 206, solve the optimization problem of edge computing task scheduling and obtain the scheduling matrix of multiple edge service providers.

[0073] Specifically: Initialize the scheduling matrix, calculate the objective function value of the scheduling matrix, and obtain the objective function change value; Construct a Lagrangian function based on the objective function and constraints of the optimization problem of edge computing task scheduling; According to the Lagrangian function and the change value of the objective function, the minimum quadratic programming problem is constructed; Adjust the Lagrange multiplier to meet the KKT condition and obtain the updated value of the Lagrange multiplier; According to the updated value of Lagrange multiplier, the least quadratic programming problem is solved to obtain the optimal solution of the change value of the objective function; Perform line search based on the optimal solution of the objective function change value to obtain the calculation step length; According to the calculation step size, the scheduling matrix is ​​updated along the search direction, and it is iterated to recalculate the objective function value of the scheduling matrix until the preset conditions are met, and the current scheduling matrix is ​​output as the scheduling matrix of the multi-edge service provider.

[0074] Among them, construct the Lagrangian function:

[0075] In the formula, is the Lagrangian function, For the The task scheduling matrix for the iteration is For the The Lagrange multipliers corresponding to the constraints are: For the Constraints; Construct a minimum quadratic programming problem:

[0076] In the formula, is the change value of the objective function, for The Hessian matrix of for The gradient of For Constraint The Jacobian matrix of .

[0077] More specifically: To solve the DSO problem, we first analyze its complexity. It is known that scheduling jobs on parallel machines (Scheduling Jobs on Identical Parallel Machines, SIPM) is NP-hard, and its specific definition is: Assign jobs to Each job runs in parallel on multiple machines. Must be on a machine with The processing time is uninterrupted, and each machine can only process one job at a time. The purpose is to minimize the processing time of all jobs. In DSO, there are Computational tasks and ESP. The size and completion time of each computing task are recorded as and , each computing task is divided into several jobs , then the corresponding completion time is Then, we perform a polynomial reduction of the SIPM problem to the DSO problem. Assume that the SIPM instance is to be assigned Homework The machines process the time to run on one machine. This means that each job can be scheduled to any machine. In this way, SIPM minimizes the processing time of all jobs, which is equivalent to DSO minimizing the completion time of all computing tasks. Moreover, this reduction process can be completed in polynomial time. Therefore, SIPM can be obtained DSO, DSO problem is also NP-hard. Moreover, DSO problem is NP-hard, so there is no polynomially solvable algorithm for DSO problem.

[0078] When solving the DSO problem, the problem involves a large number of constraints (number of ), designed the idea of ​​designing a solution algorithm starting from processing constraints, and handled a large number of constraints by constructing Lagrangian functions. Its flexibility and convergence help to realize computing power scheduling in complex edge computing environments, thereby achieving numerically stable convergence. The core idea of ​​the solution is to construct a Lagrangian function (line 8) and a least squares problem (line 9) in each iteration to find the optimal and judgement Whether the convergence threshold is met to update the current solution (line 13); in order to further ensure the optimality of the Lagrangian operator in the Lagrangian function, the KKT (Karush Kuhn Tucker) condition judgment (line 10) is set to improve the robustness of the required computing task scheduling matrix. The detailed process is shown in Table 1.

[0079] Table 1: Pseudocode for solving the DSO problem

[0080] In this step, Algorithm 1 enables the DSO problem to be solved in time complexity The solution is completed under the condition of . This is because: the time complexity of Initialize in Algorithm 1 is , its time complexity is mainly based on Iterate. ① Number of iterations: According to the iteration counter, set The algorithm ends after the update, and the time complexity it brings is ;②Construct the minimum quadratic programming problem: Algorithm 1 needs to transform each constraint when constructing the minimum quadratic programming problem. constraints, the time complexity is ; ③Problem scale: The number of variables involved in solving the DSO problem is , then the time complexity it brings is Therefore, the time complexity of Algorithm 1 is .

