Task coordination scheduling method, device and equipment

Abstract

The present application relates to the technical field of computing power optical network, and provides a task cooperative scheduling method, device and equipment, comprising responding to a data aggregation task to be scheduled, generating a candidate scheme set containing computing power nodes and transmission paths based on resource constraint conditions; determining resource cost based on resource expected consumption and unified value of resources, wherein the unified value of resources is obtained based on the base value determined based on resource capacity and the scarcity determined based on available resources; determining task utility based on expected execution performance and resource cost, and determining a target scheduling scheme based thereon; thus, by fusing static capacity and dynamic available scarcity to construct a unified value scale, the resource cost evaluation can respond to the dynamic fluctuation of network load in real time, thereby realizing the joint cooperative optimization of heterogeneous resources and improving the task scheduling efficiency and load balancing of the system.

Description

Technical Field

[0001] This invention relates to the field of computing power optical network technology, and in particular to a task collaborative scheduling method, apparatus and equipment. Background Technology

[0002] With the rapid development of emerging applications such as artificial intelligence, big data analytics, digital twins, metaverse, and generative AI, massive data processing tasks are constantly emerging. Unlike traditional centralized computing models, more and more data is distributed and stored in data centers, edge data centers, and cloud computing nodes in different regions. In recent years, optical computing networks, as a new type of network infrastructure integrating computing, storage, and optical transmission resources, have become an important platform for carrying out such tasks. Within this platform, achieving efficient and accurate task scheduling under limited resource conditions is a crucial technical requirement.

[0003] To meet these requirements, scheduling methods based on specific optimization objectives are commonly used. For example, they tend to select nodes and paths with strong computing power and low transmission latency to shorten task completion time, or they use reliability models to ensure task success rates. Some solutions also establish basic resource consumption models or attempt joint scheduling of network resources through multi-objective weighted summation.

[0004] However, when dealing with heterogeneous resources with different physical attributes, the evaluation system is relatively isolated and mainly relies on static information about the resources. Faced with real-time changes in available resources during network operation, it struggles to accurately reflect the dynamic differences of the system under different load conditions. Furthermore, when selecting feasible scheduling strategies, it is difficult to accurately pinpoint the scheduling decision with the optimal overall benefits. Summary of the Invention

[0005] This invention provides a task collaborative scheduling method, apparatus, and device to solve the technical problems of current data aggregation task scheduling methods, which cannot measure the overhead of heterogeneous resources under a unified benchmark, and the final scheduling scheme is difficult to achieve global optimization.

[0006] This invention provides a task collaborative scheduling method, comprising the following steps: In response to the data aggregation task to be scheduled, a set of candidate scheduling schemes corresponding to the data aggregation task is generated based on resource constraints; each candidate scheduling scheme in the set of candidate scheduling schemes includes the computing power node to be allocated in the computing power optical network and the transmission path; Based on the expected resource consumption of the data aggregation task under each of the candidate scheduling schemes, and the unified resource value of each computing node in the candidate scheduling schemes, the resource cost of each candidate scheduling scheme is determined; wherein, the unified resource value of the computing node is determined based on the basic resource value and resource scarcity of the computing node; the basic resource value is determined based on the resource capacity of the computing node, and the resource scarcity is determined based on the available resources of the computing node; Based on the expected execution performance of each candidate scheduling scheme and the resource cost, the task utility of each candidate scheduling scheme is determined. Based on the task utility, a target scheduling scheme is determined from the candidate scheduling scheme set to execute the data aggregation task.

[0007] According to the task collaborative scheduling method provided by the present invention, the unified resource value of the computing power node is determined in the following way: Based on the resource capacity of the computing power node and the global maximum capacity corresponding to the resource category to which the computing power node belongs, the basic resource value of the computing power node is determined. Based on the available resources of the computing power node and the total resources of the computing power node, the resource scarcity of the computing power node is determined; Based on the resource scarcity and the basic value of the resources, the unified resource value of the computing power node is determined.

[0008] According to the task collaborative scheduling method provided by the present invention, before generating a set of candidate scheduling schemes corresponding to the data aggregation task based on resource constraints in response to the currently scheduled data aggregation task, the method further includes: Based on the task data volume, task deadline, and task reliability requirements of each initial data aggregation task, the scheduling priority of each initial data aggregation task is determined. Based on the scheduling priority of each initial data aggregation task, the data aggregation task to be scheduled is determined.

[0009] According to the task collaborative scheduling method provided by the present invention, the resource constraints include computing resource constraints, storage resource constraints, bandwidth resource constraints, spectrum non-overlapping constraints, spectrum continuity constraints, and spectrum consistency constraints; the step of generating a candidate scheduling scheme set corresponding to the data aggregation task based on the resource constraints includes: Based on the computing resource constraints and the storage resource constraints, candidate computing nodes corresponding to the data aggregation task are determined; Based on each of the candidate computing power nodes, the bandwidth resource constraints, the spectrum non-overlapping constraints, the spectrum continuity constraints, and the spectrum consistency constraints, the corresponding candidate transmission paths are determined. Based on each candidate computing node and its corresponding candidate transmission path, a set of candidate scheduling schemes is generated for the data aggregation task.

[0010] According to the task collaborative scheduling method provided by the present invention, the resource categories of the computing power nodes include computing resources, storage resources, and network resources; the step of determining the resource cost of each candidate scheduling scheme based on the expected resource consumption of the data aggregation task under each candidate scheduling scheme and the unified resource value of each computing power node in the candidate scheduling scheme includes: Based on the expected resource consumption of the computing resources, storage resources and network resources by the data aggregation task, and the unified value of the resources corresponding to each resource category, the computing resource cost, storage resource cost and network resource cost are determined respectively. Based on the computing resource cost, the storage resource cost, and the network resource cost, the resource cost of each candidate scheduling scheme is determined.

[0011] According to the task collaborative scheduling method provided by the present invention, the expected execution performance includes task completion time; determining the task utility of each candidate scheduling scheme based on the expected execution performance of each candidate scheduling scheme and the resource cost includes: Based on the original data volume of the data aggregation task and the computing resources allocated to the local computing power node, the local processing time of the computing power node is determined. Based on the original data volume and data aggregation selection factor of the data aggregation task, the actual data volume of the computing power node is determined. Based on the actual amount of aggregated data and the computing resources allocated to the target computing power node, the aggregation processing time of the computing power node is determined. The task completion time of the data aggregation task is determined based on the local processing time, the aggregation processing time, and the network communication time of the transmission path in the candidate scheduling scheme. Based on the task completion time and the resource cost, the task utility of each candidate scheduling scheme is determined.

