Method and system for screening and matching candidate sets of working sub-graphs of computing resource graph

Abstract

The invention provides a working sub-graph candidate set screening and matching method and system for a computing resource graph, and the method comprises the steps: employing a secondary candidate set mode for candidate set screening, and screening out a sub-graph candidate set meeting conditions from the computing resource graph in real time through constructing a two-stage screening mechanism; according to the corresponding matching method, according to information such as node connection relations, resource configuration parameters and performance indexes stored in the candidate set, all the sub-graphs capable of executing the calculation task in an optimal mode are accurately found out by applying a specific query algorithm. According to the method, the problems of computing resource waste and unbalanced use in a big data distributed cluster can be effectively avoided while the computing speed is guaranteed to be optimal. In a big data distributed resource computing scene, the candidate set screening and matching method for the working sub-graphs of the computing resource graph, provided by the invention, shows good screening and matching efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of big data computing. Specifically, it relates to a method and system for screening and matching candidate sets of working subgraphs of a computing resource graph, and more specifically, to a method for screening and matching candidate sets of working subgraphs of a computing resource graph in a big data distributed dynamic resource scenario. Background Art

[0002] With the rapid development of information technology, big data has been widely used in various fields. In the big data distributed resource computing scenario, the computing resource graph, as an important tool for describing the computing resource structure and the connection relationships between resources, plays a key role in the effective allocation of resources, the reasonable scheduling of tasks, and the optimization of system performance.

[0003] A big data distributed cluster refers to a system composed of multiple computing devices interconnected through a network, which can collaboratively process large-scale data storage, computing, and analysis tasks. Its core features include distributed storage, parallel computing, elastic expansion, and high availability. In such scenarios, the reasonable allocation of computing resources directly determines the efficiency and cost of task execution. If the resource allocation is improper and leads to load imbalance, the task completion time will be extended several times; real-time data processing tasks are sensitive to network latency. If the link bandwidth between computing nodes is insufficient, it will cause a sharp increase in data processing latency and even lead to system crashes.

[0004] The computing resource graph can intuitively display the organizational form and interaction paths of various resources in a computing cluster. The nodes in the graph represent computing resources, such as servers, storage devices, GPU clusters, etc., and the edges represent the data transmission links, dependency relationships, or collaborative working relationships between resources. By analyzing the computing resource graph, administrators can clearly understand the resource distribution of the entire computing cluster, including information such as the type, performance parameters, and load status of resources, thereby providing an important basis for resource planning, allocation, and scheduling. In big data distributed resource computing, different business requirements and data processing characteristics pose diverse requirements for the combination and connection methods of computing resources. For example, for deep learning training tasks, a large amount of GPU resources and high-bandwidth data transmission links are required to ensure the efficiency of model training; for real-time stream data processing tasks, low-latency computing resources and fast data interaction channels are needed to ensure that data can be processed in a timely manner.

[0005] At present, in the research and application of computing resource graphs, the screening and matching of candidate sets of working subgraphs is an important research direction. A working subgraph refers to a partial subgraph in a computing resource graph that is related to a specific task or business. It contains various computing resources required to complete the task and their connection relationships. In practical applications, it is often necessary to screen out a candidate set of working subgraphs that meet specific conditions from a complex computing resource graph and match it with the actual business requirements to achieve precise resource allocation and efficient task execution. However, existing methods for screening and matching candidate sets of working subgraphs have some limitations. On the one hand, with the continuous expansion of the scale of computing clusters and the increasing complexity of data processing tasks, traditional screening and matching algorithms have high computational complexity and low efficiency when dealing with large-scale computing resource graphs, making it difficult to meet real-time requirements. For example, in a super-large-scale cloud computing center, the computing resource graph may contain hundreds of thousands of nodes and links, and using traditional algorithms for screening and matching of working subgraphs may consume a large amount of time and computing resources. On the other hand, existing methods are insufficient in considering the dynamic changes of computing resources. The resource environment in the big data distributed resource computing scenario is dynamically changing, such as the failure and repair of computing nodes, the elastic scaling of resources, and the real-time adjustment of business requirements. Traditional methods often have difficulty quickly adapting to these changes, resulting in a decline in the accuracy and effectiveness of screening and matching results. In addition, different business scenarios have diverse requirements for working subgraphs, and existing general methods are difficult to meet the personalized needs of various complex business scenarios. For example, in the high-frequency trading system in the financial field, extremely high requirements are placed on the stability, low latency, and security of computing resources, and it is necessary to screen out working subgraphs with high reliability and strong security protection attributes; while in the meteorological big data simulation scenario, more attention is paid to the large-scale parallel processing ability of computing resources, and the screening and matching criteria for working subgraphs are different from those in the financial field.

