Dynamic demand-based calculation network fusion multilayer scheduling optimization method

Abstract

The invention relates to the technical field of network resource allocation and management, and discloses a dynamic demand-based computing network integration multilayer scheduling optimization method, which comprises the following steps of: S1, collecting and preprocessing related data of computing power and network resources; s2, according to the pre-processed related data, constructing a network resource model for converting a discrete computing power and network resource allocation problem into a continuous optimization problem; and S3, applying the constructed network resource model to an actual computing network fusion system, solving a computing network fusion multi-layer scheduling optimization problem through an optimized genetic algorithm in combination with multi-level computing power and a network resource dynamic scheduling strategy, and accurately sensing and analyzing task requirements and real-time load states. And dynamic scheduling and optimal distribution of computing power and network resources are realized. According to the method, the resource utilization efficiency is improved, resource waste is reduced, and the task execution efficiency is optimized.

Description

Technical Field

[0001] The present invention relates to the technical field of network resource allocation and management, and particularly to a multi-layer scheduling optimization method for computing-network convergence based on dynamic requirements. Background Art

[0002] With the development of informatization and intelligence, the demand for computing power and network resources is increasing day by day. Currently, most network resource scheduling methods adopt static or single-layer optimization methods, which are difficult to dynamically adapt to the fluctuations of task requirements and the imbalance of resource utilization. At the same time, in complex scenarios such as demand interruption or high load, the robustness and recovery ability of the traditional method are limited, and it cannot meet the demand for efficient resource scheduling in a dynamic environment.

[0003] As a new type of technical system, computing-network convergence deeply integrates computing resources and network resources, and improves the overall resource utilization efficiency through collaborative optimization. However, the multi-layer scheduling optimization of computing-network convergence in a dynamic demand environment still faces the following challenges:

[0004] Complexity of dynamic requirements: Task requirements are time-varying and uncertain, and the resource scheduling scheme needs to be able to respond in real time.

[0005] Complexity of resource allocation: The allocation of computing power resources and network resources is restricted by various constraints such as device capacity and network bandwidth, and load balancing also needs to be considered.

[0006] Requirement for system resilience: In the case of interference such as demand interruption and high load, the robustness, flexibility and rapidity of the system need to be improved to ensure the continuity of tasks and service quality. Summary of the Invention

[0007] The present invention aims to solve the problems of slow response speed, low resource utilization efficiency, insufficient system robustness and recovery ability of traditional network resource scheduling methods in a dynamic demand environment. By proposing a multi-layer scheduling optimization method for computing-network convergence based on dynamic requirements, it realizes the accurate perception and analysis of task requirements and real-time load status, and realizes the dynamic optimal allocation of computing power and network resources.

[0008] The object of the present invention can be achieved by the following technical solutions:

[0009] In a first aspect, the present invention provides a multi-layer scheduling optimization method for computing-network convergence based on dynamic requirements, including the following steps:

[0010] S1: Collect relevant data of computing power and network resources and perform preprocessing. The relevant data includes network resource dynamic demand data, task demand data, computing power resource data and network resource data;

[0011] S2: Construct a network resource model based on the preprocessed relevant data to transform the discrete computing power and network resource allocation problem into a continuous optimization problem;

[0012] S3: Apply the constructed network resource model to the actual computing and networking convergence system, combine multi-level dynamic scheduling strategies for computing power and network resources, and solve the multi-level scheduling optimization problem of computing and networking convergence through an optimized genetic algorithm, accurately sense and analyze task requirements and real-time load status, and achieve dynamic scheduling and optimized allocation of computing power and network resources.

[0013] In some preferred embodiments, wherein step S2 further includes:

[0014] Conduct a sensitivity analysis of interference characteristics and evaluate their impact on resilience indicators;

[0015] By changing interference characteristic parameters, observe the change trend of resilience indicators to determine the sensitivity of the computing and networking convergence system to different interference characteristics;

[0016] According to the sensitivity, provide guidance for the design and adjustment of multi-level scheduling strategies.

[0017] In some preferred embodiments, wherein constructing the network resource model includes:

[0018] Determine the modeling scope, including determining the scope of network resources, computing power resources, and task requirements;

[0019] Define decision variables, where the decision variables include the allocated amount of computing power resources, the allocated amount of network bandwidth, and the computing nodes to which tasks are assigned;

[0020] Determine the constraint conditions, where the constraint conditions include computing power device capacity limitations, network bandwidth upper limits, and task execution time limitations;

[0021] Construct an objective function with the goal of minimizing resource costs while considering the satisfaction of task requirements and system load balancing.

