DAG structure task cooperative scheduling method in heterogeneous edge computing network

Abstract

The invention discloses a DAG structure task collaborative scheduling method in a heterogeneous edge computing network, relates to the technical field of edge computing, and aims to realize resource collaborative utilization and optimization of edge server utility. A DAG structure task collaborative scheduling problem in a heterogeneous edge computing network is modeled and decoupled into an alliance forming stage and an alliance task scheduling stage. In the alliance forming stage, an alliance algorithm based on a partition-type alliance game is provided based on a high-income target of an edge server. In the stage of task scheduling in the alliance, a DAG scheduling strategy is provided to schedule DAG tasks arriving in real time in the alliance. According to the invention, the effectiveness of the edge servers is maximized through an efficient cooperation mechanism, so that the competition and cooperation relationship between the edge servers is effectively coordinated in the heterogeneous edge environment with limited resources.

Description

Technical Field

[0001] This invention relates to the field of edge computing technology, and in particular to a method for collaborative scheduling of DAG structure tasks in heterogeneous edge computing networks. Background Technology

[0002] In recent years, the rapid development of IoT technology and 5G mobile communication networks has driven the widespread application of real-time applications, such as facial recognition, virtual reality (VR), and augmented reality (AR). However, most of these applications are computationally intensive and latency-sensitive, and stringent latency requirements have become an obstacle for mobile devices to perform complex tasks. Resource-constrained mobile devices struggle to provide satisfactory Quality of Service (QoS). Offloading computing tasks has become an effective means of reducing latency, and cloud computing was once considered an ideal solution. However, due to long physical distances and increased network load, it leads to high latency and high energy consumption, affecting cloud offloading efficiency. To address this issue, the European Telecommunications Standards Institute (ETS) proposed the concept of Mobile Edge Computing (MEC) in 2014. MEC pushes cloud computing capabilities down to edge nodes, providing low-latency, high-efficiency computing services, becoming a key technology supporting the era of AIoT (Artificial Intelligence of Things).

[0003] Task offloading, a hot research topic in edge computing, has attracted widespread attention from academia and industry. However, existing research often neglects the dependencies between tasks during modeling and optimization, thus limiting the applicability and effectiveness of offloading strategies in complex application scenarios. In real-world applications, mobile applications such as AR or VR can typically be divided into multiple sub-tasks. Furthermore, according to Alibaba's analysis of 4 million applications, over 75% of applications exhibit dependencies between tasks, which can be modeled as a Directed Acyclic Graph (DAG) structure. In addition, various latency-sensitive and computationally intensive applications place higher demands on system resources, which resource-constrained edge nodes still struggle to handle. To improve service quality, Collaborative Edge Computing (CEC) architecture has emerged, supporting multiple edge nodes to collaboratively share information and computing resources, achieving resource complementarity and collaborative task processing. Under this architecture, DAG tasks with dependencies can be divided into multiple subtasks and distributed to different edge nodes for parallel processing, effectively reducing overall task latency. With the dense deployment of edge servers and the improvement of network interconnection capabilities, the collaborative processing capabilities of edge nodes will be further enhanced, and the user experience is expected to continue to improve.

[0004] Nevertheless, edge nodes are not always in a state of complete collaboration; rather, a certain degree of resource competition and self-interest exists in actual deployments. At the industry level, major telecom operators are actively deploying edge computing platforms, such as Verizon's 5G Edge and China Mobile's OpenSigma. Major technology companies have also launched various edge computing service products, such as Microsoft's Azure Edge Zones, Huawei's EdgeGallery, and Amazon's Wavelength Framework. More and more operators are integrating edge computing with cloud platform applications to improve business agility. Therefore, in areas with high mobile user density, edge servers are often deployed and managed by different operators, forming a heterogeneous edge environment with multiple operators. Consequently, in this heterogeneous edge environment with multiple operators, the collaborative relationship between edge servers is often limited by their respective management strategies and profit considerations, no longer characterized by unconditional resource sharing, but rather by a certain degree of competition and self-interest.

[0005] Existing related work generally assumes unconditional cooperation among all edge nodes in the network, failing to adequately consider the impact of competition among edge nodes during task unloading. This competition can lead to resource imbalances. Edge nodes with limited computing power may become idle due to difficulty attracting task unloading requests, while edge nodes with stronger computing power may be unable to complete scheduling on time due to task overload, affecting the overall system performance. Summary of the Invention

[0006] The technical problem this invention aims to solve is to overcome the shortcomings of existing technologies and provide a collaborative scheduling method for DAG-structured tasks in heterogeneous edge computing networks. This method optimizes the task scheduling efficiency of edge servers with competitive and cooperative relationships in resource-constrained heterogeneous edge computing networks. First, for DAG-structured tasks with sequential dependencies, a task collaborative scheduling framework based on alliance formation is constructed. Second, an edge server alliance formation algorithm and a multi-DAG-structured task scheduling algorithm within the alliance are designed to achieve resource collaborative utilization and optimization of edge server efficiency.

[0007] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0008] A method for collaborative scheduling of DAG-structured tasks in a heterogeneous edge computing network, as proposed in this invention, includes:

[0009] Step S1: System modeling is performed for the DAG structure task collaborative scheduling problem in heterogeneous edge computing networks. A heterogeneous edge computing network model, a DAG structure task model, a task offloading model, a task preference model, and an edge utility model are constructed. The optimization objective of the DAG structure task collaborative scheduling problem is defined as maximizing the utility of the edge server.

[0010] Step S2: Decompose the task collaborative scheduling problem of DAG structure in heterogeneous edge computing network into two stages, and solve the two stages separately; wherein, the two stages include the first stage and the second stage, the first stage is the alliance formation stage, and the second stage is the task scheduling stage within the alliance.

[0011] As a further optimization scheme for the DAG structure task collaborative scheduling method in a heterogeneous edge computing network described in this invention, step S2 involves solving these two stages separately; specifically as follows:

[0012] Step S2.1: The first stage is the alliance formation stage, which is based on the low latency requirements of user equipment to construct the user equipment's preference function for edge servers;

[0013] Based on the high-return objective of edge servers and combined with historical statistical information of user device tasks, a preference function for edge servers to the alliance is constructed.

[0014] Finally, a consortium formation algorithm is designed. Based on the user device's preference function for edge servers and the edge server's preference function for consortiums, the edge servers in the edge computing network are divided into different consortia. Each consortium is a subset of the set of edge servers in the edge computing network, and the consortia are mutually exclusive, thereby maximizing the historical estimation utility of edge servers in competitive and cooperative scenarios.

[0015] Step S2.2: The second stage is the task scheduling stage within the alliance. Considering the deadline requirements of the tasks, the DAG structure tasks that arrive in real time in each alliance are scheduled based on the HEFT algorithm to maximize the workload of tasks that are completed on time, thereby maximizing the actual scheduling efficiency of the edge server.

