Computing power network task scheduling method and device, storage medium and program product

Abstract

The invention provides a computing power network task scheduling method and device, a storage medium and a program product, and relates to the technical field of computers.The method comprises the steps that a task scheduling optimization problem is obtained by defining an objective function between network energy consumption of a computing power network and network carbon emission of the computing power network and constraint conditions corresponding to the objective function; the optimization objective of the objective function is to reduce the network energy consumption and the network carbon emission by optimizing the computing power node, the forwarding path and the forwarding time corresponding to the computing power task; solving the task scheduling optimization problem through a preset optimization algorithm to obtain a task scheduling method of the computing power network; the task scheduling method is used for indicating a forwarding path of a computing power task in a computing power network, a computing power node for executing the computing power task and forwarding time corresponding to the computing power node; according to the method, the problems of relatively low expandability and management efficiency of task allocation can be solved; and the task allocation flexibility and management efficiency are improved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a computing network task scheduling method, device, storage medium and program product. Background Art

[0002] With the rapid development of information technology, the number of networked devices in intelligent application systems has increased dramatically, generating unprecedented big data. These data are characterized by large volume, variety, and high speed, which places higher demands on computing and network capabilities. Although the cloud-edge architecture improves communication and computing efficiency, resource allocation is mainly limited to a single domain, resulting in reduced resource utilization and limiting the development of data-intensive applications. The emergence of computing power networks (CPNs) solves these problems. By connecting and coordinating heterogeneous computing resources in the cloud, edge, and endpoints, CPNs can effectively integrate computing and network resources and perform computing task scheduling.

[0003] At present, traditional computing task scheduling methods include: task allocation and management methods based on deep reinforcement learning and software-defined networks; using deep reinforcement learning algorithms to interact with the environment to autonomously learn the best task allocation strategy to achieve optimization of energy consumption and latency; and realizing dynamic task scheduling and resource management through the centralized control characteristics of software-defined networks.

[0004] However, the increasing dynamic and unpredictable traffic in the Internet of Things makes computing task scheduling complicated, especially in real-time scheduling, which requires a highly adaptive mechanism. In addition, fog nodes usually have limited resources and are mostly battery-powered, with limited computing and energy resources, which brings challenges to task scheduling and energy management. Based on this, in a heterogeneous Internet of Things environment, it becomes very difficult to balance the relationship between network latency, scheduling accuracy, energy consumption, and response time, and there are problems with the scalability of task allocation and low management efficiency. Summary of the invention

[0005] In view of this, an embodiment of the present invention provides a computing power network task scheduling method, device, storage medium and program product to eliminate or improve one or more defects existing in the prior art.

[0006] One aspect of the present invention provides a computing power network task scheduling method, the method comprising the following steps:

[0007] Define the objective function between the network energy consumption of the computing power network and the network carbon emissions of the computing power network and the constraints corresponding to the objective function to obtain the task scheduling optimization problem; the optimization goal of the objective function is to reduce the network energy consumption and network carbon emissions by optimizing the computing power nodes, forwarding paths and forwarding time corresponding to the computing power tasks; the constraints include delay constraints, resource constraints, computing power node decision constraints and forwarding path decision constraints;

[0008] The task scheduling optimization problem is solved by a preset optimization algorithm to obtain a task scheduling method for the computing power network; the task scheduling method is used to indicate the forwarding path of the computing power task in the computing power network, the computing power node that executes the computing power task, and the forwarding time corresponding to the computing power node.

[0009] In some embodiments of the present invention, the delay constraint refers to a constraint condition constructed based on the relationship between the forwarding time and processing delay corresponding to the computing task and the tolerable delay of the computing network;

[0010] Resource constraints include constraints based on the relationship between the current load and the preset load, constraints based on the relationship between network energy consumption and the preset total energy, and constraints based on the relationship between the amount of cached data corresponding to the computing power node and the preset cache resources.

[0011] In some embodiments of the present invention, the computing power task includes a first computing power task set received in the current time slot and a second computing power task set received before the current time slot; the network energy consumption includes the transmission energy consumption and execution energy consumption corresponding to each computing power task in the first computing power task set, and the transmission energy consumption and execution energy consumption corresponding to each computing power task in the second computing power task set.

[0012] In some embodiments of the present invention, the processing delay includes the task execution delay corresponding to the computing task, the sum of the queuing delay and the task transmission delay; the current load includes the current computing network load and the current computing node load.

[0013] In some embodiments of the present invention, a task scheduling optimization problem is solved by a preset optimization algorithm to obtain a task scheduling method for a computing power network, including:

[0014] Initialize the current population; the current population includes several solutions;

[0015] Based on the current population, the objective function is calculated to obtain the network energy consumption and network carbon emissions, and based on the constraints, the constraint violation corresponding to each solution is obtained;

[0016] Based on the violation of constraint conditions, each solution in the current population is non-dominated and sorted to obtain the non-dominated sorting result;

[0017] Determine the congestion distance corresponding to each solution based on network energy consumption and network carbon emissions;

[0018] Based on the non-dominated sorting results and the crowding distance corresponding to each solution, candidate solutions are determined in the current population;

[0019] When the algorithm converges, the candidate solution is determined as the task scheduling method.