[0006] For the screening and matching of candidate sets of working subgraphs of computing resource graphs in the big data distributed resource computing scenario, a more efficient, flexible, and dynamic-change-adaptive method is needed to improve the utilization rate of computing resources and the efficiency of business processing, and meet the diverse needs of different business scenarios.

[0007] Patent document CN117648466A (application number: 202311661711.9) discloses a distributed subgraph matching method and device. The main node takes all strongly connected components in the pattern graph as a pattern subgraph, reducing the computational difficulty and resource consumption of complex pattern graph matching; the main node obtains calculation information according to the pattern subgraph and distributes it to all nodes in turn, so that all nodes search for matching subgraphs in the target graph that match the pattern subgraph, and then obtain a matching result set and upload it to the distributed storage system; the main node distributes the matching result set to the slave nodes according to the connection relationship between the pattern subgraphs for merging the matching result set, obtains a target result set and returns it to the main node, and stores it in the distributed storage system; by implementing distributed parallel processing, the computational speed and horizontal expansion ability of graph matching are improved, and it can handle ultra-large-scale data graphs and pattern graphs. Summary of the Invention

[0008] Aiming at the defects in the prior art, the purpose of the present invention is to provide a method and system for screening and matching candidate working subgraph sets of a computational resource graph.

[0009] According to a method for screening and matching candidate working subgraph sets of a computational resource graph provided by the present invention, it includes: Step S1: Analyze the connection relationship between nodes in the computational resource graph D, and based on a preset standard, preliminarily screen the nodes to generate a global candidate set GC; Step S2: Update the edge connection information for the dynamic change of network nodes in the computational resource graph D, update the global candidate set GC based on the updated edge connection information, and further screen based on the updated global candidate set GC according to the preset standard to generate a temporary candidate set LC; wherein, the temporary candidate set LC includes an index structure; Step S3: Obtain a candidate working subgraph set based on the working subgraph Q from the temporary candidate set LC, and select the optimal working subgraph based on the candidate working subgraph set according to the load balancing strategy.

[0010] Preferably, the step S1 includes: constructing a screening model based on a quantization index including neighbor label frequency; based on node attributes, connection relationships, and computational task requirements, using the screening model to preliminarily screen redundant nodes irrelevant to the task objective to obtain the global candidate set GC.

[0011] Preferably, the step S1 includes: Assign an attribute label to each node in the computational resource graph D to identify the resource type and performance parameter characteristics of the node; for any node , count the label distribution of its neighbor nodes and calculate the neighbor label frequency; assume the label of node u is A, if it has k neighbor nodes with label B, then the screening model is: the neighbor label frequency of the B label of the candidate points of u is greater than or equal to k; For nodes labeled A , use the screening model to screen for nodes that satisfy the condition If there are k or more neighbors labeled B, then the node is included in the global candidate points of the node ; Traverse all nodes in the working sub-graph Q, repeat the above steps, and form a global candidate set GC by aggregating all nodes that meet the conditions.

[0012] Preferably, the step S2 includes: updating the node-associated edge information in real time based on network dynamic change events, updating the global candidate set based on the updated node-associated edge information, performing secondary screening on the updated global candidate set using the screening model, and establishing an index structure including node topological features, resource status, and task matching degree to form a temporary candidate set LC.

[0013] Preferably, the step S2 includes: monitoring the computing resource graph D in real time, and triggering the update of the temporary candidate set when a node addition or disconnection event occurs.

[0014] Preferably, the step S3 includes: Determine the matching order of the working sub-graph Q according to the number of candidate points of each node in the working sub-graph Q in the temporary candidate set LC, and obtain a set of candidate working sub-graphs according to the matching order; select the optimal working sub-graph based on the set of candidate working sub-graphs according to the load balancing strategy.

