A distributed software research and development resource scheduling method and system

Abstract

The present application relates to the technical field of resource scheduling, in particular to a distributed software research and development resource scheduling method and system, in the present application, the multi-dimensional resource demand image of the to-be-scheduled research and development task is acquired; the multi-dimensional resource supply image of each computing node in the cluster is acquired; the multi-dimensional resource demand image of all to-be-scheduled research and development tasks and the multi-dimensional resource supply image of all computing nodes are converged to generate a unified global adaptation situation to accurately perceive the supply and demand state of the cluster resources; then the matching degree evaluation value between each task and node is dynamically calculated as a multi-dimensional dynamic adaptation factor, and the tasks are intelligently sorted and the optimal target node is selected according to the multi-dimensional dynamic adaptation factor, and task scheduling and node mapping are performed; this process effectively solves the problems of uneven resource allocation and fragmentation in the distributed research and development environment, realizes efficient and accurate matching of tasks and resources, and significantly improves the overall resource utilization rate and scheduling throughput of the cluster.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of resource scheduling, in particular to a distributed software research and development resource scheduling method and system. BACKGROUND

[0002] In a distributed software research and development environment, resource scheduling is a core link affecting system performance and stability. Traditional resource scheduling methods usually rely on static rules or single-dimensional resource allocation strategies, such as simple threshold trigger mechanisms based on CPU or memory. Such methods are difficult to cope with the complex needs of research and development tasks for multi-dimensional resources (such as computing, storage, network bandwidth, heterogeneous hardware accelerators, etc.), and often ignore the dynamic changes of the global resource state of the cluster. Due to the lack of unified modeling and real-time perception of resource supply and demand, existing technologies are prone to cause resource allocation conflicts, local node congestion, and resource fragmentation, which further leads to task execution delay, low resource utilization, and limited system scalability. In addition, traditional methods usually cannot adaptively respond to the dynamic differences of task characteristics (such as the resource demand patterns of compute-intensive and data-intensive tasks are different), further limiting the overall efficiency and flexibility of the distributed research and development environment. SUMMARY

[0003] The purpose of the present application is a distributed software research and development resource scheduling method and system to solve the problems raised in the background technology. The specific technical problems include how to realize the global dynamic adaptation of multi-dimensional resource demand and supply to solve the problems of low resource utilization efficiency and unstable task execution performance caused by single-dimensional resource scheduling, local optimization, and lack of real-time global situation awareness in the distributed research and development environment.

[0004] To achieve the above purpose, one of the purposes of the present application is a distributed software research and development resource scheduling method, comprising the following method steps:

[0005] S1. Obtain the R&D tasks to be scheduled by listening to the task submission queue creation event; parse the metadata attached to the task, including task type, dependent libraries, configuration files, and user-specified resource constraints; query the historical database to retrieve the actual resource consumption records during the historical execution of the task or similar tasks; clean, normalize, and aggregate the resource consumption records using a built-in feature extraction algorithm to transform unstructured information into a structured multidimensional resource demand vector; generate a multidimensional resource demand profile containing resource type, demand intensity, time series features, and priority weights, and store it in the global resource status database; by comprehensively parsing multi-source information such as task metadata and historical execution records, construct a refined profile that surpasses traditional CPU and memory requirements, containing the demand intensity and patterns of multiple types of resources (such as IO, bandwidth, and heterogeneous accelerators); transform the originally single, static task resource demand into a dynamic, multidimensional structured vector, laying a precise demand-side data foundation for subsequent true multidimensional resource adaptation.

[0006] S2. Data from each compute node in the cluster is acquired through monitoring agents deployed on each node; static and dynamic data of the nodes are collected in real time through underlying instructions; the static and dynamic data are cleaned and normalized, and aggregated and analyzed to generate a multi-dimensional resource supply vector; a multi-dimensional resource supply profile is generated for each compute node and synchronized to the global resource status database; through real-time monitoring, not only the static resource capacity of the nodes is collected, but also their dynamic utilization rate is captured, thus forming a supply profile that can truly reflect the real-time multi-dimensional resource surplus and comprehensive load of the nodes; the real-time resource status of each node in the cluster is accurately depicted, providing accurate supply-side input to solve scheduling deviations caused by the lack of real-time perception, and ensuring the timeliness and accuracy of adaptation decisions.

