Heterogeneous workflow task scheduling method, equipment and medium

Abstract

The invention discloses a heterogeneous workflow task scheduling method and device and a medium. The method comprises the steps of obtaining original data of sub-tasks under multiple feature dimensions according to a preset period; according to a preset scoring rule of each feature dimension, converting the original data into a quantitative score under the corresponding feature dimension; based on the original data, adopting an entropy weight method and an analytic hierarchy process to calculate the weight of the subtask under each feature dimension; calculating a total feature score of the sub-tasks based on the quantitative scores and weights of the sub-tasks under each feature dimension; and obtaining a scheduling scheme of the heterogeneous workflow task through the total feature score of each subtask. According to the method, the timeliness and adaptability of the scheduling decision are improved.

Description

Technical Field

[0001] This application relates to the field of cloud-edge-device collaboration and workflow scheduling technology, and in particular to a heterogeneous workflow task scheduling method, device and medium. Background Technology

[0002] Heterogeneous workflow tasks refer to a task chain consisting of multiple interconnected and collaborative sub-tasks designed to achieve predetermined business goals. These sub-tasks differ significantly in their core requirements and cannot be processed using the same set of rules or a single node. Cloud-edge-device collaborative scheduling is an effective path for executing heterogeneous workflow tasks. It relies on the linkage of three levels of nodes—terminal, edge nodes, and cloud—and matches the differentiated needs of heterogeneous tasks through complementary node capabilities and dynamic adaptation.

[0003] However, existing cloud-edge-device collaborative scheduling has some limitations. Most existing scheduling schemes are based on a single dimension, such as task type or priority, to make scheduling decisions, without comprehensively considering all the dimensions of task characteristics, resulting in poor adaptability of scheduling strategies to actual task requirements. Summary of the Invention

[0004] This application provides a heterogeneous workflow task scheduling method, device, and medium to solve the problem that the scheduling strategy is poorly adapted to the actual needs of the task because the prior art does not comprehensively consider all the task feature dimensions.

[0005] The embodiments of this application adopt the following technical solutions: On the one hand, embodiments of this application provide a heterogeneous workflow task scheduling method, the method comprising: Obtain raw data for subtasks across multiple feature dimensions according to a preset cycle; According to the preset scoring rules for each feature dimension, the original data is transformed into a quantitative score under the corresponding feature dimension; Based on the original data, the weights of the subtasks under each feature dimension are calculated using the entropy weight method and the analytic hierarchy process. Based on the quantitative scores and weights of the sub-tasks under each feature dimension, the total feature score of the sub-tasks is calculated. By summing the features of each subtask, a scheduling scheme for heterogeneous workflow tasks is obtained.

[0006] In one example, the calculation of the weights of the subtasks under each feature dimension using the entropy weight method and the analytic hierarchy process specifically includes: The objective weights of the subtasks under each feature dimension are calculated using the entropy weight method. The subjective weights of the subtasks under each feature dimension are calculated using the analytic hierarchy process. According to the preset fusion coefficient rules, the objective weights and subjective weights are fused using a weighted average method to obtain the weights of the subtasks under each feature dimension.

[0007] In one example, the calculation of the subjective weights of the subtasks under each feature dimension using the analytic hierarchy process specifically includes: Based on the business attributes corresponding to each feature dimension, all feature dimensions are combined in pairs. For each pair of feature dimensions, the business importance of the two feature dimensions is compared, and the comparison results of the difference in importance between the two feature dimensions are generated. The comparison results are quantified using a scaling method to generate a judgment matrix whose order is consistent with the number of feature dimensions; Based on the number of feature dimensions, the judgment matrix is ​​subjected to consistency verification; When the consistency check passes, the eigenvector corresponding to the largest eigenvalue in the judgment matrix is ​​calculated using the sum-product method; The feature vectors are normalized to obtain the subjective weights of the corresponding feature dimensions.

[0008] In one example, calculating the total feature score of the subtask based on its quantized scores and weights across each feature dimension specifically includes: The weighted score of each feature dimension is calculated by multiplying the quantitative score of each feature dimension by the weight. The weighted scores of all feature dimensions of the subtask are summed to obtain the total feature score of the subtask.