[0020] In some embodiments of the present invention, when the algorithm does not converge, the method further includes:

[0021] Perform crossover and mutation processing on candidate solutions to generate offspring populations;

[0022] After merging the current population with the offspring population, iterate the current population;

[0023] The execution is based on the current population, calculating the objective function and obtaining the constraint violation corresponding to each solution based on the constraint conditions.

[0024] In some embodiments of the present invention, before performing crossover and mutation processing on candidate solutions to generate a child population, the method further includes: dynamically adjusting the crossover probability and mutation probability corresponding to the current population based on network energy consumption and network carbon emissions.

[0025] Another aspect of the present invention provides a computing power network task scheduling device, including a processor, a memory, and a computer program / instruction stored in the memory, wherein the processor is used to execute the computer program / instruction, and when the computer program / instruction is executed, the device implements the steps of the aforementioned computing power network task scheduling method.

[0026] Another aspect of the present invention provides a computer-readable storage medium having a computer program / instruction stored thereon, which, when executed by a processor, implements the steps of the aforementioned computing power network task scheduling method.

[0027] Another aspect of the present invention provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the aforementioned computing power network task scheduling method.

[0028] The computing power network task scheduling method, device, storage medium and program product of the present invention can solve the problems of low scalability and management efficiency of task allocation; deeply integrate the computing power of computing power nodes and network resources, including energy consumption and cache resources, etc. through the computing power network task scheduling device; by defining the optimization goal as minimizing the total energy consumption and carbon emissions of the computing power network, while meeting the constraints such as tolerable delay, processing delay and load balancing of the task, combined with dynamic network load and delay requirements, a highly adaptable scheduling strategy is provided for the task allocation of big data applications to achieve efficient scheduling of computing power tasks, optimize the utilization of renewable energy, improve the overall operating efficiency of the computing power network, and at the same time improve the flexibility and management efficiency of task allocation, ensuring that the computing power network can still operate efficiently and stably under the conditions of traffic growth and resource constraints.

[0029] Additional advantages, purposes, and features of the present invention will be described in part in the following description, and will become apparent to those skilled in the art after studying the following, or may be learned from the practice of the present invention. The purposes and other advantages of the present invention may be achieved and obtained by the structures specifically indicated in the specification and the accompanying drawings.

[0030] Those skilled in the art will appreciate that the objectives and advantages that can be achieved with the present invention are not limited to the above specific description, and the above and other objectives that can be achieved by the present invention will be more clearly understood from the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The drawings described herein are used to provide a further understanding of the present invention, constitute a part of the present application, and do not constitute a limitation of the present invention. In the drawings:

[0032] Figure 1 A flowchart of a computing power network task scheduling method provided in one embodiment of the present invention.

[0033] Figure 2 A schematic diagram of the structure of a computing power network provided in one embodiment of the present invention.

[0034] Figure 3 A flowchart for solving a task scheduling optimization problem provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0035] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments and the accompanying drawings. Here, the illustrative embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0036] It should also be noted that, in order to avoid obscuring the present invention due to unnecessary details, only structures and / or processing steps closely related to the solutions according to the present invention are shown in the accompanying drawings, while other details that are not closely related to the present invention are omitted.

[0037] It should be emphasized that the term “include / comprises” when used herein refers to the presence of features, elements, steps or components, but does not exclude the presence or addition of one or more other features, elements, steps or components.

[0038] It should also be noted that, unless otherwise specified, the term “connection” herein may refer not only to a direct connection but also to an indirect connection with an intermediate.

[0039] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. In the accompanying drawings, the same reference numerals represent the same or similar components, or the same or similar steps.

[0040] Figure 1 The figure is a flow chart of a computing network task scheduling method according to an embodiment of the present invention, wherein the computing network task scheduling method is applied to a computing network task scheduling device in a computing network.

[0041] refer to Figure 2 In some embodiments of the present invention, in order to support the operation of the computing power network, the computing power network needs to obtain energy from the energy supply module. The computing power network includes but is not limited to a computing power network task scheduling device, a computing power node, a computing power network router, and a general router. Among them, the number of the computing power network task scheduling device, the computing power node, the computing power network router, and the general router is at least one, and the computing power network task scheduling device includes a server, a load balancer, or an edge computer and other devices; the computing power node includes an edge computer, a server, or a tablet computer and other devices; this embodiment does not limit the number and type of devices in the computing power network.

[0042] In some embodiments of the present invention, general routers and computing power network routers are used to receive computing power tasks sent by clients and forward them to computing power nodes; computing power nodes are used to execute computing power tasks; the computing power network task scheduling device obtains resource information of the computing power network and resource information of the energy supply module using discrete time slots, and executes the computing power network task scheduling method.

[0043] Each discrete time slot represents a fixed time unit Δt. The operation time of the computing power network is divided into fixed time periods according to the time unit Δt. In each discrete time slot, the energy value is static, and the energy values ​​corresponding to different discrete time slots are the same or different.

[0044] In some embodiments of the present invention, the set of computing nodes is represented by The router set is represented as in, represents a general router responsible for receiving and forwarding data, Represents a computing network router responsible for data transmission and forwarding.