[0022] In some preferred embodiments, the design of the multi-level dynamic scheduling strategy for computing power and network resources includes:

[0023] According to the type of computing power resources, network topology structure, as well as the scale and characteristics of tasks, divide into multiple scheduling levels and formulate scheduling strategies for each level;

[0024] Among them, the multiple scheduling levels include a global resource scheduling layer, a regional resource scheduling layer, and a local resource scheduling layer; the global resource scheduling layer is responsible for overall resource coordination and strategic planning; the regional resource scheduling layer focuses on resource optimization and task allocation within the region; the local resource scheduling layer performs precise resource scheduling for specific business scenarios.

[0025] In some preferred embodiments, the optimized genetic algorithm design includes:

[0026] S321: Encode the chromosome using integer encoding. The chromosome contains gene sequences for task assignment, resource allocation amount, and execution time. Each gene represents the relevant attributes of a task or resource. Genetic operations and constraint processing are performed through integer encoding.

[0027] S322: Consider the dynamic time window limit of network resources and the satisfaction requirement when initializing the population. Here, the population represents the set of resource allocation schemes.

[0028] S323: Sort the chromosomes according to the gene excellence value, select the chromosomes with higher gene excellence to enter the next generation population, and at the same time retain the elite individuals in the previous generation population to ensure that excellent genes are not lost and accelerate the convergence of the algorithm. Here, the gene excellence value represents the excellence degree of the resource allocation scheme, and the elite individual represents the chromosome with the highest gene excellence.

[0029] S324: Simulate the recombination process of the resource allocation scheme, and use a specific crossover method to exchange part of the chromosome paths.

[0030] S325: Perform mutation operations to avoid the algorithm falling into a local optimum, including: randomly selecting a parent chromosome, randomly selecting two gene positions on the chromosome, and exchanging the gene values at these two positions to form a mutated offspring chromosome.

[0031] S326: Solve the network-fusion multi-layer scheduling optimization problem, including: using the objective function of the constructed network resource model as the objective function of the algorithm, incorporating the constraint conditions of the constructed network resource model into the feasibility judgment of the algorithm, and solving the network-fusion multi-layer scheduling optimization problem through steps S321 to S325. By continuously iteratively optimizing the chromosome population, gradually approaching the optimal resource allocation scheme; generating a new population in each iteration, calculating the gene excellence value of each chromosome in the new population, and retaining the chromosomes with higher gene excellence values to enter the next generation until the termination condition of gene excellence value convergence is met.

[0032] In some preferred embodiments, the S322: Consider the dynamic time window limit of network resources and the satisfaction requirement when initializing the population, including:

[0033] Generate an initial feasible chromosome according to the task demand time window and resource availability to ensure that the resource allocation scheme represented by each chromosome is feasible under time and resource constraints.

[0034] Remove the chromosomes that do not meet the requirements through a screening factor, and retain the high-quality chromosomes that meet the time window and resource limitations to form the initial population.

[0035] In some preferred embodiments, the step S323: sorting the chromosomes according to the gene excellence value, selecting the chromosomes with higher gene excellence to enter the next generation population, and retaining the elite individuals in the previous generation population to ensure that excellent genes are not lost and accelerate the convergence of the algorithm, includes:

[0036] Calculating the gene excellence value of each chromosome through a gene excellence function, which comprehensively considers resource utilization rate, task completion time, and cost;

[0037] Selecting the resource allocation schemes with better gene excellence values to enter the next round of screening and retaining the elite individuals.

[0038] In some preferred embodiments, the expression of the gene excellence function is:

[0039] Value(x) = α × ResourceUtilization(x) + β × TaskCompletionRate(x) - γ × Cost(x)

[0040]

[0041] Where x is the chromosome, representing a resource allocation scheme; α, β, and γ are weight coefficients used to adjust the importance of each factor in gene excellence evaluation; ResourceUtilization(x) is the resource utilization rate; TaskCompletionRate(x) is the task completion rate, CompletedTasks(x) is the number of tasks completed on time when using the resource allocation scheme x, and TotalTasks is the total number of tasks; Cost(x) is the cost of the resource allocation scheme x, and its calculation method is to calculate the sum of resource cost and network resource cost.

[0042] In some preferred embodiments, the selecting the resource allocation schemes with better gene excellence values to enter the next round of screening and retaining the elite individuals includes:

[0043] After calculating the gene excellence value of each chromosome, sorting the chromosomes from largest to smallest according to the gene excellence value;

[0044] Calculating the selection probability of each chromosome: Where G is the population size, represents the sum of the gene excellence values of the entire population; where, Value(x g ) represents the gene excellence value of the g-th chromosome x g ;

[0045] Randomly generating a number r between 0 and 1, if then select chromosome xg Enter the next generation population while retaining elite individuals; among them, represents the sum of the selection probabilities from the 1st chromosome to the gth chromosome in the population, which is used to determine which interval the random number r falls into; represents the sum of the selection probabilities from the 1st chromosome to the (g - 1)th chromosome in the population, and is used to determine whether the currently considered gth chromosome should be selected.