[0009] In one example, the process of obtaining a scheduling scheme for heterogeneous workflow tasks by using the total feature scores of each subtask specifically includes: Based on the total feature score and the preset task label division rules, a task label for each subtask is generated; Based on the task tags and the preset tag node correspondence rules, a target execution node is assigned to each subtask.

[0010] In one example, the method further includes: The particle swarm optimization algorithm is used to iteratively calculate the task tags of each subtask and the current resource status of the candidate execution nodes to obtain the target execution node corresponding to each subtask and the proportion of resources allocated to the subtask by that node.

[0011] In one example, the method further includes: By using a pre-trained resource matching model, resource adaptability prediction is performed on the task labels of each subtask and the current resource status of the target execution node, thus obtaining the resource adaptability results between the target execution node and the subtask. When the resource adaptation result does not meet the preset standard value, the subtask is migrated from the current target execution node to other execution nodes that meet the adaptation standard.

[0012] In one example, the method further includes: Security checks are performed on execution nodes and subtasks running on those nodes in heterogeneous workflows using preset verification rules.

[0013] On the other hand, embodiments of this application provide a heterogeneous workflow task scheduling device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to execute any of the above-described heterogeneous workflow task scheduling methods.

[0014] On the other hand, embodiments of this application provide a heterogeneous workflow task scheduling non-volatile computer storage medium storing computer-executable instructions, which are capable of executing any of the heterogeneous workflow task scheduling methods described above.

[0015] The above-described technical solutions adopted in the embodiments of this application can achieve the following beneficial effects: The method provided in this application comprehensively acquires the core attribute information of subtasks, providing sufficient data source support for subsequent accurate quantitative modeling and scheduling decisions, and avoiding one-sided task characterization due to missing feature dimensions. By transforming unstructured, qualitative raw data into structured, quantitative scoring data, a computational foundation is laid for subsequent weight fusion and total score calculation. By combining objective data patterns with subjective business needs to determine weights, it avoids both ignoring business priorities with singular objective weighting and deviating from actual data with singular subjective weighting, making weight configuration more comprehensive and improving the accuracy of subsequent total score calculation. Weighted summation achieves comprehensive quantification of multi-dimensional features of subtasks, condensing the complex attributes of subtasks into a single total score indicator, providing an intuitive basis for subsequent task classification and scheduling decisions. Dynamically updated task labels ensure real-time matching between labels and actual task status, improving the timeliness and accuracy of scheduling decisions. This avoids the problem of poor adaptability between scheduling strategies and actual task needs due to insufficient consideration of comprehensive task feature dimensions. Attached Figure Description

[0016] To more clearly illustrate the technical solution of this application, some embodiments of this application will be described in detail below with reference to the accompanying drawings, in which: Figure 1 A flowchart illustrating a heterogeneous workflow task scheduling method provided in this application embodiment; Figure 2This application provides an embodiment of a cloud-edge-device dynamic workflow scheduling system with hierarchical architecture and module interaction diagram. Figure 3 This application provides a schematic diagram of a task profile construction and dynamic update process. Figure 4 A schematic diagram illustrating the dynamic scheduling timing and load migration of cloud-edge-device systems provided in this application embodiment; Figure 5 A schematic diagram illustrating a cross-domain collaborative scheduling architecture and encrypted data transmission flow provided in an embodiment of this application; Figure 6 This application provides a schematic diagram of a hierarchical fault tolerance and fault recovery state transition. Figure 7 This is a schematic diagram of the structure of a heterogeneous workflow task scheduling method device provided in an embodiment of this application. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0018] Some embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0019] Figure 1 This is a flowchart illustrating a heterogeneous workflow task scheduling method provided in an embodiment of this application. This method can be applied to different business domains, such as internet finance, e-commerce, instant messaging, gaming, and government services. The process can be executed by computing devices within the corresponding domain, and certain input parameters or intermediate results can be manually adjusted to improve accuracy.

[0020] The analysis method involved in the embodiments of this application can be implemented by a terminal device or a server, and this application does not impose any special limitations on it. For ease of understanding and description, the following embodiments are all described in detail using a server as an example.

