An edge-assisted high-efficiency cross-domain task scheduling method

Abstract

The present invention discloses an edge-assisted high-efficiency cross-domain task scheduling method, which continuously processes user batch task information arriving in real time at the edge server, the current remaining physical resource situation of the edge server, the carbon intensity of the current data centers distributed across geographical ranges accessing the power grid, and the network bandwidth information connected between the data center and the edge server, and ensures that the instantaneous decision does not violate the user service quality demand constraints in the long-term operation through the Lyapunov optimization framework, makes online task scheduling decisions, reduces the transmission cost through resource-constrained edge-assisted task execution, and makes full use of the low-carbon characteristics of edge nodes and cross-domain cloud data centers, and finally achieves the goal of overall high-efficiency scheduling. The present invention guarantees the average performance target of the system and minimizes the generation of carbon emissions.

Description

Technical Field

[0001] The present invention relates to the field of edge computing technology, and in particular to an edge-assisted high-efficiency cross-domain task scheduling method. Background Art

[0002] Edge computing moves computing and storage to the edge of the network closer to the user, uses edge servers or edge computing devices to perform tasks, reduces latency and bandwidth requirements, and improves user privacy protection capabilities. Due to its flexible computing power deployment and wide geographical distribution, the edge computing paradigm can obtain more flexible, diverse, greener and cleaner energy supply, and bring new energy-saving and carbon-reduction methods such as power consumption regulation, energy storage and energy harvesting to green computing. Research on green edge computing has shown that low-carbon edge nodes powered by green energy are completely necessary and feasible. These green nodes can be powered by renewable energy such as solar or wind power, and can provide a certain amount of computing power to support the deployment of computing tasks under the constraints of limited resources, thereby sharing the computing and traffic pressure on the user side and the cloud network side, while effectively reducing system energy consumption and carbon emissions, and achieving high-efficiency computing.

[0003] Since the number and distribution of tasks arriving at the edge are relatively random, it is very likely that the auxiliary edge computing power will be unable to support the local execution of all tasks. In order to avoid task execution timeout or loss and violate the user service agreement requirements, some tasks still need to be uploaded to the cloud for processing. For tasks uploaded to the cloud for processing, it is also necessary to consider practical methods to reduce the energy consumption of their execution process. For clouds composed of multiple data centers with large geographical distribution differences, cross-domain task scheduling has been proven to be an effective means of energy consumption control or cost reduction. By scheduling tasks across domains to data centers with high energy efficiency and low carbon emissions, the energy consumption and carbon dioxide emissions generated by the tasks can be fully reduced at the cost of equal or minimal latency. Summary of the invention

[0004] The purpose of the present invention is to provide an edge-assisted high-efficiency cross-domain task scheduling method to solve the problems raised in the above-mentioned background technology.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0006] An edge-assisted high-efficiency cross-domain task scheduling method, the steps of which include:

[0007] At the beginning of a time slot, the system parameter information of the current time slot is obtained, and the obtained system parameter information of the current time slot is uploaded to the edge node scheduler;

[0008] The edge node scheduler builds a "carbon emission-computing power" mapping relationship of (task computing amount / server computing speed)*server power consumption*power utilization*carbon emission factor=carbon emissions generated by a single task on a specific computing power, based on the system parameter information of the current time slot and for tasks with different computing amounts, by using the computing speed and power information of different servers, combined with the power utilization rate and carbon emission factor of the cloud data center;

[0009] According to the "carbon emission-computing power" mapping relationship, the constraints of minimizing the overall carbon emission of the cloud-edge system as the long-term optimization goal and the user task service time as the long-term optimization goal are determined;

[0010] Based on the optimization goal and constraints, the Lyapunov queue value of the current time slot is used to establish the instantaneous optimization sub-goal in the current time slot, and the approximate optimal strategy is obtained by solving it through an approximate algorithm;

[0011] The edge node scheduler schedules the tasks arriving at the edge server according to the approximate optimal strategy obtained, and updates the Lyapunov queue value before the start of the next time slot according to the execution time of the tasks in this time slot.

[0012] According to the above technical solution, the system parameter information of the current time slot includes the number of tasks arriving at the edge node, the task calculation amount, the task data amount, the edge node calculation rate, the regional power grid carbon intensity data of each cross-regionally distributed cloud data center, the power utilization rate, the physical server power consumption and its calculation speed, and the network bandwidth and physical layer energy consumption and carbon emission rate information connecting the edge node and the cloud data center.