[0007] S3. Aggregate the multi-dimensional resource demand profiles of all pending R&D tasks and the multi-dimensional resource supply profiles of all computing nodes to generate a unified global adaptation situation. The specific process of generating the global adaptation situation includes:

[0008] The data fusion algorithm is invoked to standardize and align the different data in the multidimensional resource demand profile and the multidimensional resource supply profile, in order to ensure that all resource dimensions have a unified unit of measurement and comparability.

[0009] Aggregation is achieved by constructing a global resource and task original matrix. The set of row vectors in this resource and task original matrix represents the multi-dimensional resource demand profile of all tasks, and the set of column vectors represents the multi-dimensional resource supply profile of all nodes.

[0010] Deep analysis is performed on the original matrix of resources and tasks, running state calculation, and dynamically outputting global resource distribution ratio, local congestion degree, resource fragmentation, and potential matching path. The process of deep analysis specifically includes:

[0011] The ratio of total demand to total supply of each type of resource is calculated to obtain the global resource distribution ratio.

[0012] For each computing node, the real-time utilization rate of each type of resource is arithmetically averaged to obtain the comprehensive load index of the node, which identifies the local congestion degree.

[0013] For resource fragmentation, the sum of squares of the ratio of the remaining amount of each resource on each node to the total remaining amount of the resource in the cluster is calculated to obtain the fragmentation index.

[0014] Element-by-element comparison is performed on the original matrix of resources and tasks. If the demand of a task for a certain resource is less than or equal to the remaining amount of the corresponding resource of the node, the resource dimension is marked as satisfied. The ratio of the number of satisfied resource dimensions to the total dimensions in each task and node pair is calculated to generate the potential matching path.

[0015] This step is the core cognitive link of global dynamic adaptation, aiming to completely solve the problem of lack of real-time global situation awareness and local optimization. Its role is to integrate all discrete supply and demand profiles into a unified matrix, and dynamically calculate quantitative indicators reflecting the global resource surplus and deficit (distribution ratio), node pressure (congestion degree), resource fragmentation (fragmentation index), and matching possibility (matching path) through deep analysis. The effect is to generate a "global situation board" that supports intelligent decision-making, enabling the scheduling system to go beyond the local perspective of individual nodes or tasks and consider the overall state of the cluster, providing a fundamental guarantee for global optimal decision-making.

[0016] S4, based on the global adaptation situation, dynamically calculates a quantitative matching degree evaluation value for each combination of a to-be-scheduled research and development task and each computing node, and uses it as a multi-dimensional dynamic adaptation factor, specifically including:

[0017] For each task in the task list and each node in the node list, extract the multi-dimensional resource demand vector of the task and the multi-dimensional resource supply vector of the node from the global adaptation situation.

[0018] Call the matching degree calculation function and use the global resource distribution ratio to weight the multi-dimensional resource demand vector of the task.

[0019] Use the discount factor calculated from the comprehensive load index of the node to scale its multi-dimensional resource supply vector.

[0020] The two vectors after weighting and scaling are input into a certainty model based on cosine similarity for operation, to calculate the cosine value of the angle between the two vectors in the space where they are located, to obtain a preliminary matching degree;

[0021] The preliminary matching degree, the ratio corresponding to the potential matching path, and the reciprocal of the node resource fragment index are weighted and arithmetically averaged to calculate a matching degree evaluation value, which is used as a multi-dimensional dynamic adaptation factor; all tasks and all nodes form a matching degree matrix, with rows corresponding to the to-be-scheduled research and development tasks and columns corresponding to the computing nodes.