[0016] As a further optimization scheme for the DAG structure task collaborative scheduling method in a heterogeneous edge computing network described in this invention, step S1 involves constructing a heterogeneous edge computing network model, including:

[0017] Heterogeneous edge computing networks consist of having heterogeneous edge servers and Composed of different DAG task types, the set of edge servers and the set of DAG structure task types are represented as follows: and Each edge server Represented as ,in and Let represent the computing power and location coordinates of the edge server, respectively; considering the periodic offloading model, once a stable alliance structure is formed, this alliance structure is maintained for a certain lifespan. And it is divided into A task unloading cycle of equal length ,Right now Each user generates a task according to a different random process at the beginning of each uninstallation cycle, symbol... For set The number of elements contained therein.

[0018] As a further optimization scheme for the DAG structure task collaborative scheduling method in a heterogeneous edge computing network described in this invention, in step S1, a DAG structure task model is constructed:

[0019] have A set of DAG structure task types , For the first This section describes various DAG structure task types. Without loss of generality, it is assumed that different types of tasks have different topological structures or generation locations. Modeled as a quadruple, ,in , This represents a set of subtasks with dependencies. Indicates the first Types of DAG structure tasks The Sub-tasks Indicates the total number of subtasks; , Indicates the first Types of DAG structure tasks The Sub-tasks Indicates a subtask Pointing to subtask The directed edge, This represents the set of directed edges between subtasks. Indicates the location where the task was generated. This indicates that the probability of a task being generated within a given unloading cycle follows a Bernoulli distribution; define an indicator variable. Used to indicate during the task unloading cycle Internal DAG structure task types Whether to generate; for each subtask Directed edge weight Indicates the first Sub-tasks Transmitted to the Sub-tasks The amount of data, , Indicates a subtask Pointing to subtask The directed edge, Subtasks The set of all direct predecessor subtasks; , Indicates a subtask Pointing to subtask The directed edge, Subtasks The set of all direct successor subtasks; introduce a virtual starting subtask with zero computational effort. and a virtual end subtask ; The output weights represent the size of the input data for the task. The input weights represent the size of the task result.

[0020] As a further optimization scheme for the DAG structure task collaborative scheduling method in a heterogeneous edge computing network described in this invention, in step S1, a task offloading model is constructed:

[0021] DAG structure generation task type Users to edge server The uplink transmission rate is ;in, For uplink channel bandwidth, It is a DAG structure task type Uplink transmission power, For noise power, For channel gain between the task and the edge server, The calculation method is as follows , It is the path loss index. , This represents the Euclidean distance between the task and the server. Represents edge server Location coordinates, Indicates the task type of the DAG structure Two-dimensional position coordinates The Euclidean norm is used to calculate the Euclidean distance between two two-dimensional coordinates; similarly, the downlink transmission rate from the edge server to the task is expressed as... , For downlink channel bandwidth, Edge server The transmit power; the external transmission delay for mission offloading to the alliance and returning results is... ,in, , Indicates the task type of the DAG structure To the Alliance uplink transmission rate, Indicates a virtual starting subtask The set of all direct successor subtasks, Indicates a virtual starting subtask direct successor subtask Similarly, the size of the input data, Indicates the virtual termination of the subtask The set of all direct predecessor subtasks, Indicates the virtual termination of the subtask direct precursor sub-task The size of the output data. , Indicates the first Alliance Task types to DAG structure downlink transmission rate; when subtask Sent to the league Edge servers within Execution, and subtasks Precursor mission Redirected to another edge server in the same consortium At that time, the amount of data transmitted between the two edge servers was Therefore, the internal transmission delay is expressed as ;in, , Represents edge server and The Euclidean distance between them Represents edge server Location coordinates, This represents the transmission delay coefficient for transmitting one unit of data per unit distance; if the server and If they are from the same device, the internal transmission delay is zero; define a binary variable. , Used to represent subtasks Was it uninstalled to the edge server? Satisfying constraints Edge server Processing subtasks The required computation delay is expressed as ,in It is a subtask The required computational workload It is an edge server The computational frequency; defining subtasks Readiness time The calculation method is as follows :in, Subtasks Completion time, Subtasks and Transmission latency between edges; each edge server maintains a task waiting queue following a first-come, first-served strategy; subtasks The start execution time is expressed as , Completion time, This indicates that the task is immediately adjacent to the subtask in the waiting queue. Previous subtasks; subtasks On the edge server The completion time is expressed as ;Task The total completion time depends on the completion time of the virtual termination subtask; therefore, the task In the league The total completion time within is expressed as , Indicates the virtual termination of the subtask The completion time.

[0022] As a further optimization scheme for the DAG structure task collaborative scheduling method in a heterogeneous edge computing network described in this invention, in step 1, a task preference model is constructed.

[0023] Users who generate a certain type of DAG structure task always make decisions about whether to offload their tasks to a particular federation based on their preference for edge servers according to that type of DAG structure task; for a given federation DAG structure task type The preference values ​​are constructed in the following way;

[0024] First, classify the DAG structure task types Treat it as a complete and indivisible task type, ignoring DAG structure task types. The dependencies between internal subtasks are then determined; subsequently, the user's DAG structure task type is estimated. Uninstall to the Alliance The total completion time is calculated, which consists of the estimated computation time, the estimated internal transmission time, and the estimated external transmission time; therefore, the estimated computation time is expressed as... , where the function This indicates the diminishing marginal returns effect caused by increased computing resources. , and This is an adjustable coefficient; task In the league The estimated internal transmission time is expressed as ;in, Indicates alliance The average distance between edge servers is calculated as follows: , Indicates server and The Euclidean distance between them; therefore, given a coalition structure ,Task Against the alliance The preference value is defined as the estimated total completion time of the task within the alliance. ,Right now In a given alliance structure Next, calculate the total completion time for each task. The preference values ​​for different alliances are listed and arranged in non-decreasing order to form a preference list. ;

[0025] Step S1.5: Constructing the edge utility model; during the alliance formation phase, edge servers Join the alliance The gains obtained are expressed as , Indicates the first In each alliance The One edge server, It is an edge server The calculation frequency, where the return According to server computing power The proportion of computing power allocated within the alliance. Indicates alliance The estimated total task load is calculated using the following formula: ,in Indicates the task type of the DAG structure The probability of arrival within a single unloading cycle. Indicates alliance The task set, Includes all Tasks that are the preferred consortiums in their preference list, and whose total computational requirements do not exceed the total computational capacity that the consortium can support; defined as ,in, Indicates task In alliance structure The alliance with the highest preference level; edge server Join the alliance The resulting communication cost is expressed as ,in, These are adjustable parameters used to control the impact of distance on communication costs; thus, the edge server... Join the Alliance The obtained historical estimated utility is expressed as During the task scheduling phase, the edge server The actual scheduling utility obtained is defined as , Indicates the first Each task unloading cycle Indicates the first step in the task unloading cycle. At a certain point in time, It is an indicator function, when the condition is met. When satisfied, indicator function The value is 1; taking into account the impact of historical experience and current actual scheduling results on utility evaluation, the function is defined. This represents the fusion function of historically estimated utility and actual cooperative utility. .

