Resource scheduling method and device, electronic equipment and storage medium

Abstract

The invention discloses a resource scheduling method and device, electronic equipment and a storage medium, and relates to the technical field of computing power scheduling. The method comprises the steps of determining semantic features of task instruction data, constructing a clockwise execution chain flow and an anticlockwise execution chain flow according to the semantic features and the task instruction data, and performing standardization processing on the clockwise execution chain flow and the anticlockwise execution chain flow to generate a chain task graph structure; determining instruction association data, determined for the task chain instruction, of a clockwise execution chain flow and an anticlockwise execution chain flow in the chain type task graph structure, and determining a coupling symmetry relation of the task chain instruction according to the instruction association data; when the coupling symmetric relationship meets a preset condition, determining a resource adjustment strategy according to a cycle category corresponding to the task chain instruction, and determining a resource matching result according to the resource adjustment strategy; and determining a configuration energy value of the quantum task in the task chain instruction of the forward execution chain flow, fusing the resource matching results according to the configuration energy value to obtain a target resource matching result, and performing hierarchical scheduling according to the target resource matching result.

Description

Technical Field

[0001] This invention relates to the field of computing power scheduling technology, and in particular to a resource scheduling method, apparatus, electronic device and storage medium. Background Technology

[0002] In heterogeneous computing networks, the smooth execution of task chains depends on the task capabilities of different heterogeneous networks. Heterogeneous networks will, to some extent, support the capability borrowing needs of the central node based on their own offloadable capabilities. After allocating offloadable capabilities to the central node, the heterogeneous network will perform load balancing of network resources according to its own balancing algorithm.

[0003] After the central network completes its task, it releases the heterogeneous computing power it holds. If this power is released directly through the original network, it disrupts the load balancing of the heterogeneous networks, causing imbalances to reappear. Although the heterogeneous networks can regain balance through load balancing, this increases network overhead. Therefore, how to rationally allocate resources has become a pressing issue. Summary of the Invention

[0004] This invention provides a resource scheduling method, apparatus, electronic device, and storage medium to solve the problem of resource imbalance in networks in the prior art.

[0005] According to one aspect of the present invention, a resource scheduling method is provided, wherein the method includes:

[0006] Obtain task instruction data, determine the semantic features of the task instruction data, construct forward execution chain flow and reverse execution chain flow based on the semantic features and the task instruction data, perform standardization processing on the forward execution chain flow and reverse execution chain flow, and generate a chain-like task graph structure;

[0007] Determine the instruction association data for the forward execution chain and the reverse execution chain in the chained task graph structure with respect to the task chain instructions, and determine the coupling symmetry relationship of the task chain instructions based on the instruction association data;

[0008] When it is determined that the coupling symmetry relationship meets the preset conditions, a resource adjustment strategy is determined according to the loop category corresponding to the task chain instruction, and a resource allocation result is determined according to the resource adjustment strategy.

[0009] Identify the quantum task in the task chain instructions of the sequential execution flow, determine the configuration energy value of the quantum task, fuse the resource allocation result according to the configuration energy value to obtain the target resource allocation result, and perform hierarchical scheduling according to the target resource allocation result.

[0010] According to another aspect of the present invention, a resource scheduling apparatus is provided, wherein the apparatus comprises:

[0011] The task graph construction module is used to acquire task instruction data, determine the semantic features of the task instruction data, construct forward execution chain flow and reverse execution chain flow based on the semantic features and the task instruction data, perform standardization processing on the forward execution chain flow and reverse execution chain flow, and generate a chain-like task graph structure.

[0012] The relationship determination module is used to determine the instruction association data of the forward execution chain and the reverse execution chain in the chained task graph structure for the task chain instructions, and to determine the coupling symmetry relationship of the task chain instructions based on the instruction association data;

[0013] The allocation determination module is used to determine a resource adjustment strategy based on the loop category corresponding to the task chain instruction when it is determined that the coupling symmetry relationship meets the preset conditions, and to determine the resource allocation result according to the resource adjustment strategy.

[0014] The resource scheduling module is used to determine the quantum task in the task chain instructions of the sequential execution chain, determine the configuration energy value of the quantum task, fuse the resource allocation result according to the configuration energy value to obtain the target resource allocation result, and perform hierarchical scheduling according to the target resource allocation result.

[0015] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0016] At least one processor; and

[0017] A memory communicatively connected to the at least one processor; wherein,

[0018] The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the resource scheduling method according to any embodiment of the present invention.

[0019] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to execute and implement the resource scheduling method described in any embodiment of the present invention.

[0020] According to another aspect of the present invention, embodiments of the present invention also provide a computer program product, the computer program product including a computer program, which, when executed by a processor, implements the resource scheduling method of any embodiment of the present invention.

[0021] The technical solution of this invention involves acquiring task instruction data, determining the semantic features of the task instruction data, constructing forward and reverse execution chains based on the semantic features and the task instruction data, standardizing the forward and reverse execution chains to generate a chained task graph structure, determining the instruction association data of the forward and reverse execution chains for the task chain instructions in the chained task graph structure, determining the coupling symmetry relationship of the task chain instructions based on the instruction association data, and when the determined coupling symmetry relationship meets preset conditions, determining a resource adjustment strategy based on the loop category corresponding to the task chain instructions, and determining the resource allocation result according to the resource adjustment strategy. The system identifies quantum tasks within the task chain instructions of the forward execution flow, determines the configuration energy value of each quantum task, and fuses the resource allocation results based on the configuration energy value to obtain the target resource allocation result. Hierarchical scheduling is then performed according to the target resource allocation result. By constructing forward and reverse execution flows, the system achieves automatic identification and separation of cyclic and acyclic tasks. Separate resource adjustment strategies are designed for cyclic and acyclic tasks, avoiding low-accuracy scheduling methods and improving the matching precision of tasks and resources. Simultaneously, a hierarchical scheduling and execution mechanism is designed at the node layer, intra-domain layer, and cross-domain layer, enabling the orderly release and scheduling of resources across multiple heterogeneous networks.

[0022] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart of a resource scheduling method provided in Embodiment 1 of the present invention;

[0025] Figure 2 This is a flowchart of a resource scheduling method provided according to Embodiment 2 of the present invention;

[0026] Figure 3 This is a flowchart of a resource scheduling method provided according to Embodiment 3 of the present invention;

[0027] Figure 4 This is a flowchart of a method for determining resource allocation results according to Embodiment 3 of the present invention;

[0028] Figure 5This is a schematic diagram of the structure of a resource scheduling device according to Embodiment 3 of the present invention;

[0029] Figure 6 This is a schematic diagram of the structure of an electronic device that implements a resource scheduling method according to an embodiment of the present invention. Detailed Implementation

[0030] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0031] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0032] Example 1

[0033] Figure 1 This is a flowchart of a resource scheduling method according to Embodiment 1 of the present invention. This embodiment is applicable to resource scheduling in a computing network system. The method can be executed by a resource scheduling device, which can be implemented in hardware and / or software and can be configured in an electronic device, such as a computing network system. Figure 1 As shown, the method includes:

[0034] S110. Obtain task instruction data, determine the semantic features of the task instruction data, construct forward execution chain flow and reverse execution chain flow based on the semantic features and task instruction data, standardize the forward execution chain flow and reverse execution chain flow, and generate a chain-like task graph structure.

[0035] Task instruction data can be understood as a structured data set containing key information such as attributes, execution requirements, resource consumption, and quantum behavior of each instruction in the task chain. Generally, task instruction data can be sent by the upper-level task scheduling center. In one embodiment, task instruction data may include task type labels, such as quantum factorization, behavior classification, and key distribution; task scale index vectors, including parameter quantity, communication span, and time window; performance constraint vectors, such as maximum allowable latency, fault tolerance level, and bandwidth requirements; data structure descriptions, such as input tensor dimensions and node dependency graphs; and a set of control instructions, such as prefer_quantum=true and max_latency=5ms. Semantic features can be understood as the core business logic and execution requirement information carried in the task instruction data that can be recognized and parsed by the system. Generally, semantic features may include control type and constraint values. The forward execution chain flow, with the positive dependency relationship of task instructions as its core, represents the execution link from the starting node to the ending node, indicating the data flow path between task nodes when the task proceeds normally as originally planned. The reverse execution chain is a fallback path centered on reverse dependencies, which only occurs when an instruction in the forward execution chain fails. The chained task graph structure is a visualization and formal modeling of the forward execution chain, the reverse execution chain, and the dependencies between task instructions. It uses a directed graph as its core structure, abstracting task instructions as nodes and the dependencies between instructions as directed edges.