[0021] It should be noted that the server can be a single device or a system composed of multiple devices, i.e., a distributed server. This application does not make any specific limitations on this.

[0022] Figure 1 The process includes the following steps: S101. Obtain the original data of the subtask under multiple feature dimensions according to the preset cycle.

[0023] Among these, the feature dimensions may include, for example, task type, real-time requirements, data scale, reliability requirements, resource requirements, and related task characteristics.

[0024] Task types may include industrial control, data acquisition, model inference, etc. Real-time requirements may include task response time limits, data scale may include the volume level of data associated with the task, reliability requirements may include the core nature of the business to which the task belongs, resource requirements may include the task's requirements for CPU, memory, and bandwidth, etc. Associated task characteristics may include whether the task has pre- or post-task dependencies.

[0025] S102. Based on the preset scoring rules for each feature dimension, the original data is transformed into quantitative scores under the corresponding feature dimensions.

[0026] Among them, the pre-set scoring rules for each feature dimension can be, for example, pre-set corresponding quantitative scoring rules from 0 to 100 points. The scoring rules are formulated in combination with the characteristics of business scenarios and task requirements.

[0027] For example, the preset scoring rules for the real-time requirement dimension can be set as follows: 30 points for a response time of less than or equal to 100 milliseconds, 20 points for 1-5 seconds, and 10 points for more than 5 seconds.

[0028] S103. Based on the original data, the entropy weight method and the analytic hierarchy process are used to calculate the weights of the subtasks under each feature dimension.

[0029] In some embodiments of this application, the entropy weight method and the analytic hierarchy process (AHP) are used to calculate the weights of subtasks under each feature dimension, specifically including: The objective weights of subtasks under each feature dimension are calculated using the entropy weight method. The subjective weights of subtasks under each feature dimension are calculated using the analytic hierarchy process (AHP). Based on the preset fusion coefficient rules, the objective and subjective weights are fused using a weighted average method to obtain the weights of subtasks under each feature dimension.

[0030] Objective weights reflect the dispersion and information value of the data itself; the greater the data variation, the higher the objective weight of the corresponding dimension. Subjective weights reflect the importance and priority of each dimension in the business scenario.

[0031] For example, the subjective weights of subtasks under each feature dimension are calculated using the analytic hierarchy process (AHP), specifically including: Based on the business attributes corresponding to each feature dimension, all feature dimensions are paired. For each pair of feature dimensions in each pair, their business importance is compared, generating a comparison result of the difference in importance between the two feature dimensions. The comparison result is quantified using a scaling method, generating a judgment matrix with an order consistent with the number of feature dimensions. The judgment matrix is ​​then validated for consistency based on the number of feature dimensions. If the consistency validation passes, the eigenvector corresponding to the largest eigenvalue in the judgment matrix is ​​calculated using a sum-product method. The eigenvector is then normalized to obtain the subjective weights of the corresponding feature dimensions.

[0032] Among them, the pairwise comparison of feature dimensions refers to combining all six feature dimensions in pairs according to the importance of the business attributes corresponding to each feature dimension, comparing the business importance of the two dimensions in each pair, and clarifying the difference in importance between the two.

[0033] The judgment matrix can be generated using a 1-9 scale, for example. In the 1-9 scale, 1 indicates that the two dimensions are equally important, 3 indicates slightly higher, 5 indicates significantly higher, 7 indicates strongly higher, and 9 indicates extremely higher. 2, 4, 6, and 8 are the intermediate values ​​of adjacent scales and satisfy reciprocity. That is, if the quantization value of dimension A to dimension B is 'a', then the quantization value of dimension B to dimension A is 1 / a. The quantization comparison results generate a judgment matrix consistent with the number of feature dimensions. That is, when the number of feature dimensions is 6, a 6th-order judgment matrix is ​​generated.

[0034] Consistency verification involves querying the corresponding random consistency index RI based on the number of feature dimensions. The consistency index CI of the judgment matrix can be calculated, for example, by the following formula (1). (1) To determine the maximum eigenvalue of a matrix and the consistency ratio CR, for example, the following formula (2) can be used. (2) For example, when CR < 0.1, the judgment matrix passes the consistency check, ensuring that there is no logical contradiction in the subjective judgment.