[0026] As a further optimization scheme of the DAG structure task collaborative scheduling method in a heterogeneous edge computing network described in this invention, in step 1, the optimization objective of the DAG structure task collaborative scheduling problem in a heterogeneous edge computing network is defined as maximizing the utility of each edge server within the lifecycle of the consortium structure; wherein, maximizing the utility of each edge server within the lifecycle of the consortium structure specifically means:

[0027] The input to the DAG-structured task cooperative scheduling method in a heterogeneous edge computing network is defined as: the set of edge servers. DAG structure task type set Edge servers computing power and position coordinates DAG structure task Subtask set The set of directed edges between subtasks Location where the task is generated The probability of a task being generated within a given unloading cycle. Transmission delay coefficient for transmitting one bit of data per unit distance The lifecycle of the alliance structure, and the task unloading cycle;

[0028] The output of the DAG structure task cooperative scheduling method in heterogeneous edge computing networks is defined as: having Stable alliance structure of each alliance , Indicates alliance structure The first in One alliance, The value range is 1 to Unloading decision for DAG structure tasks ;

[0029] The optimization objective of the DAG structure task cooperative scheduling method in heterogeneous edge computing networks is as follows:

[0030]

[0031]

[0032]

[0033]

[0034]

[0035] Among them, constraint C1 means that a subtask can be offloaded to at most one edge server for computation, constraint C2 means that the total completion time of the task does not exceed the task offloading cycle, and constraint C3 means that the subtask is completely offloaded to the edge server for computation.

[0036] As a further optimization scheme for the DAG structure task collaborative scheduling method in a heterogeneous edge computing network described in this invention, step S2.1 includes the following steps:

[0037] Step S2.1.1: Alliance Structure Initialization; First, each edge server forms an independent alliance, resulting in the initial alliance structure. The current alliance structure Updated to According to the formula Calculate each task under the current alliance structure Initial preference list Given a consortium structure, the expression for the edge server's preference function for the consortium is: The result of this expression is used to characterize the utility of edge servers joining a consortium, where, Indicates alliance structure Next Alliance Edge servers in Historical estimated utility, case 1 indicates that the alliance admission conditions are met and the current alliance structure was not formed before this, i.e., it satisfies and ,in Indicates alliance structure Next Alliance Edge servers in Historical estimated utility Indicates alliance structure Lower edge server From the Alliance After leaving the edge server Historical estimated utility Represents edge server It is used to store a historical set of alliances that have been joined in the past. This indicates that only edge servers are included. The set, case 2 indicates that the alliance structure size is 1, that is... Each edge server calculates the initial utility for the given federation structure based on the task's preference list. ;

[0038] Step S2.1.2: Alliance Preference Assessment; Based on the current alliance structure, in the current alliance structure Based on this, the alliance Each edge server within The process will involve assessing the potential benefits of joining other alliances and constructing appropriate temporary alliance structures. ,in , Indicates server Join the Alliance The new alliance after that, , Indicates server Leave the league The remaining alliance;

[0039] Step S2.1.3: Alliance structure adjustment; if the edge server Join another league The utility gained is greater than that of an edge server. Join the alliance If the utility of the alliance is satisfied, then the attempt to join the alliance is fulfilled. Time Edge Server Will withdraw from its current alliance , Represents edge server Join the alliance The subsequent alliance structure, Represents edge server Join another league Back Edge Server The effect, Indicates the current alliance structure Lower edge server Join the alliance The utility of this, and updating the alliance structure, to obtain ;

[0040] Step S2.1.4: Repeat steps S2.1.2 and S2.1.3 until no edge server can further improve its utility by joining other alliances, i.e., the current alliance division is complete. A stable state has been reached.

[0041] As a further optimization scheme for the DAG structure task collaborative scheduling method in a heterogeneous edge computing network described in this invention, step S2.2 includes the following steps:

[0042] Step S2.2.1: Obtain the set of tasks to be scheduled and construct a priority queue; after a stable alliance structure is formed on the edge servers, each alliance obtains the set of tasks it has competed for in each task unloading cycle. In this step, all DAG tasks within the alliance are treated as independent tasks and their priorities are determined separately.

[0043] First, calculate the priority index for each DAG task. It is defined as the ratio of the total computational load of the task to the total output data volume. ,in, This represents the total computational cost of the DAG task. The total amount of data transmitted for the task; based on Sort all DAG tasks to form a task queue arranged from high to low priority;

[0044] Step S2.2.2: Calculate the rank value of subtasks based on the topology; for each DAG task obtained in step S2.2.1, calculate the rank value of its internal subtasks according to its topology; using the HEFT algorithm, calculate the rank value of each subtask in the DAG structure task based on its average computation time. With average communication time Perform recursive calculations to obtain the level values ​​of the subtasks. ,in, This represents the average computation time of the subtask. This represents the average communication time of the subtask; The set of all direct successor subtasks of the subtask; express Each set of direct successor subtasks in the set; all the resulting subtasks are sorted from highest to lowest priority value to form a subtask priority queue;

[0045] Step S2.2.3: Select the optimal edge server to schedule subtasks based on completion time; execute scheduling for each sorted subtask sequentially, targeting the current subtask. It collects the available computing resource status of all edge servers within the alliance in real time and based on the formula Calculate the earliest completion time and select the edge server that can complete the subtask earliest. Execute the task; for each subtask scheduled, immediately select the optimal server dynamically based on the real-time resource status; the scheduling process continues until all subtasks of all DAG tasks have been scheduled.

[0046] A computer device includes a memory, a processor, and a computer program stored in the memory. The processor executes the computer program to implement the above-described method for collaborative scheduling of DAG structure tasks in a heterogeneous edge computing network.

[0047] Compared with the prior art, the present invention, employing the above technical solution, has the following technical effects:

[0048] The method of this invention decomposes the DAG structure task collaborative scheduling problem in heterogeneous edge computing networks into two stages, decoupling decision variables while reducing problem complexity;

[0049] The first stage aims to maximize the historical estimated utility of edge servers. In this stage, a partitioned alliance formation algorithm based on user preferences and the high-return objective of edge servers is proposed, which has convergence and Nash stability.