[0036] In this embodiment, task instruction data can be received, control instructions within the task instruction data can be determined, and the corresponding control type and constraint value of the control instructions can be determined. The control type and constraint value are then used as semantic features of the task instruction data. In actual operation, semantic features may include, but are not limited to, resource preferences and priority selection of quantum nodes, latency limits to a maximum latency of 5ms, and node affinity restrictions to execution on E1 or E2 nodes. The dependencies between task instruction data are then determined, and forward and reverse execution chains are constructed according to the dependencies and semantic features. In one embodiment, semantic features can be used as constraint information for the forward and reverse execution chains. The task node format, edge format, and constraint information format of the forward and reverse execution chains are unified, and a directed graph is constructed as a chain-like task graph structure.

[0037] S120. Determine the instruction association data for the forward and reverse execution flow in the chained task graph structure, and determine the coupling symmetry relationship of the task chain instructions based on the instruction association data.

[0038] In this context, task chain instructions are the basic units that constitute the forward execution flow, reverse execution flow, and chained task graph. They are the smallest executable units that carry core information such as task execution logic, resource requirements, and quantum characteristics. Generally, task chain instructions can include loop instructions and non-loop instructions. Instruction-associated data can be understood as data associated with task chain instructions. In actual operation, instruction-associated data can include, but is not limited to, input data vectors, output data vectors, and state change vectors of the forward and reverse execution flows. The coupling-symmetry relationship can be understood as coupling and symmetry.

[0039] In the embodiments, the input and output data vectors of the forward and reverse execution chain flows in the chained task graph structure, as well as the state change vectors of the forward and reverse execution chain flows, can be determined. Coupling is determined based on the input and output data vectors, and symmetry is determined based on the state change vectors of the forward and reverse execution chain flows. In actual operation, for loop instructions in the task chain, the input and output data vectors of the loop instruction in each loop can be determined. The estimated value of the linear / nonlinear data processing function for the input data vector can be determined. The Euclidean norm of the output data vector and the estimated value of the data processing function in each loop can be used as a stability vector. The rate of change of the stability vector with respect to loop cycles can be determined. If the rate of change is greater than a preset jump threshold, the trend can be determined as a jump. The absolute value of the difference in stability vector between the critical loop where the jump occurs and the next loop after the critical loop is determined. When the absolute value is greater than the jump sensitivity constant, it is determined that a second-order phase transition coupling segment exists near the critical loop of the task chain instruction. For non-looping instructions in the task chain, the input and output data vectors at different time points can be determined. The estimated value of the linear / nonlinear data processing function of the input data vector can be determined. The Euclidean norm of the output data vector and the estimated value of the data processing function in each round can be determined as the stability vector. The rate of change of the stability vector with respect to the loop rounds can be determined. If the rate of change is greater than the preset jump threshold, the trend can be determined as a jump. The absolute value of the difference between the stability vector at the time point of the jump and the next time point can be determined. When the absolute value is greater than the jump sensitivity constant, it is determined that there is a second-order phase transition coupling segment near the time point of the jump in the task chain instruction.

[0040] Next, determine the state change vectors of the same task chain instructions in the forward and reverse execution flows, i.e., the resource state change vector on the forward execution path and the resource recovery vector on the reverse rollback path. Then, determine the Euclidean norm product of the resource state change vector on the forward execution path and the resource recovery vector on the reverse rollback path, respectively. The ratio of the stability vector to the product is then used as the path symmetric vector. The closer the threshold is to 1, the more symmetrical the path; the closer it is to -1, the more contradictory the paths, requiring rescheduling. Generally, multiple thresholds can be set for symmetric path vectors. For example, when... The symmetry is determined to be perfect symmetry; when The symmetry is determined to be a fundamental symmetry; when The symmetry is determined to be asymmetric; when The symmetry can be set to complete opposition, etc., according to user needs. After determining the second-order phase transition coupling segment and symmetry, the coupling symmetry relationship can be determined according to the second-order phase transition coupling segment and symmetry. In actual operation, when the instruction does not have a second-order phase transition coupling segment, if the symmetry is satisfied ( If the symmetry is not satisfied, it is determined that the coupling symmetry relationship is satisfied; if the symmetry is not satisfied ( The coupling symmetry relationship is not satisfied when the instruction contains a second-order phase transition coupling segment. ), is determined to be a coupled symmetric relationship that is satisfied; if the symmetry is not satisfied ( This is considered an indication that the coupling symmetry relationship is not satisfied. Alternatively, if it is determined that the coupling has a second-order phase transition coupling, and the symmetry satisfies the first symmetry interval within the preset symmetry interval (e.g., ...), then... The coupling symmetry relationship is determined to satisfy the preset conditions; when it is determined that there is no second-order phase transition coupling, if the symmetry satisfies the second symmetry interval in the preset symmetry interval (e.g., The determination of the coupled symmetric relationship is based on the preset conditions, and the judgment rules can be set according to user requirements.

[0041] S130. When the coupling symmetry relationship is determined to meet the preset conditions, the resource adjustment strategy is determined according to the loop category corresponding to the task chain instruction, and the resource allocation result is determined according to the resource adjustment strategy.

[0042] Resource adjustment strategies refer to the strategies for adjusting cross-domain heterogeneous computing resources (such as Graphics Processing Units (GPUs), Central Processing Units (CPUs), and edge nodes). Generally, the resource adjustment strategies differ for different loop categories. Loop categories can include loop instructions and non-loop instructions. Resource allocation results can include the consumption ratio of task instructions on each type of heterogeneous computing resource and the associated strategy set. Generally, resource allocation results can include a resource allocation vector, which numerically represents the consumption ratio of each resource type by the task (summing to 1). Generally, resource allocation results can include a resource allocation vector and scheduling rules bound to the resource allocation results. Scheduling rules bound to the resource allocation results include: prioritizing resources (resources with the highest proportion in the allocation vector); and discarding resources (resources marked as uncontrollable with a high proportion in non-loop instructions).

[0043] In this embodiment, when the coupling symmetry relationship is satisfied, a preset condition is met, and the loop category corresponding to the task chain instruction can be determined. For loop instructions, the heterogeneous computing power resource consumption structure corresponding to their instruction execution can be extracted to construct a resource allocation vector. For non-loop instructions, resource consumption paths with low symmetry are eliminated and redundant resources are released according to the symmetry index. Specifically, for loop instructions, the heterogeneous computing power resource consumption vector of each round of instructions can be determined, and the average value of the resource consumption vectors of all rounds can be used as the stable resource allocation vector of the loop instruction in different resource types. The resource with the highest distribution in the stable resource allocation vector is determined as the preferred resource, and the stable resource allocation vector and the preferred resource are used as the resource allocation result. For non-loop instructions, when it is determined that the non-loop instruction has a second-order phase transition coupling, a stability risk is determined. In the case of a stability risk, a resource component with a high proportion but uncontrollable value is identified as the target resource, and the target resource is discarded. The remaining resources after removing the discarded resource are normalized to obtain the adjusted resource allocation vector of the non-loop instruction. The resource allocation vector and the discarded resource are used as the resource allocation result.

[0044] S140. Determine the quantum task in the task chain instructions of the sequential execution chain, determine the configuration energy value of the quantum task, fuse the resource allocation results according to the configuration energy value to obtain the target resource allocation result, and perform hierarchical scheduling according to the target resource allocation result.