[0081] Step 208, scheduling edge computing power of multiple edge service providers according to the scheduling matrix of multiple edge service providers.

[0082] In this step, how to schedule edge computing power according to the scheduling matrix belongs to the existing technology and will not be repeated here.

[0083] The above-mentioned edge computing power scheduling method based on latency-aware optimization has the following beneficial effects: 1) A multi-edge service provider computing power sharing framework (CoEdge) is designed to form a cooperative relationship between multiple ESPs, thereby forming a cross-ESP service user model, so that CoEdge supports the scheduling and adaptation of computing tasks, realizes the interconnection of multiple ESPs, avoids the imbalance of resource utilization due to excess or shortage of resources, and thus improves the user service experience. For ESP, by sharing computing power resources, on the one hand, it can flexibly schedule and expand resources according to actual needs, and on the other hand, it can collaboratively serve widely distributed terminals and process massive requests without having to deploy servers on a large scale to achieve comprehensive service coverage; for end users, thanks to the cross-ESP service model, on the one hand, they can obtain service support that is difficult to achieve in the traditional independent service model, expanding the scope of accessible resources, and on the other hand, they can enjoy responsive and high-quality services, and can focus more on their core business. The CoEdge proposed in this application can provide a reference model for building a new generation of computing power sharing, and is expected to become a win-win solution for ESPs and end users.

[0084] 2) In response to the challenge of limited resources of a single ESP in CoEdge, an edge computing task scheduling optimization problem (DSO) was established and solved with the goal of minimizing service latency to avoid performance degradation, response delays, and even service interruptions that are very likely to occur during peak service hours, achieve efficient computing resource sharing and efficient edge computing task scheduling, and provide users with responsive services.

[0085] It should be understood that although Figure 2 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 2 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0086] In a specific embodiment, this application is called CoEdge optimal service. Taking service delay as an indicator, 35 ESPs and 55 ESPs are selected to form CoEdge, which are respectively recorded as 35-CoEdge and 55-CoEdge, and the performance difference between this application (CoEdge) and the prior art (UnCoEdge) is compared.

[0087] Firstly, ESP is taken as the research object.

[0088] In terms of service latency, Figure 6 The minimum service delay of each ESP group in CoEdge and UnCoEdge is shown. It can be found that compared with UnCoEdge, the minimum service delay of each ESP group is reduced in CoEdge, which shows that in terms of service delay, the ESP cooperative service method is more advantageous than the independent operation method, which proves the sharing of CoEdge. In addition, in terms of the degree of reduction in service delay, 35-CoEdge is reduced by about 1.61~2.97 times; 55-CoEdge is reduced by about 2.87~4.15 times. And as Figure 6 As shown in (a), the maximum minimum service delay in 35-CoEdge is 0.650 milliseconds. Figure 6 As shown in (b), the maximum minimum service delay in 55-CoEdge is 0.423 milliseconds, which means that overall, 55-CoEdge has a lower service delay than 35-CoEdge, and can therefore respond to user requests more quickly.

[0089] Then telecom operators are taken as the research objects.

[0090] In terms of service latency, Figure 7 The distribution of the minimum service delay of each telecom operator in CoEdge and UnCoEdge is shown. It can be found that compared with UnCoEdge, the minimum service delay of each telecom operator is reduced in CoEdge, and the minimum service delay of the ESP belonging to each telecom operator is not much different from each other. This shows that in terms of service delay, CoEdge is beneficial to all telecom operators and can better balance the service delay of different ESPs in each telecom operator. In addition, in terms of the degree of reduction in service delay, 35-CoEdge is reduced by about 3.71~5.96 times, such as Figure 7 (a) shows that 55-CoEdge is reduced by about 6.23 to 8.87 times, as shown in Figure 7 (b). This shows that the number of deployed ESPs has a positive optimization effect on telecom operators in reducing service delay. In addition, it can be clearly found that the minimum service delay of Rogers in 35-CoEdge is reduced less than that in 55-CoEdge, which is 4.05 times and 6.23 times respectively. The reason is that the average local resource volume of Rogers in 35-CoEdge is less than that in 55-CoEdge, which is 36.18GB and 43.11GB respectively. This shows that the telecom operators with more local resources have a greater degree of optimization in service delay in CoEdge.