[0050] The second stage aims to maximize the actual scheduling efficiency of edge servers and proposes an improved HEFT algorithm to schedule DAG structure tasks that arrive in real time in each federation.

[0051] This invention decouples a complex problem into two sub-stages through an innovative method, achieving coordinated resource utilization and maximizing the utility of edge servers. It also provides convergence and stability guarantees for the proposed algorithm, ensuring the feasibility of the method. Attached Figure Description

[0052] Figure 1 This is a schematic diagram of the overall system model of a heterogeneous edge computing network provided in an embodiment of the present invention;

[0053] Figure 2 This is the overall framework of the DAG structure task collaborative scheduling method in heterogeneous edge computing networks provided by the embodiments of the present invention;

[0054] Figure 3 This is a detailed flowchart of the alliance formation algorithm provided according to an embodiment of the present invention. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0056] To alleviate the problems mentioned in the background, edge servers belonging to different entities can form alliances to jointly compete for tasks and collaboratively share the tasks of high-load servers, thereby improving the overall task processing capacity of the system. Therefore, this invention, considering the competition and cooperation relationships between heterogeneous edge nodes, designs a DAG structure task collaborative scheduling method in a heterogeneous edge computing network.

[0057] This invention provides a task collaborative scheduling method based on a DAG structure formed by alliances in a heterogeneous edge computing network, which performs the following steps S1-S2 to optimize the utility of edge servers:

[0058] Step S1: System modeling is performed for the task collaborative scheduling problem of DAG (Directed acyclic graph) structure in heterogeneous edge computing networks. A heterogeneous edge computing network model, a DAG structure task model, a task offloading model, a task preference model, and an edge utility model are constructed. The optimization objective of the problem is defined as maximizing the utility of the edge server.

[0059] The specific steps of step S1 are as follows:

[0060] Step S1.1: Construct a heterogeneous edge computing network model; the heterogeneous edge computing network consists of... heterogeneous edge servers and The system consists of several different DAG structure task types, and the overall system model diagram is shown below. Figure 1 As shown. The set of edge servers and the set of DAG structure task types are represented as follows: and Considering the heterogeneity of edge servers, each edge server... Represented as a pair ,in and These represent the computing power and location coordinates of the edge server, respectively. Considering the periodic offloading model, once a stable alliance structure is formed, it maintains this structure for a certain lifespan. And it is divided into An equal-length unloading cycle ,Right now Each user generates a task according to a different random process at the beginning of each uninstallation cycle, symbol... For set The number of elements contained therein. Each user offloads computationally intensive tasks to a chosen consortium within the edge computing network to achieve low latency and high privacy;

[0061] Step S1.2: Construct a DAG structure task model; the set of DAG structure task types is denoted as... Without loss of generality, it is assumed that different types of tasks typically have different topologies or originating locations. For example, face recognition tasks and vehicle navigation tasks differ in their topologies: the former often occurs in indoor architectural environments, while the latter typically occurs in road scenarios. Therefore, classifying tasks based on their originating location is reasonable. Tasks of the same type are identical in terms of topology, computational workload, and data transfer scale. For the first... Types of DAG structure tasks It can be modeled as a quadruple ,in Represents a set of subtasks with dependencies. Indicates the first Types of DAG structure tasks The Sub-tasks Indicates the total number of subtasks; This represents the set of directed edges between subtasks. Indicates the first Types of DAG structure tasks The Sub-tasks Indicates the first Types of DAG structure tasks The Sub-tasks Indicates a subtask Pointing to subtask The directed edge, Indicates the location where the task was generated. This represents the probability of a task being generated within a given unloading cycle, and this probability follows a Bernoulli distribution. Define an indicator variable. Used to indicate the task unloading cycle Internal mission Whether to generate. For each subtask. , Indicates the required computational workload, edge This represents the dependencies between subtasks, and their weights. Indicates the first Sub-tasks Transmitted to the Sub-tasks Data volume, collection This represents the set of all direct predecessor subtasks of this subtask. Indicates a subtask Pointing to subtask Directed edges, set This represents the set of all direct successor subtasks of this subtask. Indicates a subtask Pointing to subtask The directed edges. After the task calculation is completed, the task result will be returned to the user. To represent the start and end points of the task, a virtual starting subtask with zero computational workload is introduced. and Virtual starting subtask The output weights represent the size of the input data for the task, and the subtask is virtually terminated. The input weights represent the size of the task result.

[0062] Step S1.3: Construct the task offloading model; When a task is offloaded to a consortium, it is first transmitted by the user to the edge server within the consortium via a wireless link. After the task is processed, the calculation result is returned to the user via a wireless link. These two transmission processes are collectively referred to as task-related external transmission. Within the consortium, the various subtasks of the task are transmitted and processed between edge servers via wired links; this process is called internal transmission. For the external transmission process, the edge servers occupy orthogonal wireless channels to avoid mutual interference between uplink communications from different users. To prevent conflicts when multiple users send data to the same edge server, a time-division multiple access protocol is used to coordinate uplink transmission. Generate DAG structure task types. Users to edge server The uplink transmission rate is .in For uplink channel bandwidth, It is a task Uplink transmission power, For noise power, For channel gain between the task and the edge server, The calculation method is as follows ,in It is the path loss index. , This represents the Euclidean distance between the task and the server, where Represents edge server Location coordinates, Indicates the task type of the DAG structure Location coordinates This represents the Euclidean norm, used to calculate the Euclidean distance between two coordinates. Similarly, the downlink transmission rate from the edge server to the task can be expressed as... ,in For downlink channel bandwidth, Edge server The transmit power. To minimize wireless communication latency, when the DAG structure task type... Assigned to the league During processing, the user always transmits task input data to the nearest edge server within the consortium. After computation, the result is returned to the user. Therefore, the external transmission latency for task offloading to the consortium and returning the result is... , Indicates the task type of the DAG structure To the Alliance uplink transmission rate, Indicates a virtual starting subtask The set of all direct successor subtasks, Indicates a virtual starting subtask direct successor subtask Similarly, the size of the input data, Indicates the virtual termination of the subtask The set of all direct predecessor subtasks, Indicates the virtual termination of the subtask direct precursor sub-task The size of the output data. , Indicates the task type of the DAG structure To the Alliance The uplink transmission rate. When the subtask Sent to the league Edge servers within Execution, and its predecessor subtask Redirected to another edge server in the same consortium At that time, the amount of data transmitted between the two edge servers was Therefore, the internal transmission delay can be expressed as .in, Indicates server and The Euclidean distance between them Represents edge server Position coordinates; constant This represents the transmission delay coefficient for transmitting one unit of data per unit distance. (If the edge server...) and If it's the same device, the internal transmission latency is zero. Assume each subtask is completely offloaded to an edge server for execution, and that the subtasks are not further broken down. Define binary variables. , used to represent subtasks Was it uninstalled to the edge server? Satisfying constraints Edge server Processing subtasks The required computation time can be expressed as ,in It is a subtask The required computational workload It is an edge server The computation frequency; Considering the dependencies between subtasks, subtasks Only after all its predecessor subtasks have been completed and all relevant data has been received can it be used on the edge server. Execution begins. Its ready time is defined. That is, the earliest time that can be executed after the above conditions are met, calculated as follows: :in, Subtasks Completion time, Subtasks and Transmission latency between them. Each edge server can only process one subtask at any given time and follows a first-come, first-served policy, therefore each edge server maintains a task waiting queue. This indicates that the task is immediately adjacent to the subtask in the waiting queue. The previous subtask had a completion time of Then subtask The start execution time can be expressed as Subtasks On the edge server The completion time can be expressed as .Task The total completion time depends on the completion time of the virtual termination subtask. Therefore, the task In the league The total completion time within can be expressed as , Indicates the virtual termination of the subtask Completion time;