[0015] A working sub-graph candidate set screening and matching system for a computing resource graph according to the present invention includes: Module M1: Analyze the connection relationship between nodes in the computing resource graph, perform preliminary screening on the nodes based on a preset standard, and generate a global candidate set GC; Module M2: Update the edge connection information for dynamic changes of network nodes in the computing resource graph, update the global candidate set GC based on the updated edge connection information, perform further screening on the updated global candidate set GC based on a preset standard, and generate a temporary candidate set LC; wherein, the temporary candidate set LC includes an index structure; Module M3: Obtain a set of candidate working sub-graphs based on the working sub-graph Q and the temporary candidate set LC, and select the optimal working sub-graph based on the set of candidate working sub-graphs according to the load balancing strategy.

[0016] Preferably, the module M1 includes: constructing a screening model based on a quantization index including neighbor label frequencies; using the screening model to preliminarily screen redundant nodes that are not relevant to the task objective based on node attributes, connection relationships, and computing task requirements, and obtaining a global candidate set GC; The module M1 includes: Attribute labels are assigned to each node in the computing resource graph D to identify the resource type and performance parameter characteristics of the node; for any node , the label distribution of its neighbor nodes is counted, and the neighbor label frequency is calculated; let the label of node u be A, if it has k neighbors with label B, then the screening model is: the neighbor label frequency of the B label of the candidate points of u is greater than or equal to k; For the node with label A , use the screening model to screen and satisfy the node There are k or more neighbors with label B, then the node is included in the global candidate points of the node ; Traverse all nodes in the working sub-graph Q, repeat and trigger the above steps, and couple all the nodes that meet the conditions into the global candidate set GC.

[0017] Preferably, the module M2 includes: real-time updating of node association edge information based on network dynamic change events, updating the global candidate set based on the updated node association edge information, performing secondary screening on the updated global candidate set using the screening model, and establishing an index structure including node topological characteristics, resource status, and task matching degree to form a temporary candidate set LC; The module M2 includes: real-time monitoring of the computing resource graph D, and triggering the update of the temporary candidate set when a node addition or disconnection event occurs.

[0018] Preferably, the module M3 includes: Determine the matching order of the working sub-graph Q according to the number of candidate points of each node in the working sub-graph Q in the temporary candidate set LC, obtain the candidate working sub-graph set according to the matching order; select the optimal working sub-graph based on the candidate working sub-graph set according to the load balancing strategy.

[0019] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention can effectively avoid the problems of computing resource waste and uneven use while ensuring that the computing speed reaches the optimum; 2. In the large-scale distributed resource computing scenario, the candidate set screening and matching method for the computing resource graph working sub-graph proposed by the present invention shows good matching efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] By reading the detailed description of the non-restrictive embodiments with reference to the following drawings, other features, objects, and advantages of the present invention will become more apparent: Figure 1 It is a flow chart of the candidate set screening and matching method for the working sub-graph of the computing resource graph.

[0021] Figure 2aIt is a diagram obtained after resource abstraction of a big data distributed system.

[0022] Figure 2b It is a resource diagram required for calculation.

[0023] Figure 3 It is Figure 2a A schematic diagram after deleting an edge based on it.

[0024] Figure 4 It is Figure 3 A schematic diagram after adding an edge based on it.

[0025] Figure 5 It is a schematic diagram of the change of the global candidate set GC.

[0026] Among them, N represents NPU, C represents CPU, and G represents GPU. Specific implementation manners

[0027] The present invention will be described in detail below with reference to specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but do not limit the present invention in any form. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several changes and improvements can still be made. These all belong to the protection scope of the present invention.

[0028] Embodiment 1 According to a method for screening and matching candidate working subgraph sets of a computing resource graph provided by the present invention, as Figure 1 shown, it includes: Step S1: Analyze the connection relationship between nodes of the computing resource graph D, and preliminarily screen the nodes based on a preset standard to generate a global candidate set GC; Step S2: Dynamically update the edge connection information for network nodes in the computing resource graph D, update the global candidate set GC based on the updated edge connection information, and further screen based on the updated global candidate set GC according to the preset standard to generate a temporary candidate set LC; wherein, the temporary candidate set LC includes an index structure; Step S3: Obtain a candidate working subgraph set based on the working subgraph Q from the temporary candidate set LC, and select the optimal working subgraph based on the candidate working subgraph set according to the load balancing strategy.

[0029] Specifically, the step S1 includes: constructing a screening model based on a quantization index including neighbor label frequency; using the screening model to preliminarily screen redundant nodes irrelevant to the task objective based on node attributes, connection relationships, and computing task requirements to obtain the global candidate set GC.