[0022] This step is the core of the decision-making algorithm for global dynamic adaptation, aiming to convert multi-dimensional and global situation information into quantifiable decision-making basis; its role is to design a comprehensive matching degree function that organically integrates global resource distribution (to solve the allocation of scarce resources), node load (to avoid congestion), resource structure similarity (cosine similarity), and fragment integration; a single score (adaptation factor) is generated that can measure multi-dimensional resource matching quality and global benefits, converting the complex multi-objective optimization problem into a sortable and comparable numerical value, providing an accurate mathematical model for finding the optimal solution in the global range.

[0023] S5, based on the global adaptation situation, calculating the highest multi-dimensional dynamic adaptation factor value that each to-be-scheduled research and development task can obtain, and sorting all to-be-scheduled research and development tasks according to the value, to prioritize the research and development task with the highest value, specifically including:

[0024] Traverse each row in the matching degree matrix, with each row corresponding to the multi-dimensional dynamic adaptation factor of a to-be-scheduled research and development task and all computing nodes;

[0025] For each task, identify and extract the item with the highest value from all associated multi-dimensional dynamic adaptation factors;

[0026] Weighted sum of the highest adaptation factor value calculated and the ratio corresponding to the potential matching path of the task to obtain a final scheduling priority score;

[0027] Sort all to-be-scheduled tasks in descending order according to the final scheduling priority score to generate a global task scheduling priority sequence.

[0028] Based on the global adaptation situation, find and determine the target computing node that can make the multi-dimensional dynamic adaptation factor value of the current to-be-scheduled research and development task reach the highest, and map the task to the target computing node for execution, specifically including:

[0029] Traverse each row of the matching degree matrix. Each row corresponds to a multi-dimensional dynamic adaptation factor between a research and development task to be scheduled and all computing nodes.

[0030] For each task, identify and extract the item with the highest value from all the values ​​of its associated multidimensional dynamic adaptation factors;

[0031] The highest fit factor value calculated is weighted and summed with the ratio of the potential matching path for the task to obtain a final scheduling priority score.

[0032] All tasks to be scheduled are sorted in descending order based on their final scheduling priority scores to generate a global task scheduling priority sequence.

[0033] For the highest priority R&D task to be scheduled in the global task scheduling priority sequence, its corresponding row vector is retrieved from the matching degree matrix based on its unique identifier; the row vector is traversed and the maximum value is located through comparison operation; the column index corresponding to the maximum value uniquely identifies the target computing node.

[0034] This step is the final execution of the global dynamic adaptation strategy, aiming to solve local optimization problems and ultimately improve resource efficiency and task performance. Its function is to implement a two-stage strategy of "global review and optimal decision-making" based on the global situation and matching degree matrix. This involves first sorting (prioritizing the scheduling of tasks that best utilize globally optimal resources), and then mapping (selecting the globally most suitable node for the task). This ensures that every scheduling decision is made in the current global environment, maximizing both task execution performance and cluster resource utilization. By cyclically executing this process, it can continuously and dynamically adapt to changes, ultimately systematically solving the problems of low resource utilization efficiency and unstable task execution performance.

[0035] The second objective of this invention is to provide a system for a distributed software development resource scheduling method, comprising a task profile construction module, a node profile construction module, a global situation generation module, a multi-dimensional dynamic adaptation factor calculation module, and a scheduling decision module, wherein:

[0036] The task profiling module obtains a multi-dimensional resource requirement profile of the R&D tasks to be scheduled.

[0037] The node profiling module obtains a multi-dimensional resource supply profile for each computing node in the cluster.

[0038] The global situation generation module gathers multi-dimensional resource demand profiles of all R&D tasks to be scheduled and multi-dimensional resource supply profiles of all computing nodes to generate a unified global adaptation situation, which fully represents the current cluster's resource demand and supply status.