[0012] According to the task collaborative scheduling method provided by the present invention, the expected execution performance further includes reliability satisfaction; the step of determining the task utility of each candidate scheduling scheme based on the task completion time and the resource cost includes: Based on the task completion time and the deadline of the data aggregation task, determine the time satisfaction level; Based on the failure rate distribution of the computing nodes and the transmission path in the candidate scheduling scheme, the reliability satisfaction is determined. Based on the time satisfaction, the reliability satisfaction, and the resource cost, the task utility of each candidate scheduling scheme is determined.

[0013] According to the task collaborative scheduling method provided by the present invention, determining a target scheduling scheme from the candidate scheduling scheme set based on the task utility to execute the data aggregation task includes: Based on the task acceptance status and task rejection status of multiple data aggregation tasks, determine the task rejection penalty items corresponding to the multiple data aggregation tasks; Based on the task utility of each data aggregation task in the task reception state and the task rejection penalty, the global scheduling evaluation benefit is determined. With the goal of maximizing the benefits of the global scheduling evaluation, a target scheduling scheme is determined from the candidate scheduling scheme set to execute the data aggregation task.

[0014] The present invention also provides a task collaborative scheduling device, comprising: The scheme generation module is used to generate a set of candidate scheduling schemes corresponding to the data aggregation task in response to the data aggregation task, based on resource constraints; each candidate scheduling scheme in the set of candidate scheduling schemes includes the computing power node to be allocated in the computing power optical network and the transmission path; The cost determination module is used to determine the resource cost of each candidate scheduling scheme based on the expected resource consumption of the data aggregation task under each candidate scheduling scheme and the unified resource value of each computing node in the candidate scheduling scheme; wherein, the unified resource value of the computing node is determined based on the basic resource value and resource scarcity of the computing node; the basic resource value is determined based on the resource capacity of the computing node, and the resource scarcity is determined based on the available resources of the computing node; The utility determination module is used to determine the task utility of each candidate scheduling scheme based on the expected execution performance of each candidate scheduling scheme and the resource cost. An execution module is used to determine a target scheduling scheme from the candidate scheduling scheme set based on the task utility, so as to execute the data aggregation task.

[0015] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the program to implement the task cooperative scheduling method as described above.

[0016] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the task cooperative scheduling method as described above.

[0017] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the task cooperative scheduling method as described above.

[0018] The task collaborative scheduling method, apparatus, and equipment provided by this invention realize the comprehensive evaluation of heterogeneous resources in the same value space by constructing a unified value scale based on the basic value and scarcity of resources. This enables resource cost assessment to respond in real time to the dynamic fluctuations of network load, thereby realizing the joint collaborative optimization of heterogeneous resources and improving the task scheduling efficiency and load balancing of the system. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the present invention or the conventional ones, the drawings used in the embodiments or conventional descriptions will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of the computing power optical network provided in an embodiment of the present invention.

[0021] Figure 2 This is a flowchart illustrating the task collaborative scheduling method provided in an embodiment of the present invention.

[0022] Figure 3 This is a schematic diagram of the task collaborative scheduling device provided in an embodiment of the present invention.

[0023] Figure 4 This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0025] It should be noted that in the description of this invention, the terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element. The terms "upper," "lower," etc., indicating orientation or positional relationships based on the orientation or positional relationships shown in the accompanying drawings, are only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the system or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0026] The terms "first," "second," etc., used in this invention are used to distinguish similar objects, not to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class, without limiting the number of objects; for example, a first object can be one or more. Furthermore, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0027] See Figure 1 As shown, the computing power optical network targeted by the embodiments of this application is represented as follows: G ( V , E ),,in V Represents a set of network nodes. E Represents a set of fiber optic links. Network node. V There are two types of nodes. The first type is network nodes that have computing power node functions. V CC The first type is nodes that possess both data processing and network forwarding capabilities; the second type is exchange nodes that do not have computing power node functions. Vs This refers to nodes that only perform network data exchange and forwarding functions. These computing nodes are interconnected via fiber optic links, forming a network architecture that supports the distributed execution of data aggregation tasks.

[0028] In this embodiment of the invention, a data aggregation task refers to a computational task that requires multiple distributed nodes to preprocess or perform local computations on local data before transmitting the intermediate results to a target node for fusion analysis. Data aggregation tasks include, but are not limited to, distributed artificial intelligence inference, federated learning, large-scale log analysis, and multi-source perception fusion. The execution process of a data aggregation task typically involves three stages: local processing of source data nodes, data transmission in the optical network, and fusion processing at the target aggregation node.

[0029] Figure 2 This is a flowchart illustrating the task collaborative scheduling method provided in an embodiment of the present invention, as shown below. Figure 2 As shown, the method includes the following steps 210 to 240.

[0030] Step 210: In response to the data aggregation task to be scheduled, a set of candidate scheduling schemes corresponding to the data aggregation task is generated based on resource constraints; each candidate scheduling scheme in the set of candidate scheduling schemes includes computing power nodes to be allocated and transmission paths in the computing power optical network.

[0031] When a data aggregation task arrives in the scheduling queue, the system responds to the task and, based on the current resource status of each computing node and network link in the computing power optical network, generates a set of candidate scheduling schemes corresponding to the data aggregation task according to preset resource constraints.

[0032] Specifically, the candidate scheduling scheme set includes several candidate scheduling schemes. Each candidate scheduling scheme describes the resource allocation decisions required to complete the data aggregation task, including the computing power nodes to be allocated and the transmission paths in the computing power optical network.

[0033] Among them, computing power nodes refer to nodes in the computing power optical network that can provide computing processing capabilities, storage caching capabilities, or network transmission capabilities for data aggregation tasks, such as target aggregation nodes with computing power processing capabilities; transmission paths refer to the sequence of optical fiber links used to carry data transmission from each source data node to the target aggregation node, as well as the spectrum time slot resources allocated on the links.