[0091] The experimental results show that: CoEdge performs outstandingly in service latency (compared to UnCoEdge, the best improvement is about 2.87~4.15 times) and can balance the differences in ESP.

[0092] The present application also provides an edge computing power scheduling device based on latency-aware optimization, such as Figure 8 As shown, in one embodiment, it includes: an acquisition module 802, a modeling module 804, a solution module 806 and a scheduling module 808, wherein: An acquisition module 802 is used to acquire an edge computing environment and build a computing power sharing framework for multiple edge service providers; A modeling module 804 is used to establish an optimization problem of edge computing task scheduling based on a multi-edge service provider computing power sharing framework, with the goal of minimizing service delay, and with the amount of computing tasks undertaken by the edge service provider, the scheduling decision of the edge service provider, the scheduling matrix of the edge service provider, and the amount of computing tasks processed by the edge service provider as constraints; A solution module 806 is used to solve the optimization problem of edge computing task scheduling and obtain a scheduling matrix of multiple edge service providers; The scheduling module 808 is used to schedule the edge computing power of multiple edge service providers according to the scheduling matrix of the multiple edge service providers.

[0093] For the specific limitations of the edge computing power scheduling device based on latency-aware optimization, please refer to the limitations of the edge computing power scheduling method based on latency-aware optimization above, which will not be repeated here. Each module in the above device can be implemented in whole or in part by software, hardware and a combination thereof. The above modules can be embedded in or independent of the processor in the computer device in the form of hardware, or stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0094] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Fig. 9 As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for scheduling edge computing power based on latency-aware optimization is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0095] Those skilled in the art will understand that Fig. 9 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0096] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method in the above embodiment when executing the computer program.

[0097] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method in the above embodiment are implemented.

[0098] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0099] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0100] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. The edge computing power scheduling method based on latency-aware optimization is characterized by: include: Obtain edge computing environment and build a computing power sharing framework for multiple edge service providers; Based on the computing power sharing framework of multiple edge service providers, with the goal of minimizing service latency, the optimization problem of edge computing task scheduling is established with the amount of computing tasks undertaken by edge service providers, the scheduling decisions of edge service providers, the scheduling matrix of edge service providers, and the amount of computing tasks processed by edge service providers as constraints; Solve the optimization problem of edge computing task scheduling and obtain the scheduling matrix of multiple edge service providers; The edge computing power of multiple edge service providers is scheduled according to the scheduling matrix of multiple edge service providers.

2. The edge computing power scheduling method based on latency-aware optimization according to claim 1 is characterized in that: The optimization issues of edge computing task scheduling include: ； in, ； In the formula, For ESP In the period The scheduling strategy, For ESP In the period The scheduling matrix, For an ESP, It is a collection of edge computing servers belonging to ESP or ESP. For ESP In the period The service delay, For ESP The amount of local computing resources, For ESP In the period The amount of computing tasks undertaken within For ESP The unit calculation delay is For ESP The unit upload / download transmission delay, For the period ESP Dispatch to ESP The computational tasks For ESP To ESP The transmission delay between For ESP The unit upload / download transmission delay, For ESP The unit calculation delay is For ESP In the period Scheduling strategy.

3. The edge computing power scheduling method based on latency-aware optimization according to claim 2 is characterized in that: The constraints are based on the amount of computing tasks undertaken by the edge service provider, the scheduling decision of the edge service provider, the scheduling matrix of the edge service provider, and the amount of computing tasks processed by the edge service provider, including: ； ； ； ； ； ； In the formula, For ESP In the period The total amount of computing tasks undertaken The proportion of For another ESP, For ESP In the period The amount of computing tasks that need to be scheduled, For ESP In the period The amount of computational tasks processed internally.