[0039] The multi-dimensional dynamic adaptation factor calculation module dynamically calculates a quantitative matching degree evaluation value for each combination of a to-be-scheduled research and development task and each computing node based on the global adaptation situation, and uses the value as the multi-dimensional dynamic adaptation factor;

[0040] The scheduling decision module calculates the highest multi-dimensional dynamic adaptation factor value that each to-be-scheduled research and development task can obtain based on the global adaptation situation, sorts all to-be-scheduled research and development tasks according to the values, and preferentially schedules the research and development task with the highest value; and finds and determines the target computing node that can make the multi-dimensional dynamic adaptation factor value of the current to-be-scheduled research and development task reach the highest value based on the global adaptation situation, and maps the task to the target computing node for execution.

[0041] Compared with the prior art, the present application has the following beneficial effects:

[0042] By constructing a multi-dimensional resource requirement portrait of a task and a real-time supply portrait of a node, and generating a unified adaptation situation that integrates global resource distribution, node load, resource fragmentation, and potential matching paths, a matching degree evaluation value that comprehensively considers multi-objective optimization is dynamically calculated for each task-node combination based on the situation, and finally, global optimal decision of task sorting and node mapping is realized according to the value, thereby improving the cluster resource utilization efficiency and improving the stability of task execution performance. BRIEF DESCRIPTION OF DRAWINGS

[0043] Figure 1 The figure is a schematic diagram of the overall method steps of the present application;

[0044] Figure 2 The figure is a schematic diagram of the core process of step S5 of the present application;

[0045] Figure 3 The figure is a schematic diagram of the overall module of the present application.

[0046] In the figure: 100, task portrait construction module; 200, node portrait construction module; 300, global situation generation module; 400, multi-dimensional dynamic adaptation factor calculation module; 500, scheduling decision module. DETAILED DESCRIPTION

[0047] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0048] Next, please refer to Figure 1 One of the purposes of the present embodiment is a distributed software research and development resource scheduling method, which comprises the following method steps:

[0049] S1, obtain the multi-dimensional resource demand profile of the to-be-scheduled research and development task by listening to the task submission queue or receiving a task creation event from the upper management platform. Once a new task enters the to-be-scheduled state, the construction process of the multi-dimensional resource demand profile is triggered, which specifically includes:

[0050] First, the computing engine parses the metadata attached to the task, including the task type, dependent library, configuration file, and user-specified resource constraint requirements; at the same time, the historical database is queried to retrieve the actual resource consumption records in the historical execution process of the task or similar tasks, including peak computing load, memory occupation trend, storage input / output size, and network transmission characteristics.

[0051] Subsequently, the resource consumption records are cleaned, normalized, and aggregated by the built-in feature extraction algorithm, and the unstructured information is converted into a structured multi-dimensional resource demand vector. The vector quantifies the task resource demand from multiple dimensions, and finally generates a multi-dimensional resource demand profile containing resource type, demand intensity, time series characteristics, and priority weight, and stores it in the global resource state library to provide standardized input for global situation awareness.

[0052] S2, periodically and actively obtain the multi-dimensional resource supply profile of each computing node in the cluster through the monitoring agent deployed on each node or the centralized resource manager, which specifically includes:

[0053] First, the static data and dynamic data of the node are collected in real time through the underlying instructions. The static data includes the total number of processor cores, the total capacity of memory, the type and size of storage media, the network interface bandwidth, and the type and number of dedicated hardware accelerators. The dynamic data includes the current processor idle rate, the memory remaining amount, the storage input / output throughput, the network bandwidth instantaneous utilization rate, and the health status and real-time load of various hardware resources;

[0054] Subsequently, the computing engine cleans, normalizes, and aggregates the static data and dynamic data to generate a structured and standardized multi-dimensional resource supply vector. The multi-dimensional resource supply vector quantifies the real-time resource supply capability and remaining situation of the node from multiple dimensions, and finally generates a multi-dimensional resource supply profile for each computing node and synchronizes it to the global resource state library to provide accurate and real-time supply end data basis for building a unified global adaptation situation.