[0046] In some preferred embodiments, the algorithm performance evaluation metrics include resilience metrics, computing power resource utilization rate, task completion rate, average task completion time, and system cost;

[0047] The resilience metrics include robustness, redundancy, flexibility, and rapidity, and are used to measure the resilience of the system under interference; among them,

[0048] Robustness, used to measure the ability of the system to resist interference, is calculated as the ratio of the decline in system performance during interference to the performance without interference;

[0049] Redundancy, used to reflect the substitutability of system elements, is measured by calculating the proportion of substitutable elements in the system or the availability of backup resources;

[0050] Flexibility, used to evaluate the system's ability to diagnose and solve problems, is quantified according to the adjustment of the resource allocation strategy and the re - planning of the task execution path when the system faces interference;

[0051] Rapidity, used to reflect the timeliness of the system's recovery function, measures the time required for the system to recover from interference to the normal service level.

[0052] Compared with the prior art, the beneficial effects of the present invention are:

[0053] (1) Through multi - level scheduling strategies and optimized genetic algorithms, the present invention can more accurately match task requirements and resource supply, avoid resource idleness and waste, and thus significantly improve the utilization efficiency of computing power and network resources.

[0054] (2) By defining and evaluating resilience metrics, the present invention improves the robustness and recovery ability of the system in complex scenarios such as demand interruption and high load. The system can quickly adjust the resource allocation strategy and task execution path when facing interference, ensuring task continuity and service quality.

[0055] (3) The present invention can real - time sense and analyze task requirements and resource load status, realize dynamic optimization allocation of computing power and network resources, contribute to more quickly responding to task requirements, reducing task waiting time and execution time, and improving task completion rate and average completion time.

[0056] (4) By constructing an objective function with the goal of minimizing resource costs, the present invention can reduce the total cost of the system while meeting task requirements; optimizing resource allocation and utilization reduces unnecessary resource waste, thereby reducing the usage costs of computing and network resources.

[0057] (5) By improving resource utilization efficiency, optimizing task execution efficiency, and enhancing system robustness, the present invention can improve user experience and service quality. Through precise task scheduling and resource matching, delays and errors during task execution are reduced, improving user satisfaction.

[0058] (6) The multi-level scheduling strategy and optimized genetic algorithm proposed by the present invention have good adaptability and scalability, and can handle computing-network fusion systems of different scales and complexities. By adjusting the scheduling strategy and algorithm parameters, different scenarios and requirements can be adapted, improving the flexibility and customizability of the system.

[0059] In summary, by proposing a multi-level scheduling optimization method for computing-network fusion based on dynamic requirements, the present invention significantly improves resource utilization efficiency, enhances system robustness and recovery ability, optimizes task execution efficiency, reduces system costs, and improves user experience and service quality. At the same time, this method also has good adaptability and scalability, and can meet the requirements of computing-network fusion systems of different scales and complexities. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] The above and / or additional aspects and advantages of the present invention will become apparent and be readily understood from the following description of the embodiments in conjunction with the accompanying drawings, where:

[0061] Figure 1 is a schematic flowchart of a multi-level scheduling optimization method for computing-network fusion based on dynamic requirements provided by an embodiment of the present invention;

[0062] Figure 2 is a flowchart of a multi-level scheduling optimization method for computing-network fusion based on dynamic requirements provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0063] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0064] In the description of the present invention, it should be understood that for the orientation description, such as the orientation or positional relationship indicated by up, down, front, back, left, right, etc., it is based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation to the present invention.

[0065] In the description of the present invention, the meaning of several is one or more, the meaning of multiple is more than two, greater than, less than, exceeding, etc. are understood as not including the present number, above, below, within, etc. are understood as including the present number. If the first and second are described only for the purpose of distinguishing technical features, they cannot be understood as indicating or implying relative importance or implicitly indicating the quantity of the indicated technical features or implicitly indicating the sequence relationship of the indicated technical features.

[0066] The present invention proposes a multi-layer scheduling optimization method for computing-network convergence based on dynamic requirements. By designing multi-level scheduling strategies and precisely perceiving and analyzing task requirements and real-time load status, dynamic optimal allocation of computing power and network resources is achieved. Overall resources are planned and coordinated at the global resource scheduling layer, refined adjustments are made according to local conditions at the regional resource scheduling layer, and precise matching is carried out for specific business scenarios at the local resource scheduling layer, thereby improving the utilization efficiency of resources at different levels and scenarios and reducing resource waste.

[0067] By calculating custom evaluation metrics, such as resilience metrics (such as robustness, redundancy, flexibility, and rapidity, etc.), the resilience of the system under interference such as demand interruption is evaluated and enhanced. For example, in an enterprise's distributed computing platform, when some computing nodes fail (a case of demand interruption), the system can ensure the continued execution of critical tasks through redundant computing resources or quickly adjusting the task allocation strategy (reflecting redundancy and flexibility), reduce the decline in system performance (reflecting robustness), and quickly resume to the normal service level (reflecting rapidity).