[0045] The network structure of the computing power network is an undirected connected graph. express.

[0046] in, It represents the set of computing nodes, general routers and computing network routers in the computing network, which is expressed by the following formula:

[0047]

[0048] in, Represents a collection of computing power nodes; Represents a set of routers; M represents the number of computing nodes or computing network routers; I represents the number of general routers.

[0049] The set of network links in the computing network can be expressed as follows:

[0050]

[0051] in, Represents the direct network connection between the computing power node, the general router and the computing power network router. The network connection weight is non-negative; Represents the collection of computing nodes, general routers, and computing network routers in the computing network.

[0052] For any discrete time slot Maximum network link capacity Λ max , can be expressed by the following formula:

[0053]

[0054] in, represents the maximum network link capacity of network link l; It is a matrix that represents the network link capacity and network transmission rate corresponding to each node in the computing power network.

[0055] The network transmission rate U(t) can be expressed by the following formula:

[0056]

[0057] Among them, u l (t) represents the network transmission rate of network link l; It is a matrix that represents the network link capacity and network transmission rate corresponding to each node in the computing power network.

[0058] In the network link In the case of , the maximum network link capacity of network link l is greater than 0, and the network transmission rate of network link l is greater than 0; otherwise, the maximum network link capacity of network link l is equal to 0, and the network transmission rate of network link l is equal to 0.

[0059] In some embodiments of the present invention, there are multiple transmission paths between the router and the computing power node, which can be expressed by the following formula:

[0060]

[0061] Among them, S i,m represents the set of transmission paths between the general router i and the computing power node m; Indicates the sth transmission path.

[0062] The computing tasks received by the computing network refer to the tasks sent by the user through the client, which are received and forwarded by the routers in the computing network. At discrete time slot T = 0, the first computing task set (i.e., “new” tasks) arriving at the general router is expressed as

[0063] in, Indicates that it has reached a general router New task set; Indicates arrival The nth new task. Among them, Represents the computing task ω i,n The corresponding data volume (in bits); Represents the computing task ω i,n The corresponding computational effort (i.e., the number of CPU cycles required to perform the task); Represents the computing task ω i,n When a general router receives a computing task, it can temporarily put the task into its cache, and the computing network task scheduling device will ensure that Process the task during the period.

[0064] The second computing task set (i.e., the “historical” task set) that arrives at the general router before the discrete time slot T = 0. Although the computing network has made a scheduling decision for the historical tasks, due to the forwarding delay mechanism, the historical tasks may be forwarded later.

[0065] Computing networks consume a lot of computing resources in big data processing, resulting in high energy consumption, which greatly increases the carbon footprint of computing networks and exacerbates global warming. Based on this, in order to mitigate the negative impact on the environment, computing networks urgently need to reduce their dependence on non-renewable energy and use renewable energy with almost no carbon emissions as the main energy supply. In this context, task scheduling solutions that can improve energy efficiency and maximize the use of renewable energy will become a key technology to achieve sustainable green computing networks and promote the development of big data.

[0066] In traditional computing network task scheduling systems, the characteristics of computing networks and renewable energy are mostly ignored. Specifically, computing network mandatory scheduling schemes jointly consider the impact of heterogeneous computing nodes and complex network paths on latency and energy consumption, but due to the geographical diversity and unpredictability of big data applications and renewable energy, renewable energy computing networks are highly dynamic as time and location change. In this case, traditional computing network task scheduling systems ignore network considerations, and spatiotemporal dynamics may not be suitable for green computing networks.

[0067] In order to solve the above technical problems, the computing power network task scheduling method provided by the present invention fully considers the integration of network and computing power, as well as the spatiotemporal dimensions of big data applications and renewable energy, and aims to minimize energy consumption and carbon emissions while meeting delay constraints and long-term load balancing. The computing power network task scheduling method provided in the embodiment of the present application is explained in detail below. The computing power network task scheduling method may include the following steps S101 to S102:

[0068] Step S101, define the objective function between the network energy consumption of the computing power network and the network carbon emissions of the computing power network and the constraints corresponding to the objective function to obtain the task scheduling optimization problem.

[0069] The network energy consumption of the computing network refers to the electrical energy consumed for data transmission, processing and storage in the computing network. The network carbon emissions of the computing network refer to the carbon dioxide (CO2) emissions generated by electricity consumption during the operation of the entire computing network.

[0070] In some embodiments of the present invention, the objective function optimization goal is to reduce network energy consumption and network carbon emissions by optimizing the computing nodes, forwarding paths and forwarding time corresponding to the computing tasks. Specifically, the objective function can be expressed by the following formula:

[0071]

[0072] Among them, E total Represents the network energy consumption of the computing network; C total It represents the network carbon emissions of the computing power network; X represents the computing power node decision, which is used to indicate the computing power node that executes the computing power task; Y represents the forwarding path decision, which is used to indicate the forwarding path of the computing power task in the computing power network; Z represents the forwarding time of the computing power task.