[0034] Resource constraints are the resource capacity constraints that computing nodes must meet when accepting task scheduling, and are used to ensure the feasibility of candidate scheduling schemes at the physical resource level.

[0035] In some embodiments, resource constraints may include constraints on the computing power of computing nodes, constraints on the storage resources of computing nodes, constraints on the bandwidth resources of transmission paths, and constraints on the non-overlapping, continuity, and consistency of optical network spectrum time slot allocation, so as to ensure that the generated candidate scheduling scheme can be actually executed under the current network state.

[0036] In this embodiment, firstly, the computing power nodes in the computing power optical network are screened according to resource constraints to determine candidate computing power nodes that meet the constraints. Then, for each candidate computing power node, multiple candidate transmission paths from each source data node to the target computing power node are searched, and it is verified whether each transmission path meets the constraints. Finally, feasible combinations of candidate computing power nodes and candidate transmission paths that meet all constraints are recorded as candidate scheduling schemes and added to the candidate scheduling scheme set to form a set of schemes to be evaluated. Φ j .

[0037] Step 220: Based on the expected resource consumption of the data aggregation task under each of the candidate scheduling schemes, and the unified resource value of each computing node in the candidate scheduling schemes, determine the resource cost of each candidate scheduling scheme; wherein, the unified resource value of the computing node is determined based on the basic resource value and resource scarcity of the computing node; the basic resource value is determined based on the resource capacity of the computing node, and the resource scarcity is determined based on the available resources of the computing node.

[0038] Here, the expected resource consumption refers to the amount of resources that the data aggregation task is expected to occupy on various computing nodes under a certain candidate scheduling scheme, such as the expected amount of computing processing resources, storage cache resources, and network bandwidth resources, which can be determined based on the task data scale, processing requirements, and transmission requirements.

[0039] The unified value of resources is a quantitative representation of the comprehensive value of each computing node in a computing power optical network. It is used to map heterogeneous resources in the computing power optical network to a unified value space, so that different types of resources can be compared and collaboratively decided on the same scale.

[0040] Specifically, the unified value of resources is determined based on the basic value of resources and the scarcity of resources. The basic value of resources represents the inherent capability level of the computing node itself, which is determined based on the resource capacity of the computing node. The larger the resource capacity, the higher the basic value of resources. The scarcity of resources represents the tightness of the currently available resources of the computing node, which is determined based on the available resources of the computing node. The fewer available resources, the higher the scarcity of resources.

[0041] In this embodiment, after determining the unified resource value of each computing node, the resource cost of the candidate scheduling scheme is calculated by combining the expected resource consumption of each computing node by the data aggregation task under the candidate scheduling scheme, that is, the total value of resources consumed for task execution.

[0042] In some embodiments, resource cost can be the weighted sum of the uniform value of resources of each computing node and the expected consumption of corresponding resources, thereby quantifying the overall resource consumption cost of the candidate scheduling scheme under a uniform value scale.

[0043] Step 230: Based on the expected execution performance of each candidate scheduling scheme and the resource cost, determine the task utility of each candidate scheduling scheme.

[0044] Expected execution performance refers to the performance quality metric that a data aggregation task is expected to achieve under a given candidate scheduling scheme. In some embodiments, expected execution performance may include task completion time, i.e., the time required from the start of task execution to the completion of all data aggregation processing; a shorter task completion time indicates better execution performance. In some embodiments, expected execution performance may also include metrics reflecting the stability of task execution, such as task reliability.

[0045] In this embodiment, the expected performance gains and resource costs of the task are considered together to obtain the task utility of each candidate scheduling scheme; the better the expected performance and the lower the resource cost, the higher the task utility.

[0046] In some embodiments, task utility can be obtained based on the difference between task benefits and resource costs, wherein task benefits are quantified based on the satisfaction level with expected performance. Step 240: Based on the task utility, determine the target scheduling scheme in the candidate scheduling scheme set to execute the data aggregation task.

[0047] The target scheduling scheme is the final scheduling decision selected from the set of candidate scheduling schemes, used to guide the actual resource allocation and execution process of the data aggregation task. Specifically, after obtaining the task utility of each candidate scheduling scheme in the set of candidate scheduling schemes, the candidate scheduling schemes are sorted or compared according to their task utility, and the candidate scheduling scheme with the highest task utility is determined as the target scheduling scheme.

[0048] In some embodiments, the candidate scheduling scheme set can be traversed. Φ j All candidate scheduling schemes are considered, the task utility of each scheme is calculated, and the candidate scheduling scheme with the highest task utility is selected as the target scheduling scheme. After the target scheduling scheme is determined, the resource allocation is completed according to the computing power nodes and transmission paths specified in the target scheduling scheme, thereby executing the data aggregation task.

[0049] The task collaborative scheduling method provided in this invention constructs a unified value scale based on the basic value and scarcity of resources, thereby realizing the comprehensive evaluation of heterogeneous resources in the same value space. This enables resource cost assessment to respond in real time to the dynamic fluctuations of network load, thus achieving joint collaborative optimization of heterogeneous resources and improving the task scheduling efficiency and load balancing of the system.

[0050] In some embodiments, the unified resource value of the computing node is determined in the following manner: Based on the resource capacity of the computing power node and the global maximum capacity corresponding to the resource category to which the computing power node belongs, the basic resource value of the computing power node is determined. Based on the available resources of the computing power node and the total resources of the computing power node, the resource scarcity of the computing power node is determined; Based on the resource scarcity and the basic value of the resources, the unified resource value of the computing power node is determined.

[0051] Resource base value characterizes the capability level of a computing node relative to similar resources in the computing power optical network. Specifically, by comparing the resource capacity of a computing node with the global maximum capacity of the corresponding resource category, the problem of inconsistent units of measurement caused by differences in physical attributes between different types of resources is eliminated, thereby mapping the capability levels of different types of resources to a unified dimensionless numerical space.

[0052] In this embodiment, the resource categories include computing resources, storage resources, and network resources, which correspond to the data processing capability of the computing node, the data caching capability of the computing node, and the data transmission capability of the fiber optic link, respectively.

[0053] For computing resources, their fundamental value is mainly reflected in the amount of data that a node can process per unit of time. The fundamental value of computing resources is defined as: ; in, C r Represents computing power nodes r Data processing capability, that is, the amount of data that can be processed per unit of time. C max This represents the maximum data processing capacity of all computing nodes in the optical computing network. V c ( r )∈[0,1].