[0068] The discrete problem of allocating computing power and network resources is transformed into a continuous optimization problem, and the multi-layer scheduling optimization problem of computing-network convergence is solved through an optimized genetic algorithm, reducing the total system cost and achieving economical and efficient resource scheduling.

[0069] Please refer to Figure 1 and Figure 2 As shown, a multi-layer scheduling optimization method for computing-network convergence based on dynamic requirements includes the following steps:

[0070] S1: Collect relevant data on computing power and network resources and perform preprocessing. The relevant data includes network resource dynamic demand data, task demand data, computing power resource data, and network resource data;

[0071] In this step S1, it specifically includes:

[0072] S11: Collect dynamic demand data of network resources, task demand data, computing power resource data (such as computing device performance parameters, current available computing power, etc.) and network resource data (network metrics such as bandwidth, latency, packet loss rate, etc., different network node states, etc.).

[0073] S12: Clean and organize the collected data, remove outliers and incorrect data to ensure the accuracy and integrity of the data.

[0074] S2: Construct a network resource model based on the preprocessed relevant data, which is used to transform the discrete computing power and network resource allocation problem into a continuous optimization problem;

[0075] In this step S2, it is used to transform the discrete computing power and network resource allocation problem into a continuous optimization problem, which is convenient for dealing with large-scale complex systems. By estimating parameters and decision variables through continuous density functions, effective decisions can be made when accurate plans cannot be obtained but operating costs need to be considered.

[0076] In the allocation of computing power and network resources, the original problem may be discrete, that is, the matching relationship between resources (such as computing nodes, network links) and tasks (such as computing tasks, data transmission tasks) is clear and discontinuous.

[0077] In order to better deal with large-scale complex systems, this problem can be transformed into a continuous optimization problem. That is to say, in the present invention, the matching relationship between resources and tasks is no longer regarded as discrete points, but as a function or variable that can be optimized in a continuous space. Through this transformation, continuous optimization algorithms (such as gradient descent, genetic algorithms, etc.) can be used to solve the problem, and these algorithms are usually more suitable for large-scale problems than discrete optimization algorithms. In the embodiments of the present invention, the optimized genetic algorithm is specifically used as an example for illustration, as will be described later.

[0078] Network resource modeling aims to transform the discrete computing power and network resource allocation problem into a continuous optimization problem, so as to facilitate the use of mathematical and algorithm tools for solution. Through modeling, the modeling scope, decision variables, constraint conditions and objective function can be clearly determined, providing a basis for subsequent scheduling optimization.

[0079] Among them, the process of constructing the network resource model includes:

[0080] (1) Determine the modeling scope

[0081] Network resources: including network bandwidth, latency, topology, etc.

[0082] Computing power resources: including CPU, memory, storage, etc.

[0083] Task requirements: including the size, execution time, priority, etc. of the task.

[0084] (2) Define decision variables:

[0085] Computing power resource allocation amount: Represents the amount of computing power resources allocated to each task. Specifically, let x ij Represents the allocation amount of task i to computing node j (such as computing resource allocation amount or data traffic allocation amount)

[0086] Network bandwidth allocation amount: Represents the network bandwidth allocated to each task. Specifically, let y jk Represents the bandwidth allocation amount of computing node j to network link k (or other network resource allocation amount).

[0087] Computing node to which the task is allocated: Represents which computing node each task is allocated to for execution. Specifically, let the computing node j to which task i is allocated.

[0088] (3) Determine the constraint conditions, which include computing power device capacity limit, network bandwidth upper limit, task execution time limit, etc.;

[0089] Computing power device capacity limit: where n is the total number of tasks, and H j is the capacity of computing node j (such as computing power upper limit, storage capacity upper limit, etc.).

[0090] Network bandwidth upper limit: m is the total number of computing nodes, and B k is the bandwidth upper limit of network link k.

[0091] Task execution time limit: and are the minimum and maximum allowed execution times of task i respectively, and t i is the actual execution time of task i.

[0092] (4) Construct the objective function, aiming to minimize the resource cost, while considering the task requirement satisfaction degree and system load balance.

[0093]

[0094] Among them, C is the total resource cost, and c j is the unit resource cost of computing node j (such as computing resource usage cost, energy consumption cost, etc.), and d k$c_k$ is the unit bandwidth cost of network link $k$, and $l$ is the total number of network links. At the same time, the objective function can also consider the degree of task demand satisfaction (such as setting a penalty term to increase the cost when the task is not completed on time or does not meet a certain quality of service) and system load balancing (such as measuring by calculating indicators such as the load variance of computing nodes or network links).

[0095] Specifically, in the modeling process, mathematical modeling languages (such as MATLAB, Python, etc.) are used to describe decision variables, constraints, and the objective function. Graph theory and network analysis methods are applied to handle the network topology, calculate the paths and bandwidths between computing nodes, and combined with optimization algorithms such as genetic algorithms to solve the continuous optimization problem after modeling.