[0030] Specifically, the step S1 includes: The computing resource graph includes: ; where D represents the computing resource graph; V represents the set of nodes, which are computing resource entities, , and each node represents a physical or virtual computing resource, including servers, GPUs, storage devices, etc.; E represents the set of edges, which are communication links between nodes, , and each edge represents that there is a direct communication path between ; L represents the label function, which assigns one or more labels to each node to describe the resource type or applicable task type; T represents the time function, which records the status update time of nodes or edges to support dynamic change perception.

[0031] Attribute labels are assigned to each node in the computing resource graph D to identify the resource type and performance parameter characteristics of the nodes; for any node , the label distribution of its neighbor nodes is statistically analyzed, and the neighbor label frequency is calculated; assuming that the label of node u is A, if it has k neighbors with the label B, the screening model is: the neighbor label frequency of the B label of the candidate points of u is greater than or equal to k; For the node with the label A , the screening model is used to screen out nodes that meet the condition that there are k or more neighbors with the label B, and then the node is included in the global candidate points of the node ; Traverse all the nodes in the working sub-graph Q, repeat and trigger the above steps, and couple all the nodes that meet the conditions into the global candidate set GC.

[0032] Specifically, the step S2 includes: updating the node-associated edge information in real time based on the network dynamic change event, updating the global candidate set based on the updated node-associated edge information, performing secondary screening on the updated global candidate set using the screening model, and establishing an index structure including node topological features, resource status, and task matching degree to form a temporary candidate set LC.

[0033] Specifically, the step S2 includes: monitoring the computing resource graph in real time, and triggering the update of the temporary candidate set when a node addition or disconnection event occurs.

[0034] Specifically, the step S3 includes: Determining the matching order of the working sub-graph Q according to the number of candidate points of each node in the working sub-graph Q in the temporary candidate set LC, obtaining the candidate working sub-graph set according to the matching order; selecting the optimal working sub-graph based on the candidate working sub-graph set according to the load balancing strategy.

[0035] Example 2 A method and system for screening and matching candidate sets of working subgraphs of a computing resource graph provided by the present invention achieve precise scheduling of computing resources and rapid execution of tasks through a two-level screening mechanism and an efficient matching strategy.

[0036] Among them, in terms of two-level candidate set screening, a hierarchical filtering architecture is adopted: the first-level screening is based on node attributes, connection relationships, and computing task requirements, constructs a screening model through quantitative indicators such as neighbor label frequencies, quickly eliminates redundant nodes irrelevant to the task target, and obtains a global candidate set; the second-level screening is based on network dynamic change events, including node addition and link disconnection, updates node-associated edge information in real time, and at the same time further updates the global candidate set, performs deep pruning on the updated global candidate set in combination with the screening model, and establishes an index structure including node topological features, resource status, and task matching degree to form a refined temporary candidate set, greatly compressing the invalid search space.

[0037] The matching stage adopts an efficient graph matching strategy based on index driving, including: by parsing the topological structure features, resource configuration parameters, and computing task constraint conditions stored in the candidate set index, constructing a priority matching model, preferentially selecting subgraphs with high correlation and low load for matching, and realizing the rapid positioning and screening of working subgraphs. Finally, based on the subgraph load balancing evaluation model, the optimal working subgraph is dynamically selected to execute the computing task, effectively improving the task scheduling efficiency and resource utilization rate in the distributed resource computing scenario.

[0038] The present invention also provides a system for screening and matching candidate sets of working subgraphs of a computing resource graph. The system for screening and matching candidate sets of working subgraphs of the computing resource graph can be implemented by executing the process steps of the method for screening and matching candidate sets of working subgraphs of the computing resource graph. That is, those skilled in the art can understand the method for screening and matching candidate sets of working subgraphs of the computing resource graph as a preferred implementation manner of the system for screening and matching candidate sets of working subgraphs of the computing resource graph.

[0039] Example 3 Example 3 is a preferred example of Example 1 A method for screening and matching candidate sets of working subgraphs of a computing resource graph provided by the present invention includes: The method is applied to a big data distributed resource computing environment, and involves a fully distributed resource network topology graph, a resource subgraph (working subgraph) Q required for current computing, a global candidate set GC, a temporary candidate set LC, a set M of all currently matched working subgraphs, and a set of matched working subgraphs that disappear / appear newly after updating the edges Δ M. Through a two-level candidate set screening mechanism and an efficient matching algorithm, rapid screening and precise matching of working subgraphs in the network topology graph are achieved; Specifically, it includes the following steps: Global candidate set construction step: Based on the neighbor label frequency (NLF) metric criterion, perform coarse-grained screening on the network topology graph to generate a global candidate set GC; Temporary candidate set update step: For the dynamic changes of network nodes, including node addition / disconnection, update the edge connection information, update the global candidate set GC based on the updated edge connection information, and perform secondary pruning based on the updated global candidate set to generate a temporary candidate set LC .