[0055] S3, real-time extraction of all to-be-scheduled task multi-dimensional resource demand profiles and all computing node multi-dimensional resource supply profiles from the global resource state library, based on these data points, generate a unified global adaptation situation, which specifically includes:

[0056] By invoking data fusion algorithms, the different data in the multi-dimensional resource demand profile and the multi-dimensional resource supply profile are standardized and aligned to ensure that all resource dimensions (such as computing, memory, storage, network, etc.) have a unified unit of measurement and comparability.

[0057] Aggregation is achieved by constructing a global resource and task matrix. The set of row vectors in this matrix represents the multi-dimensional resource demand profile of all tasks, while the set of column vectors represents the multi-dimensional resource supply profile of all nodes. Based on this, a deep analysis is performed on the matrix, and further situational calculations are performed to dynamically output the global resource distribution ratio, local congestion level, resource fragmentation, and potential matching paths. The deep analysis process specifically includes:

[0058] The ratio of total demand to total supply for each type of resource is calculated to obtain the global resource distribution ratio. For each computing node, the real-time utilization rate of each type of resource (e.g., processor utilization rate minus idle rate, memory utilization rate minus remaining memory percentage) is arithmetically averaged to obtain the node's comprehensive load index, which indicates the degree of local congestion. For resource fragmentation, the sum of the squares of the ratios of the remaining quantity of each resource on each node to the total remaining quantity of that resource in the cluster is calculated to obtain the fragmentation index. Simultaneously, the original resource and task matrices are compared element-by-element. If the resource demand of a task is less than or equal to the corresponding remaining quantity of the node, that resource dimension is marked as satisfyable. The ratio of the number of satisfyable resource dimensions to the total number of dimensions in each task-node pair is calculated to generate potential matching paths.

[0059] This series of automated operations outputs a unified data view as a global adaptive situation. By incorporating global situational information of the cluster into the original resource and task matrix, it fully and structurally represents the real-time status of the current overall resource demand and supply of the cluster, providing a core situational awareness foundation for subsequent scheduling decisions.

[0060] S4. Based on the global adaptation status, a deterministic and quantitative calculation process is executed to dynamically generate a matching degree evaluation value for each combination of the R&D task to be scheduled and each computing node, which serves as a multi-dimensional dynamic adaptation factor, specifically including:

[0061] For each task in the task list and each node in the node list, first extract the multi-dimensional resource demand vector of the task and the multi-dimensional resource supply vector of the node from the global adaptation situation.

[0062] Subsequently, a matching degree calculation function is called, which calculates the global resource distribution ratio to weight the multi-dimensional resource demand vector of the task, highlighting the importance of scarce resources, and uses the discount factor calculated by the comprehensive load index of the node (local congestion level) to scale its multi-dimensional resource supply vector to avoid congested nodes;

[0063] The two vectors after weighting and scaling are input into a deterministic model based on cosine similarity for operation. The model calculates the cosine value of the angle between the two vectors in the space they are in, and the value range is zero to one. This numerical result is the preliminary matching degree;

[0064] Finally, the preliminary matching degree is weighted and arithmetically averaged with the ratio corresponding to the potential matching path calculated in step S3 and the reciprocal of the node resource fragmentation index, and a unique, quantitative multi-dimensional dynamic adaptation factor is finally output. This factor comprehensively represents the matching degree between the task's demand and the node's current supply under the global resource situation. All multi-dimensional dynamic adaptation factors of all tasks and all nodes together form a matching degree matrix, with the rows corresponding to the to-be-scheduled research and development tasks and the columns corresponding to the computing nodes.