[0035] The sum-product method is used to calculate the eigenvector corresponding to the largest eigenvalue in the judgment matrix that has passed the verification. The eigenvector is then normalized to obtain the subjective weight of each feature dimension. The subjective weight reflects the importance and priority of each dimension in the business scenario.

[0036] S104. Calculate the total feature score of the subtask based on the quantitative scores and weights of the subtasks under each feature dimension.

[0037] In some embodiments of this application, the total feature score of the subtask is calculated based on the quantized scores and weights of the subtask in each feature dimension, specifically including: The weighted score of each feature dimension is calculated by multiplying the quantified score by its weight. The weighted scores of all feature dimensions of the subtask are then summed to obtain the total feature score of the subtask.

[0038] In a specific example, the quantitative scores for each dimension of the subtask are as follows: task type 10 points, real-time requirement 30 points, data scale 10 points, reliability requirement 25 points, resource requirement 20 points, and related task features 5 points, with corresponding weights of 0.1, 0.3, 0.1, 0.25, 0.2, and 0.05, respectively. Then, the weighted scores for each dimension are 10×0.1=1 point, 30×0.3=9 points, 10×0.1=1 point, 25×0.25=6.25 points, 20×0.2=4 points, and 5×0.05=0.25 points, respectively. The total feature score is 1+9+1+6.25+4+0.25=21.5 points.

[0039] S105. By summing the features of each subtask, a scheduling scheme for heterogeneous workflow tasks is obtained.

[0040] In some embodiments of this application, a scheduling scheme for heterogeneous workflow tasks is obtained by summing the feature scores of each subtask, specifically including: Based on the total feature score and the preset task label division rules, a task label is generated for each subtask. Then, a target execution node is assigned to each subtask according to the task label and the preset label node correspondence rules.

[0041] For example, task tags can be generated by dividing subtasks into three categories based on the total feature score and preset task tag division rules: Category A (high-priority core tasks, corresponding to high-score segments), Category B (medium-priority regular tasks, corresponding to medium-score segments), and Category C (low-priority non-real-time tasks, corresponding to low-score segments). A profile update mechanism is deployed to dynamically update tags according to a preset period, such as 10 seconds, based on task execution status and environmental changes. Based on edge priority and hierarchical scheduling principles, preset tag-node correspondence rules are used. Category A tasks are preferentially assigned to the edge nodes closest to the data source. Category B tasks are assigned according to the edge node load status; if the edge load is sufficient, they are executed locally; otherwise, they are assigned to the regional cloud. Category C tasks are uniformly scheduled by the regional cloud or the core cloud, matching the corresponding target execution node according to the task tag.

[0042] Furthermore, to optimize resource allocation, the method also includes: The particle swarm optimization algorithm is used to iteratively calculate the task tags of each subtask and the current resource status of the candidate execution nodes to obtain the target execution node corresponding to each subtask and the proportion of resources allocated to the subtask by that node.

[0043] For example, an improved particle swarm optimization (PSO) algorithm is used, with the optimization objectives of minimizing scheduling latency, minimizing resource consumption, and maximizing task completion rate. The algorithm takes subtask labels and the current resource status of candidate execution nodes as input, such as CPU utilization, memory usage, and bandwidth availability, and performs iterative calculations. The improved PSO algorithm's inertia weights employ an adaptive adjustment strategy; for example, they are dynamically adjusted based on the scheduling architecture load. When the load is high, the algorithm prioritizes improving convergence speed, while when the load is low, it prioritizes exploring the optimal solution. Individual cognitive learning factors and social cognitive learning factors are set. After a preset number of iterations, the algorithm outputs the optimal target execution node for each subtask and the proportion of resources allocated to that node for the subtask.

[0044] Furthermore, to address the issue that existing cloud-edge-device resource allocation and task scheduling paths rely on pre-configuration, which cannot dynamically adjust based on task characteristics, and where edge nodes often experience resource shortages under high load and idle resources under low load, the method also includes: A pre-trained resource matching model is used to predict the resource suitability of each subtask's task label and the current resource status of the target execution node, thus obtaining the resource suitability result between the target execution node and the subtask. When the resource suitability result does not meet the preset standard value, the subtask is migrated from the current target execution node to another execution node that meets the suitability standard.