[0040] Working subgraph matching step: Utilize the index information of the temporary candidate set and quickly locate the set of subgraphs that need to be updated through an optimized graph matching algorithm ΔM , and update the candidate working subgraph set M, and then select the optimal subgraph on the candidate working subgraph set M based on the load balancing strategy to execute the computing task

[0041] Specifically, the global candidate set construction step includes: Define node label and neighbor label frequency sub-step: Assign an attribute label to each node in the network topology graph to identify characteristics such as the resource type and performance parameters of the node. For any node , count the label distribution of its neighbor nodes and calculate the neighbor label frequency (NLF). Suppose the label of node u is A. If it has k neighbors with label B, then the NLF of the B label of the candidate points of u must be greater than or equal to k

[0042] Filter global candidate points sub-step: For a node with label A , if it meets the following conditions, include it in the global candidate points of node u: Node v has k or more neighbors with label B (where k is the NLF of the B label of node u

[0043] Generate global candidate set sub-step: Traverse all nodes in the working subgraph Q, repeat the above steps, and form the global candidate set GC with all the nodes that meet the conditions

[0044] Specifically, the temporary candidate set update step includes: Monitor the network topology graph in real time. When a node addition ins(v1, v2) or disconnection del(v1, v2) event occurs, trigger the temporary candidate set update process

[0045] Update the edge connection information for the dynamic changes of network nodes in the computing resource graph Update the global candidate set GC based on the updated edge connection information. The updated global candidate set is subjected to secondary screening using the same screening method, and an index structure including node topology characteristics, resource status, and task matching degree is established to form a temporary candidate set LC

[0046] The working sub-graph matching step includes: For the temporarily obtained candidate set LC, determine the matching order of the working sub-graphs according to the number of candidate points in the temporarily obtained candidate set LC, and obtain the set Δ M of the changed working sub-graphs according to the matching order, and then perform Δ changes to M to obtain a new M. More specifically, if adding edges, it should be M←M∪ΔM; if deleting edges, it should be M←M - ΔM; Finally, when calculating to execute, select the most suitable working sub-graph from M according to the load balancing strategy for execution.

[0047] Embodiment 4 Embodiment 4 is a preferred example of Embodiment 1 In a super-large-scale big data computing cluster, usually hundreds of thousands of heterogeneous computing resources are deployed. These resources exhibit significant heterogeneous characteristics: CPUs are suitable for general computing tasks; GPUs are good at parallel computing and vector operations; NPUs are designed specifically for neural network acceleration. At the same time, there is a collaborative adaptation relationship between hardware resources. For example, high-performance CPUs need to be paired with large-capacity memories to give full play to the advantages of compute-intensive tasks, while low-speed CPUs combined with high-capacity storage are more suitable for I / O-intensive tasks. In addition, the connection status (bandwidth, latency, reliability) of each resource node in the network topology also affects the task execution efficiency.

[0048] The above-mentioned computing resources are abstracted as nodes in the distributed computing resource topology graph, and the node labels identify the task types suitable for the resources (such as CPU - general computing, GPU - deep learning, NPU - model inference), and the edges between nodes represent the direct communication links between resources. The global resource graph is maintained by the global management node with a master-slave architecture, where: Master node: Responsible for resource screening, task allocation, and real-time update of the global resource graph; Slave node: Achieves hot backup through the data synchronization mechanism, and at the same time undertakes redundant computing and load balancing tasks to ensure the high availability of the system.

[0049] When big data computing tasks such as Spark / Flink are submitted to the management node, the system will generate a corresponding computing graph, that is, the working sub-graph Q, based on the task characteristics. The generation process needs to parse the key parameters of the task, such as data scale (input / output data volume); data storage location (data warehouse, data lake); operation type (batch processing, real-time stream computing, machine learning training, etc.); performance requirements (concurrency, response time requirements, data transmission bandwidth requirements), etc.