[0054] For storage resources, their fundamental value is mainly reflected in the node's ability to receive and cache external data. The fundamental value of storage resources is defined as follows: ; in, S r Represents computing power nodes r Available storage resources S max This represents the maximum available storage resources across all computing nodes in the optical computing network. V s ( r )∈[0,1].

[0055] For network resources, their fundamental value is mainly reflected in the optical link's ability to support data transmission and aggregation. The fundamental value of network resources is defined as follows: ; in, B r,r’ Indicates link ( r,r’ The bandwidth resources of the link, i.e., the number of spectrum time slots available for the link. B max This represents the maximum link bandwidth in a computing-powered optical network. V b ( r,r’ )∈[0,1].

[0056] Resource scarcity characterizes the resource utilization of a computing node under the current network load. Resource scarcity is determined based on the ratio of the available resources of the computing node to the total resources. The fewer the available resources and the higher the proportion of total resource utilization, the greater the resource scarcity.

[0057] For computing resources, their scarcity is defined as: ; in, Represents computing power nodes r The amount of computing resources currently available for data processing Represents computing power nodes r The total amount of inherent computing resources.

[0058] For storage resources, their scarcity is defined as: ; in, Represents computing power nodes r Currently available storage resources For computing power nodes r The total amount of inherent storage resources.

[0059] For network resources, their scarcity is defined as: ; in, For link ( r,r’ The number of available spectrum time slots, To represent a link ( r,r’ The inherent number of spectrum time slots.

[0060] After completing the normalization of basic resource value and the modeling of dynamic resource scarcity, the unified resource value of computing nodes is determined by integrating the two. The general definition of the unified resource value is... ,in, This is a scarcity sensitivity coefficient used to adjust the strength of the response of the uniform value of a resource to changes in scarcity.

[0061] Specifically, computing nodes r The unified value of computing processing resources is Computing nodes r The unified value of storage resources is ;link( r,r’ The unified value of bandwidth resources is .

[0062] After determining the unified resource value of each computing node, the data aggregation task... j Neutron mission k The resource cost is defined as: ; in, , , These are respectively represented as subtasks k The expected consumption of computing, storage, and network resources.

[0063] Furthermore, further, data aggregation tasks j The total resource cost is defined as: ; in, For data aggregation tasks j Total number of subtasks included.

[0064] The task collaborative scheduling method provided in this invention obtains the basic value of resources by normalizing the resource capacity of computing nodes, transforms the dynamic occupancy of available resources into resource scarcity, and then integrates the two to obtain a unified value of resources, thereby realizing the quantitative representation of computing resources, storage resources and network resources in a unified value space.

[0065] In some embodiments, before generating a set of candidate scheduling schemes corresponding to the data aggregation task based on resource constraints in response to the currently scheduled data aggregation task, the method further includes: Based on the task data volume, task deadline, and task reliability requirements of each initial data aggregation task, the scheduling priority of each initial data aggregation task is determined. Based on the scheduling priority of each initial data aggregation task, the data aggregation task to be scheduled is determined.

[0066] It should be understood that the initial data aggregation task refers to all data aggregation tasks currently entering the scheduling system and waiting to be processed, including all tasks in the scheduling queue that have not yet been allocated scheduling resources. In actual optical computing networks, multiple data aggregation tasks often arrive at the scheduling system simultaneously. If they are scheduled in an arbitrary order, urgent tasks, tasks with large data volumes, or tasks with high reliability requirements may not be able to be scheduled in a timely manner due to resource contention, thereby affecting the overall task completion rate and service quality of the system.

[0067] Based on this, in this embodiment, before executing the scheduling process, the scheduling priority of each task is pre-calculated according to the attribute characteristics of each initial data aggregation task, so as to determine the order in which each task enters the scheduling process.

[0068] Specifically, the task data volume based on each initial data aggregation task D j Task deadline T j and mission reliability requirements Rreq j This determines the scheduling priority of each initial data aggregation task. Here, the task data volume represents the data aggregation task. j The larger the original data volume involved in the task, the higher the resource requirements for scheduling and processing; resources should be allocated to it first. The task deadline indicates the data aggregation task. j The latest time a task must be completed, the more pressing the deadline, the higher its priority for scheduling; task reliability requirements indicate the data aggregation task... j The minimum requirements for system stability and reliability during execution; the higher the reliability requirement, the more sensitive the task is to system stability.

[0069] In this embodiment, a data aggregation task is defined by combining the above three dimensions of data. j The scheduling priority is: ; in, w 1, w 2, w 3 represents the amount of task data. D j Task deadline T j and mission reliability requirements Rreq j Adjustable weights D max This represents the maximum amount of data for all tasks in the current scheduling queue. The task has inherent time constraints.

[0070] After determining the scheduling priority of each initial data aggregation task, all initial data aggregation tasks are sorted in descending order of scheduling priority to generate an ordered queue of tasks to be scheduled.

[0071] In some embodiments, when multiple initial data aggregation tasks have the same scheduling priority value, the order in which the tasks arrive at the scheduling system can be used as a secondary sorting criterion to ensure the orderliness of the task queue.

[0072] In this embodiment, after the target scheduling scheme for the current task to be scheduled is determined, the next task is taken out from the task queue to be scheduled, and the above scheduling process is repeated until all initial data aggregation tasks are scheduled.

[0073] The task collaborative scheduling method provided in this invention combines a multi-dimensional scheduling priority mechanism that considers the comprehensive task data volume, task deadline, and task reliability requirements. Before formal scheduling, the processing order of tasks is determined in advance, ensuring that tasks with large data volumes, tight deadlines, or high reliability requirements can obtain resource allocation first. This improves the overall task processing efficiency under the condition of limited system resources.

[0074] In some embodiments, the resource constraints include computational resource constraints, storage resource constraints, bandwidth resource constraints, spectrum non-overlapping constraints, spectrum continuity constraints, and spectrum consistency constraints; generating a set of candidate scheduling schemes corresponding to the data aggregation task based on the resource constraints includes: Based on the computing resource constraints and the storage resource constraints, candidate computing nodes corresponding to the data aggregation task are determined; Based on each of the candidate computing power nodes, the bandwidth resource constraints, the non-overlapping constraints, the spectrum continuity constraints, and the spectrum consistency constraints, the corresponding candidate transmission paths are determined. Based on each candidate computing node and its corresponding candidate transmission path, a set of candidate scheduling schemes is generated for the data aggregation task.