[0045] For example, a resource matching model based on a backpropagation neural network is trained, with training samples covering various combinations of tasks and resource states. The model is input with subtask labels and the current resource state of the target execution node to predict resource suitability and obtain a matching result, i.e., determining whether the target node's resources meet the task execution requirements. If the matching result does not meet a preset standard value, the subtask is migrated from the current target execution node to another execution node that meets the suitability standard. The migration strategy is implemented differently according to task type; for example, type A tasks are migrated to redundant edge nodes in the same region, type B tasks are migrated to the regional cloud, and type C tasks are paused to release resources or migrated based on priority.

[0046] When a task is a cross-regional task, the core scheduling center coordinates resources in each region and distributes tasks according to the principles of proximity and load balancing. Cross-domain data transmission adopts a dual-link access mode, and task data and scheduling instructions are encrypted and transmitted using corresponding encryption algorithms to ensure the security and stability of cross-domain scheduling.

[0047] Furthermore, to ensure the accuracy of subtask execution, methods also include: Security checks are performed on execution nodes and subtasks running on those nodes in heterogeneous workflows using preset verification rules.

[0048] For example, edge nodes and regional cloud, and regional nodes and core scheduling center exchange heartbeat packets at fixed intervals, including information such as node ID, task execution progress, and resource status. At the same time, a check code for the task execution result is generated at a preset interval. The consistency between the heartbeat packet reception and the check code is verified to determine whether the execution node is faulty or whether the task execution is abnormal.

[0049] If a fault or anomaly is detected, a tiered fault tolerance strategy is implemented based on the task tag. Specifically, Category A tasks utilize a 1+1 redundant backup of edge nodes, locking a portion of the edge node's computing power to ensure execution during network outages. Category B tasks employ a dual-active deployment of edge and regional cloud, enabling breakpoint resumption in case of failure. Category C tasks are paused with their progress encrypted and saved, resuming execution according to priority after fault recovery.

[0050] The method provided in this application comprehensively acquires the core attribute information of subtasks, providing ample data source support for subsequent accurate quantitative modeling and scheduling decisions. By transforming unstructured, qualitative raw data into structured, quantitative scoring data, a computational foundation is laid for subsequent weight fusion and total score calculation. Weights are determined by combining objective data patterns with subjective business needs, resulting in more comprehensive weight configuration and improved accuracy in subsequent total score calculation. Weighted summation achieves comprehensive quantification of multi-dimensional features of subtasks, condensing the complex attributes of subtasks into a single total score indicator, providing an intuitive basis for subsequent task classification and scheduling decisions. Dynamic task label updates ensure real-time matching between labels and actual task status, improving the timeliness of scheduling decisions.

[0051] It should be noted that, although the embodiments in this application are based on... Figure 1 Steps S101 to S105 will be described sequentially, but this does not mean that steps S101 to S105 must be performed in a strict order. The reason this embodiment follows this order is... Figure 1 The order in which steps S101 to S105 are described is provided to facilitate understanding of the technical solutions of the embodiments of this application by those skilled in the art. In other words, in the embodiments of this application, the order of steps S101 to S105 can be appropriately adjusted according to actual needs.

[0052] based on Figure 1 In addition to the method described herein, this specification also provides some specific implementation schemes and extension schemes of this method, which will be further explained below.

[0053] Figure 2 This is an example of the hierarchical architecture and module interaction diagram of a cloud-edge-device dynamic workflow scheduling system provided in this application. Figure 2As shown, the architecture comprises a four-tiered structure: terminal layer, edge layer, cloud layer, and cross-domain layer. The terminal layer includes IoT devices such as sensors, AGV controllers, business terminals such as sorting equipment controllers, and operating applications. It is used to initiate heterogeneous workflow tasks, collect raw task data, and receive scheduling instructions from the upper layer; it is the system's data input and task initiation source.

[0054] The edge layer is deployed with 12 edge nodes divided into sub-parks, with 4 nodes in each sub-park. Each node integrates a profile generation unit, a resource awareness unit, and a fault tolerance unit to execute Class A high real-time tasks and some Class B routine tasks. At the same time, it completes the initial generation of local task profiles, real-time collection of resource status, and basic fault tolerance processing.