[0050] After the working sub-graph Q is generated, the system executes the following resource screening and matching process: 1. Global candidate set construction: Based on the Neighbor Label Frequency (NLF) algorithm, screen out the initial candidate node set GC that matches Q in the global resource graph; 2. Working subgraph matching: Further screen and generate the candidate working subgraph set M that meets the task requirements from GC; 3. Load balancing scheduling: Through the load evaluation model (considering metrics such as CPU load, memory usage, network latency, etc.), select the optimal working subgraph from M to execute the task.

[0051] To cope with the dynamic changes of resources, the management node monitors the network topology changes (such as node failures, link interruptions) and resource status fluctuations (such as resource expansion, task load changes) in real time. When the global resource graph is updated: 1. Candidate set synchronous update: Trigger the incremental update of the global candidate set GC; 2. Temporary candidate set calculation: Combine the updated GC with the change information to generate the temporary candidate set LC; 3. Working subgraph dynamic adjustment: Calculate the changed working subgraph set ΔM through LC and merge and update it with the existing candidate working subgraph set M.

[0052] When subsequent identical or similar tasks are submitted, the system can directly select a suitable working subgraph from the updated M based on the load balancing strategy, significantly reducing the resource screening time. Through the above mechanism, the present invention realizes the dynamic perception, efficient screening and intelligent scheduling of resources in a large-scale computing cluster, greatly improving the resource utilization rate and task execution efficiency of big data computing.

[0053] Embodiment 5 Figure 2a It is a distribution map of big data computing resources, where C, N, and G represent CPU, NPU, and GPU respectively, which are abstracted as labels in the figure, and the edges between points represent that two computing resources can communicate directly. Figure 2b It is the working subgraph required by a computing task. It can be seen in Figures 2a to 2b : is 's candidate node; is 's candidate node; is 's candidate node; is 's candidate node; that is: .

[0054] Therefore, at this time, if you want to perform a calculation, you can select the subgraph {v1, v3, v5, v7} for calculation.

[0055] Figure 3It represents that the edge (v5, v7) is deleted, that is, the two computing resources v5 and v7 can no longer communicate directly, and the overall communication link will change. The changed communication links are: ; If it is an edge deletion, it should be M←M - ΔM; therefore, after the edge is deleted, M is updated to an empty set, and there is no matching resource in the graph to satisfy the execution of this calculation.

[0056] Figure 4 It represents that (v2, v7) is reconnected. At this time, the changed communication links are: .

[0057] If it is an edge addition, it should be M←M ∪ ΔM. Since the original M is an empty set, the current M is updated to: ; Among them, M represents the set of candidate working subgraphs, represents the set of candidate working subgraphs to be updated, represents the nodes on the computing resource graph, represents the nodes on the working subgraph.

[0058] At this time, the computing task can select the subgraph {v2, v3, v6, v7} for calculation.

[0059] During this period, the change of the global candidate set is as Figure 5 shown.

[0060] Those skilled in the art know that in addition to implementing the system, device and its various modules provided by the present invention in the form of pure computer-readable program code, the method steps can be logically programmed to make the system, device and its various modules provided by the present invention be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, embedded microcontrollers, etc. to achieve the same program. Therefore, the system, device and its various modules provided by the present invention can be regarded as a kind of hardware component, and the modules included therein for implementing various programs can also be regarded as the structures within the hardware component; the modules for implementing various functions can also be regarded as either software programs for implementing the method or the structures within the hardware component.

[0061] The specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the above specific implementation manners, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. Without conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. A method for screening and matching candidate sets of working subgraphs of a computing resource graph, characterized in that, Including: Step S1: Analyze the connection relationships between nodes in the computing resource graph D, and preliminarily screen the nodes based on a preset standard to generate a global candidate set GC; Step S2: Update the edge connection information for the dynamic changes of network nodes in the computing resource graph D, update the global candidate set GC based on the updated edge connection information, and further screen the updated global candidate set GC based on the preset standard to generate a temporary candidate set LC; wherein, the temporary candidate set LC includes an index structure; Step S3: Obtain a candidate working subgraph set based on the working subgraph Q and the temporary candidate set LC, and select an optimal working subgraph based on the candidate working subgraph set according to the load balancing strategy.

2. The method for screening and matching candidate sets of working subgraphs of a computing resource graph according to claim 1, wherein The step S1 includes: constructing a screening model based on a quantization index including neighbor label frequencies; using the screening model to preliminarily screen redundant nodes irrelevant to the task objective based on node attributes, connection relationships, and computing task requirements to obtain the global candidate set GC.