[0065] S5, please refer to Figure 2 Based on the global adaptation situation, the highest multi-dimensional dynamic adaptation factor value that each to-be-scheduled research and development task can obtain is calculated, and all to-be-scheduled research and development tasks are sorted according to the value, and the research and development task with the highest value is preferentially scheduled. The specific process includes:

[0066] The process starts with scanning and extracting the existing matching degree matrix in the global resource state library. The scheduling engine traverses each row in the matching degree matrix, and each row corresponds to the multi-dimensional dynamic adaptation factor of a to-be-scheduled research and development task and all computing nodes. For each task, the system performs a maximum value calculation operation to identify and extract the highest value from all the multi-dimensional dynamic adaptation factors associated with the task. This maximum value represents the theoretically optimal resource matching level that the task can achieve under the current cluster resource situation;

[0067] Subsequently, according to the highest adaptation factor value calculated, the highest adaptation factor value is weighted and summed with the ratio corresponding to the potential matching path of the task calculated in S3 (i.e. the average degree to which the demand can be met by the nodes in the cluster), to obtain a final scheduling priority score, and all the tasks to be scheduled are sorted in descending order according to the final scheduling priority score to generate a global task scheduling priority sequence; this sorting logic aims to prioritize scheduling tasks that have both high resource matching (high adaptation factor) and extensive feasibility (high matching path ratio), thereby improving resource utilization efficiency while reducing the risk of scheduling failure, thereby improving the overall utilization efficiency of cluster resources and the throughput of the scheduling system.

[0068] Based on the global adaptation situation, the target computing node that can make the multi-dimensional dynamic adaptation factor value of the task node of the current to-be-scheduled research and development task reach the highest is found and determined, and the task is mapped to the node for execution, and the specific process includes:

[0069] The process is initiated for the to-be-scheduled research and development task with the highest priority in the global task scheduling priority sequence, and the corresponding row vector of the task is retrieved from the matching matrix according to the unique identifier of the task, which contains the multi-dimensional dynamic adaptation factor value of the task paired with each computing node in the cluster; the maximum value in the row vector is located by comparison operation; the column index corresponding to the maximum value uniquely identifies the target computing node, and the current multi-dimensional resource supply capability of the target computing node is the highest in the multi-dimensional resource demand portrait of the to-be-scheduled task; after the target node is determined, the scheduling engine generates resource allocation and task deployment instructions, maps all data, dependencies and configuration context required for task execution to the target node, and starts the task execution process; the mapping operation ensures that the task obtains the optimal resource matching in the global scope, thereby realizing the local optimization of this scheduling decision, and helping to optimize the overall resource distribution of the cluster and reduce resource fragmentation.

[0070] Please refer to Figure 3 The second purpose of the embodiment is to provide a distributed software research and development resource scheduling method system, which comprises a task portrait construction module 100, a node portrait construction module 200, a global situation generation module 300, a multi-dimensional dynamic adaptation factor calculation module 400 and a scheduling decision module 500, wherein:

[0071] The task portrait construction module 100 obtains the multi-dimensional resource demand portrait of the to-be-scheduled research and development task;

[0072] The node portrait construction module 200 obtains the multi-dimensional resource supply portrait of each computing node in the cluster;

[0073] The global situation generation module 300 gathers the multi-dimensional resource demand portraits of all the to-be-scheduled research and development tasks and the multi-dimensional resource supply portraits of all the computing nodes, and generates a unified global adaptive situation, wherein the global adaptive situation completely characterizes the resource demand and supply states of the current cluster;

[0074] The multi-dimensional dynamic adaptive factor calculation module 400 dynamically calculates a quantitative matching degree evaluation value for each combination of the to-be-scheduled research and development task and the computing node based on the global adaptive situation, and takes the value as the multi-dimensional dynamic adaptive factor;

[0075] The scheduling decision module 500 calculates the highest multi-dimensional dynamic adaptive factor value that each to-be-scheduled research and development task can obtain based on the global adaptive situation, and sorts all the to-be-scheduled research and development tasks according to the values, and preferentially schedules the research and development task with the highest value; and finds and determines the target computing node that can make the multi-dimensional dynamic adaptive factor value of the current to-be-scheduled research and development task reach the highest based on the global adaptive situation, and maps the task to the target computing node for execution.