4. The edge computing power scheduling method based on latency-aware optimization according to any one of claims 1 to 3, characterized in that: Solve the optimization problem of edge computing task scheduling and obtain the scheduling matrix of multiple edge service providers, including: Initialize the scheduling matrix, calculate the objective function value of the scheduling matrix, and obtain the objective function change value; Construct a Lagrangian function based on the objective function and constraints of the optimization problem of edge computing task scheduling; According to the Lagrangian function and the change value of the objective function, the minimum quadratic programming problem is constructed; Adjust the Lagrange multiplier to meet the KKT condition and obtain the updated value of the Lagrange multiplier; According to the updated value of Lagrange multiplier, the least quadratic programming problem is solved to obtain the optimal solution of the change value of the objective function; Perform line search based on the optimal solution of the objective function change value to obtain the calculation step length; According to the calculation step size, the scheduling matrix is ​​updated along the search direction, and it is iterated to recalculate the objective function value of the scheduling matrix until the preset conditions are met, and the current scheduling matrix is ​​output as the scheduling matrix of the multi-edge service provider.

5. The edge computing power scheduling method based on latency-aware optimization according to claim 4 is characterized in that: According to the objective function and constraints of the optimization problem of edge computing task scheduling, a Lagrangian function is constructed, including: In the formula, is the Lagrangian function, For the The task scheduling matrix for the iteration is For the The Lagrange multipliers corresponding to the constraints are: For the constraints.

6. The edge computing power scheduling method based on latency-aware optimization according to claim 5 is characterized in that: According to the Lagrangian function and the change value of the objective function, the minimum quadratic programming problem is constructed, including: In the formula, is the change value of the objective function, for The Hessian matrix of for The gradient of For Constraint The Jacobian matrix of .

7. The edge computing power scheduling method based on latency-aware optimization according to any one of claims 1 to 3, characterized in that: A computing power sharing framework for multiple edge service providers, including: a user layer, an edge layer, and a cloud layer connected in sequence, wherein the user layer is also connected to the cloud layer; The user layer includes multiple terminal devices that dynamically generate computing tasks and hope to obtain services; The edge layer includes multiple edge service providers that provide responsive computing power for computing tasks; The cloud layer includes multiple computing centers, providing the user layer with greater responsive computing power than the edge layer.

8. The edge computing power scheduling method based on latency-aware optimization according to claim 7 is characterized in that: The edge layer also includes: a service analyzer, a central optimizer, and a policy executor connected in sequence; the service analyzer is also connected to the user layer, the service analyzer is also connected to each edge service provider, and the policy executor is also connected to each edge service provider; The service analyzer continuously analyzes the computing tasks at the user level, describes the service characteristics and outputs them to the central optimizer; The central optimizer uses the service features output by the service analyzer as input, executes the optimization algorithm for resource sharing and service revenue, generates a scheduling matrix for computing tasks and service pricing for each edge service provider, and thus forms a computing power sharing strategy and outputs it to the strategy executor; The policy executor redirects each computing task to the best edge service provider based on the computing power sharing strategy output by the central optimizer.

9. The edge computing power scheduling device based on latency perception optimization is characterized by: include: The acquisition module is used to obtain the edge computing environment and build a computing power sharing framework for multiple edge service providers; A modeling module is used to establish an optimization problem for edge computing task scheduling based on a multi-edge service provider computing power sharing framework, with the goal of minimizing service latency, and with the amount of computing tasks undertaken by edge service providers, the scheduling decisions of edge service providers, the scheduling matrix of edge service providers, and the amount of computing tasks processed by edge service providers as constraints; The solution module is used to solve the optimization problem of edge computing task scheduling and obtain the scheduling matrix of multiple edge service providers; The scheduling module is used to schedule the edge computing power of multiple edge service providers according to the scheduling matrix of multiple edge service providers.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

Images