[0075] Among these constraints, the computing resource constraint requires that the available computing power of the candidate target computing nodes must meet the minimum computing power requirements of the data aggregation task, ensuring that the computing nodes can complete the data aggregation process within the deadline; the storage resource constraint requires that the available storage resources of the candidate target computing nodes must meet the minimum cache capacity requirements of the data aggregation task, ensuring that the computing nodes can receive and cache intermediate result data transmitted from each source data node; the bandwidth resource constraint requires that the available bandwidth of each fiber link on the candidate transmission path must meet the minimum bandwidth requirements of the data aggregation task, ensuring that the data can be transmitted from the source data node to the target computing node within the specified time; the spectrum non-overlapping constraint, spectrum continuity constraint, and spectrum consistency constraint are physical constraints unique to elastic optical networks, requiring that the spectrum time slots allocated to the data aggregation task for all fiber links on the same transmission path must meet the requirements of non-overlapping, continuity, and consistency, to ensure the physical establishment of the end-to-end optical path.

[0076] In this embodiment, all computing nodes with computing power processing capabilities in the computing power optical network are first obtained. V CC The current resource status, including the available computing power of each node. With available storage resources .

[0077] For the currently scheduled data aggregation task j Based on the data size and processing requirements, determine the minimum required computing power threshold and minimum storage resource threshold. Traverse... V CC All nodes in, will Not less than the minimum computing power required for the task and Nodes with storage resources not less than the minimum required for the task should be added to the candidate computing node set. Ω j As a data aggregation task j The candidate target aggregation node.

[0078] For candidate computing power node set Ω j Each candidate computing node in r' From each source data node r Departure, in the computing power optical network G(V, E) Searching from source data nodes r To candidate computing power nodes r' of K 10 candidate paths are selected to form a candidate path set. P s,r = { p 1 , p 2, ..., p K}

[0079] For each candidate path, it is further examined whether it simultaneously satisfies the bandwidth resource constraints and the constraints of non-overlapping, continuity, and consistency of spectrum. If all fiber links on the path have sufficient available bandwidth and non-overlapping, continuous, and consistent idle spectrum time slot blocks can be found on all links of the path, then the path is a candidate transmission path that satisfies the constraints; otherwise, the path is excluded.

[0080] For the candidate computing power node set Ω j For each candidate computing power node, and all valid candidate transmission paths from each source data node to that candidate computing power node, the combination of the candidate computing power node and each candidate transmission path is recorded as a candidate scheduling scheme and added to the candidate scheduling scheme set. Φ j Each candidate scheduling scheme in the candidate scheduling scheme set includes the identifier of the target aggregate computing power node, the data transmission path from each source data node to the target computing power node, and the spectrum time slot information allocated on each transmission path.

[0081] The task collaborative scheduling method provided in this invention combines computing resource constraints, storage resource constraints, bandwidth resource constraints, and spectrum non-overlapping, continuity, and consistency constraints. This reduces the complexity of candidate scheme generation while ensuring that the selected candidate scheduling schemes are feasible in all dimensions of computing, storage, bandwidth, and spectrum non-overlapping, continuity, and consistency.

[0082] In some embodiments, the resource categories of the computing power nodes include computing resources, storage resources, and network resources; determining the resource cost of each candidate scheduling scheme based on the expected resource consumption of the data aggregation task under each candidate scheduling scheme and the uniform resource value of each computing power node in the candidate scheduling scheme includes: Based on the expected resource consumption of the computing resources, storage resources and network resources by the data aggregation task, and the unified value of the resources corresponding to each resource category, the computing resource cost, storage resource cost and network resource cost are determined respectively. Based on the computing resource cost, the storage resource cost, and the network resource cost, the resource cost of each candidate scheduling scheme is determined.

[0083] In this embodiment, the computational resource cost is the data aggregation task. j Neutron mission k At the target computing power node r The value consumed by the use of computing resources is defined as the uniform value of computing resources. UVc With expected consumption of computing resources The product of storage resources and data aggregation tasks; storage resource costs are the cost of data aggregation tasks. j Neutron mission k At the target computing power node r The value consumed by occupying storage resources is defined as the uniform value of storage resources. UV s With expected consumption of storage resources The product of; network resource costs are the data aggregation task j Neutron mission k The value consumed by occupying network resources on the transmission path is defined as the uniform value of network resources. UV b Expected consumption of network resources The product of.

[0084] After determining the unified resource value of each computing node, the data aggregation task... j Neutron mission k The resource cost is defined as: ; Furthermore, data aggregation tasks j The total resource cost is defined as: ; in, For data aggregation tasks j Total number of subtasks included.

[0085] The task collaborative scheduling method provided in this invention quantifies the contribution of computing resources, storage resources, and network resources to the overall scheduling cost by summarizing the costs of three types of heterogeneous resources under a unified value scale.

[0086] In some embodiments, the expected execution performance includes task completion time; determining the task utility of each candidate scheduling scheme based on the expected execution performance of each candidate scheduling scheme and the resource cost includes: Based on the original data volume of the data aggregation task and the computing resources allocated to the local computing power node, the local processing time of the computing power node is determined. Based on the original data volume and data aggregation selection factor of the data aggregation task, the actual data volume of the computing power node is determined. Based on the actual amount of aggregated data and the computing resources allocated to the target computing power node, the aggregation processing time of the computing power node is determined. The task completion time of the data aggregation task is determined based on the local processing time, the aggregation processing time, and the network communication time of the transmission path in the candidate scheduling scheme. Based on the task completion time and the resource cost, the task utility of each candidate scheduling scheme is determined.

[0087] The execution process of a data aggregation task includes two types of data processing stages. The first stage is that each source data node preprocesses its local raw data, and the amount of data processed is the same as the amount of raw data. The second stage is that the target aggregation node merges the intermediate result data transmitted from each source data node.

[0088] In this embodiment, at the computing node r Data aggregation task j Local processing time Defined as: ; in, Indicates at computing power nodes r Data aggregation tasks in internal storage j The amount of raw data, Represents computing power nodes r Its data processing capabilities.