[0073] In some embodiments of the present invention, the network energy consumption of the computing network includes computing task ω i,n Via network path Routing transmission energy consumption and processing power tasksω i,n The execution energy consumption of computing node m. The computing tasks include the first computing task set received in the current time slot and the second computing task set received before the current time slot; accordingly, the network energy consumption includes the transmission energy consumption and execution energy consumption corresponding to each computing task in the first computing task set, and the transmission energy consumption and execution energy consumption corresponding to each computing task in the second computing task set.

[0074] Specifically, the transmission energy consumption is expressed by the following formula:

[0075]

[0076] in, Represents the computing task ω i,n The corresponding data volume; Indicates the path along the network The router set; express; express; Represents a general router The energy consumption of the transponder during the photoelectric / electro-optical conversion; Represents computing power network router The energy consumption of the transponder during the photoelectric / electro-optical conversion; Represents a general router The energy consumption of electronic processing.

[0077] Processing computing tasksω i,n The execution energy consumption of computing node m is expressed by the following formula:

[0078]

[0079] In the formula, ζ m represents the effective capacitance parameter of computing power node m of the computing chipset; f m is the computing power resource (i.e., CPU computing frequency) of computing power node m; Indicates the amount of computation (i.e., the number of CPU cycles required to perform the task).

[0080] The network carbon emissions of the computing power network include the carbon emissions of the computing power network and computing power nodes. Computing power nodes (such as data centers, servers, etc.) consume a lot of electricity when performing calculations. The way electricity is produced (such as coal, natural gas, renewable energy, etc.) directly affects carbon emissions. If electricity comes from non-renewable energy sources such as fossil fuels, carbon emissions will increase significantly. At the same time, data transmission in the computing power network also consumes electricity, especially under high load conditions. The network equipment required for transmission (such as routers, switches, etc.) also requires electricity.

[0081] Renewable energy includes solar energy and wind energy. For renewable energy, carbon-free emissions are considered. Based on this, the carbon emissions of the computing power network and computing power node m in time slot t can be expressed as:

[0082]

[0083] Among them, C B represents the carbon intensity of non-renewable energy; Represents the carbon emissions of the computing power network corresponding to time slot t; and Both represent the carbon emissions of computing node m in time slot t.

[0084] in, It is expressed by the following formula:

[0085]

[0086] In the formula, represents the transmission energy consumption of the historical task corresponding to time slot t; represents the transmission energy consumption of the new task corresponding to time slot t; The supply of renewable energy corresponding to the computing power network at time slot t is expressed as follows:

[0087]

[0088] In the formula, represents the solar power generation of the renewable energy supplier's solar panels matched with the computing network at time slot t. Specifically, It is expressed as: In the formula, κ net represents the area of ​​the solar panel, represents the conversion rate of solar radiation into electrical energy, γ net represents the solar irradiance at time slot t.

[0089] represents the wind power generation of the renewable energy supplier's wind turbine generator set matched with the computing power network at time slot t. Specifically, It is expressed as: In the formula, α net represents the wind power conversion rate, represents the conversion rate of solar radiation into electrical energy, ζ net represents the air density at the wind power site, ∈ net represents the wind turbine rotor area, v net Represents the wind speed of the wind turbine at time slot t.

[0090] It is expressed by the following formula:

[0091]

[0092] In the formula, Represents the computational energy consumption of the historical task corresponding to time slot t on computing node m; represents the computing energy consumption of the new task corresponding to time slot t on computing node m; The supply of renewable energy corresponding to the computing power network at time slot t is expressed as follows:

[0093]

[0094] In the formula, represents the solar power generation of the renewable energy supplier's solar panels matched with the computing power node m at time slot t. Specifically, It is expressed as: In the formula, κ m represents the area of ​​the solar panel, represents the conversion rate of solar radiation into electrical energy, γ m represents the solar irradiance at time slot t.

[0095] represents the wind power generation of the renewable energy supplier's wind turbine generator set matched with the computing power node m at time slot t. Specifically, Expressed as In the formula, α m represents the wind power conversion rate, ζ m represents the air density at the wind power site, ∈ m represents the wind turbine rotor area, v m Represents the wind speed of the wind turbine at time slot t.

[0096] In some embodiments of the present invention, the constraints include latency constraints, resource constraints, computing node decision constraints, and forwarding path decision constraints.

[0097] Among them, delay constraints refer to constraints constructed based on the relationship between the forwarding time and processing delay corresponding to the computing task and the tolerable delay of the computing network; resource constraints include constraints constructed based on the relationship between the current load and the preset load, constraints constructed based on the relationship between network energy consumption and the preset total energy, and constraints constructed based on the relationship between the cache data volume corresponding to the computing node and the preset cache resources.

[0098] In the traditional task allocation and management methods based on deep reinforcement learning and software defined network (SDN), the deep reinforcement learning algorithm can autonomously learn the best task allocation strategy through interaction with the environment to optimize energy consumption and latency. The centralized control feature of SDN supports dynamic task scheduling and resource management. When a new task arrives, the SDN controller obtains the status information of all fog nodes, and the deep reinforcement learning algorithm selects the optimal node for processing based on the remaining energy, current load and task execution requirements of the node. At the same time, the SDN controller analyzes the network and node status in real time to ensure that the task is assigned to the fog node closest to the source node and with the lowest latency. In addition, in order to improve energy efficiency, tasks are preferentially assigned to nodes with sufficient energy.