[0045] In this context, a quantum task refers to a specific operational task that relies on the principles of quantum mechanics (superposition, entanglement, interference), requires the participation of quantum resources (qubits, quantum gates, quantum nodes, etc.), and is executed through a pure quantum or quantum-classical hybrid paradigm to solve high-dimensional, combinatorial optimization, or nonlinear problems that are difficult to process efficiently with classical computing power. The configuration energy value is a numerical value obtained by weighting, quantizing, and normalizing the historical quantum behavior indicators of the quantum task (number of state superpositions, quantum jump amplitudes, effective interference frequency, etc.). Generally, the configuration energy value reflects the task's dependence on quantum nodes; the higher the configuration energy, the stronger the quantum interference, the greater the computational complexity, and the higher the priority requirement for quantum window scheduling. The target resource allocation result can be understood as the resource allocation result after adjusting the attitude energy value. Generally, the target resource allocation result can include the preferred / discarded resources and the corrected resource proportions for the quantum task.

[0046] In this embodiment, within the operator's computing power network scheduling scenario, the sequential execution chain contains multiple highly complex task instructions that need to be bound to quantum nodes for execution. These instructions can be extracted as quantum tasks. The quantum behaviors of these tasks, such as state superposition, quantum jump amplitude, and interference frequency, can be analyzed during their historical execution to calculate their corresponding configuration energy level as the configuration energy value. The configuration energy value is normalized to obtain the target configuration energy value. The product of the target configuration energy value and a preset adjustment factor is determined as the first product. The sum of the first product and a preset value is used as a correction factor. The product of the correction factor and the resource allocation result is determined as the target resource allocation result. Generally, if the configuration energy value is high, the priority for quantum computing resources (such as a central GPU cluster) is increased in the resource allocation result, reducing the dependence on ordinary CPUs or edge nodes. If the configuration energy value is low, it indicates that it can accept looser resource matching, and the system can prioritize the execution of more available edge nodes, thereby alleviating the pressure on central resources.

[0047] Based on the target resource allocation results, the release and scheduling of multi-layer heterogeneous network resources are implemented, constructing hierarchical scheduling execution strategies at the node layer, intra-domain layer, and cross-domain layer. Generally, priority can be given to completing the release of computing power and node binding in the local resource pool. The scheduling center maps each task to GPU, CPU, or edge execution unit according to its modified resource preference: if the current node resources meet the allocation requirements, the task is directly scheduled for execution; if the node's computing resources are insufficient, or the current time window is unavailable, the task is marked as a node-level scheduling failure and passed to the next scheduling layer for further processing. The node-level feedback marking matrix constructed here will serve as one of the input data for intra-domain scheduling. When a task fails to be successfully bound at the local node layer, the system enters the resource pool within the city or operator's region for intra-domain scheduling. Task scheduling at this stage strictly relies on the execution flags and failure reasons returned by the node layer: if a task fails only due to insufficient resources, a GPU or CPU node with higher computing power and stable connection is searched within the region for supplementary scheduling; if the failure is due to task execution window conflicts or link blockage, the task window period is dynamically adjusted, and a feasible path is regenerated based on the intra-domain node state graph; then, the system is upgraded to a cross-domain scheduling mechanism. When a task cannot find suitable resources within the current computing network domain, the system triggers the computing network federated scheduling mechanism to perform cross-domain task release in a wider range of operator resource pools. The scheduling decision at this stage relies on the intra-domain scheduling failure reason matrix and task urgency score: the system prioritizes finding remote computing power domains that are highly matched with the current task in terms of task attributes, resource type, latency sensitivity, etc.; a resource mapping request is initiated through the federated coordinator deployed in the computing network center, and the optimal migration path is calculated based on the network topology and load status; if cross-domain scheduling is successful, the system updates the global resource state graph and synchronizes available resource change information to the node layer and intra-domain layer, forming a dynamic resource return and state sharing mechanism.

[0048] In this embodiment of the invention, task instruction data is acquired, and its semantic features are determined. Based on these semantic features and the task instruction data, forward and reverse execution chains are constructed. These chains are then standardized to generate a chained task graph structure. The instruction association data between the forward and reverse execution chains and the task chain instructions is determined. The coupling symmetry relationship between the task chain instructions is determined based on this instruction association data. When the coupling symmetry relationship meets preset conditions, a resource adjustment strategy is determined based on the loop category corresponding to the task chain instructions. The resource allocation result is then determined according to the resource adjustment strategy, and the forward execution chain is finalized. The system executes quantum tasks within the task chain instructions of the execution flow, determines the configuration energy value of the quantum tasks, and fuses the resource allocation results based on the configuration energy value to obtain the target resource allocation result. Hierarchical scheduling is then performed according to the target resource allocation result. By constructing forward and reverse execution flow chains, the system achieves automatic identification and separation of cyclic and acyclic tasks. Separate resource adjustment strategies are designed for cyclic and acyclic tasks, avoiding low-accuracy scheduling methods and improving the matching precision of tasks and resources. Simultaneously, a hierarchical scheduling execution mechanism is designed at the node layer, intra-domain layer, and cross-domain layer, enabling the orderly release and scheduling of resources across multiple heterogeneous networks.

[0049] In one embodiment, after determining the instruction association data for the forward and reverse execution flow in the chained task graph structure relative to the task chain instructions, and determining the coupling symmetry relationship of the task chain instructions based on the instruction association data, the method further includes:

[0050] If the coupling symmetry relationship does not meet the preset conditions, delete any task instruction data and reconstruct the forward execution chain and the reverse execution chain.

[0051] In an embodiment, if it is determined that the coupling symmetry relationship does not meet the preset conditions, any task instruction data can be deleted, and the forward execution chain and reverse execution chain can be reconstructed until the coupling symmetry relationship meets the preset conditions.

[0052] Example 2

[0053] Figure 2 This is a flowchart of a resource scheduling method according to Embodiment 2 of the present invention. This embodiment is a further optimization and extension based on the above embodiments, and can be combined with various optional technical solutions in the above embodiments. Figure 2 As shown, the method includes:

[0054] S201. Receive task instruction data, determine the control instructions in the task instruction data, and split the control instructions into instruction prefixes and instruction values.

[0055] In this embodiment, task instruction data can be received, and control instructions can be extracted from the task instruction data. The control instructions can be split into instruction prefixes and instruction values ​​using the field information in the control instructions.

[0056] S202. Match the control type associated with the instruction prefix, determine the parsing rules associated with the control type, extract the constraint values ​​of the control instructions according to the parsing rules, and use the control type and constraint values ​​as semantic features of the task instruction data.

[0057] Control type is a classification dimension for control instructions. For example, control type can include resource preference, latency limit, and node affinity. Constraint values ​​can include, but are not limited to, prioritizing quantum nodes, a maximum latency of 5ms, and limiting execution to E1 or E2 nodes.

[0058] In this embodiment, the control type associated with the instruction prefix can be determined, the constraint value of the control instruction can be extracted according to the control type determination parsing rules, and the control type and constraint value can be used as semantic features of the task instruction data.

[0059] S203. Determine the dependencies between task requests, construct forward and reverse execution chains according to the dependencies, and use semantic features as constraint information for task requests in the forward and reverse execution chains.

[0060] In this embodiment, dependencies may include data dependencies, resource dependencies, and timing dependencies, etc. Based on these dependencies, forward execution flow and reverse execution flow can be constructed respectively. For example, the forward execution flow is: The reverse execution chain is as follows: (like (In case of failure) The system uses data read / write dependencies between tasks, state transition instructions, and rollback trigger conditions. Semantic features are then used as constraint information for task requests in both the forward and reverse execution flows.

[0061] S204. Standardize the forward and reverse execution flow to generate a directed graph as a chained task graph structure.

[0062] In this embodiment, the forward and reverse execution chains can be standardized by unifying the format of task nodes, edges, and constraint information. The execution status (success, ready, rollback, exception) of each task node on the chain is dynamically labeled, generating a directed graph as a chain-like task graph structure. In actual operation, the chain-like task graph structure records information including: task node status, compatibility with computing resource types, feasibility of time window binding, and whether quantum node window collaborative computation is involved.

[0063] S205. Identify task chain instructions in the forward execution chain and the reverse execution chain, and determine the input data vector and output data vector of each round of instructions in the loop instructions in the task chain, as well as the input data vector and output data vector of non-loop instructions in the task chain at different time points during execution.