[0063] Step S1.4: Construct a task preference model; When users choose which federation to offload a DAG structure task type to, they always make decisions based on their preference for edge servers for that DAG structure task type. This is mainly because users cannot know the real-time computing resource status and workload of each edge server in the network; at the same time, each user's offloading decision is made independently, and decision information cannot be shared with each other in real time. Therefore, users can only form preferences based on the estimated completion time of tasks in different federations and make offloading decisions accordingly. For a given federation DAG structure task type The preference values ​​are constructed in the following way. First, the DAG structure task types are... This is treated as a complete and indivisible task type, ignoring the dependencies between its internal subtasks. Subsequently, the user's understanding of the DAG structure task type is estimated. Uninstall to Alliance The total completion time is calculated by considering the estimated computation time, estimated internal transmission time, and estimated external transmission time. Intuitively, the more computing resources a consortium possesses, the lower the processing latency. Therefore, the estimated computation time can be expressed as... , where the function This represents the diminishing marginal returns effect caused by increased computing resources. It can be expressed as... .in and This is an adjustable coefficient used to control for diminishing returns from resource increases. Generally, this expression is sufficient to characterize the marginal effect of edge server computing power; of course, other equivalent forms can also be used. The estimated internal transmission time is related to data transmission between edge servers within the consortium, and the DAG structure task type... In the league The estimated internal transmission time can be expressed as: .in, Indicates alliance The average distance between edge servers is calculated as follows: Therefore, given an alliance structure DAG structure task type Against the alliance The preference value can be defined as the estimated total completion time of the task within the alliance, i.e. In a given alliance structure Based on the estimated total completion time, the task type of each DAG structure can be calculated. The preference values ​​for different alliances are then arranged in ascending order to form a preference list. The assumption is that edge servers within the alliance can perceive each other's preference information. This assumption holds true because information within the system is highly transparent, and servers can infer the preferences of other nodes through signaling or cooperation mechanisms. Based on this preference information, each edge server will form an alliance structure based on its own interests. As the alliance structure dynamically adjusts, the task's preference for each alliance will also change accordingly. Furthermore, without affecting the subsequent alliance formation and scheduling results, the preference model can be replaced by any other equivalent estimation model.

[0064] Step S1.5: Constructing the Edge Utility Model; In an edge computing network, edge servers gain revenue by processing tasks. Because each edge server is self-interested and rational, they will form alliances based on historical information to compete for more tasks and higher revenue. Typically, servers tend to join alliances with larger task volumes and higher computational loads. However, the actual revenue of an edge server depends not only on the timing of joining an alliance but also on the number of tasks successfully completed within that alliance. Specifically, during the alliance formation phase, edge servers… Join the alliance The gains obtained can be expressed as , Indicates the first In each alliance The One edge server, It is an edge server Calculation frequency, revenue According to server computing power The proportion of computing power allocated within the alliance. Indicates alliance The estimated total task load is calculated using the following formula: ,in Indicates the task type of the DAG structure The probability of arrival within a single unloading cycle. (Set) Indicates alliance The task set, Includes all Tasks that are the preferred alliances in its preference list, and whose total computational requirements do not exceed the total computational capacity that the alliance can support. Defined as ,in, Indicates task In alliance structure The alliance with the highest preference level. Data migration between edge servers relies on fiber optic communication links, and as the number of servers in the alliance increases, so does the communication overhead. Since the data transmission rate of the wired backhaul link is fixed, the main factor affecting communication costs is the physical distance between the servers. Therefore, edge servers... Join the alliance The resulting communication cost can be expressed as ,in, These are adjustable parameters used to control the impact of distance on communication costs. Therefore, the edge server... Join the alliance The historically estimated utility obtained can be expressed as: During the task scheduling phase, the edge server The actual cooperative utility obtained can be defined as... , Indicates the first Each task unloading cycle Indicates the first step in the task unloading cycle. At a certain point in time, It is an indicator function, when the condition is met. When satisfied, indicator function The value is 1. Historically estimated utility reflects predictions of alliance cooperation performance during the alliance formation phase, while actual cooperation utility provides real feedback on alliance performance during the task scheduling phase. Therefore, considering the combined impact of historical experience and current actual scheduling results on utility evaluation, a function is defined. This represents the fusion function of historically estimated utility and actual cooperative utility. .

[0065] Step S1.6: Define the optimization objective of the DAG structure task cooperative scheduling method in heterogeneous edge computing networks as maximizing the utility of each edge server throughout the lifecycle of the alliance structure, specifically:

[0066] The input to the DAG-structured task cooperative scheduling method in a heterogeneous edge computing network is defined as: the set of edge servers. DAG structure task type set Edge servers computing power and position coordinates DAG structure task Subtask set The set of directed edges between subtasks Location where the task is generated The probability of a task being generated within a given unloading cycle. Transmission delay coefficient for transmitting one bit of data per unit distance The lifecycle of the alliance structure, and the task unloading cycle;

[0067] The output of the DAG-structured task cooperative scheduling method in heterogeneous edge computing networks is defined as: a stable alliance structure. , Indicates alliance structure The first in One alliance, The value range is 1 to Unloading decision for DAG structure tasks ;

[0068] The optimization objective of the model is as follows:

[0069]

[0070]

[0071]

[0072]

[0073] Where C1 indicates that a subtask is offloaded to at most one edge server for computation, C2 indicates that the total completion time of the task does not exceed the task offloading cycle, and C3 indicates that the subtask is completely offloaded to the edge server for computation.