[0099] However, the increasing dynamic and unpredictable traffic in the Internet of Things makes computing task scheduling complicated, especially in real-time scheduling, which requires a highly adaptive mechanism. In addition, fog nodes usually have limited resources and are mostly battery-powered, with limited computing and energy resources, which brings challenges to task scheduling and energy management. Based on this, in a heterogeneous Internet of Things environment, it becomes very difficult to balance the relationship between network latency, scheduling accuracy, energy consumption, and response time, and there are problems with the scalability of task allocation and low management efficiency.

[0100] In order to solve the above problems, in some embodiments of the present invention, the computing power of the computing nodes and network resources, including energy consumption and cache resources, are deeply integrated through the computing network task scheduling device. By defining the optimization goal as minimizing the total energy consumption and carbon emissions of the computing network, while satisfying the constraints such as the tolerable delay, processing delay and load balancing of the task, combined with the dynamic network load and delay requirements, a highly adaptable scheduling strategy is provided for the task allocation of big data applications to achieve efficient scheduling of computing tasks, optimize the use of renewable energy, improve the overall operating efficiency of the computing network, and at the same time improve the flexibility and management efficiency of task allocation, ensuring that the computing network can still operate efficiently and stably under the conditions of traffic growth and resource constraints.

[0101] In some embodiments of the present invention, the processing delay includes the sum of the task execution delay corresponding to the computing task, the queuing waiting delay and the task transmission delay.

[0102] Among them, the task transmission delay is expressed by the following formula:

[0103]

[0104] In the formula, Represents the computing task ω i,n The corresponding data volume; l represents the network link; represents the sth transmission path between the general router and the computing power node; u l Indicates the network transmission rate; Represents the computing task ω i,n The corresponding forwarding timestamp can be expressed as follows:

[0105]

[0106] In the formula, is a non-negative integer; Represents virtual delay. The computing task ω arriving at time slot 0 i,n Will be in the time slot Forward.

[0107] Computing Taskω i,nThe average queuing waiting delay of can be expressed as the computing task ω i,n The average waiting time of the historical tasks that arrived before, and the computing power task ω i,n The average queue waiting delay of historical tasks that arrive at the same time, and the sum of the average queue waiting delay caused by other new tasks. Specifically, the computing task ω i,n The average queue waiting delay can be expressed as follows:

[0108]

[0109] In the formula, f m Indicates the computing power resources (i.e. CPU computing frequency) of computing power node m; Representation and computing task ω i,n The total amount of computing power of other new computing tasks that arrive at the same time; Representation and computing task ω i,n The computational requirements of asynchronously arriving historical tasks; Representation and computing task ω i,n The computational requirements of historical tasks that arrived synchronously.

[0110] Task execution delay refers to the time it takes for computing node m to execute computing task ω i,n The delay generated during the process can be expressed by the following formula:

[0111]

[0112] In the formula, f m Indicates the computing power resources of computing node m; Indicates the amount of computation, that is, the number of CPU cycles required to perform a task.

[0113] Specifically, the delay constraint constructed based on the relationship between the forwarding time and processing delay corresponding to the computing task and the tolerable delay of the computing network is expressed as follows:

[0114]

[0115] In the formula, Represents the computing task ω i,n The corresponding forwarding timestamp; D total (ω i,n ) represents the computing task ω i,n The corresponding processing delay; Represents the computing task ω i,n The corresponding tolerable delay; Represents the first computing power task set.

[0116] Maintaining long-term load balance is essential to improving resource utilization and ensuring system stability. In some embodiments of the present invention, the current load includes network load balancing corresponding to the computing network and node load balancing corresponding to the computing nodes. Specifically, the current load can be expressed by the following formula:

[0117]

[0118] In the formula, and Represents the weight factor; L CPN (t) indicates node load balancing; L net (t) indicates network load balancing.

[0119] In some embodiments of the present invention, the load balance of all computing nodes is evaluated by standard deviation. Specifically, the node load balance L CPN (t) can be expressed by the following formula:

[0120]

[0121] Where, L m (t) represents the load of computing node m at discrete time slot t, expressed as in, Indicates the maximum cache resource corresponding to the cache queue maintained by computing node m, B m (t) represents the total amount of data cached in the computing node m at discrete time slot t; Indicates the average load of the computing nodes in the entire computing network.

[0122] Load balancing model L of computing power network net (t) can be expressed by the following formula:

[0123]

[0124] Where, L l (t) represents the load of network link l at discrete time slot t, expressed as Among them, Γ l (t) represents the amount of data on network link l at discrete time slot t, Indicates the average load of the entire computing network; Represents the collection of network links in the computing network.

[0125] Specifically, the constraint constructed by the relationship between the current load and the preset load can be expressed by the following formula:

[0126]

[0127] Where, L total(t) represents the current load corresponding to the discrete time slot t, including the network load balancing corresponding to the computing network and the node load balancing corresponding to the computing node; T represents the number of discrete time slots; L up Indicates the preset load.