[0096] (5) Model verification and optimization:

[0097] Verify the sensitivity and stability of the model through sensitivity analysis.

[0098] Optimize and adjust the model according to the verification results to improve the accuracy and practicality of the model.

[0099] Specifically, conduct sensitivity analysis on interference characteristics (such as the duration, intensity, frequency, etc. of demand interruption), and evaluate its impact on resilience indicators (see step S4 for details);

[0100] By changing the interference characteristic parameters, observe the changing trend of resilience indicators to determine the sensitivity of the computing and networking integrated system to different interference characteristics, provide a basis for formulating targeted countermeasures, and also help identify the key factors that have the greatest impact on the system's resilience, providing an important basis for subsequent strategy adjustment. For example, it is found through analysis that the system is more sensitive to long-term and high-intensity demand interruptions, and key attention should be paid to strengthening resource reserve and scheduling strategy optimization in such situations.

[0101] In addition, sensitivity analysis can help understand the impact of different resource allocation strategies on the system's resilience. By simulating different resource allocation schemes, the optimal resource allocation strategy under specific interference conditions can be found, thereby optimizing the overall performance of the system.

[0102] S3: Apply the constructed network resource model to the actual computing and networking integrated system, combine multi-level dynamic scheduling strategies for computing power and network resources, and solve the multi-level scheduling optimization problem of computing and networking integration through an optimized genetic algorithm, accurately sense and analyze task demands and real-time load status, and achieve dynamic scheduling and optimal allocation of computing power and network resources.

[0103] In this step S3, the following steps are included:

[0104] S31: Design of multi-level dynamic scheduling strategies for computing power and network resources

[0105] This step S31 specifically includes:

[0106] S311: Determination of the hierarchical architecture

[0107] In the computing-network convergence environment, in order to efficiently respond to the changes in dynamic demands, the present invention carefully divides multiple scheduling levels according to the types of computing power resources (such as high-performance computing clusters, edge computing nodes, cloud servers, etc.), network topologies (different levels of networks such as core networks, access networks, edge networks, etc.), and the scale and characteristics of tasks: the global resource scheduling layer, the regional resource scheduling layer, and the local resource scheduling layer.

[0108] Global resource scheduling layer: As the highest-level scheduling layer, it is responsible for the coordination and strategic planning of overall resources. It comprehensively considers the total amount of all available resources, the overall demands of all current tasks, and the importance weights of different regions and services. The global layer needs to have a highly global perspective, be able to foresee the trends of resource demands, and formulate corresponding resource reserve and scheduling strategies.

[0109] Regional resource scheduling layer: Located between the global layer and the local layer, it plays a connecting role. According to the global resource allocation plan, combined with the real-time load of resources within the region (such as network traffic peaks, utilization rates of computing power devices, tightness of storage resources, etc.) and the aggregation characteristics of tasks within the region, it further refines and adjusts the amount of resources allocated to tasks within this region. The regional layer needs to closely monitor the changes in resources within the region and timely adjust the scheduling strategy to ensure the balanced utilization of resources within the region.

[0110] Local resource scheduling layer: The closest to the specific business scenario, it is responsible for precisely scheduling resources for the detailed task requirements in a specific business scenario. It considers special requirements such as real-time video stream processing for network bandwidth and nearby computing power nodes, and combines the surrounding adjustable resources (such as adjacent edge computing nodes, specific network links, storage devices, etc.) for resource scheduling matching. The local layer needs to have a high degree of flexibility and response speed and be able to quickly adapt to the changes in the business scenario.

[0111] At the same time, the present invention clarifies the responsibilities and management scopes of each level to ensure that each level collaborates with each other and performs their respective duties. The global layer is responsible for the macro planning and strategic decision-making of overall resources; the regional layer is responsible for the optimal allocation of resources and task assignment within the region; the local layer is responsible for resource scheduling and execution in specific business scenarios.

[0112] S312: Formulation of scheduling strategies for each level

[0113] (1) Global resource scheduling layer strategy

[0114] Comprehensively consider resource requirements: At the global level, it is necessary to comprehensively consider the total amount of all available resources and the overall requirements of all current tasks, and through multi-objective optimization, ensure the balance between resource supply and demand.

[0115] Importance weight allocation: According to the importance weights of different regions and businesses, reasonably allocate resources, and prioritize ensuring the resource requirements of key businesses and important regions.

[0116] Optimize the application of genetic algorithm: Based on the optimized genetic algorithm (see step S32 for details), formulate a preliminary resource allocation plan. This algorithm can consider various factors, such as resource utilization rate, task completion time, cost, etc., and seek the optimal resource allocation scheme.

[0117] Balance the use of resources: While ensuring tasks with high priorities and high requirements for resource timeliness, also take into account the overall balance of resource utilization, support cross-domain resource scheduling, and avoid excessive tension or idleness of local resources.