[0013] According to the above technical solution, the long-term optimization goal is:

[0014]

[0015] Where T represents the total number of time slots, E represents the number of edge servers, and J e,t represents the number of jobs that arrive at edge node e at time slot t, D represents the number of data centers, and x e,j,d,t is the decision of whether task j arriving at edge computing node e in time slot t should be transmitted to cloud data center d, C e,j,d,t is the carbon emission expected to be generated when task j arriving at edge computing node e in time slot t is transmitted to cloud data center d, that is, C e,j,d,t is the calculation carbon of the task plus the transmission carbon, a e,j,t represents the computational effort of task j arriving at edge computing node e in time slot t, s e,j,t Indicates the amount of data required for the task to be transmitted, v d,t represents the computing speed of a server (assuming the servers are homogeneous) in data center d during time slot t, P d,tIndicates its power, PUE d,t Indicates the power utilization rate, ci d,t represents the average carbon intensity factor of the power grid, cr e,d,t It represents the carbon emissions generated by transmitting each unit of data on the link between edge node e and cloud data center d.

[0016] According to the above technical solution, the constraints of the long-term optimization goal include:

[0017] The average completion time constraint for all tasks is:

[0018]

[0019] in,

[0020] The bandwidth constraints for each physical link connecting the edge node and the cloud data center are:

[0021]

[0022] Constraints on the decision variables for each task:

[0023]

[0024] In the formula, T0 represents the system constraint on the average delay. The task delay is calculated by calculating the transmission delay and calculation time of each task. The T0 value can be set by the system administrator or automatically set according to the task requirements. t represents the average delay of all tasks executed in time slot t. The task delay is calculated by calculating the transmission delay and calculation time of each task. e,j,t represents the computational effort of task j arriving at edge computing node e in time slot t, s e,j,t Indicates the amount of data required for the task to be transmitted, v d,t represents the computing speed of a server (assuming the servers are homogeneous) in data center d during time slot t, b e,j,d,t It is represented by the bandwidth allocated between the task j arriving at the edge computing node e and the cloud data center d in time slot t, B e,d,t Indicates the upper limit of the link bandwidth between edge node e and cloud data center d.

[0025] According to the above technical solution, the approximate optimal strategy solving step includes:

[0026] The long-term optimization goal is decomposed into each time slot through the Lyapunov optimization framework, and the long-term optimization goal is converted into an instantaneous optimization sub-goal, and the constraints of the instantaneous optimization sub-goal are determined;

[0027] At the beginning of each time slot, a random algorithm based on the Markov approximation process is used to solve the approximate optimal strategy. The specific steps include: selecting a default scheduling strategy at the beginning, randomly changing the scheduling strategy of a task in each iteration, and calculating the target value of the instantaneous optimization sub-goal of the current time slot under this scheduling strategy, calculating the difference between this target value and the target value before the change, and calculating the probability of changing the original strategy to this new strategy based on this difference Δ The state is transferred according to the calculated probability η, and the currently explored optimal solution is recorded in each iteration. When the set maximum number of iterations is reached or the state has not changed after multiple iterations, the iteration ends and the previously recorded optimal solution is output;

[0028] Where r is a parameter used to adjust the effect of the difference on the probability (that is, used to adjust the algorithm's tendency to change to a new state to explore more solution space).

[0029] According to the above technical solution, the instantaneous optimization sub-objective is:

[0030]

[0031] Where V is a parameter used to adjust the trade-off between carbon emissions and delay constraints, Q t is the value of the Lyapunov queue at time slot t, which is 0 at the beginning, E represents the number of edge servers, and J e,t represents the number of jobs that arrive at edge node e at time slot t, D represents the number of data centers, and x e,j,d,t is the decision of whether task j arriving at edge computing node e in time slot t should be transmitted to cloud data center d, C e,j,d,t is the estimated carbon emission generated when task j arriving at edge computing node e in time slot t is transmitted to cloud data center d.

[0032] According to the above technical solution, the constraint conditions of the instantaneous optimization sub-goal are consistent with the constraint conditions of the long-term optimization goal.

[0033] According to the above technical solution, the Lyapunov queue value Q is updated before the next time slot starts. t+1 :

[0034] Q t+1 =max(0,Q t +T t -T0);

[0035] In the formula, Q t is the value of the Lyapunov queue at time slot t, T0 represents the system constraint on average delay, T t It represents the average execution delay of all tasks in time slot t.