[0128] In the traditional task allocation method for real-time applications in heterogeneous sensor networks, tasks are divided into synchronous tasks and simple tasks according to the number of parent tasks and child tasks. After task division, the entire task graph is converted into a task partition graph, in which each branch is scheduled independently, thereby reducing the complexity of scheduling. After the task division is completed, the overall deadline will be assigned to each task partition to ensure that the tasks of each partition can be completed within their respective sub-deadlines, ensuring that the overall execution time of the final task does not exceed the global deadline, and minimizing energy consumption while ensuring that the deadline is met.

[0129] However, in the above traditional methods, while reducing energy consumption, the algorithm must balance between energy consumption and real-time performance. How to find the best balance point without affecting either side is very complicated. It is difficult to balance energy consumption and service quality. At the same time, there are usually dependencies between tasks. While minimizing energy consumption, if the task assigned to the fast node needs to wait for the dependent task assigned to the slow node, it may cause unnecessary delays and reduce efficiency.

[0130] In order to solve the above problems, in some embodiments of the present invention, by introducing delay constraints and long-term load balancing strategies, the balance problem between energy consumption and real-time performance in traditional methods is effectively solved. This method not only ensures the reliability and efficiency of task processing, but also reduces the delay caused by dependencies through a dynamic scheduling mechanism, avoiding blocking tasks of fast nodes on slow nodes. In addition, while reducing energy consumption, carbon emissions are effectively reduced, providing strong support for the realization of green computing. This solution is not only suitable for current complex big data application scenarios, but also provides important technical guarantees for future intelligent energy management and promotes the goal of sustainable development.

[0131] In some embodiments of the present invention, the constraint condition constructed based on the relationship between the network energy consumption and the preset total energy amount can be expressed by the following formula:

[0132] E total ≤Q total

[0133] Among them, E total represents network energy consumption; Q total In some embodiments of the present invention, Q total Refers to the renewable energy energy that the energy supply module can provide.

[0134] The constraint condition constructed based on the relationship between the cache data volume corresponding to the computing power node and the preset cache resources is expressed by the following formula:

[0135]

[0136] In the formula, B m (t) represents the total amount of data cached in the computing node m at discrete time slot t; Indicates the preset cache resources of computing node m.

[0137] The computing power node decision constraints and forwarding path decision constraints are expressed by the following formula:

[0138]

[0139] In the formula, S i,m represents the set of transmission paths between the general router i and the computing power node m; Indicates S i,m The sth transmission path in the calculation; X represents the computing node decision. When X is 1, the computing node indicated by the computing node decision is computing node m; Y represents the forwarding path decision. When Y is 1, the forwarding path is S i,m The sth transmission path in ω i,n Represents a computing task; W represents a computing task set.

[0140] Step S102, solving the task scheduling optimization problem through a preset optimization algorithm to obtain a task scheduling method for the computing power network; the task scheduling method is used to indicate the forwarding path of the computing power task in the computing power network, the computing power node that executes the computing power task, and the forwarding time corresponding to the computing power node.

[0141] Among them, the task scheduling optimization problem can be expressed by the following formula:

[0142]

[0143] C3:E total ≤Q total

[0144]

[0145] In the formula, E total Represents the network energy consumption of the computing network; C total Indicates the network carbon emissions of the computing power network; X represents the computing power node decision, which is used to indicate the computing power node that executes the computing power task; Y represents the forwarding path decision, which is used to indicate the forwarding path of the computing power task in the computing power network; Z represents the forwarding time of the computing power task; Represents the computing task ω i,nThe corresponding forwarding timestamp; D total (ω i,n ) represents the computing task ω i,n The corresponding processing delay; Represents the computing task ω i,n The corresponding tolerable delay; represents the first computing power task set; L total (t) represents the current load corresponding to the discrete time slot t, including the network load balancing corresponding to the computing network and the node load balancing corresponding to the computing node; T represents the number of discrete time slots; L up Indicates preset load; Q total Indicates the preset total energy; B m (t) represents the total amount of data cached in the computing node m at discrete time slot t; Represents the preset cache resources of computing node m; S i,m represents the set of transmission paths between the general router i and the computing power node m; Indicates S i,m The sth transmission path in .

[0146] In some embodiments of the present invention, the task scheduling optimization problem is solved by a preset optimization algorithm, wherein the preset optimization algorithm refers to a pre-set multi-objective optimization algorithm, including but not limited to a non-dominated sorting genetic algorithm (NSGA-II), a genetic algorithm (GA) or a particle swarm optimization (PSO), etc.

[0147] When the preset optimization algorithm is a non-dominated sorting genetic algorithm, the algorithm is optimized based on the original non-dominated sorting genetic algorithm to adapt to its specific needs and constraints, including initializing the population, non-dominated sorting, crowding distance calculation, selection crossover and mutation processing, and merging and re-selection processing.

[0148] The initialization population refers to a population that randomly generates a set of solutions. represents the population of generation t, where {X,Y,Z} j,k ∈P j represents the kth individual in the jth generation, N pop Indicates the overall size, each individual includes computing node decision, forwarding path and forwarding time. At the same time, the initialization also includes population-related parameters, including the maximum number of generations N gen , crossover probability P c , mutation probability P M .