[0064] The input data vector refers to the structured numerical set of all input data required during the execution of the task chain instructions, including raw data, intermediate results, control parameters, etc. The output data vector refers to the structured numerical set of all result data output after the task chain instructions are completed, including business results, status feedback, resource change records, etc.

[0065] In this embodiment, the input data vector and output data vector of each round of instructions in the loop instructions of the task chain instructions can be collected, as well as the input data vector and output data vector of non-loop instructions in the task chain instructions at different time points during execution.

[0066] S206. Determine the objective function processing estimate of the input data vector, and use the output data vector of the task chain instructions belonging to the same round / same time point and the Euclidean norm of the objective function processing estimate as the stability vector.

[0067] The objective function processing estimate can be understood as the function estimate of the input data vector after linear / nonlinear data processing. The stability vector is a structured numerical vector describing the fluctuation of key indicators such as resource consumption in multiple rounds (or multiple stages of execution) of a single task instruction or the entire execution chain.

[0068] In the embodiments, the objective function processing estimate of the input data vector can be determined separately. For loop instructions, the output data vector of the task chain instructions belonging to the same round and the Euclidean norm of the objective function processing estimate can be used as the stability vector. For non-loop instructions, the output data vector of the task chain instructions belonging to the same time point and the Euclidean norm of the objective function processing estimate can be used as the stability vector.

[0069] S207. Determine the trend of change according to the stability vector and the preset mutation threshold, and determine the coupling according to the trend of change and the jump sensitivity constant.

[0070] The trend of change can include jumps and non-jumps. Coupling can include the presence of second-order phase transition coupling and the absence of second-order phase transition coupling. The preset mutation threshold can be a pre-set critical value for judging whether a jump has occurred, and the jump sensitivity constant is a critical value used to determine the coupling. Generally speaking, the preset mutation threshold and jump sensitivity constant can be set according to business requirements.

[0071] In this embodiment, for loop instructions in the task chain, the rate of change of the stability vector with respect to the loop iterations can be determined. If the rate of change is greater than a preset jump threshold, the trend can be determined as a jump. The absolute value of the difference in stability vector between the critical iteration where the jump occurs and the next iteration is determined. When the absolute value is greater than the jump sensitivity constant, it is determined that a second-order phase transition coupling segment exists near the critical iteration loop of the task chain instruction. For non-loop instructions in the task chain, the rate of change of the stability vector with respect to the loop iterations can be determined. If the rate of change is greater than a preset jump threshold, the trend can be determined as a jump. The absolute value of the difference in stability vector between the time point where the jump occurs and the next time point is determined. When the absolute value is greater than the jump sensitivity constant, it is determined that a second-order phase transition coupling segment exists near the time point loop where the jump occurs in the task chain instruction.

[0072] S208. Determine the state change vectors of the same task chain instructions in the forward and reverse execution flow, determine the path symmetry vector according to the state change vector and the stability vector, and determine the symmetry of the task chain instructions according to the preset symmetry interval.

[0073] The state change vector includes resource state change vectors on the forward execution path and resource recovery vectors on the reverse rollback path. The number of preset symmetrical intervals can be one or more; for example, the preset symmetrical intervals can be... .

[0074] In this embodiment, the Euclidean norms of the resource state change vector on the forward execution path and the resource recovery vector on the reverse rollback path can be determined separately. The product of these Euclidean norms is then determined, and the ratio of the stability vector to the Euclidean norm is used as the path-symmetric vector. Generally, the path-symmetric vector... When the path symmetry vector satisfies the preset symmetry interval, the task chain instructions can be determined to satisfy symmetry.

[0075] S209. Determine the coupling-symmetry relationship according to coupling and symmetry.

[0076] In one embodiment, determining the coupling symmetry relationship according to coupling and symmetry includes:

[0077] When the coupling is determined to be a second-order phase transition coupling, if the symmetry is satisfied, the coupling symmetry relationship is determined to satisfy the preset condition; if the symmetry is not satisfied, the coupling symmetry relationship is determined to not satisfy the preset condition.

[0078] When it is determined that there is no second-order phase transition coupling, if the symmetry is satisfied, the coupling symmetry relationship is determined to satisfy the preset condition; if the symmetry is not satisfied, the coupling symmetry relationship is determined to not satisfy the preset condition.

[0079] In the embodiments, if the coupling is a second-order phase transition coupling and satisfies symmetry, the coupling symmetry relationship can be determined to satisfy the preset condition; if the coupling is a second-order phase transition coupling and satisfies symmetry but does not satisfy symmetry, the coupling symmetry relationship can be determined to not satisfy the preset condition; if the coupling is a second-order phase transition coupling and satisfies symmetry, the coupling symmetry relationship can be determined to satisfy the preset condition; if the coupling is a second-order phase transition coupling and does not satisfy symmetry, the coupling symmetry relationship can be determined to not satisfy the preset condition.

[0080] S210, Determine the loop type of the task chain instructions.

[0081] The loop category includes loop instructions and non-loop instructions.

[0082] S211. For cyclic instructions, determine the heterogeneous computing power resource consumption vector for each round of instructions, take the average value of the resource consumption vectors for all rounds, obtain the stable resource allocation vector of cyclic instructions in different resource types, determine the resource with the highest distribution in the stable resource allocation vector as the preferred resource, and use the stable resource allocation vector and the preferred resource as the resource allocation result.

[0083] In this embodiment, when the task chain instruction is determined to be a cyclic instruction, the average value of the heterogeneous computing power resource consumption vector of each round of instructions can be calculated as the stable resource allocation vector of the cyclic instruction under different resource types. The resource with the highest distribution in the stable resource allocation vector is determined as the preferred resource, and the stable resource allocation vector and the preferred resource are determined as the resource allocation result.

[0084] S212. For non-cyclic instructions, determine the stability risk according to the coupling of the non-cyclic instructions. When there is a stability risk, identify the target resource, treat the target resource as the discarded resource, normalize the remaining resources after removing the discarded resource, and obtain the resource allocation vector after the non-cyclic instruction is adjusted. Use the resource allocation vector and the discarded resource as the resource allocation result.

[0085] In this embodiment, when the task chain instruction is determined to be a non-cyclic instruction, if it is determined that there is a second-order phase transition coupling, it can be determined that a stability risk exists. At this time, the resource component with a high proportion but uncontrollable value can be identified as the target resource and discarded. Then, the remaining resources after removing the discarded resources are normalized to obtain the resource allocation vector after the non-cyclic instruction is adjusted. The resource allocation vector and the discarded resources are used as the resource allocation result.

[0086] S213. Extract the quantum task from the task chain instructions in the sequential execution flow, determine the historical quantum behavior index of the quantum task, and determine the configuration energy value according to the historical quantum behavior index.

[0087] In the embodiment, quantum tasks in the task chain instructions of the sequential execution flow can be extracted, and historical quantum behavior indicators such as state superposition, quantum jump amplitude, and interference frequency of the quantum tasks in the historical operation process can be determined. The corresponding configuration energy level can be calculated as the configuration energy value.

[0088] S214. The target configuration energy value is obtained by normalizing the configuration energy value.

[0089] In this embodiment, the configuration energy value can be normalized to determine the target configuration energy value.

[0090] S215. Determine the product of the target configuration energy value and the preset adjustment factor as the first product, use the sum of the first product and the preset value as the correction factor, determine the product of the correction factor and the resource allocation result as the target resource allocation result, and perform hierarchical scheduling according to the target resource allocation result.

[0091] In this embodiment, the product of the target configuration energy value and the preset adjustment factor can be calculated as the first product, the sum of the first product and the preset value can be determined as the correction factor, the product of the correction factor and the resource allocation result can be used as the target resource allocation result, and hierarchical scheduling can be performed according to the target resource allocation result.