[0036] Compared with the prior art, the beneficial effects achieved by the present invention are as follows: the present invention makes online task scheduling decisions based on the user batch task information arriving in real time at the edge server, the current remaining physical resource status of the edge server, the carbon intensity of the grid access of the current data centers distributed across geographical ranges, and the network bandwidth information connected between the data center and the edge server. It reduces the transmission cost by executing resource-limited edge auxiliary tasks, and makes full use of the low-carbon characteristics of edge nodes and cross-domain cloud data centers, ultimately achieving the goal of overall high-efficiency scheduling. In the case where the system parameters such as the task status and grid carbon intensity at future moments cannot be accurately predicted, the present invention can ensure that instantaneous decisions do not violate the user service quality demand constraints in the long-term operation through the Lyapunov optimization framework, while ensuring the system average performance target and effectively reducing the generation of carbon emissions. The algorithm proposed in the present invention has theoretical performance guarantees, and its performance advantages can be verified through simulation experiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0038] Figure 1 This is a flowchart of the steps of an edge-assisted high-efficiency cross-domain task scheduling method according to an embodiment;

[0039] Figure 2 is a structural diagram of an edge-assisted high-efficiency cross-domain task scheduling system of an embodiment;

[0040] Figure 3 Schematic diagram of carbon emissions of the present invention and several comparative algorithms under different task quantity scales in the embodiment;

[0041] Figure 4 It is a schematic diagram of the average task completion time of the present invention and several comparative algorithms under different task quantity scales;

[0042] Figure 5 is the influence of system parameter r on the experimental results;

[0043] Figure 6 is the influence of system parameter V on the experimental results;

[0044] Figure 7 are the system parameters T0 and R max The impact on the experimental results. DETAILED DESCRIPTION

[0045] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0046] Example 1

[0047] The present invention provides a technical solution, an edge-assisted high-efficiency cross-domain task scheduling method, the steps of which include:

[0048] S1. At the beginning of a time slot, obtain the system parameter information of the current time slot, and upload the obtained system parameter information of the current time slot to the edge node scheduler; the system parameter information of the current time slot includes the number of tasks arriving at the edge node, the task calculation amount, the task data amount, the edge node calculation rate, the regional power grid carbon intensity data of each cross-regionally distributed cloud data center, the power utilization rate, the power consumption of the physical server and its calculation speed, and the network bandwidth and physical layer energy consumption and carbon emission rate information between the edge node and the cloud data center.

[0049] S2. The edge node scheduler builds a "carbon emission-computing power" mapping relationship for the task based on the system parameter information of the current time slot. The specific steps are as follows:

[0050] At the beginning of each time slot (the length of the time slot is fixed as a system configuration), the scheduler obtains the current power supply components of each regional data center and calculates the carbon emission factor of the power components of each data center, and calculates the carbon intensity of the power of each data center;

[0051] For tasks with different computing amounts, by using the computing speed and power information of different servers, combined with the power utilization rate and carbon emission factor of the cloud data center, a "carbon emission-computing power" mapping relationship is constructed: (task computing amount / server computing speed)*server power consumption*power utilization*carbon emission factor=carbon emissions generated by a single task on a specific computing power;

[0052] S3. According to the “carbon emission-computing power” mapping relationship, determine the constraints of minimizing the overall carbon emission of the cloud-edge system as the long-term optimization goal and the user task service time as the long-term optimization goal;

[0053] Long-term optimization goals:

[0054]

[0055] Where T represents the total number of time slots, E represents the number of edge servers, and J e,t represents the number of jobs that arrive at edge node e at time slot t, D represents the number of data centers, and xe,j,d,t is the decision of whether task j arriving at edge computing node e in time slot t should be transmitted to cloud data center d, C e,j,d,t is the estimated carbon emission generated when task j arriving at edge computing node e in time slot t is transmitted to cloud data center d.

[0056] The constraints of the long-term optimization goal include:

[0057] The average completion time constraint for all tasks is:

[0058]

[0059] The bandwidth constraints for each physical link connecting the edge node and the cloud data center are:

[0060]

[0061] Constraints on the decision variables for each task:

[0062]

[0063] Where, T t represents the average delay of all tasks executed in time slot t, T0 represents the system constraint on the average delay, and b e,j,d,t It is represented by the bandwidth allocated between the task j arriving at the edge computing node e and the cloud data center d in time slot t, B e,d,t Indicates the upper limit of the link bandwidth between edge node e and cloud data center d.