3. The method for screening and matching candidate sets of working subgraphs of a computing resource graph according to claim 2, wherein The step S1 includes: Attribute labels are assigned to each node in the computing resource graph D to identify the resource type and performance parameter characteristics of the node; for any node , the label distribution of its neighbor nodes is counted, and the neighbor label frequency is calculated; let the label of node u be A, if it has k neighbors with label B, then the screening model is: the neighbor label frequency of the B label of the candidate points of u is greater than or equal to k; For nodes with label A , use the screening model to screen for nodes that have k or more neighbors with label B, then include the node in the global candidate points of the node ; Traverse all nodes in the working subgraph Q, repeatedly trigger the above steps, and couple all qualified node sets into the global candidate set GC.

4. The method for screening and matching candidate sets of working subgraphs of a computing resource graph according to claim 1, characterized in that, The step S2 includes: updating the node associated edge information in real time based on network dynamic change events, updating the global candidate set based on the updated node associated edge information, performing secondary screening on the updated global candidate set using the screening model, and establishing an index structure including node topological features, resource status, and task matching degree to form the temporary candidate set LC.

5. The method for screening and matching candidate sets of working subgraphs of a computing resource graph according to claim 4, wherein The step S2 includes: monitoring the computing resource graph in real time, and triggering the update of the temporary candidate set when a node addition or disconnection event occurs.

6. The method for screening and matching candidate sets of working subgraphs of a computing resource graph according to claim 1, characterized in that, The step S3 includes: Determine the matching order of the working subgraph Q according to the number of candidate points of each node in the working subgraph Q in the temporary candidate set LC, obtain the candidate working subgraph set according to the matching order; select an optimal working subgraph based on the candidate working subgraph set according to the load balancing strategy.

7. A system for screening and matching candidate sets of working subgraphs of a computing resource graph, characterized in that Including: Module M1: Analyze the connection relationships between nodes in the computing resource graph, and preliminarily screen the nodes based on a preset standard to generate a global candidate set GC; Module M2: Update the edge connection information for the dynamic changes of network nodes in the computing resource graph, update the global candidate set GC based on the updated edge connection information, and further screen the updated global candidate set GC based on the preset standard to generate a temporary candidate set LC; wherein, the temporary candidate set LC includes an index structure; Module M3: Obtain a candidate working subgraph set based on the working subgraph Q and the temporary candidate set LC, and select an optimal working subgraph based on the candidate working subgraph set according to the load balancing strategy.

8. The working sub-graph candidate set screening and matching system for a computing resource graph according to claim 7, characterized in that The module M1 includes: constructing a screening model based on a quantization index including neighbor label frequencies; using the screening model to preliminarily screen redundant nodes irrelevant to the task objective based on node attributes, connection relationships, and computing task requirements to obtain the global candidate set GC; The module M1 includes: Attribute labels are assigned to each node in the computing resource graph D to identify the resource type and performance parameter characteristics of the node; for any node , the label distribution of its neighbor nodes is statistically analyzed, and the neighbor label frequency is calculated; assume that the label of node u is A, if it has k neighbors with label B, then the screening model is: the neighbor label frequency of the B label of the candidate points of u is greater than or equal to k; For nodes with label A , use the screening model to screen for nodes that have k or more neighbors with label B, then include the node in the global candidate points of the node ; Traverse all nodes in the working subgraph Q, repeatedly trigger the above steps, and couple all qualified node sets into the global candidate set GC.

9. The working sub-graph candidate set screening and matching system for a computing resource graph according to claim 7, wherein The module M2 includes: real-time updating node associated edge information based on network dynamic change events, updating the global candidate set based on the updated node associated edge information, performing secondary screening on the updated global candidate set using a screening model, and establishing an index structure including node topological features, resource status, and task matching degree to form a temporary candidate set LC; The module M2 includes: real-time monitoring of the computing resource graph, and triggering the update of the temporary candidate set when node addition or disconnection events occur.

10. The working sub-graph candidate set screening and matching system for a computing resource graph according to claim 7, characterized in that The module M3 includes: Determining the matching order of the working subgraph Q in the temporary candidate set LC according to the number of candidate points of each node in the working subgraph Q, and obtaining a candidate working subgraph set according to the matching order; selecting an optimal working subgraph based on the candidate working subgraph set according to the load balancing strategy.

Images