[0118] (2) Strategies for the regional resource scheduling layer

[0119] Refine resource allocation: According to the global resource allocation plan, combined with the real-time load of resources and the characteristics of task aggregation in the region, further refine and adjust the amount of resources allocated to tasks within this region.

[0120] Dynamic weight allocation: Adopt a dynamic weight allocation strategy, and adjust the weights in real time according to factors such as the priority of tasks within the region and resource utilization rate. Appropriately tilt resources towards urgent or important tasks within the region to improve the utilization efficiency of regional resources.

[0121] Load balancing: Closely monitor the usage of resources within the region, and adjust the scheduling strategy in a timely manner to ensure the load balancing of resources within the region, and further emphasize the dynamic collaborative optimization between levels. For example, the regional layer and the local layer can exchange two-way information flows to share their respective resource usage status and task requirements in real time. In this way, once the local layer detects a resource bottleneck, the regional layer can promptly allocate resources to avoid the impact of local resource shortages on global task execution and avoid local resource overload or idleness.

[0122] (3) Strategies for the local resource scheduling layer

[0123] Precise resource matching: For the detailed task requirements in specific business scenarios, combined with the surrounding adjustable resources, conduct precise resource scheduling and matching. For example, real-time video stream processing tasks need to select computing power nodes and network bandwidth resources nearby to ensure low latency and high bandwidth.

[0124] Real-time Monitoring and Feedback: Establish a real-time monitoring and feedback mechanism to continuously monitor the task execution status and resource usage. Through real-time data feedback, adjust the scheduling strategy in a timely manner, strengthen the fault tolerance and recovery of network failures or resource anomalies, and ensure the smooth execution of tasks and the optimal utilization of resources.

[0125] Flexible Adaptation to Changes: The local layer needs to have a high degree of flexibility and response speed and be able to quickly adapt to changes in business scenarios. For example, when the task requirements change or resources fail, it can quickly adjust the scheduling strategy and incorporate a resource dynamic migration mechanism. When the resource scheduling at a certain level is unbalanced or the task load changes drastically, the system can quickly execute cross-level or cross-regional resource migrations. For instance, during high loads, some edge computing tasks can be dynamically migrated to the cloud for processing, or some tasks can switch from low-latency links to higher-bandwidth links to ensure task continuity and stability.

[0126] According to the above step S31, through the design and implementation of a multi-level scheduling strategy, the present invention can achieve dynamic optimization of the allocation of computing power and network resources, improve the utilization efficiency of resources at different levels and in different scenarios, reduce resource waste, and enhance the resilience of the system in cases such as demand interruptions.

[0127] S32: Design of an Optimized Genetic Algorithm

[0128] This step S32 specifically includes:

[0129] S321: Encode the chromosome using an integer coding method. The chromosome contains gene sequences for task allocation, resource allocation amount, and execution time. Each gene represents the relevant attributes of a task or resource, and genetic operations and constraint processing are performed through integer coding.

[0130] S322: Consider the dynamic time window limit of network resources and the satisfaction requirement when initializing the population; wherein, the population represents a set of resource allocation schemes.

[0131] In some preferred embodiments, step S322: Consider the dynamic time window limit of network resources and the satisfaction requirement when initializing the population, including:

[0132] Generate an initial feasible chromosome according to the task demand time window and resource availability to ensure that the resource allocation scheme represented by each chromosome is feasible under time and resource constraints.

[0133] Remove the chromosomes that do not meet the requirements through a screening factor, and retain the high-quality chromosomes that meet the time window and resource limitations to form an initial population.

[0134] S323: Sort the chromosomes according to the gene excellence value, select the chromosomes with higher gene excellence to enter the next generation population, and at the same time retain the elite individuals in the previous generation population to ensure that excellent genes are not lost and accelerate the convergence of the algorithm; wherein, the gene excellence value represents the excellence degree of the resource allocation scheme, and the elite individual represents the chromosome with the highest gene excellence.

[0135] In some preferred embodiments, wherein, step S323 specifically includes:

[0136] S3231: Calculate the gene excellence value of each chromosome through the gene excellence function. The gene excellence function comprehensively considers factors such as resource utilization rate, task completion time, and cost. The higher the gene excellence, the better the resource allocation scheme represented by the chromosome.

[0137] Among them, the expression of the gene excellence function is:

[0138] Value(x) = α × ResourceUtilization(x) + β × TaskCompletionRate(x) - γ × Cost(x)

[0139]

[0140] Among them, x is the chromosome, which represents a resource allocation scheme; α, β, γ are weight coefficients used to adjust the importance of each factor in the gene excellence evaluation; ResourceUtilization(x) is the resource utilization rate; TaskCompletionRate(x) is the task completion rate, CompletedTasks(x) is the number of tasks completed on time when using the resource allocation scheme x, and TotalTasks is the total number of tasks; Cost(x) is the cost of the resource allocation scheme x, and its calculation method is to calculate the sum of the resource cost and the network resource cost.