[0076] The basic principles, main features and advantages of the present application are shown and described above. It should be understood by those skilled in the art that the present application is not limited by the above-mentioned embodiments, and the above-mentioned embodiments and descriptions in the specification are only preferred examples of the present application and are not intended to limit the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. A distributed software development resource scheduling method, characterized in that, The methods and steps include the following: S1. Obtain a multi-dimensional resource requirement profile of the R&D tasks to be scheduled; S2. Obtain a multi-dimensional resource supply profile for each computing node in the cluster; S3. Aggregate the multi-dimensional resource demand profiles of all R&D tasks to be scheduled and the multi-dimensional resource supply profiles of all computing nodes to generate a unified global adaptation status, in which the global adaptation status fully represents the current resource demand and supply status of the cluster. S4. Based on the global adaptation situation, dynamically calculate a quantitative matching degree evaluation value for each combination of R&D task to be scheduled and each computing node, and use it as a multi-dimensional dynamic adaptation factor. S5. Based on the global adaptation situation, calculate the highest value of the multidimensional dynamic adaptation factor that each R&D task to be scheduled can obtain, and sort all the R&D tasks to be scheduled according to the value, and prioritize scheduling the R&D task with the highest value. Based on the global adaptation situation, the target computing node that can make the multidimensional dynamic adaptation factor value of the R&D task to be scheduled the highest is found and determined, and the task is mapped to the target computing node for execution. The process of generating the global adaptation situation specifically includes: The data fusion algorithm is invoked to standardize and align the different data in the multidimensional resource demand profile and the multidimensional resource supply profile, in order to ensure that all resource dimensions have a unified unit of measurement and comparability. Aggregation is achieved by constructing a global resource and task original matrix. The set of row vectors in this resource and task original matrix represents the multi-dimensional resource demand profile of all tasks, and the set of column vectors represents the multi-dimensional resource supply profile of all nodes. Perform in-depth analysis of the original resource and task matrix, calculate operational status, and dynamically output global resource distribution ratio, local congestion level, resource fragmentation, and potential matching paths. The deep analysis process specifically includes: Calculate the ratio of total demand to total supply for each type of resource to obtain the global resource distribution ratio; For each computing node, the real-time utilization rate of its various resources is arithmetically averaged to obtain the comprehensive load index of that node, which is used to indicate the degree of local congestion. For resource fragmentation, the fragmentation index is obtained by summing the squares of the ratios of the remaining quantity of each resource on each node to the total remaining quantity of that resource in the cluster. The original resource and task matrices are compared element by element. If a resource requirement of a task is less than or equal to the remaining resource of the corresponding node, the resource dimension is marked as satisfyable. The ratio of the number of satisfyable resource dimensions to the total number of dimensions in each task-node pair is calculated to generate potential matching paths.

2. The distributed software development resource scheduling method according to claim 1, characterized in that, The generation process of the multidimensional dynamic adaptation factor includes: For each task in the task list and each node in the node list, extract the multi-dimensional resource demand vector of the task and the multi-dimensional resource supply vector of the node from the global adaptation situation. Call the matching degree calculation function and use the global resource distribution ratio to weight the multidimensional resource demand vector of the task; The discount factor calculated using the node's comprehensive load index is used to scale its multidimensional resource supply vector. The two vectors, after being weighted and scaled, are input into a deterministic model based on cosine similarity for computation. The cosine value of the angle between the two vectors in their respective spaces is calculated to obtain the preliminary matching degree. The matching degree evaluation value is calculated by taking the weighted arithmetic mean of the ratio of the initial matching degree to the potential matching path and the reciprocal of the node resource fragmentation index, and is used as a multidimensional dynamic adaptation factor. The multidimensional dynamic adaptation factors of all tasks and all nodes together form a matching degree matrix. The rows of the matching degree matrix correspond to the R&D tasks to be scheduled, and the columns correspond to the computing nodes.