[0089] Target aggregation node r' Aggregation processing time Defined as: ; in, Dm j,r This indicates the amount of data in the target aggregation node. S a This represents the data selection factor during the data aggregation process. The purpose of aggregation is to reduce the amount of data transmitted over the network, therefore, it is set... S a <1.

[0090] It should be understood that network communication time mainly includes transmission time, propagation time, processing time, and queuing time. Transmission time refers to the time required to deliver data from a node to the network link, which is greatly affected by the data volume and network bandwidth. Propagation time refers to the time required for data to propagate along the link, influenced by the distance between the source and destination nodes. Queuing and processing time at network nodes, specifically at routers / switches, is much shorter than transmission and propagation time. Therefore, in this embodiment, the source data node is determined based on the transmission and propagation times within the network latency. r and target aggregation node r’ The network communication time between them is: ; in, v This represents the time it takes for light to travel a unit distance. v =5us / km; Indicates the source data node r With the target aggregation node r' The total physical distance along the transmission path between them This represents the link bandwidth allocated on the transmission path. The data transmission time represents the time when the data is sent. Data volume through bandwidth The time required for the link to complete transmission.

[0091] Furthermore, data aggregation tasks j Task completion time Defined as the sum of local processing time, maximum network communication time, and aggregated processing time: ; in, g a This represents the set of target computing power nodes.

[0092] Determine the task completion time With resource costs Cost j Then, based on time satisfaction Q t and reliability satisfaction Q r Building Task Benefits Benefit j The task utility is obtained by dividing the task benefit by the resource cost. U j ,in, , τ j Indicates task j The deadline; , R j Indicates task j Reliability satisfaction R min This represents the minimum reliability threshold for receiving the task.

[0093] The task collaborative scheduling method provided in this embodiment of the invention achieves a comprehensive quantitative evaluation of the performance and cost of candidate scheduling schemes through the above-described method.

[0094] In some embodiments, the expected performance also includes reliability satisfaction; determining the task utility of each candidate scheduling scheme based on the task completion time and the resource cost includes: Based on the task completion time and the deadline of the data aggregation task, determine the time satisfaction level; Based on the failure rate distribution of the computing nodes and the transmission path in the candidate scheduling scheme, the reliability satisfaction is determined. Based on the time satisfaction, the reliability satisfaction, and the resource cost, the task utility of each candidate scheduling scheme is determined.

[0095] In this embodiment, time satisfaction Q t This is used to quantify the degree to which candidate scheduling schemes meet the task deadline requirements in terms of time performance. Here, time satisfaction is... Q t The calculation was performed using the exponential decay model described above, and will not be elaborated further here. Q t The larger the value is (∈(0, 1]), the more time the task can be completed relative to the deadline, and the better the time performance; the smaller the value is, the closer the task is to or beyond the deadline, and the worse the time performance.

[0096] Reliability satisfaction Q r This is used to quantify the degree to which candidate scheduling schemes meet the minimum reliability requirements of the task in terms of reliability. In this embodiment, the failure rates of computing nodes and optical links are set to follow an exponential distribution, meaning that the failure rate of a resource in time... t Reliability of normal internal operation R(t) Defined as: ; in, This indicates the failure rate; the higher the value, the less reliable the resource.

[0097] If the data needs to be processed at the computing power nodes, then the computing power nodes... r The node reliability is: ; in, Represents computing power nodes i The failure rate.

[0098] Data from computing nodes r Transmitted to computing nodes r’ ,but( r,r’ The reliability of a link is defined as follows: ; A path consists of multiple optical links connected in series; therefore, the path reliability is: ; in, For the failure rate of the optical link, path p r,r’ ={ l 1, l 2,⋯ , l m}

[0099] Data aggregation task j Overall reliability Defined as all subtasks in a task k The product of the reliability of the source data node, the reliability of the target aggregation node, and the reliability of the transmission path is involved. ; Determining overall reliability Afterwards, the reliability and effectiveness of the task. Q r Defined as: ; in, R min This represents the minimum acceptable reliability threshold for data aggregation tasks. Q r ∈[0, 1], when R j equal R min hour, Q r = 0, indicating that the scheme just barely meets the minimum reliability requirements; when R j When equal to 1, Q r = 1 indicates that the scheme has the highest reliability.

[0100] Finally, a multidimensional utility function is constructed. (Data aggregation task) j Subtasks in k Utility function: ; in, λ 1 and λ 2 These are the weighting coefficients for time satisfaction and reliability satisfaction, respectively, which can be adjusted and configured according to the emphasis on time performance and reliability requirements in actual application scenarios.

[0101] The task collaborative scheduling method provided in this invention constructs a comprehensive task utility function covering three dimensions: time performance, system reliability, and resource consumption cost. This enables scheduling decisions to simultaneously weigh multiple performance objectives under a single utility index, thereby improving the overall quality and system reliability of the scheduling scheme.

[0102] In some embodiments, determining a target scheduling scheme from the candidate scheduling scheme set based on the task utility to execute the data aggregation task includes: Based on the task acceptance status and task rejection status of multiple data aggregation tasks, determine the task rejection penalty items corresponding to the multiple data aggregation tasks; Based on the task utility of each data aggregation task in the task reception state and the task rejection penalty, the global scheduling evaluation benefit is determined. With the goal of maximizing the benefits of the global scheduling evaluation, a target scheduling scheme is determined from the candidate scheduling scheme set to execute the data aggregation task.

[0103] In practical optical computing networks, when there are many tasks to be scheduled or network resources are scarce, the scheduling controller may not be able to find a feasible scheduling scheme that meets the resource constraints for all data aggregation tasks. At this time, some tasks will be in a task rejection state, that is, the task fails to be scheduled and is rejected for execution; correspondingly, tasks that successfully obtain scheduling resource allocation are in a task receiving state.

[0104] For the current scheduling period, the total M A data aggregation task, introducing task receiving state variables. x j To describe the scheduling result of each task: ; The task rejection penalty is a quantitative representation of the system efficiency loss caused by task rejection in the scheduling system, defined as follows: ;in, This is the task rejection penalty coefficient, used to adjust the severity of the penalty imposed by the scheduling system on task rejection behavior. This represents the total number of tasks rejected during the current scheduling period. The scheduling system generates a new schedule for each rejected task. The loss of benefits.