[0141] Among them, the setting of the weight coefficients α, β, γ needs to be adjusted according to the requirements and goals of the specific problem. These coefficients can be determined through experiments or expert experience. For example, if the system values resource utilization rate more, α can be set higher; if it values task completion time more, β can be set higher; if it values cost more, γ can be set higher. Suppose the system values resource utilization rate and task completion time more, and α = 0.5, β = 0.4, γ = 0.1 can be set, indicating that in the gene excellence evaluation, the resource utilization rate accounts for 50% of the weight, the task completion time accounts for 40% of the weight, and the cost accounts for 10% of the weight.

[0142] S3232: Select the resource allocation schemes with better performance according to the gene excellence value to enter the next round of screening, and retain the elite individuals. Specifically, it includes:

[0143] After calculating the gene excellence value of each chromosome, sort the chromosomes in descending order according to the gene excellence value;

[0144] Calculate the probability of each chromosome being selected: where G is the population size, represents the sum of the gene excellence values of the entire population; among them, Value(x g ) represents the gene excellence value of the g-th chromosome x g ;

[0145] Randomly generate a number r between 0 and 1. If then select chromosome x g to enter the next-generation population while retaining elite individuals; among them, represents the sum of the selection probabilities from the first chromosome to the g-th chromosome in the population, which is used to determine which interval the random number r falls into; represents the sum of the selection probabilities from the first chromosome to the (g - 1)-th chromosome in the population, which is used to determine whether the currently considered g-th chromosome should be selected. It is part of the calculation of the cumulative probability distribution and is used to compare with the random number r to determine whether the currently considered g-th chromosome should be selected.

[0146] S324: Simulate the recombination process of the resource allocation scheme, and use a specific crossover method to exchange part of the chromosome paths;

[0147] In this step S324, simulate the recombination process of the resource allocation scheme, and use a specific crossover method to exchange part of the paths of the parent and offspring chromosomes. Specifically, randomly select two parent chromosomes, determine the crossover point according to the crossover probability, and exchange part of the gene sequences at the crossover point to form two offspring chromosomes, which helps to introduce new excellent chromosome combinations and increase population diversity.

[0148] Let the crossover probability be p u , for two parent chromosomes parent1 and parent2, randomly generate a number r between 0 and 1. If r < p u , then perform the crossover operation.

[0149] Among them, the selection of the crossover probability usually depends on the complexity of the problem and the population size. For complex problems, the crossover probability can be slightly higher to increase the generation speed of new individuals. A commonly used empirical value is between 0.6 and 0.9. For example, when initializing the genetic algorithm of the present invention, the crossover probability is set to 0.7 to ensure that the algorithm explores new solutions while retaining certain excellent genes.

[0150] Randomly determine the crossover point u (1 ≤ u ≤ U, where U is the chromosome length). After crossover, obtain the offspring chromosomes child1 and child2. The first u genes of child1 come from parent1, and the last U - u genes come from parent2; child2 is the opposite.

[0151] S325: Perform mutation operations to prevent the algorithm from falling into a local optimum, including: randomly select a parent chromosome, randomly select two gene positions on the chromosome, and exchange the gene values at these two positions to form a mutated offspring chromosome;

[0152] Let the mutation probability be p uu , for each chromosome x, randomly generate a number r between 0 and 1. If r < p uu , then perform the mutation operation.

[0153] The mutation probability is usually set relatively low to avoid destroying excellent genes. Commonly used empirical values are between 0.001 and 0.1. For example, when initializing the genetic algorithm of the present invention, the mutation probability is set to 0.01 to ensure that the algorithm can maintain the population diversity while not losing excellent solutions due to excessive mutation.

[0154] Randomly select two gene positions pos1 and pos2 (1 ≤ pos1, pos2 ≤ U), and exchange the gene values of chromosome x at these two positions to obtain the mutated offspring chromosome x'.

[0155] S326: Solve the computing-network fusion multi-layer scheduling optimization problem, including: taking the objective function of the constructed network resource model as the objective function of the algorithm, incorporating the constraint conditions of the constructed network resource model into the feasibility judgment of the algorithm, and solving the computing-network fusion multi-layer scheduling optimization problem through steps S321 to S325. By continuously iteratively optimizing the chromosome population, gradually approaching the optimal resource allocation scheme; generating a new population in each iteration, calculating the gene excellence value of each chromosome in the new population, and retaining the chromosomes with higher gene excellence values to enter the next generation until the termination condition of gene excellence value convergence is met.

[0156] Preferably, a computing-network fusion multi-layer scheduling optimization method based on dynamic demand according to an embodiment of the present invention further includes:

[0157] Step S4: Algorithm performance evaluation and optimization

[0158] The present invention customizes a set of evaluation index systems that conform to the current algorithm, including resilience indexes (such as robustness, redundancy, flexibility, and rapidity, etc.), computing power resource utilization rate, task completion rate (the ratio of the number of tasks completed on time to the total number of tasks), average task completion time, system cost (including computing power resource cost, network resource cost, etc.).