[0149] Non-dominated sorting refers to performing non-dominated sorting on the population and identifying different dominance levels. Crowding distance calculation refers to calculating the crowding of each solution to maintain the diversity of solutions.

[0150] In some embodiments of the present invention, by calculating the objective function E total and C total , and the violation of the constraints, all individuals are assigned to different non-dominated queues. At the same time, in order to ensure the diversity of the population, according to E total and C total Estimate each solution {X,Y,Z} j,k The crowding distance is used to select candidate solutions through non-dominated sorting and crowding distance.

[0151] The selection crossover and mutation process refers to the generation of a new population after the selection based on the non-dominated hierarchy and crowding distance. After that, the parent and child generations are merged through the merging and re-selection process, and the non-dominated sorting is performed again to select the optimal solution.

[0152] Specifically, the task scheduling optimization problem is solved by a preset optimization algorithm, and the task scheduling method of the computing power network is obtained, including: initializing the current population; the current population includes several solutions; based on the current population, the objective function is calculated to obtain the network energy consumption and network carbon emissions, and based on the constraints, the constraint violation corresponding to each solution is obtained; based on the constraint violation, each solution in the current population is non-dominated sorted to obtain a non-dominated sorting result; based on the network energy consumption and network carbon emissions, the crowding distance corresponding to each solution is determined; based on the non-dominated sorting result and the crowding distance corresponding to each solution, a candidate solution is determined in the current population; when the algorithm converges, the candidate solution is determined as the task scheduling method. Among them, the candidate solution refers to a solution with a higher dominance level and a shorter crowding distance that is preferentially selected.

[0153] When the algorithm has not converged, the candidate solutions are cross-processed and mutated. The scheduling strategies of multiple solutions are combined at the intersection, and some scheduling strategies are randomly changed through mutation processing to increase the diversity of solutions. The above steps are then repeated until the algorithm converges.

[0154] Specifically, when the algorithm has not converged, it also includes: performing crossover and mutation processing on candidate solutions to generate a child population; merging the current population with the child population, and iterating the current population; executing the steps of calculating the objective function based on the current population and obtaining the constraint violation situation corresponding to each solution based on the constraint conditions.

[0155] In addition, in the computing power network, since network energy consumption and carbon emissions may change over time, in order to improve the practicality of the task scheduling method, before performing crossover processing and mutation processing, it is necessary to dynamically adjust the crossover probability and mutation probability according to the network energy consumption and network carbon emissions; at the same time, unnecessary calculations can be reduced and energy consumption can be reduced.

[0156] Specifically, the candidate solutions are subjected to crossover and mutation processing before the generation of the offspring population, and the process also includes: dynamically adjusting the crossover probability and mutation probability corresponding to the current population based on the network energy consumption and the network carbon emission.

[0157] refer to Figure 3 ,The solution process of task scheduling optimization problem includes at least the following steps:

[0158] Step S301, initializing a population P and population parameters, wherein the population parameters include population generation number, crossover probability and mutation probability;

[0159] Step S302, estimating the optimization objective function and the constraints of the population P, and performing a non-dominated sorting on the population P;

[0160] Step S303, calculating the crowding distance corresponding to the population P;

[0161] Step S304, determining candidate solutions based on crowding distance and non-dominated sorting;

[0162] Step S305, dynamically updating the crossover probability and mutation probability;

[0163] Step S306, performing crossover and mutation processing on the candidate solutions, generating a progeny population, and calculating constraint conditions;

[0164] Step S307, judging whether the current generation is greater than the population generation, if the current generation is greater than the population generation, executing step S309, otherwise executing step S308;

[0165] Step S308, integrating and classifying the offspring population with the population P to obtain a new population P; executing step S302;

[0166] Step S309: output candidate solutions as task scheduling methods.

[0167] In summary, the computing power network task scheduling method provided by the present invention obtains a task scheduling optimization problem by defining an objective function between the network energy consumption of the computing power network and the network carbon emissions of the computing power network and constraints corresponding to the objective function; the objective function optimization goal is to reduce network energy consumption and network carbon emissions by optimizing the computing power nodes, forwarding paths and forwarding times corresponding to the computing power tasks; the constraints include delay constraints, resource constraints, computing power node decision constraints and forwarding path decision constraints; the task scheduling optimization problem is solved by a preset optimization algorithm to obtain a task scheduling method for the computing power network; the task scheduling method is used to indicate the forwarding path of the computing power task in the computing power network, the computing power node that executes the computing power task and the forwarding time corresponding to the computing power node; it can Solve the problems of low scalability and management efficiency of task allocation; deeply integrate the computing power of computing nodes and network resources, including energy consumption and cache resources, through the computing network task scheduling device; by defining the optimization goal as minimizing the total energy consumption and carbon emissions of the computing network, while meeting the constraints such as tolerable delay, processing delay and load balancing of the task, combined with dynamic network load and delay requirements, provide a highly adaptable scheduling strategy for the task allocation of big data applications, so as to achieve efficient scheduling of computing tasks, optimize the use of renewable energy, improve the overall operating efficiency of the computing network, and at the same time improve the flexibility and management efficiency of task allocation, ensuring that the computing network can still operate efficiently and stably under the conditions of traffic growth and resource constraints.