[0092] In this embodiment of the invention, by receiving task instruction data, the control instructions in the task instruction data are determined. The control instructions are split into instruction prefixes and instruction values. The control type associated with the instruction prefix is ​​matched, and the parsing rule associated with the control type is determined. The constraint value of the control instruction is extracted according to the parsing rule. The control type and constraint value are used as semantic features of the task instruction data to determine the dependency relationship between task requests. According to the dependency relationship, a forward execution chain and a reverse execution chain are constructed. The semantic features are used as constraint information for task requests in the forward execution chain and the reverse execution chain. The forward execution chain and the reverse execution chain are standardized to generate a directed graph as a chain-like task graph structure, thereby achieving accurate establishment of the forward execution chain and the reverse execution chain. By identifying task chain instructions in the forward execution chain and the reverse execution chain, the input data vector and output data vector of each round of instructions in the loop instructions in the task chain instructions are determined, as well as the input data vector and output data vector of non-loop instructions in the task chain instructions at different time points during execution. The objective function processing estimate of the input data vector is determined, and the output data vectors of task chain instructions belonging to the same round / same time point are classified. The Euclidean norm of the estimated value processed by the objective function is used as the stability vector. The change trend is determined according to the stability vector and the preset mutation threshold. The coupling is determined according to the change trend and the jump sensitivity constant. The state change vectors of the same task chain instructions in the forward and reverse execution flow are determined. The path symmetry vector is determined according to the state change vector and the stability vector. The symmetry of the task chain instructions is determined according to the preset symmetry interval. The coupling symmetry relationship is determined according to the coupling and symmetry. The loop category of the task chain instructions is determined. The resource allocation result is determined. The quantum tasks in the task chain instructions of the forward execution flow are extracted. The historical quantum behavior index of the quantum tasks is determined. The configuration energy value is determined according to the historical quantum behavior index. The configuration energy value is normalized to obtain the target configuration energy value. The product of the target configuration energy value and the preset adjustment factor is determined as the first product. The sum of the first product and the preset value is determined as the correction factor. The product of the correction factor and the resource allocation result is determined as the target resource allocation result. Hierarchical scheduling is performed according to the target resource allocation result. The resource allocation result is constructed and the configuration energy value correction mechanism is integrated to achieve high-fitness scheduling and allocation.

[0093] Example 3

[0094] Figure 3 This is a flowchart of a resource scheduling method according to Embodiment 3 of the present invention. Taking the in-loop resource allocation model and the non-loop resource discarding strategy as examples of resource allocation results, the resource allocation value as the resource allocation result, the quantum configuration energy as the configuration energy value as an example, and the forward and reverse execution chain flows as examples of forward and reverse execution chain flows, further illustrating a resource scheduling method, such as... Figure 3 As shown, the method includes:

[0095] Step 1: Extract task instruction data from the task flow of the central node and construct the forward execution chain and the reverse execution chain.

[0096] (1) The computing network system is widely deployed between telecommunications operator data centers, edge nodes, and quantum nodes, facing heterogeneous computing power scheduling requests from multiple industries and different users. Therefore, before scheduling, it is necessary to connect with the task request stream and perform structured parsing of the task's semantics, structure, and scheduling preferences to improve resource matching accuracy and scheduling path intelligence. The computing network system receives the i-th task instruction data (task request stream). Its general structure is defined as follows: .in, Label the task type, such as quantum factorization, behavior classification, key distribution, etc. This is a vector of task size metrics, including the number of parameters, communication span, time window, etc. This is a performance constraint vector, such as maximum allowable latency, fault tolerance level, and bandwidth requirements; For data structure description, such as input tensor dimensions, node dependency graph, etc.; This is a set of control instructions, such as prefer_quantum=true and max_latency=5ms.

[0097] based on Constructing semantic features to describe the coupling between task instruction data and computing resources Its semantic content is derived from existing semantic extraction algorithms. Obtained after extraction and processing. From the task request stream. Extract and transform the instructions using the following mapping function: ;in Indicates the control type (such as "resource constraints", "goal preferences", "security level"); Specific constraint values ​​or priorities are shown in Table 1.

[0098] Table 1 Semantic Feature Description Table

[0099]

[0100] (2) Construction of forward and reverse execution chains: Combining the dependency graph structure (dependency relationship) between task instruction data, the system constructs two execution paths. Forward execution chain (ForwardChain) represents the flow path of data between task nodes when the task proceeds normally as originally planned; Reverse execution chain (ReverseChain) derives the sequence of tasks that need to be rolled back and its scope of impact when a task fails or resources are abnormally released.

[0101] The sequential execution chain is as follows: The reverse execution chain is as follows: (like (In case of failure) The system uses the data read / write dependencies between tasks, state transition instructions, and rollback trigger conditions to automatically generate a sequential execution flow. With reverse execution chain It also dynamically marks the execution status (success, ready, rollback, exception) of each task node on the chain.

[0102] (3) Chain flow structure standardization and scheduling input generation: To support a unified input representation for the resource release scheduling model, the system standardizes the execution sequences of the forward and reverse execution chains to generate a chain-like task graph structure. This graph structure records: task node status; compatibility with computing resource types; feasibility of time window binding; and whether quantum node window collaborative computation is involved. The forward and reverse chain graphs output by the system serve as the core inputs for subsequent window alignment, resource release inference, and quantum balance scheduling models.

[0103] Step 2: Compare the coupling and symmetry of the data corresponding to the instructions under different loops in the forward and reverse execution flow. Determine whether the instructions under the loop and non-loop conditions conform to the second-order phase transition coupling and symmetry relationship. If they do not conform, guide and correct the forward and reverse execution flow respectively.

[0104] This step aims to build upon the sequential execution chain constructed in step one. Reverse execution chain The system identifies loop structures in task chains and analyzes the data coupling degree and state symmetry characteristics of task execution within these loops under the context of quantum resource release scheduling. Based on this, the system determines whether tasks within the chain are in a second-order phase transition coupling state and uses the results to drive subsequent adaptive scheduling strategy selection and quantum window migration strategy generation.

[0105] (1) It can identify the loop body structure and construct the instruction data mapping table. The system executes the chain flow sequentially. With reverse execution chain The algorithm scans the instruction chain structure, identifies data-dependent loops, and processes each instruction in each loop iteration. The data involved in the execution process is collected. Definition: For the first Wheel of Command The input data vector; Instructions in round r The output data vector. Then the instruction. The data mapping stability vector within the loop body is: ;in, For estimating linear / nonlinear data processing functions, For the Euclidean norm, through analysis exist By observing the changing trends, the system can determine whether the instruction is stable in different loop cycles and whether there is linear drift or critical mutation behavior.

[0106] (2) Perform a second-order phase transition coupling comparison on the mapped data. If the instruction... In a certain round A dramatic shift in data mapping began to occur, namely... exist Nearby conditions meet: ,and ;in To preset mutation threshold, If the jump sensitivity constant is used, the system identifies this loop segment as a second-order phase transition coupled segment, indicating that the task state has changed from linearly schedulable to a resource-sensitive coupled state, which is common in scenarios such as quantum window conflicts and state drift caused by resource release. At this time, the rollback behavior in its reverse execution chain will exhibit asymmetric characteristics, that is, the forward execution path cannot be mirrored into the reverse rollback path, and a nonlinear repair model needs to be used.

[0107] (3) Perform symmetric vector construction and scheduling strategy adaptation and adaptive optimization guidance on the mapped data.

[0108] To quantify the symmetry of the execution path, the system compares the state transition trajectories of corresponding nodes in the forward and reverse execution chains of the task, and defines each instruction. Path symmetry vector ;in Resource state change vector along the forward execution path; Resource recovery vector on the reverse rollback path; The closer to 1, the more symmetrical the path; the closer to -1, the more contradictory the path, requiring rescheduling.

[0109] Step 3: If the symmetry and second-order phase transition coupling in Step 2 do not satisfy the conventional mean, then delete any one of the instructions in Step 1, reconstruct the forward execution chain and the reverse execution chain, and execute Step 2 again.

[0110] Step 4: Based on the coupling determination results, perform task resource screening and allocation, and construct a resource allocation model within the loop and a non-loop resource discarding strategy.

[0111] In one embodiment, Figure 4 This is a flowchart of a method for determining resource allocation results according to Embodiment 3 of the present invention, as shown below. Figure 4As shown, this step, based on the completed construction of the forward and reverse execution chains of the task and the determination of coupling and symmetry, filters, allocates, models, and removes some cross-domain heterogeneous computing resources (such as GPUs, CPUs, edge nodes, etc.) according to the characteristics of the task structure and execution behavior.