3. The distributed software development resource scheduling method according to claim 1, characterized in that, The sorting process for the R&D tasks to be scheduled specifically includes: Traverse each row of the matching degree matrix. Each row corresponds to a multi-dimensional dynamic adaptation factor between a research and development task to be scheduled and all computing nodes. For each task, identify and extract the item with the highest value from all the values ​​of its associated multidimensional dynamic adaptation factors; The highest fit factor value calculated is weighted and summed with the ratio of the potential matching path for the task to obtain a final scheduling priority score. All tasks to be scheduled are sorted in descending order based on their final scheduling priority scores to generate a global task scheduling priority sequence.

4. The distributed software development resource scheduling method according to claim 1, characterized in that, The process of determining the target computing node specifically includes: For the highest priority R&D task to be scheduled in the global task scheduling priority sequence, its corresponding row vector is retrieved from the matching degree matrix based on its unique identifier; Traverse the row vector and locate the maximum value through comparison operations; The column index corresponding to this maximum value uniquely identifies the target computation node.

5. The distributed software development resource scheduling method according to claim 1, characterized in that, The execution process of the target computing node includes: The scheduling engine generates resource allocation and task deployment instructions, maps all the data, dependencies and configuration context required for task execution to the target node, and starts the task execution process.

6. The distributed software development resource scheduling method according to claim 1, characterized in that, The process of obtaining the multi-dimensional resource demand profile specifically includes: Obtain the R&D tasks to be scheduled by listening to the task submission queue creation event; Parse the metadata accompanying the task, including the task type, dependent libraries, configuration files, and user-specified resource constraints; Query the historical database to retrieve the actual resource consumption records during the historical execution of this task or similar tasks; By using a built-in feature extraction algorithm to clean, normalize, and aggregate resource consumption records, unstructured information is transformed into a structured multidimensional resource demand vector. Generate a multi-dimensional resource demand profile that includes resource type, demand intensity, time series characteristics, and priority weights, and store it in the global resource status database.

7. The distributed software development resource scheduling method according to claim 1, characterized in that, The process of obtaining the multi-dimensional resource supply profile includes: Data from each compute node in the cluster is obtained through monitoring agents deployed on each node; Static and dynamic data of nodes are collected in real time through underlying commands; Static and dynamic data are cleaned and normalized, and then aggregated and analyzed to generate a multidimensional resource supply vector. A multi-dimensional resource provision profile is generated for each computing node and synchronized to the global resource status database.

8. A system using the distributed software development resource scheduling method according to any one of claims 1-7, characterized in that, It includes a task profile construction module (100), a node profile construction module (200), a global situation generation module (300), a multi-dimensional dynamic adaptation factor calculation module (400), and a scheduling decision module (500), wherein: The task profile building module (100) obtains a multi-dimensional resource requirement profile of the R&D task to be scheduled; The node profiling module (200) obtains a multi-dimensional resource supply profile of each computing node in the cluster; The global situation generation module (300) gathers the multi-dimensional resource demand profiles of all scheduled R&D tasks and the multi-dimensional resource supply profiles of all computing nodes to generate a unified global adaptation situation, in which the global adaptation situation fully represents the current cluster's resource demand and supply status. The multidimensional dynamic adaptation factor calculation module (400) calculates a quantitative matching degree evaluation value for each combination of the R&D task to be scheduled and each computing node based on the global adaptation situation, and uses it as a multidimensional dynamic adaptation factor. The scheduling decision module (500) calculates the highest multidimensional dynamic adaptation factor value that each R&D task to be scheduled can obtain based on the global adaptation situation, and sorts all R&D tasks to be scheduled according to the value, prioritizing the scheduling of the R&D task with the highest value; and based on the global adaptation situation, finds and determines the target computing node that can make its multidimensional dynamic adaptation factor value reach the highest for the current R&D task to be scheduled, and maps the task to the target computing node for execution.

Images