[0105] Finally, the global scheduling evaluation benefit is defined as the sum of the task utilities of all accepted tasks minus the task rejection penalty: ; in, The total utility generated by all received tasks characterizes the scheduling system's performance in terms of task quality.

[0106] Based on this, the benefits are evaluated using global scheduling. F To maximize the optimization objective, we traverse the entire set of candidate scheduling schemes for all tasks to be scheduled, and compare the results of different scheme combinations. F The value of determines the FThe scheduling decision that reaches the maximum value will determine the corresponding candidate scheduling scheme as the target scheduling scheme.

[0107] The task collaborative scheduling method provided in this embodiment of the invention maximizes the overall scheduling efficiency of the system while improving the reception rate of data aggregation tasks, and realizes the efficient collaborative utilization of computing resources, storage resources and network resources in the computing power optical network.

[0108] Based on any of the above embodiments, the present invention also provides a task collaborative scheduling device, with reference to... Figure 3 The device includes: The scheme generation module 310 is used to generate a set of candidate scheduling schemes corresponding to the data aggregation task in response to the data aggregation task, based on resource constraints; each candidate scheduling scheme in the set of candidate scheduling schemes includes the computing power node to be allocated in the computing power optical network and the transmission path; The cost determination module 320 is used to determine the resource cost of each candidate scheduling scheme based on the expected resource consumption of the data aggregation task under each candidate scheduling scheme and the unified resource value of each computing node in the candidate scheduling scheme; wherein, the unified resource value of the computing node is determined based on the basic resource value and resource scarcity of the computing node; the basic resource value is determined based on the resource capacity of the computing node, and the resource scarcity is determined based on the available resources of the computing node; The utility determination module 330 is used to determine the task utility of each of the candidate scheduling schemes based on the expected execution performance of each candidate scheduling scheme and the resource cost. The execution module 340 is used to determine the target scheduling scheme in the candidate scheduling scheme set based on the task utility, so as to execute the data aggregation task.

[0109] The task collaborative scheduling device provided by this invention realizes the comprehensive evaluation of heterogeneous resources in the same value space by constructing a unified value scale based on the basic value and scarcity of resources. This enables resource cost assessment to respond in real time to the dynamic fluctuations of network load, thereby realizing the joint collaborative optimization of heterogeneous resources and improving the task scheduling efficiency and load balancing of the system.

[0110] Figure 4 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 4As shown, the electronic device may include: a processor 410, a communications interface 420, a memory 430, and a communication bus 440, wherein the processor 410, the communications interface 420, and the memory 430 communicate with each other via the communication bus 440. The processor 410 can call logical instructions in the memory 430 to execute a task cooperative scheduling method, which includes: In response to a data aggregation task to be scheduled, a set of candidate scheduling schemes corresponding to the data aggregation task is generated based on resource constraints; each candidate scheduling scheme in the set of candidate scheduling schemes includes a computing power node to be allocated in the computing power optical network and a transmission path; Based on the expected resource consumption of the data aggregation task under each of the candidate scheduling schemes, and the unified resource value of each computing node in the candidate scheduling schemes, the resource cost of each candidate scheduling scheme is determined; wherein, the unified resource value of the computing node is determined based on the basic resource value and resource scarcity of the computing node; the basic resource value is determined based on the resource capacity of the computing node and the global maximum capacity corresponding to the resource category to which the computing node belongs; and the resource scarcity is determined based on the available resources of the computing node and the total resources of the computing node. Based on the expected execution performance of each candidate scheduling scheme and the resource cost, the task utility of each candidate scheduling scheme is determined. Based on the task utility, a target scheduling scheme is determined from the candidate scheduling scheme set to execute the data aggregation task.

[0111] Furthermore, the logical instructions in the aforementioned memory 430 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the conventional, or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc., each of which can store program code.

[0112] On the other hand, embodiments of the present invention also provide a computer program product, the computer program product including a computer program, the computer program being stored on a non-transitory computer-readable storage medium, and when the computer program is executed by a processor, the computer is able to execute the task cooperative scheduling method provided in each of the above embodiments, the method including: In response to a data aggregation task to be scheduled, a set of candidate scheduling schemes corresponding to the data aggregation task is generated based on resource constraints; each candidate scheduling scheme in the set of candidate scheduling schemes includes a computing power node to be allocated in the computing power optical network and a transmission path; Based on the expected resource consumption of the data aggregation task under each of the candidate scheduling schemes, and the unified resource value of each computing node in the candidate scheduling schemes, the resource cost of each candidate scheduling scheme is determined; wherein, the unified resource value of the computing node is determined based on the basic resource value and resource scarcity of the computing node; the basic resource value is determined based on the resource capacity of the computing node and the global maximum capacity corresponding to the resource category to which the computing node belongs; and the resource scarcity is determined based on the available resources of the computing node and the total resources of the computing node. Based on the expected execution performance of each candidate scheduling scheme and the resource cost, the task utility of each candidate scheduling scheme is determined. Based on the task utility, a target scheduling scheme is determined from the candidate scheduling scheme set to execute the data aggregation task.

[0113] In another aspect, embodiments of the present invention also provide a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the task cooperative scheduling method provided in each of the above embodiments, the method comprising: In response to a data aggregation task to be scheduled, a set of candidate scheduling schemes corresponding to the data aggregation task is generated based on resource constraints; each candidate scheduling scheme in the set of candidate scheduling schemes includes a computing power node to be allocated in the computing power optical network and a transmission path; Based on the expected resource consumption of the data aggregation task under each of the candidate scheduling schemes, and the unified resource value of each computing node in the candidate scheduling schemes, the resource cost of each candidate scheduling scheme is determined; wherein, the unified resource value of the computing node is determined based on the basic resource value and resource scarcity of the computing node; the basic resource value is determined based on the resource capacity of the computing node and the global maximum capacity corresponding to the resource category to which the computing node belongs; and the resource scarcity is determined based on the available resources of the computing node and the total resources of the computing node. Based on the expected execution performance of each candidate scheduling scheme and the resource cost, the task utility of each candidate scheduling scheme is determined. Based on the task utility, a target scheduling scheme is determined from the candidate scheduling scheme set to execute the data aggregation task.

[0114] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0115] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the conventional, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0116] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in each of the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of each embodiment of the present invention.