[0074] Step S2: Decompose the collaborative scheduling of DAG structure tasks in heterogeneous edge computing networks into two stages and solve them separately. The overall framework of the collaborative scheduling method for DAG structure tasks in heterogeneous edge computing networks is as follows: Figure 2 As shown, it includes the following steps:

[0075] Step S2.1: The first stage is the alliance formation stage, which optimizes the historical estimated utility of edge servers. First, based on the low latency requirements of user devices, a preference function for edge servers by user devices is constructed. Then, based on the high-return objective of edge servers and combined with historical statistical information of user device tasks, a preference function for alliances by edge servers is constructed. Finally, an alliance formation algorithm is designed. Based on the user device preference function and the edge server preference function, the edge servers in the edge computing network are divided into different alliances. Each alliance is a subset of the set of edge servers in the edge computing network, and the alliances are mutually exclusive, thereby maximizing the historical estimated utility of edge servers in both competitive and cooperative scenarios. The flowchart of this algorithm is as follows: Figure 3 As shown.

[0076] Step S2.1 includes the following steps:

[0077] Step S2.1.1: Alliance Structure Initialization; First, each edge server forms an independent alliance, resulting in the initial alliance structure. The current alliance structure Updated to According to the formula Calculate each task under the current alliance structure Initial preference list Given a consortium structure, the expression for the edge server's preference function for the consortium is: The result of this expression is used to characterize the utility of edge servers joining a consortium, where, Indicates alliance structure Next Alliance Edge servers in Historical estimated utility, case 1 indicates that the alliance admission conditions are met and the current alliance structure was not formed before this, i.e., it satisfies and ,in Indicates alliance structure Next Alliance Edge servers in Historical estimated utility Indicates alliance structure Lower edge server From the Alliance After leaving the edge server Historical estimated utility Represents edge server It is used to store a historical set of alliances that have been joined in the past. This indicates that only edge servers are included. The set, case 2 indicates that the alliance structure size is 1, that is... Each edge server calculates the initial utility for the given federation structure based on the task's preference list. ;

[0078] Step S2.1.2: Alliance Preference Assessment; Based on the current alliance structure, in the current alliance structure Based on this, the alliance Each edge server within The process will involve assessing the potential benefits of joining other alliances and constructing appropriate temporary alliance structures. ,in , Indicates server Join the Alliance The new alliance after that, , Indicates server Leave the league The remaining alliance;

[0079] Step S2.1.3: Alliance structure adjustment; if the edge server Join another league The utility gained is greater than that of an edge server. Join the alliance If the utility of the alliance is satisfied, then the attempt to join the alliance is fulfilled. Time Edge Server Will withdraw from its current alliance , Represents edge server Join the alliance The subsequent alliance structure, Represents edge server Join another league Back Edge Server The effect, Indicates the current alliance structure Lower edge server Join the alliance The utility of this, and updating the alliance structure, to obtain ;

[0080] Repeat steps S2.1.2 and S2.1.3 until no edge server can further improve its utility by joining other alliances, i.e., the current alliance division. A stable state has been reached.

[0081] Step S2.2: The second stage is the task scheduling stage within the consortium, which optimizes the actual scheduling efficiency of the edge servers. Considering the deadline requirements of the tasks, the HEFT algorithm is used to schedule the DAG structure tasks that arrive in real time in each consortium. First, all subtasks are sorted from high to low according to their priority values, and scheduling starts from the subtask with the highest priority. During the scheduling process, the edge server that can provide the earliest completion time is selected according to the real-time computing resource status to execute the current subtask, and this process is continuously iterated until all subtasks are completed.

[0082] Step S2.2 includes the following steps:

[0083] Step S2.2.1: Obtain the set of tasks to be scheduled and construct a priority queue; after a stable alliance structure is formed on the edge servers, each alliance obtains the set of tasks it has competed for in each task unloading cycle. In this step, all DAG tasks within the consortium are treated as independent tasks, and their priorities are determined individually. To improve the overall computational efficiency of the consortium, the priority index for each DAG task is first calculated. It is defined as the ratio of the total computational load of the task to the total output data volume. The numerator represents the total computational cost of the DAG task, and the denominator represents the total data transmission volume of the task. A higher priority value indicates a smaller proportion of communication overhead and higher computational intensity. According to... Sort all DAG tasks to form a task queue arranged from high to low priority;

[0084] Step S2.2.2: Calculate the rank value of subtasks based on the topology; for each DAG task obtained in step S2.2.1, calculate the rank value of its internal subtasks according to its topology. This step adopts the HEFT (Heterogeneous Earliest Finish Time) algorithm, which calculates the rank value of each subtask within the DAG based on its average computation time. With average communication time Perform recursive calculations to obtain the level value. ,in This represents the average computation time of the subtask. This represents the average communication time of the subtask; The set of all direct successor nodes of the subtask. express Each set contains a set of direct successor subtasks. All resulting subtasks are sorted from highest to lowest priority value, forming a subtask priority queue.

[0085] Step S2.2.3: Select the optimal edge server to schedule subtasks based on completion time; execute scheduling for each sorted subtask sequentially, targeting the current subtask. It collects the available computing resource status of all edge servers within the alliance in real time and based on the formula Calculate the earliest completion time and select the edge server that can complete the subtask earliest. Execute the task. This selection process is carried out throughout the entire scheduling process; that is, for each subtask scheduled, the optimal server is dynamically selected based on the real-time resource status. The scheduling process continues until all subtasks of all DAG tasks have been scheduled.

[0086] This invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory. The processor executes the computer program to implement the aforementioned task collaborative scheduling method based on a DAG structure formed by alliances in a heterogeneous edge computing network.

[0087] This invention also provides a computer-readable storage medium storing a computer program / instructions thereon, which, when executed by a processor, implements the aforementioned task collaborative scheduling method based on a DAG structure formed by alliances in a heterogeneous edge computing network.

[0088] This invention also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the aforementioned task collaborative scheduling method based on a DAG structure formed by alliances in a heterogeneous edge computing network.

[0089] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for collaborative scheduling of DAG-structured tasks in a heterogeneous edge computing network, characterized in that, include: Step S1: System modeling is performed for the DAG structure task collaborative scheduling problem in heterogeneous edge computing networks. A heterogeneous edge computing network model, a DAG structure task model, a task offloading model, a task preference model, and an edge utility model are constructed. The optimization objective of the DAG structure task collaborative scheduling problem is defined as maximizing the utility of the edge server. Step S2: Decompose the task collaborative scheduling problem of DAG structure in heterogeneous edge computing network into two stages, and solve the two stages separately; wherein, the two stages include the first stage and the second stage, the first stage is the alliance formation stage, and the second stage is the task scheduling stage within the alliance.