[0112] (1) Determine whether the phase change coupling conditions are met or not, and extract the task resources in the loop body according to the ratio.

[0113] For cyclic task instruction data that meets the phase transition coupling condition, the heterogeneous computing resource consumption structure corresponding to its instruction execution is extracted, and a resource allocation vector is constructed. For non-cyclic task instruction data that does not meet the phase transition coupling condition, low-symmetry resource consumption paths are eliminated according to the symmetry index, and redundant resources are released. This diversion strategy ensures that the subsequent scheduling model processes task resource units that are structurally stable, mappable, and adaptable, significantly improving resource utilization efficiency and scheduling controllability.

[0114] For loops that satisfy the coupling and jump criteria, the system extracts the heterogeneous network resource consumption vectors for each round, defining the following resource type set: traditional cloud / centralized computing resources (CPU / GPU) and edge node computing resources (Edge). For each task instruction... Resource consumption in round r The system applies to all By averaging, we obtain a stable resource allocation vector for each instruction in the loop body across different resource types: ;in, The number of cycles, Indication of instructions The average resource consumption distribution throughout the entire cycle is shown in Table 2, for example.

[0115] Table 2 Average Resource Consumption Distribution

[0116]

[0117] The system uses the stable resource allocation vector as the resource allocation result, which is the core part of the task scheduling structure. It embeds a resource binding model to implement a resource affinity scheduling priority strategy (such as tasks with high GPU ratios entering the GPU window scheduling queue first).

[0118] (2) Based on the allocation results in (1), remove and mark low symmetry resources in non-cyclic tasks to obtain the adjusted resource allocation results.

[0119] For non-cyclic task chains that do not meet the second-order phase transition coupling condition, the system performs the following screening mechanism based on the path symmetry index and jump strength calculated in step two: If The system then considers the task to have stability risks within the scheduling window; further analysis of its resource usage records identifies the high-performing but uncontrollable resource components. If the failure rate of GPU-edge nodes increases due to high access latency, then: The system will discard the associated resource channels of the instruction during the scheduling cycle and automatically trigger the lightweight task migration mechanism to release the relevant resources in the computing power pool, thereby reducing resource waste. The resource allocation results are shown in Table 3.

[0120] Table 3 Resource Allocation Results

[0121]

[0122] Step 5: Based on the adjusted resource allocation values Then, by combining the sequential quantum configuration energy value under the sequential execution chain, the ratio is subjected to quantum correction processing.

[0123] Based on the resource allocation structure (resource allocation result) output in step four, this step introduces the configuration energy level corresponding to the quantum task in the forward execution chain. Through the mechanism of quantum energy state influencing scheduling preference, the resource allocation value is adaptively quantum corrected to achieve dynamic matching between task resources and quantum computing capabilities, ensuring that the scheduling strategy achieves the optimal solution between resource utilization, execution efficiency and quantum load balance.

[0124] (1) Obtain the quantum configuration energy in the sequential execution chain.

[0125] In the scenario of network scheduling for telecom operators, there are multiple high-complexity task instructions that need to be bound to quantum nodes for execution in the sequential execution flow. The system analyzes the quantum behaviors of these tasks during their historical execution, such as state superposition, quantum jump amplitude, and interference frequency, to calculate their corresponding configuration energy level, which serves as a physical metric for the task's computational load. The configuration energy value reflects the degree of dependence of the task on quantum nodes; the higher the configuration energy, the stronger the quantum interference, the greater the computational complexity, and the higher the priority requirement for quantum window scheduling.

[0126] The system integrates the original cross-domain resource allocation (resource allocation result) [GPU, CPU, Edge] of the task instruction with its corresponding configuration energy value to adaptively adjust the resource scheduling direction. The correction rules are as follows: if the task configuration energy value is high, the priority of quantum computing resources (such as the central GPU cluster) is increased in the resource allocation, and the dependence on ordinary CPUs or edge nodes is reduced; if the energy value is low, it indicates that it can accept a looser resource matching, and the system can prioritize the execution of edge nodes with higher availability, thereby alleviating the pressure on central resources. The system uses the following simplified correction function to adjust the original resource allocation value. Adjustments will be made: ;in, This indicates the corrected quantum node resource allocation. This is the normalized configuration energy value. As an adjustment factor (which can be customized in different operator deployment environments), after correction, the system normalizes the resource allocation of each task instruction and generates an updated scheduling resource configuration set.

[0127] (2) Adapting scheduling preference label output to path score.

[0128] The quantum-corrected allocation result (target resource allocation result) is embedded into the task's scheduling metadata, serving as a crucial input for subsequent path scoring and scheduling decisions. The system also records the resource tilt magnitude before and after correction to determine whether an abnormal state rollback mechanism is triggered, ensuring a stable, continuous, and controllable scheduling process.

[0129] By introducing a quantum energy state-driven correction strategy, the system can dynamically identify "high quantum-dependent tasks," bind them to quantum nodes with stronger computing power and more suitable resources, reduce scheduling failure rate, improve execution efficiency, and release unnecessary non-quantum resource redundancy.

[0130] Step Six: Based on the revised resource allocation results (target resource allocation results) It implements the release and scheduling of multi-layered heterogeneous network resources, and constructs hierarchical scheduling execution strategies for the node layer, intra-domain layer, and cross-domain layer.

[0131] This step, as the final stage of the entire quantum balance scheduling process, executes resource release and computing power scheduling path distribution operations across multiple levels of heterogeneous networks based on the task resource allocation correction results (target resource allocation results) already completed in the previous steps. The entire resource distribution structure follows the three-layer architecture of the operator's computing network system: node layer, intra-domain layer, and cross-domain layer, performing hierarchical mapping and scheduling to maximize resource utilization while achieving distributed convergence of scheduling paths and elastic regional coverage.

[0132] (1) Node layer scheduling: Taking the resource correction result as input, perform local binding and fast distribution.

[0133] At the bottom of the scheduling chain, the system uses the corrected allocation ratio (target resource allocation result) generated in step four. Prioritize releasing computing power and binding nodes within the local resource pool. The scheduling center maps each task to GPU, CPU, or edge execution units according to its modified resource preference: if the current node's resources meet the allocation requirements, the task is directly scheduled for execution.

[0134] If a node lacks sufficient computing resources or the current time window is unavailable, the task is marked as "node-level scheduling failed" and passed to the next scheduling layer for further processing. The "node-level feedback marking matrix" constructed here will serve as one of the input data for intra-domain scheduling.

[0135] (2) Intra-domain scheduling: Receive node layer markings and perform regional scheduling and task redistribution.

[0136] When a task fails to bind successfully at the local node layer, the system enters the resource pool within the city or operator's region for intra-domain scheduling. Task scheduling at this stage strictly depends on the execution flag and failure reason returned by the node layer: if the task fails only due to insufficient resources, a GPU or CPU node with higher computing power and stable connection is sought within the region for supplementary scheduling; if the failure is due to task execution window conflicts or link blockage, the task window period is dynamically adjusted, and a feasible path is regenerated based on the intra-domain node state graph; otherwise, the system upgrades to a cross-domain scheduling mechanism.

[0137] (3) Cross-domain layer scheduling: Inherit the intra-domain layer scheduling records and activate the federated mapping and migration mechanism.

[0138] When a task cannot find suitable resources within the current computing network domain, the system triggers the computing network federated scheduling mechanism to release the cross-domain task from a wider range of operator resource pools. The scheduling decision at this stage relies on the intra-domain scheduling failure reason matrix and the task urgency score.

[0139] The system prioritizes finding remote computing domains that are highly compatible with the current task in terms of task attributes, resource type, and latency sensitivity; it initiates a resource mapping request through the federated coordinator deployed in the computing network center and calculates the optimal migration path based on the network topology and load status; if cross-domain scheduling is successful, the system updates the global resource status graph and synchronizes available resource change information to the node layer and intra-domain layer, forming a dynamic resource return and status sharing mechanism.