Claims

1. A task collaborative scheduling method, characterized in that, include: In response to the currently scheduled data aggregation task, a set of candidate scheduling schemes corresponding to the data aggregation task is generated based on resource constraints; Each candidate scheduling scheme in the candidate scheduling scheme set includes the computing power nodes to be allocated and the transmission path in the computing power optical network; Based on the expected resource consumption of the data aggregation task under each of the candidate scheduling schemes, and the unified resource value of each computing node in the candidate scheduling schemes, the resource cost of each candidate scheduling scheme is determined; wherein, the unified resource value of the computing node is determined based on the basic resource value and resource scarcity of the computing node; the basic resource value is determined based on the resource capacity of the computing node, and the resource scarcity is determined based on the available resources of the computing node; Based on the expected execution performance of each candidate scheduling scheme and the resource cost, the task utility of each candidate scheduling scheme is determined. Based on the task utility, a target scheduling scheme is determined from the candidate scheduling scheme set to execute the data aggregation task.

2. The task collaborative scheduling method according to claim 1, characterized in that, The unified resource value of the computing power node is determined in the following way: Based on the resource capacity of the computing power node and the global maximum capacity corresponding to the resource category to which the computing power node belongs, the basic resource value of the computing power node is determined. Based on the available resources of the computing power node and the total resources of the computing power node, the resource scarcity of the computing power node is determined; Based on the resource scarcity and the basic value of the resources, the unified resource value of the computing power node is determined.

3. The task collaborative scheduling method according to claim 1, characterized in that, Before generating a set of candidate scheduling schemes corresponding to the data aggregation task based on resource constraints in response to the currently scheduled data aggregation task, the method further includes: Based on the task data volume, task deadline, and task reliability requirements of each initial data aggregation task, the scheduling priority of each initial data aggregation task is determined. Based on the scheduling priority of each initial data aggregation task, the data aggregation task to be scheduled is determined.

4. The task collaborative scheduling method according to claim 1, characterized in that, The resource constraints include computational resource constraints, storage resource constraints, bandwidth resource constraints, spectrum non-overlapping constraints, spectrum continuity constraints, and spectrum consistency constraints; the step of generating a set of candidate scheduling schemes for the data aggregation task based on the resource constraints includes: Based on the computing resource constraints and the storage resource constraints, candidate computing nodes corresponding to the data aggregation task are determined; Based on each of the candidate computing power nodes, the bandwidth resource constraints, the spectrum non-overlapping constraints, the spectrum continuity constraints, and the spectrum consistency constraints, the corresponding candidate transmission paths are determined. Based on each candidate computing node and its corresponding candidate transmission path, a set of candidate scheduling schemes is generated for the data aggregation task.

5. The task collaborative scheduling method according to claim 1, characterized in that, The resource categories of the computing power nodes include computing resources, storage resources, and network resources; based on the expected resource consumption of the data aggregation task under each of the candidate scheduling schemes, and the uniform resource value of each computing power node in the candidate scheduling schemes, the resource cost of each candidate scheduling scheme is determined, including: Based on the expected resource consumption of the computing resources, storage resources and network resources by the data aggregation task, and the unified value of the resources corresponding to each resource category, the computing resource cost, storage resource cost and network resource cost are determined respectively. Based on the computing resource cost, the storage resource cost, and the network resource cost, the resource cost of each candidate scheduling scheme is determined.

6. The task collaborative scheduling method according to claim 1, characterized in that, The expected execution performance includes task completion time; determining the task utility of each candidate scheduling scheme based on the expected execution performance and resource cost includes: Based on the original data volume of the data aggregation task and the computing resources allocated to the local computing power node, the local processing time of the computing power node is determined. Based on the original data volume and data aggregation selection factor of the data aggregation task, the actual data volume of the computing power node is determined. Based on the actual amount of aggregated data and the computing resources allocated to the target computing power node, the aggregation processing time of the computing power node is determined. The task completion time of the data aggregation task is determined based on the local processing time, the aggregation processing time, and the network communication time of the transmission path in the candidate scheduling scheme. Based on the task completion time and the resource cost, the task utility of each candidate scheduling scheme is determined.

7. The task collaborative scheduling method according to claim 6, characterized in that, The expected execution performance also includes reliability satisfaction; determining the task utility of each candidate scheduling scheme based on the task completion time and the resource cost includes: Based on the task completion time and the deadline of the data aggregation task, determine the time satisfaction level; Based on the failure rate distribution of the computing nodes and the transmission path in the candidate scheduling scheme, the reliability satisfaction is determined. Based on the time satisfaction, the reliability satisfaction, and the resource cost, the task utility of each candidate scheduling scheme is determined.

8. The task collaborative scheduling method according to claim 1, characterized in that, The step of determining a target scheduling scheme from the candidate scheduling scheme set based on the task utility to execute the data aggregation task includes: Based on the task acceptance status and task rejection status of multiple data aggregation tasks, determine the task rejection penalty items corresponding to the multiple data aggregation tasks; Based on the task utility of each data aggregation task in the task reception state and the task rejection penalty, the global scheduling evaluation benefit is determined. With the goal of maximizing the benefits of the global scheduling evaluation, a target scheduling scheme is determined from the candidate scheduling scheme set to execute the data aggregation task.

9. A task collaborative scheduling device, characterized in that, include: The scheme generation module is used to generate a set of candidate scheduling schemes corresponding to the data aggregation task to be scheduled, based on resource constraints. Each candidate scheduling scheme in the candidate scheduling scheme set includes the computing power nodes to be allocated and the transmission path in the computing power optical network; The cost determination module is used to determine the resource cost of each candidate scheduling scheme based on the expected resource consumption of the data aggregation task under each candidate scheduling scheme and the unified resource value of each computing node in the candidate scheduling scheme; wherein, the unified resource value of the computing node is determined based on the basic resource value and resource scarcity of the computing node; the basic resource value is determined based on the resource capacity of the computing node, and the resource scarcity is determined based on the available resources of the computing node; The utility determination module is used to determine the task utility of each candidate scheduling scheme based on the expected execution performance of each candidate scheduling scheme and the resource cost. An execution module is used to determine a target scheduling scheme from the candidate scheduling scheme set based on the task utility, so as to execute the data aggregation task.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the task cooperative scheduling method as described in any one of claims 1 to 8.

Images