2. The method for collaborative scheduling of DAG structure tasks in a heterogeneous edge computing network according to claim 1, characterized in that, In step S2, the solutions for these two stages are performed separately; the details are as follows: Step S2.1: The first stage is the alliance formation stage, which is based on the low latency requirements of user equipment to construct the user equipment's preference function for edge servers; Based on the high-return objective of edge servers and combined with historical statistical information of user device tasks, a preference function for edge servers to the alliance is constructed. Finally, a consortium formation algorithm is designed. Based on the user device's preference function for edge servers and the edge server's preference function for consortiums, the edge servers in the edge computing network are divided into different consortia. Each consortium is a subset of the set of edge servers in the edge computing network, and the consortia are mutually exclusive, thereby maximizing the historical estimation utility of edge servers in competitive and cooperative scenarios. Step S2.2: The second stage is the task scheduling stage within the alliance. Considering the deadline requirements of the tasks, the DAG structure tasks that arrive in real time in each alliance are scheduled based on the HEFT algorithm to maximize the workload of tasks that are completed on time, thereby maximizing the actual scheduling efficiency of the edge server.

3. The method for collaborative scheduling of DAG structure tasks in a heterogeneous edge computing network according to claim 2, characterized in that, In step S1, a heterogeneous edge computing network model is constructed, including: Heterogeneous edge computing networks consist of having heterogeneous edge servers and Composed of different DAG task types, the set of edge servers and the set of DAG structure task types are represented as follows: and Each edge server Represented as ,in and Let represent the computing power and location coordinates of the edge server, respectively; considering the periodic offloading model, once a stable alliance structure is formed, this alliance structure is maintained for a certain lifespan. And it is divided into A task unloading cycle of equal length ,Right now Each user generates a task according to a different random process at the beginning of each uninstallation cycle, symbol... For set The number of elements contained therein.

4. The method for collaborative scheduling of DAG structure tasks in a heterogeneous edge computing network according to claim 3, characterized in that, In step S1, construct the DAG structure task model: have A set of DAG structure task types , For the first This section describes various DAG structure task types. Without loss of generality, it is assumed that different types of tasks have different topological structures or generation locations. Modeled as a quadruple, ,in , This represents a set of subtasks with dependencies. Indicates the first Types of DAG structure tasks The Sub-tasks Indicates the total number of subtasks; , Indicates the first Types of DAG structure tasks The Sub-tasks Indicates a subtask Pointing to subtask The directed edge, This represents the set of directed edges between subtasks. Indicates the location where the task was generated. This indicates that the probability of a task being generated within a given unloading cycle follows a Bernoulli distribution; define an indicator variable. Used to indicate during the task unloading cycle Internal DAG structure task types Whether to generate; for each subtask Directed edge weight Indicates the first Sub-tasks Transmitted to the Sub-tasks The amount of data, , Indicates a subtask Pointing to subtask The directed edge, Subtasks The set of all direct predecessor subtasks; , Indicates a subtask Pointing to subtask The directed edge, Subtasks The set of all direct successor subtasks; introduce a virtual starting subtask with zero computational effort. and a virtual end subtask ; The output weights represent the size of the input data for the task. The input weights represent the size of the task result.

5. The method for collaborative scheduling of DAG structure tasks in a heterogeneous edge computing network according to claim 4, characterized in that, In step S1, the task unloading model is constructed: DAG structure generation task type Users to edge server The uplink transmission rate is ;in, For uplink channel bandwidth, It is a DAG structure task type Uplink transmission power, For noise power, For channel gain between the task and the edge server, The calculation method is as follows , It is the path loss index. , This represents the Euclidean distance between the task and the server. Represents edge server Location coordinates, Indicates the task type of the DAG structure Two-dimensional position coordinates The Euclidean norm is used to calculate the Euclidean distance between two two-dimensional coordinates; similarly, the downlink transmission rate from the edge server to the task is expressed as... , For downlink channel bandwidth, Edge server The transmit power; the external transmission delay for mission offloading to the alliance and returning results is... ,in, , Indicates the task type of the DAG structure To the Alliance uplink transmission rate, Indicates a virtual starting subtask The set of all direct successor subtasks, Indicates a virtual starting subtask direct successor subtask Similarly, the size of the input data, Indicates the virtual termination of the subtask The set of all direct predecessor subtasks, Indicates the virtual termination of the subtask direct precursor sub-task The size of the output data. , Indicates the first Alliance Task types to DAG structure downlink transmission rate; when subtask Sent to the league Edge servers within Execution, and subtasks Precursor mission Redirected to another edge server in the same consortium At that time, the amount of data transmitted between the two edge servers was Therefore, the internal transmission delay is expressed as ;in, , Represents edge server and The Euclidean distance between them Represents edge server Location coordinates, This represents the transmission delay coefficient for transmitting one unit of data per unit distance; if the server and If they are from the same device, the internal transmission delay is zero; define a binary variable. , Used to represent subtasks Was it uninstalled to the edge server? Satisfying constraints Edge server Processing subtasks The required computation delay is expressed as ,in It is a subtask The required computational workload It is an edge server The computational frequency; defining subtasks Readiness time The calculation method is as follows :in, Subtasks Completion time, Subtasks and Transmission latency between edges; each edge server maintains a task waiting queue following a first-come, first-served strategy; subtasks The start execution time is expressed as , Completion time, This indicates that the task is immediately adjacent to the subtask in the waiting queue. Previous subtasks; subtasks On the edge server The completion time is expressed as ;Task The total completion time depends on the completion time of the virtual termination subtask; therefore, the task In the league The total completion time within is expressed as , Indicates the virtual termination of the subtask The completion time.