[0140] This embodiment proposes a novel forward / reverse execution chain coupling mechanism to achieve vectorized semantic expression and time-controllable modeling of complex task structures. It also compares the coupling and symmetry of the data corresponding to instructions under different loops in the forward and reverse execution chain flow to determine whether the instructions under loops and non-loops conform to the second-order phase transition coupling and symmetry relationship. Finally, based on the coupling determination result, it performs task resource filtering and diversion, and constructs a resource allocation model within loops and a non-loop resource discarding strategy.

[0141] By constructing a forward execution flow and a reverse tracing path for task instructions, and introducing second-order phase transition coupling and symmetry metrics, the automatic identification and separation of loop and non-loop structures within a task are achieved. Compared to traditional structured semantic extraction, this chain-like execution structure can more accurately capture the temporal dependencies and computational offset logic of tasks during computational scheduling, and adaptively quantum-correct resource allocation values ​​to achieve dynamic adaptation between task resources and quantum computing capabilities. This ensures that the scheduling strategy achieves the optimal solution between resource utilization, execution efficiency, and quantum load balance.

[0142] A cross-domain GPU / CPU heterogeneous resource allocation model is constructed and integrated with a quantum energy state correction mechanism to achieve highly adaptive scheduling and allocation. Through resource allocation vector modeling and normalization strategies, the system can dynamically learn the resource occupancy of GPUs, CPUs, and edge computing in task computation behavior. Simultaneously, for the first time, quantum configuration energy parameters are introduced to nonlinearly correct resource preferences, constructing a scheduling tilt mechanism adapted to the needs of quantum computing.

[0143] Meanwhile, based on the revised target resource allocation results , to release and schedule multi-layer heterogeneous network resources, and to construct hierarchical scheduling execution of node layer, intra-domain layer and cross-domain layer: For (1) node layer scheduling: take the resource correction result as input, execute local binding and fast distribution; For (2) intra-domain layer scheduling: take the node layer mark, execute regional scheduling and task redistribution; For (3) cross-domain layer scheduling: take the intra-domain layer scheduling record, activate federated mapping and migration mechanism.

[0144] Example 4

[0145] Figure 5 This is a schematic diagram of a resource scheduling device according to Embodiment 3 of the present invention. Figure 5 As shown, the device includes: a task graph construction module 51, a relationship determination module 52, a result determination module 53, and a resource scheduling module 54.

[0146] Among them, the task graph construction module 51 is used to acquire task instruction data, determine the semantic features of the task instruction data, construct forward execution chain flow and reverse execution chain flow based on the semantic features and task instruction data, perform standardization processing on the forward execution chain flow and reverse execution chain flow, and generate a chain-like task graph structure.

[0147] The relationship determination module 52 is used to determine the instruction association data of the forward execution chain and the reverse execution chain in the chained task graph structure for the task chain instructions, and to determine the coupling symmetry relationship of the task chain instructions based on the instruction association data.

[0148] The result determination module 53 is used to determine the resource adjustment strategy according to the loop category corresponding to the task chain instruction when the coupling symmetry relationship is determined to meet the preset conditions, and to determine the resource allocation result according to the resource adjustment strategy.

[0149] The resource scheduling module 54 is used to determine the quantum task in the task chain instructions of the sequential execution chain, determine the configuration energy value of the quantum task, fuse the resource allocation results according to the configuration energy value to obtain the target resource allocation result, and perform hierarchical scheduling according to the target resource allocation result.

[0150] The technical solution of this invention involves acquiring task instruction data through a task graph construction module, determining the semantic features of the task instruction data, constructing forward and reverse execution chains based on the semantic features and the task instruction data, standardizing the forward and reverse execution chains to generate a chained task graph structure, and determining the instruction association data between the forward and reverse execution chains and the task chain instructions in the chained task graph structure through a relationship determination module. Based on the instruction association data, the coupling symmetry relationship of the task chain instructions is determined. When the determined coupling symmetry relationship meets preset conditions, the allocation determination module determines a resource adjustment strategy based on the loop category corresponding to the task chain instructions, and determines the resource allocation according to the resource adjustment strategy. Based on the source allocation results, the resource scheduling module determines the quantum tasks in the task chain instructions of the forward execution flow, determines the configuration energy value of the quantum tasks, and fuses the resource allocation results according to the configuration energy value to obtain the target resource allocation result. Hierarchical scheduling is then performed according to the target resource allocation result. By constructing forward and reverse execution flows, automatic identification and separation of cyclic and non-cyclic tasks are achieved. Resource adjustment strategies are designed separately for cyclic and non-cyclic tasks, avoiding low-accuracy scheduling methods and improving the matching accuracy of tasks and resources. Simultaneously, a hierarchical scheduling execution mechanism is designed at the node layer, intra-domain layer, and cross-domain layer, realizing the orderly release and scheduling of multi-layered heterogeneous network resources.

[0151] In one embodiment, the task graph construction module 51 includes:

[0152] The instruction splitting unit is used to receive task instruction data, determine the control instructions in the task instruction data, and split the control instructions into instruction prefixes and instruction values.

[0153] The feature extraction unit is used to match the control type associated with the instruction prefix, determine the parsing rules associated with the control type, extract the constraint values ​​of the control instruction according to the parsing rules, and use the control type and constraint values ​​as semantic features of the task instruction data.

[0154] The execution chain construction unit is used to determine the dependencies between task requests, construct forward and reverse execution chains according to the dependencies, and use semantic features as constraint information for task requests in the forward and reverse execution chains.

[0155] The task graph construction unit is used to standardize the forward and reverse execution chains and generate a directed graph as a chain-like task graph structure.

[0156] In one embodiment, the relationship determination module 52 includes:

[0157] The vector extraction unit is used to identify task chain instructions in the forward execution chain and the reverse execution chain, and to determine the input data vector and output data vector of each round of instructions in the loop instructions in the task chain, as well as the input data vector and output data vector of non-loop instructions in the task chain at different time points during the execution process.

[0158] The stability determination unit is used to determine the objective function processing estimate of the input data vector. The output data vector of the task chain instructions belonging to the same round / same time point and the Euclidean norm of the objective function processing estimate are used as the stability vector.

[0159] The coupling determination unit is used to determine the change trend according to the stability vector and the preset mutation threshold, and to determine the coupling according to the change trend and the jump sensitivity constant.

[0160] The symmetry determination unit is used to determine the state change vectors of the same task chain instructions in the forward and reverse execution flow, determine the path symmetry vector according to the state change vector and the stability vector, and determine the symmetry of the task chain instructions according to the preset symmetry interval.

[0161] Symmetric relation units are used to determine coupled symmetric relations based on coupling and symmetry.

[0162] In one embodiment, the symmetrical relational unit is specifically used for:

[0163] When the coupling is determined to be a second-order phase transition coupling, if the symmetry is satisfied, the coupling symmetry relationship is determined to satisfy the preset condition; if the symmetry is not satisfied, the coupling symmetry relationship is determined to not satisfy the preset condition.

[0164] When it is determined that there is no second-order phase transition coupling, if the symmetry is satisfied, the coupling symmetry relationship is determined to satisfy the preset condition; if the symmetry is not satisfied, the coupling symmetry relationship is determined to not satisfy the preset condition.

[0165] In one embodiment, the resource scheduling device further includes:

[0166] The execution chain reconstruction module is used to delete any task instruction data and reconstruct the forward and reverse execution chain flows when the coupling symmetry relationship does not meet the preset conditions.

[0167] In one embodiment, the result determination module 53 includes:

[0168] A category determination unit is used to determine the loop category of task chain instructions; wherein, the loop category includes loop instructions and non-loop instructions;

[0169] The first determining unit is used to determine the heterogeneous computing power resource consumption vector of each round of instructions for the cyclic instructions, take the average value of the resource consumption vector of all rounds to obtain the stable resource allocation vector of the cyclic instructions in different resource types, determine the resource with the highest distribution in the stable resource allocation vector as the preferred resource, and use the stable resource allocation vector and the preferred resource as the resource allocation result.

[0170] The second determining unit is used to determine the stability risk according to the coupling of the non-cyclic instruction for the non-cyclic instruction, identify the target resource when there is a stability risk, treat the target resource as the discarded resource, normalize the remaining resources after removing the discarded resource, obtain the resource allocation vector after the non-cyclic instruction is adjusted, and use the resource allocation vector and the discarded resource as the resource allocation result.