[0055] The cloud layer consists of a cluster of 8 servers, including 4 scheduling nodes, 3 storage nodes, and 1 profile management node. It is used to coordinate Class B tasks in the region, execute Class C low-priority non-real-time tasks, train and optimize the task profile and resource supply matching model, manage the full amount of task profile data, and store core business data.

[0056] The cross-domain core scheduling center is equipped with three servers for dual-machine hot and cold standby. It is used to integrate resources from multiple regions, realize the overall scheduling of cross-regional tasks, and at the same time, as the global decision-making center, it links nodes at all levels to ensure the reliability of cross-domain collaboration.

[0057] The tags generated by the task profiling module are synchronized to the dynamic scheduling module. The scheduling module, based on the tags and the resource status collected by the resource awareness module, coordinates with the resource adaptation module to allocate resources. The fault tolerance module monitors the status of nodes and tasks at each level in real time. Upon detecting a fault, it triggers the fault tolerance mechanism, and data from all modules is synchronized to the cloud storage node for backup throughout the process.

[0058] Figure 3 This is a schematic diagram illustrating a task profile construction and dynamic update process provided in an embodiment of this application. Figure 3 As shown, the data input consists of the raw task data uploaded by the receiving terminal layer. This includes, for example, task type, response time limit, data size, and business attributes such as reliability level and resource requirement standards, providing a basic data source for profile building.

[0059] Dimension quantification involves converting raw data containing feature dimensions, such as task type, real-time requirements, data size, reliability requirements, resource requirements, and related task characteristics, into a quantitative score of 0-100 according to preset rules, thus standardizing unstructured data. Weight calculation shows that by using entropy weighting and the analytic hierarchy process (AHP), combined with objective data patterns and subjective business needs, the weights for each dimension are determined as follows: real-time requirements 0.3, reliability requirements 0.25, resource requirements 0.2, task type 0.1, data size 0.1, and related task characteristics 0.05.

[0060] For example, the total score can be calculated using a weighted summation formula, i.e., The total score for the task profile is obtained by multiplying the quantitative scores of each dimension by their corresponding weights and then summing them up.

[0061] Tag generation can be based on the total score to divide tasks into three levels: Category A (90-100 points, high priority core), Category B (70-89 points, medium priority regular), and Category C (0-69 points, low priority non-real-time). These tags serve as the basis for scheduling decisions.

[0062] Dynamic updates can be performed at fixed 10-second intervals, collecting data on task execution status and environmental changes such as fluctuations in resource requirements and adjustments in business priorities. This allows for the recalculation of the profile's total score and tags, ensuring that the profile matches the actual task status in real time.

[0063] Output synchronization involves synchronizing the final generated or updated task profile tags to the dynamic scheduling module and resource adaptation module to support subsequent scheduling and resource allocation decisions.

[0064] Figure 4 This is a schematic diagram illustrating a cloud-edge-device dynamic scheduling timing and load migration process provided in an embodiment of this application. Figure 4 As shown, the timeline serves as a guide, presenting the sequential logic of resource awareness and tag generation, scheduling scheme decision-making, task execution, and load migration / cross-domain scheduling, clearly defining the time nodes and triggering conditions for each stage. Specifically, the resource awareness and tag generation phase occurs within the time range of T0-T50 milliseconds (ms). The resource awareness module collects the resource status of edge and cloud nodes at a frequency of 50ms / time, including metrics such as CPU utilization, memory usage, and bandwidth availability. Simultaneously, the task profiling module generates or updates task tags.

[0065] Within the time range of T50-T130ms, the scheduling scheme decision-making stage is in which the improved particle swarm algorithm module takes task labels and resource status data as input, and aims to minimize scheduling delay, minimize resource consumption, and maximize task completion rate. After iterative calculation, it outputs the optimal scheduling scheme, such as assigning task A to the edge node closest to the data source.

[0066] During the task execution phase, which occurs within the time range of T130-T200ms, the target execution node (edge ​​node or cloud node) receives scheduling instructions, allocates corresponding resources (e.g., central processing unit, memory quota), and initiates task execution.