[0064] S4: Based on the optimization goal and constraints, the Lyapunov queue value of the current time slot is used to establish the instantaneous optimization sub-goal in the current time slot, and the approximate optimal strategy is obtained by solving the approximate algorithm. The specific steps include:

[0065] The long-term optimization goal is decomposed into each time slot through the Lyapunov optimization framework, and the long-term optimization goal is converted into an instantaneous optimization sub-goal, and the constraints of the instantaneous optimization sub-goal are determined; the constraints of the instantaneous optimization sub-goal are consistent with the constraints of the long-term optimization goal;

[0066] At the beginning of each time slot, a random algorithm based on the Markov approximation process is used to solve the approximate optimal strategy. Specifically, a default scheduling strategy is selected at the beginning, and the scheduling strategy of a task is randomly changed in each iteration. The target value of the instantaneous optimization sub-goal of the current time slot under this scheduling strategy is calculated, and the difference between this target value and the target value before the change is calculated. The probability η of changing the original strategy to the new strategy is calculated based on this difference Δ, and the state is transferred according to the calculated probability η. The currently explored optimal solution is recorded in each iteration, and the iteration is terminated when the set maximum number of iterations is reached or the state is not changed in multiple iterations, and the previously recorded optimal solution is output.

[0067] Among them, the instantaneous optimization sub-goal is:

[0068]

[0069] Where V is a parameter used to adjust the trade-off between carbon emissions and delay constraints, Q t is the value of the Lyapunov queue at time slot t, which is 0 at the beginning, E represents the number of edge servers, and J e,t represents the number of jobs that arrive at edge node e at time slot t, D represents the number of data centers, and x e,j,d,t is the decision of whether task j arriving at edge computing node e in time slot t should be transmitted to cloud data center d, C e,j,d,t is the estimated carbon emission generated when task j arriving at edge computing node e in time slot t is transmitted to cloud data center d.

[0070] S5. The edge node scheduler schedules the tasks arriving at the edge server according to the approximate optimal strategy obtained, and updates the Lyapunov queue value before the start of the next time slot according to the execution time of the tasks in this time slot, specifically:

[0071] In the time slot, the scheduler sends the approximate optimal strategy to the edge node, and the edge node schedules the task using the approximate optimal strategy, and uploads the task to the cloud data center, waits for the processing result of the cloud data center, or directly starts to execute it locally;

[0072] During the time slot, the scheduler collects the local running results of the edge server or the data results transmitted back by the cloud data center through the backbone network, sends them back to the user, and records the actual execution time of these tasks;

[0073] At the end of the time slot, update the Lyapunov queue value Q t+1 :

[0074] Q t+1 =max(0,Q t +T t -T0);

[0075] In the formula, Q t is the value of the Lyapunov queue at time slot t, T0 represents the system constraint on average delay, T t It represents the average execution delay of all tasks in time slot t.

[0076] Example 2

[0077] In a task execution system that includes edge servers and multiple geographically distributed cloud data centers (also called an edge-assisted high-efficiency cross-domain task scheduling system such as Figure 2 ) is an edge-assisted high-efficiency task scheduling method, the system includes: edge computing nodes, cloud data centers, schedulers, and networks connecting each cloud data center with edge computing nodes, each user with edge computing nodes, the user's diverse tasks can be transmitted to the edge server through the edge gateway via the edge network, and the edge node can upload the task to the data center through the backbone network; the resources on each edge computing node are limited, while the cloud data center can be approximately considered unlimited due to its more computing resources; the scheduler module is run on the control node, and at the beginning of the time slot, the current power supply components of each regional data center can be obtained and the carbon emission factor of each data center power, the current resource status of each edge node and cloud, the computing capacity and power information can be calculated, and the task parameters arriving at the edge node can be obtained, including its computing amount (measured by the number of instructions), the amount of data required for transmission (measured by bits), the delay requirement and the bandwidth requirement, etc., and the decision on the number of tasks to be processed by each edge computing node and the cloud data center is made online; for tasks that need to be uploaded to the data center, the upload is transmitted to the data center via the backbone network and the core switch, and after the data center calculates, it is transmitted back to the edge server through the return line, and is transmitted back to the user through the edge network like the tasks processed on the edge node. Users of the multiple edge gateways connected to each edge computing node can be mobile devices, laptops, etc. connected to the edge computing device.

[0078] In this system, the scheduler collects system parameter information uploaded by edge gateways, cloud data centers, and edge computing nodes, and interacts with edge computing nodes periodically. When it is impossible to accurately know user task information at future moments and the carbon intensity of the data center power grid, the scheduler calculates the Lyapunov drift function by combining the actual execution time of the user task in each time slot with the deviation value of the delay constraint, and guides the online scheduler to make a trade-off between efficiency and user service quality, so as to achieve the optimization goal of maximizing the efficiency of the cloud-edge system while meeting the user service quality.