[0171] In one embodiment, the resource scheduling module 54 includes:

[0172] The energy value determination unit is used to extract quantum tasks from the task chain instructions in the sequential execution flow, determine the historical quantum behavior indicators of the quantum tasks, and determine the configuration energy value according to the historical quantum behavior indicators.

[0173] The normalization unit is used to normalize the configuration energy value to obtain the target configuration energy value;

[0174] The resource scheduling unit is used to determine the product of the target configuration energy value and the preset adjustment factor as the first product, the sum of the first product and the preset value as the correction factor, and the product of the correction factor and the resource allocation result as the target resource allocation result.

[0175] The resource scheduling device provided in the embodiments of the present invention can execute the resource scheduling method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the method execution.

[0176] Example 5

[0177] Figure 6This is a schematic diagram of an electronic device implementing a resource scheduling method according to an embodiment of the present invention. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (such as helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0178] like Figure 6 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0179] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0180] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as a resource scheduling method.

[0181] In some embodiments, a resource scheduling method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the resource scheduling method described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to perform a resource scheduling method by any other suitable means (e.g., by means of firmware).

[0182] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0183] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0184] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0185] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0186] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0187] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through a communication network. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0188] In one embodiment, the present invention further includes a computer program product, which includes a computer program that, when executed by a processor, implements a resource scheduling method according to any embodiment of the present invention.

[0189] In implementing the computer program product, computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0190] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0191] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A resource scheduling method, characterized in that, include: Obtain task instruction data, determine the semantic features of the task instruction data, construct forward execution chain flow and reverse execution chain flow based on the semantic features and the task instruction data, perform standardization processing on the forward execution chain flow and reverse execution chain flow, and generate a chain-like task graph structure; Determine the instruction association data for the forward execution chain and the reverse execution chain in the chained task graph structure with respect to the task chain instructions, and determine the coupling symmetry relationship of the task chain instructions based on the instruction association data; When it is determined that the coupling symmetry relationship meets the preset conditions, a resource adjustment strategy is determined according to the loop category corresponding to the task chain instruction, and a resource allocation result is determined according to the resource adjustment strategy. Identify the quantum task in the task chain instructions of the sequential execution flow, determine the configuration energy value of the quantum task, fuse the resource allocation result according to the configuration energy value to obtain the target resource allocation result, and perform hierarchical scheduling according to the target resource allocation result.

2. The method according to claim 1, characterized in that, The process of acquiring task instruction data, determining the semantic features of the task instruction data, constructing a forward execution chain and a reverse execution chain based on the semantic features and the task instruction data, standardizing the forward execution chain and the reverse execution chain, and generating a chained task graph structure includes: Receive task instruction data, determine the control instructions in the task instruction data, and split the control instructions into instruction prefixes and instruction values; Match the control type associated with the instruction prefix, determine the parsing rule associated with the control type, extract the constraint value of the control instruction according to the parsing rule, and use the control type and the constraint value as semantic features of the task instruction data; Determine the dependencies between task requests, construct a forward execution chain and a reverse execution chain according to the dependencies, and use the semantic features as constraint information for task requests in the forward execution chain and the reverse execution chain; The forward and reverse execution chains are standardized to generate a directed graph as a chain-like task graph structure.

3. The method according to claim 1, characterized in that, The step of determining the instruction association data for the forward execution chain and the reverse execution chain in the chained task graph structure with respect to the task chain instructions, and determining the coupling symmetry relationship of the task chain instructions based on the instruction association data, includes: In the forward execution chain and the reverse execution chain, task chain instructions are identified, and the input data vector and output data vector of each round of instructions in the loop instructions in the task chain instructions are determined, as well as the input data vector and output data vector of non-loop instructions in the task chain instructions at different time points during execution. Determine the objective function processing estimate of the input data vector, and use the output data vector of the task chain instructions belonging to the same round / same time point and the Euclidean norm of the objective function processing estimate as the stability vector; The change trend is determined according to the stability vector and the preset mutation threshold, and the coupling is determined according to the change trend and the jump sensitivity constant. Determine the state change vectors of the same task chain instructions in the forward and reverse execution flow, determine the path symmetry vector according to the state change vector and the stability vector, and determine the symmetry of the task chain instructions according to the preset symmetry interval; The coupling symmetry relationship is determined according to the stated coupling and the stated symmetry.

4. The method according to claim 3, characterized in that, Determining the coupling symmetry relationship according to the coupling and the symmetry includes: When it is determined that the coupling is a second-order phase transition coupling, if the symmetry is satisfied, the coupling symmetry relationship is determined to satisfy a preset condition; if the symmetry is not satisfied, the coupling symmetry relationship is determined to not satisfy the preset condition. When it is determined that there is no second-order phase transition coupling, if the symmetry is satisfied, the coupling symmetry relationship is determined to satisfy the preset condition; if the symmetry is not satisfied, the coupling symmetry relationship is determined to not satisfy the preset condition.

5. The method according to claim 1, characterized in that, After determining the instruction association data for the forward execution chain and the reverse execution chain in the chained task graph structure, and determining the coupling symmetry relationship of the task chain instructions based on the instruction association data, the method further includes: If it is determined that the coupling symmetry relationship does not meet the preset conditions, delete any task instruction data and reconstruct the forward execution chain and the reverse execution chain.

6. The method according to claim 1, characterized in that, The step of determining the resource adjustment strategy based on the loop category corresponding to the task chain instruction, and determining the resource allocation result according to the resource adjustment strategy, includes: Determine the loop category of the task chain instructions; wherein, the loop category includes loop instructions and non-loop instructions; For the loop instruction, the heterogeneous computing power resource consumption vector of each round instruction is determined, the average value of the resource consumption vector of all rounds is taken to obtain the stable resource allocation vector of the loop instruction in different resource types, the resource with the highest distribution in the stable resource allocation vector is determined as the preferred resource, and the stable resource allocation vector and the preferred resource are used as the resource allocation result. For the non-cyclic instruction, the stability risk is determined according to the coupling of the non-cyclic instruction. When the stability risk exists, the target resource is identified and discarded. The remaining resources after removing the discarded resource are normalized to obtain the resource allocation vector adjusted by the non-cyclic instruction. The resource allocation vector and the discarded resource are used as the resource allocation result.

7. The method according to claim 1, characterized in that, The process of determining the quantum task in the task chain instructions of the sequential execution flow, determining the configuration energy value of the quantum task, fusing the resource allocation result with the configuration energy value to obtain the target resource allocation result, and performing hierarchical scheduling according to the target resource allocation result includes: Extract the quantum task from the task chain instructions of the sequential execution flow, determine the historical quantum behavior index of the quantum task, and determine the configuration energy value according to the historical quantum behavior index; The target configuration energy value is obtained by normalizing the configuration energy value. The product of the target configuration energy value and the preset adjustment factor is determined as the first product. The sum of the first product and the preset value is used as the correction factor. The product of the correction factor and the resource allocation result is determined as the target resource allocation result.

8. A resource scheduling device, characterized in that, include: The task graph construction module is used to acquire task instruction data, determine the semantic features of the task instruction data, construct forward execution chain flow and reverse execution chain flow based on the semantic features and the task instruction data, perform standardization processing on the forward execution chain flow and reverse execution chain flow, and generate a chain-like task graph structure. The relationship determination module is used to determine the instruction association data of the forward execution chain and the reverse execution chain in the chained task graph structure for the task chain instructions, and to determine the coupling symmetry relationship of the task chain instructions based on the instruction association data; The allocation determination module is used to determine a resource adjustment strategy based on the loop category corresponding to the task chain instruction when it is determined that the coupling symmetry relationship meets the preset conditions, and to determine the resource allocation result according to the resource adjustment strategy. The resource scheduling module is used to determine the quantum task in the task chain instructions of the sequential execution chain, determine the configuration energy value of the quantum task, fuse the resource allocation result according to the configuration energy value to obtain the target resource allocation result, and perform hierarchical scheduling according to the target resource allocation result.

9. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the resource scheduling method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute the resource scheduling method of any one of claims 1-7.

Images