[0067] Within the T200ms timeframe, the system is in a dynamic adjustment phase, monitoring the resource load of execution nodes in real time. When the load is greater than or equal to 85%, load migration is triggered. Specifically, Category A tasks are migrated to redundant edge nodes in the same region, while Category B tasks are migrated to the regional cloud. For cross-domain tasks, the core scheduling center coordinates global resources and completes the cross-domain scheduling instruction issuance and task migration within 250ms to ensure continuous task execution.

[0068] Figure 5 This is a schematic diagram illustrating a cross-domain collaborative scheduling architecture and encrypted data transmission flow, provided as an embodiment of this application. Figure 5 As shown, the architecture is structured around a core dispatch center as the central hub, connecting multiple sub-campus layers. Each sub-campus includes an edge layer and a local cloud layer, forming a cross-domain dispatch system encompassing the core, regional, and edge domains to achieve global resource coordination. The sub-campus layers and the core dispatch center utilize dual-link access via 5G and a Virtual Private Network (VPN). VPN serves as the primary transmission link, while 5G acts as a redundant link. If the VPN link is interrupted, the system automatically switches to the 5G link within 50ms to prevent cross-domain transmission interruptions.

[0069] Data transmission flows are divided into uplink and downlink. Uplink transmission consists of task profile data and resource status data from the sub-campus edge layer. This data is encrypted using the national standard SM4 block cipher algorithm at the local cloud layer and then uploaded to the core scheduling center via dual links. The core scheduling center then completes resource integration and decision-making. Downlink transmission consists of cross-domain scheduling commands generated by the core scheduling center. These commands are encrypted using elliptic curve public key cryptography and then sent to the cloud layer of the target sub-campus via dual links before being forwarded to the edge layer execution nodes, ensuring secure command transmission. Furthermore, firewalls and intrusion detection systems are deployed at both the core scheduling center and the cloud layers of each sub-campus. Combining the national standard SM4 block cipher algorithm and elliptic curve public key cryptography algorithm for encrypted transmission, a triple cross-domain security system of link redundancy, data encryption, and security protection is constructed.

[0070] Figure 6 This is a schematic diagram illustrating the hierarchical fault tolerance and fault recovery state transition provided in an embodiment of this application. Figure 6As shown, S0 indicates the normal state, where edge nodes and the cloud, and regional nodes and the core scheduling center exchange heartbeat packets at 50ms intervals. Tasks execute normally according to the scheduling plan, and core data is synchronized to storage nodes in real time. S1 indicates fault detection, where if three consecutive heartbeat packets are not received or the CRC32 checksum of the task execution result is inconsistent, a node failure or task execution failure is determined within 150ms, and the fault tolerance process is initiated. S2 indicates tiered fault tolerance, namely: Category A tasks: Enable 1+1 redundant backup of edge nodes, switch to redundant edge nodes in the same region within 50ms, and lock 80% of the edge node's computing power during network outages to ensure uninterrupted core tasks. Category B tasks: Adopt a dual-active deployment of edge and regional cloud, switch to a backup edge node or regional cloud, and enable breakpoint resume from the most recent checkpoint to avoid duplicate data calculations. Category C tasks: Suspend task execution, encrypt and save the current progress to local storage, and resume execution according to priority after the fault is recovered.

[0071] S3 indicates fault recovery detection, meaning the system scans the status of the faulty node at 1-second intervals. Once the node or network is confirmed to be back to normal, the recovery process begins. S4 indicates recovery status, meaning that Category A tasks complete status synchronization between the faulty and redundant nodes (such as task progress and resource configuration) within 10 seconds. Category B tasks resume execution within 20 seconds. Category C tasks resume execution according to priority within 30 seconds. S5 indicates data synchronization, meaning core data such as task execution results and path planning data are synchronized through triple backups at the edge local, regional cloud, and core cloud to ensure zero data loss. After synchronization is complete, the system returns to normal status S0.

[0072] Based on the same idea, some embodiments of this application also provide devices and non-volatile computer storage media corresponding to the above methods.

[0073] Figure 7 A schematic diagram of the structure of a heterogeneous workflow task scheduling method device provided in this application embodiment includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform any of the above-described heterogeneous workflow task scheduling methods.