[0079] Example 3

[0080] In the same scenario, the All-Cloud algorithm (ACloud): a time-delay greedy algorithm in which all tasks are uploaded to the nearest data center, Random Load Balance (RLB): a load balancing algorithm for random task allocation, Carbon-optimal Markov (CoM): a Markov approximation algorithm that uses carbon emissions as the target value without delay constraints, and Delay-optimal Markov (DoM): a Markov approximation algorithm that uses delay as the target value without carbon emission constraints, and an edge-assisted high-efficiency cross-domain task scheduling method (All Cloud) of the present invention are used respectively to compare the carbon emissions and overall task completion time generated by several algorithms for different task scales under specific data center and edge node configurations. Figure 3 Schematic diagram of carbon emissions of the present invention and several comparative algorithms under different task quantity scales in the embodiment; Figure 4 Schematic diagram of average task completion time of the present invention and several comparison algorithms under different task quantity scales in the embodiment.

[0081] from Figure 3 and Figure 4 It can be observed that compared with the traditional cloud computing paradigm, the edge-assisted high-efficiency cross-domain task scheduling method proposed in the present invention effectively reduces time and carbon emissions, and compared with the random load balancing algorithm, the present invention also has significant advantages. CoM and DoM can be regarded as the two extremes of the energy efficiency-time trade-off explored by the present invention. It can be observed that although these two algorithms perform best in one of the data, the performance of the other is unacceptable. On the contrary, the present invention effectively balances the relationship between the two and fully realizes the goal of maximizing the energy efficiency of the data center under the premise of meeting the user service quality.

[0082] Example 4: Several parameter settings in the system during actual deployment

[0083] Figure 5 It shows the process that the optimization target value gradually approaches the minimum value as the number of code iterations increases, where the parameter r of different curves is Figure 3 The system parameters set in the algorithm affect the probability η of state change in the Markov approximation algorithm. The larger the parameter r, the greater the probability η, which means that the system state is easier to change and the algorithm has a larger space to explore, but it is also more difficult to converge to a fixed value, and sometimes even misses the minimum value; but if the parameter r is too small, the system state is difficult to change, and it may be difficult to find the optimal solution. Figure 5 It can be found that the algorithm works best when the parameter r takes an intermediate value of 0.1.

[0084] V affects the trade-off between carbon emissions and Lyapunov queues. The size of the Lyapunov queue actually represents the deviation from the delay constraint. Therefore, V controls the trade-off between carbon emissions and delay. When V is larger, the trade-off is more biased towards carbon emissions. Therefore, the Lyapunov queue will be larger, indicating that the delay deviates more from the constraint, which is consistent with Figure 6 actual experimental results.

[0085] like Figure 7 As shown in Figure 1, T0 also directly controls the trade-off between latency and carbon emissions. When T0 is small, it means that the latency constraint is tight and it is difficult to optimize carbon emissions. When T0 is large, it means that the latency constraint is loose and carbon emissions can be optimized to a greater extent. max is the iteration number parameter. Obviously, as the number of iterations increases, the approximate solution obtained by the system will gradually tend to the theoretical optimal solution, but it will lead to more algorithm execution time. The degree of proximity to the optimal solution and the algorithm execution time are also a trade-off.

[0086] In summary, it can be seen that the setting of the system parameters in the present invention can effectively control the system optimization goal, and is convenient for users to adjust by themselves to achieve energy efficiency-time trade-off.