6. The method for collaborative scheduling of DAG structure tasks in a heterogeneous edge computing network according to claim 5, characterized in that, In step 1, a task preference model is constructed; Users who generate a certain type of DAG structure task always make decisions about whether to offload their tasks to a particular federation based on their preference for edge servers according to that type of DAG structure task; for a given federation DAG structure task type The preference values ​​are constructed in the following way; First, classify the DAG structure task types Treat it as a complete and indivisible task type, ignoring DAG structure task types. Dependencies between internal subtasks; Subsequently, the user will estimate the type of DAG structure task. Uninstall to the Alliance The total completion time is calculated, which consists of the estimated computation time, the estimated internal transmission time, and the estimated external transmission time; therefore, the estimated computation time is expressed as... , where the function This indicates the diminishing marginal returns effect caused by increased computing resources. , and This is an adjustable coefficient; task In the league The estimated internal transmission time is expressed as ;in, Indicates alliance The average distance between edge servers is calculated as follows: , Indicates server and The Euclidean distance between them; therefore, given a coalition structure ,Task Against the alliance The preference value is defined as the estimated total completion time of the task within the alliance. ,Right now In a given alliance structure Next, calculate the total completion time for each task. The preference values ​​for different alliances are listed and arranged in non-decreasing order to form a preference list. ; Step S1.5: Constructing the edge utility model; during the alliance formation phase, edge servers Join the alliance The gains obtained are expressed as , Indicates the first In each alliance The One edge server, It is an edge server The calculation frequency, where the return According to server computing power The proportion of computing power allocated within the alliance. Indicates alliance The estimated total task load is calculated using the following formula: ,in Indicates the task type of the DAG structure The probability of arrival within a single unloading cycle. Indicates alliance The task set, Includes all Tasks that are the preferred consortiums in their preference list, and whose total computational requirements do not exceed the total computational capacity that the consortium can support; defined as ,in, Indicates task In alliance structure The alliance with the highest preference level; edge server Join the alliance The resulting communication cost is expressed as ,in, These are adjustable parameters used to control the impact of distance on communication costs; thus, the edge server... Join the Alliance The obtained historical estimated utility is expressed as During the task scheduling phase, the edge server The actual scheduling utility obtained is defined as , Indicates the first Each task unloading cycle Indicates the first step in the task unloading cycle. At a certain point in time, It is an indicator function, when the condition is met. When satisfied, indicator function The value is 1; taking into account the impact of historical experience and current actual scheduling results on utility evaluation, the function is defined. This represents the fusion function of historically estimated utility and actual cooperative utility. 。 7. The method for collaborative scheduling of DAG structure tasks in a heterogeneous edge computing network according to claim 6, characterized in that, In step 1, the optimization objective of the DAG structure task collaborative scheduling problem in a heterogeneous edge computing network is defined as maximizing the utility of each edge server within the lifecycle of the consortium structure; specifically, maximizing the utility of each edge server within the lifecycle of the consortium structure means: The input to the DAG-structured task cooperative scheduling method in a heterogeneous edge computing network is defined as: the set of edge servers. DAG structure task type set Edge servers computing power and position coordinates DAG structure task Subtask set The set of directed edges between subtasks Location where the task is generated The probability of a task being generated within a given unloading cycle. Transmission delay coefficient for transmitting one bit of data per unit distance The lifecycle of the alliance structure, and the task unloading cycle; The output of the DAG structure task cooperative scheduling method in heterogeneous edge computing networks is defined as: having Stable alliance structure of each alliance , Indicates alliance structure The first in One alliance, The value range is 1 to Unloading decision for DAG structure tasks ; The optimization objective of the DAG structure task cooperative scheduling method in heterogeneous edge computing networks is as follows: ； ， ； ； ； Among them, constraint C1 means that a subtask can be offloaded to at most one edge server for computation, constraint C2 means that the total completion time of the task does not exceed the task offloading cycle, and constraint C3 means that the subtask is completely offloaded to the edge server for computation.

8. The method for collaborative scheduling of DAG structure tasks in a heterogeneous edge computing network according to claim 7, characterized in that, Step S2.1 includes the following steps: Step S2.1.1: Alliance Structure Initialization; First, each edge server forms an independent alliance, resulting in the initial alliance structure. The current alliance structure Updated to According to the formula Calculate each task under the current alliance structure Initial preference list ; Given a federation structure, the expression for the edge server's preference function for the federation is: The result of this expression is used to characterize the utility of edge servers joining a consortium, where, Indicates alliance structure Next Alliance Edge servers in Historical estimated utility, case 1 indicates that the alliance admission conditions are met and the current alliance structure was not formed before this, i.e., it satisfies and ,in Indicates alliance structure Next Alliance Edge servers in Historical estimated utility Indicates alliance structure Lower edge server From the Alliance After leaving the edge server Historical estimated utility Represents edge server It is used to store a historical set of alliances that have been joined in the past. This indicates that only edge servers are included. The set, case 2 indicates that the alliance structure size is 1, that is... Each edge server calculates the initial utility for the given federation structure based on the task's preference list. ; Step S2.1.2: Alliance Preference Assessment; Based on the current alliance structure, in the current alliance structure Based on this, the alliance Each edge server within The process will involve assessing the potential benefits of joining other alliances and constructing appropriate temporary alliance structures. ,in , Indicates server Join the Alliance The new alliance after that, , Indicates server Leave the league The remaining alliance; Step S2.1.3: Alliance structure adjustment; if the edge server Join another league The utility gained is greater than that of an edge server. Join the alliance If the utility of the alliance is satisfied, then the attempt to join the alliance is fulfilled. Time Edge Server Will withdraw from its current alliance , Represents edge server Join the alliance The subsequent alliance structure, Represents edge server Join another league Back Edge Server The effect, Indicates the current alliance structure Lower edge server Join the alliance The utility of this, and updating the alliance structure, to obtain ; Step S2.1.4: Repeat steps S2.1.2 and S2.1.3 until no edge server can further improve its utility by joining other alliances, i.e., the current alliance division is complete. A stable state has been reached.

9. A method for collaborative scheduling of DAG structure tasks in a heterogeneous edge computing network according to claim 8, characterized in that, Step S2.2 includes the following steps: Step S2.2.1: Obtain the set of tasks to be scheduled and construct a priority queue; after a stable alliance structure is formed on the edge servers, each alliance obtains the set of tasks it has competed for in each task unloading cycle. In this step, all DAG tasks within the alliance are treated as independent tasks and their priorities are determined separately. First, calculate the priority index for each DAG task. It is defined as the ratio of the total computational load of the task to the total output data volume. ,in, This represents the total computational cost of the DAG task. The total amount of data transmitted for the task; based on Sort all DAG tasks to form a task queue arranged from high to low priority; Step S2.2.2: Calculate the rank value of subtasks based on the topology; for each DAG task obtained in step S2.2.1, calculate the rank value of its internal subtasks according to its topology; using the HEFT algorithm, calculate the rank value of each subtask in the DAG structure task based on its average computation time. With average communication time Perform recursive calculations to obtain the level values ​​of the subtasks. ,in, This represents the average computation time of the subtask. This represents the average communication time of the subtask; The set of all direct successor subtasks of the subtask; express Each set of direct successor subtasks in the set; all the resulting subtasks are sorted from highest to lowest priority value to form a subtask priority queue; Step S2.2.3: Select the optimal edge server to schedule subtasks based on completion time; execute scheduling for each sorted subtask sequentially, targeting the current subtask. It collects the available computing resource status of all edge servers within the alliance in real time and based on the formula Calculate the earliest completion time and select the edge server that can complete the subtask earliest. Execute the task; for each subtask scheduled, immediately select the optimal server dynamically based on the real-time resource status; the scheduling process continues until all subtasks of all DAG tasks have been scheduled.

10. A computer device comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes a computer program to implement the DAG structure task collaborative scheduling method in a heterogeneous edge computing network as described in any one of claims 1-9.

Images