[0168] Corresponding to the above method, the present invention also provides a computing power network task scheduling device, which includes a computer device, the computer device includes a processor and a memory, the memory stores computer instructions, and the processor is used to execute the computer instructions stored in the memory. When the computer instructions are executed by the processor, the device implements the steps of the computing power network task scheduling method as described above.

[0169] The embodiment of the present invention also provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the steps of the aforementioned computing network task scheduling method are implemented. The computer-readable storage medium can be a tangible storage medium, such as a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a floppy disk, a hard disk, a removable storage disk, a CD-ROM, or any other form of storage medium known in the technical field.

[0170] An embodiment of the present invention also provides a computer program product, including a computer program / instruction, which implements the steps of the aforementioned computing power network task scheduling method when executed by a processor.

[0171] It should be understood by those skilled in the art that the exemplary components, systems and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software or a combination of the two. Whether it is performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention. When implemented in hardware, it can be, for example, an electronic circuit, an application specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, etc. When implemented in software, the elements of the present invention are programs or code segments used to perform the required tasks. The program or code segment can be stored in a machine-readable medium, or transmitted on a transmission medium or a communication link via a data signal carried in a carrier.

[0172] It should be clear that the present invention is not limited to the specific configuration and processing described above and shown in the figures. For the sake of simplicity, a detailed description of the known method is omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between the steps after understanding the spirit of the present invention.

[0173] In the present invention, features described and / or illustrated for one embodiment may be used in the same or similar manner in one or more other embodiments, and / or combined with features of other embodiments or replace features of other embodiments.

[0174] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the embodiments of the present invention may have various modifications and variations. 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. A computing network task scheduling method, characterized in that: Applied to a computing power network, the method includes: An objective function between the network energy consumption of the computing power network and the network carbon emissions of the computing power network and constraints corresponding to the objective function are defined to obtain a task scheduling optimization problem; the optimization goal of the objective function is to reduce the network energy consumption and the network carbon emissions by optimizing the computing power nodes, forwarding paths and forwarding time corresponding to the computing power tasks; the constraints include delay constraints, resource constraints, computing power node decision constraints and forwarding path decision constraints; The task scheduling optimization problem is solved by a preset optimization algorithm to obtain a task scheduling method for the computing power network; the task scheduling method is used to indicate the forwarding path of the computing power task in the computing power network, the computing power node that executes the computing power task, and the forwarding time corresponding to the computing power node.

2. The method according to claim 1, characterized in that: The delay constraint refers to a constraint condition constructed based on the relationship between the forwarding time and processing delay corresponding to the computing task and the tolerable delay of the computing network; The resource constraints include constraints constructed based on the relationship between the current load and the preset load, constraints constructed based on the relationship between the network energy consumption and the preset total energy, and constraints constructed based on the relationship between the cache data volume corresponding to the computing power node and the preset cache resources.

3. The method according to claim 2, characterized in that The computing power tasks include a first computing power task set received in the current time slot and a second computing power task set received before the current time slot; the network energy consumption includes the transmission energy consumption and execution energy consumption corresponding to each computing power task in the first computing power task set, and the transmission energy consumption and execution energy consumption corresponding to each computing power task in the second computing power task set.

4. The method according to claim 2, characterized in that: The processing delay includes the task execution delay corresponding to the computing task, the sum of the queuing waiting delay and the task transmission delay; the current load includes the current computing network load and the current computing node load.

5. The method according to claim 1, characterized in that The method of solving the task scheduling optimization problem by using a preset optimization algorithm to obtain the task scheduling method of the computing power network includes: Initializing a current population; the current population includes a plurality of solutions; Based on the current population, the objective function is calculated to obtain the network energy consumption and the network carbon emission, and based on the constraint condition, the constraint violation corresponding to each solution is obtained; Based on the violation of the constraint condition, each solution in the current population is non-dominated and sorted to obtain a non-dominated sorting result; Determining a congestion distance corresponding to each solution based on the network energy consumption and the network carbon emissions; Determine a candidate solution in the current population based on the non-dominated sorting result and the crowding distance corresponding to each solution; When the algorithm converges, the candidate solution is determined as the task scheduling method.

6. The method according to claim 5, characterized in that In the case where the algorithm does not converge, the method further comprises: Performing crossover and mutation processing on the candidate solutions to generate a progeny population; After merging the current population and the offspring population, iterating the current population; The step of calculating the objective function based on the current population and obtaining the violation of the constraint conditions corresponding to each solution based on the constraint conditions is executed.

7. The method according to claim 6, characterized in that Before performing crossover and mutation processing on the candidate solutions to generate a child population, the method further includes: dynamically adjusting a crossover probability and a mutation probability corresponding to the current population based on the network energy consumption and the network carbon emissions.

8. A computing network task scheduling device, comprising a processor, a memory, and a computer program / instruction stored in the memory, characterized in that: The processor is used to execute the computer program / instructions. When the computer program / instructions are executed, the device implements the steps of the method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instructions are executed by a processor, the steps of the method as claimed in any one of claims 1 to 7 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images