[0087] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0088] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An edge-assisted high-efficiency cross-domain task scheduling method, characterized in that: The steps include: At the beginning of a time slot, the system parameter information of the current time slot is obtained, and the obtained system parameter information of the current time slot is uploaded to the edge node scheduler; The edge node scheduler builds a "carbon emission-computing power" mapping relationship for tasks based on the system parameter information of the current time slot. Specifically, according to tasks with different computing amounts, by using the computing speed and power information of different servers, combined with the power utilization rate and carbon emission factor of the cloud data center, a "carbon emission-computing power" mapping relationship is built: task computing amount / server computing speed*server power consumption*power utilization*carbon emission factor=carbon emissions generated by a single task on a specific computing power. According to the "carbon emission-computing power" mapping relationship, the long-term optimization goal is to minimize the overall carbon emission of the cloud-edge system and the user task service time is determined as the constraint condition of the long-term optimization goal; the long-term optimization goal is: Where T represents the total number of time slots, E represents the number of edge servers, and J e,t represents the number of jobs that arrive at edge node e at time slot t, D represents the number of data centers, and x e,j,d,t is the decision of whether task j arriving at edge computing node e in time slot t should be transmitted to cloud data center d, C e,j,d,t is the carbon emission expected to be generated when task j arriving at edge computing node e in time slot t is transmitted to cloud data center d, that is, C e,j,d,t is the calculation carbon of the task plus the transmission carbon, a e,j,t represents the computational effort of task j arriving at edge computing node e in time slot t, s e,j,t Indicates the amount of data required for the task to be transmitted, v d,t represents the computing speed of a server in data center d in time slot t, P d,t Indicates its power, PUE d,t Indicates the power utilization rate, ci d,t represents the average carbon intensity factor of the power grid, cr e,d,t It represents the carbon emission generated by transmitting each unit of data on the link between edge node e and cloud data center d; The constraints of the long-term optimization goal include: The average completion time constraint for all tasks is: in, The bandwidth constraints for each physical link connecting the edge node and the cloud data center are: Constraints on the decision variables for each task: In the formula, T0 represents the system constraint on the average delay. The task delay is calculated by calculating the transmission delay and calculation time of each task. The T0 value can be set by the system administrator or automatically set according to the task requirements. t represents the average delay of all tasks executed in time slot t. The task delay is calculated by calculating the transmission delay and calculation time of each task. e,j,t represents the computational effort of task j arriving at edge computing node e in time slot t, s e,j,t Indicates the amount of data required for the task to be transmitted, v d,t represents the computing speed of a server in data center d in time slot t, b e,j,d,t It is represented by the bandwidth allocated between the task j arriving at the edge computing node e and the cloud data center d in time slot t, B e,d,t Indicates the upper limit of the link bandwidth between edge node e and cloud data center d; Based on the optimization goal and the constraint condition, the instantaneous optimization sub-goal in the current time slot is established by using the Lyapunov queue value of the current time slot, and the approximate optimal strategy is obtained by solving the approximate optimal strategy through the approximate algorithm; the approximate optimal strategy solving step comprises: decomposing the long-term optimization goal into each time slot through the Lyapunov optimization framework, converting the long-term optimization goal into the instantaneous optimization sub-goal, and determining the constraint condition of the instantaneous optimization sub-goal; at the beginning of each time slot, a random algorithm based on the Markov approximation process is used to solve and obtain the approximate optimal strategy; The edge node scheduler schedules the tasks arriving at the edge server according to the approximate optimal strategy obtained, and updates the Lyapunov queue value before the start of the next time slot according to the execution time of the tasks in this time slot.

2. According to claim 1, an edge-assisted high-efficiency cross-domain task scheduling method is characterized in that: The system parameter information of the current time slot includes the number of tasks arriving at the edge node, the task calculation amount, the task data amount, the edge node calculation rate, the regional power grid carbon intensity data of each cross-regionally distributed cloud data center, the power utilization rate, the physical server power consumption and its calculation speed, and the network bandwidth and physical layer energy consumption and carbon emission rate information between the edge node and the cloud data center.

3. The edge-assisted high-efficiency cross-domain task scheduling method according to claim 1 is characterized in that: The instantaneous optimization sub-goal is: Where V is a parameter used to adjust the trade-off between carbon emissions and delay constraints, Q t is the value of the Lyapunov queue at time slot t. The Lyapunov queue is 0 at the beginning. E represents the number of edge servers. J e,t represents the number of jobs that arrive at edge node e at time slot t, D represents the number of data centers, and x e,j,d,t is the decision of whether task j arriving at edge computing node e in time slot t should be transmitted to cloud data center d, C e,j,d,t is the estimated carbon emissions generated by task j arriving at edge computing node e in time slot t when it is transmitted to cloud data center d, T t It represents the average execution delay of all tasks in time slot t.

4. The edge-assisted high-efficiency cross-domain task scheduling method according to claim 1 is characterized in that: The constraints of the instantaneous optimization sub-objective are consistent with the constraints of the long-term optimization objective.

5. The edge-assisted high-efficiency cross-domain task scheduling method according to claim 1 is characterized in that: Update the Lyapunov queue value Q before the next time slot starts t+1 : Q t+1 =max(0,Q t +T t -T0); In the formula, Q t is the value of the Lyapunov queue at time slot t, T0 represents the system constraint on average delay, T t It represents the average execution delay of all tasks in time slot t.

Images