[0074] Some embodiments of this application provide a heterogeneous workflow task scheduling method using a non-volatile computer storage medium storing computer-executable instructions, which are capable of executing any of the above-described heterogeneous workflow task scheduling methods.

[0075] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the device and medium embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the description of the method embodiments.

[0076] The devices and media provided in this application are one-to-one with the methods. Therefore, the devices and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be repeated here.

[0077] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0078] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0079] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0080] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0081] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0082] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0083] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0084] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0085] The above description is merely an embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the technical principles of this application should fall within the protection scope of this application.

Claims

1. A heterogeneous workflow task scheduling method, characterized in that, The method comprises: acquiring original data of a subtask in multiple feature dimensions according to a preset period; translating the original data into quantized scores in the corresponding feature dimensions according to a preset scoring rule of each feature dimension; calculating the weight of the subtask in each feature dimension based on the original data by using an entropy weight method and an analytic hierarchy process; calculating a feature total score of the subtask based on the quantized scores and the weight of the subtask in each feature dimension; obtaining a scheduling scheme of a heterogeneous workflow task through the feature total scores of each subtask.

2. The method of claim 1, wherein, The calculation of the weight of the subtask in each feature dimension based on the entropy weight method and the analytic hierarchy process specifically comprises: calculating the objective weight of the subtask in each feature dimension by using the entropy weight method; calculating the subjective weight of the subtask in each feature dimension by using the analytic hierarchy process; fusing the objective weight and the subjective weight by using a weighted average method to obtain the weight of the subtask in each feature dimension according to a preset fusion coefficient rule.

3. The method of claim 2, wherein, The calculation of the subjective weight of the subtask in each feature dimension by using the analytic hierarchy process specifically comprises: combing all feature dimensions in full amount two by two according to the business attributes corresponding to each feature dimension, comparing the business importance of two feature dimensions in each group of combination, and generating a comparison result of the importance difference of two feature dimensions; quantizing the comparison result by using a scale method to generate a judgment matrix with the same number of orders as the number of feature dimensions; performing consistency check on the judgment matrix according to the number of feature dimensions; when the consistency check passes, calculating the characteristic vector corresponding to the maximum eigenvalue in the judgment matrix by using the sum product method; normalizing the characteristic vector to obtain the subjective weight of the corresponding feature dimension.

4. The method of claim 1, wherein, The calculation of the feature total score of the subtask based on the quantized scores and the weight of the subtask in each feature dimension specifically comprises: multiplying the quantized scores and the weight of each feature dimension respectively to obtain the weighted score of the subtask in each feature dimension; accumulating the weighted scores of all feature dimensions of the subtask to obtain the feature total score of the subtask.

5. The method of claim 1, wherein, The obtaining of the scheduling scheme of the heterogeneous workflow task through the feature total scores of each subtask specifically comprises: generating the task label of each subtask according to the feature total score and a preset task label division rule; allocating a target execution node to each subtask according to the task label and a preset label node corresponding rule.

6. The method of claim 1, wherein, The method further comprises: iteratively calculating the task label and the current resource state of the candidate execution node of each subtask by using a particle swarm algorithm to obtain the target execution node corresponding to each subtask and the resource proportion allocated to the subtask by the node.

7. The method of claim 1, wherein, The method further comprises: performing resource adaptability prediction on the task label and the current resource state of the target execution node of each subtask by using a pre-trained resource matching model to obtain a resource adaptation result of the target execution node and the subtask; when the resource adaptation result does not meet a preset standard value, migrating the subtask from the current target execution node to another execution node that meets the adaptation standard.

8. The method of claim 1, wherein, The method further comprises: The execution node in the heterogeneous workflow and the subtask running on the node are checked by preset verification rules.

9. A heterogeneous workflow task scheduling device, characterized by, Comprise: At least one processor; And, The memory is in communication connection with the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the heterogeneous workflow task scheduling method of any one of claims 1-8.

10. A heterogeneous workflow task scheduling non-volatile computer storage medium, storing computer-executable instructions, characterized in that, The computer executable instructions can execute the heterogeneous workflow task scheduling method of any one of claims 1-8.

Images