[0159] Among them, the resilience indicators include robustness, redundancy, flexibility, and rapidity, which are used to measure the resilience of the system under interference, and the specific definitions are as follows:

[0160] Robustness: Measures the ability of the system to resist interference, calculated as the ratio of the decline in system performance during interference to the performance without interference, that is, the ability of the system to maintain the original service level when facing interference such as demand interruption.

[0161]

[0162] Where P before is the performance indicator of the system without interference (such as task processing ability, service response time, etc.), and P during is the performance indicator of the system during interference.

[0163] Redundancy: Reflects the substitutability of system elements, measured by calculating the proportion of substitutable elements in the system or the availability of backup resources. The higher the redundancy, the more quickly the system can switch to backup resources when some resources fail and maintain normal operation.

[0164]

[0165] Among them, R is the number of substitutable elements (or backup resources) in the system, N is the total number of system elements (or resources), Available backup is the available amount of backup resources, and Total resource is the total amount of resources.

[0166] Resourcefulness: Evaluates the ability of the system to diagnose and solve problems, quantified based on aspects such as the adjustment of resource allocation strategies and the re-planning of task execution paths when the system faces interference. A system with high resourcefulness can adapt to changing demands and environments more quickly.

[0167]

[0168] Among them, t response is the time when the system responds to interference and starts to adjust resource allocation, and t recovery is the time when the system restores the task execution efficiency to a certain proportion (such as 90%) through adjusting resource allocation.

[0169] Rapidity: Reflects the timeliness of the system's recovery function, measuring the time required for the system to recover from interference to the normal service level. The shorter the recovery time, the better the rapidity.

[0170]

[0171] Among them, T recovery is the time required for the system to recover from interference to the normal service level (for example, the time from the decline in task processing ability to the recovery of normal processing ability, or the time from the increase in network latency to the recovery of normal latency, etc.).

[0172] To measure and evaluate the above resilience indicators, specific test scenarios can be designed. For example, interference situations such as demand interruption and high load can be simulated, and the performance changes of the system under these situations can be observed. At the same time, some specific quantitative indicators can be introduced to evaluate resilience. For example, to evaluate robustness, a demand interruption scenario can be simulated, the decline in system performance during interference can be recorded, and the ratio of its performance without interference can be calculated. To evaluate redundancy, the proportion of replaceable elements in the system or the availability of backup resources can be counted. To evaluate flexibility, the time for the system to adjust resource allocation strategies and task execution paths in the face of interference can be recorded. To evaluate rapidity, the time required for the system to recover from interference to the normal service level can be measured.

[0173] The definition of computing resource utilization (ResourceUtilization) is as follows:

[0174]

[0175] Among them, UsedComputeResource q is the amount of computing resources used by task q, which can be specifically broken down into CPU time, memory usage, number of storage I / O operations, GPU computing time, etc.; TotalComputeResource is the total computing resources of the system, which also includes the total CPU time, total memory, total storage I / O capacity, total GPU computing power, etc.; Q is the total number of tasks.

[0176] The definition of task completion rate (TaskCompletionRate) is as follows:

[0177]

[0178] Among them, Number completedTasks is the number of tasks completed on time, and Q is the total number of tasks.

[0179] The definition of average task completion time (AverageTaskCompletionTime) is as follows:

[0180]

[0181] Among them, CompletionTime q is the completion time of task q, and Q is the total number of tasks.

[0182] SystemCost is defined as:

[0183] SystemCost = ComputeCost + NetworkCost

[0184] Among them, ComputeCost is the cost of computing resource usage (calculated based on the computing resource usage amount and the unit computing resource cost), and NetworkCost is the cost of network resource usage (calculated based on the network bandwidth usage amount and the unit bandwidth cost).

[0185] According to the method for optimizing multi - layer scheduling of computing and network integration based on dynamic demand in the above embodiments of the present invention, by designing a multi - level dynamic scheduling strategy for computing power and network resources, and optimizing the allocation according to task requirements and real - time load status, the maximization of resource utilization efficiency is achieved, and the resilience of the system in case of demand interruption and other situations is improved.

[0186] The comparison between the method for optimizing multi - layer scheduling of computing and network integration based on dynamic demand of the present invention and the traditional method is as follows:

[0187] Select an actual computing and network integration system (such as a cloud computing data center network, a large - scale distributed computing platform, etc.) as the case study object, and collect detailed information such as task request data, computing power resource usage data, network traffic data, etc. of the system within a certain period of time. According to the collected data, determine the resource parameters (such as the performance of computing power devices, network bandwidth capacity, etc.), task demand characteristics (task type, task volume, time requirements, etc.) and demand uncertainty in the case, so as to provide real - data support for the application of models and algorithms.

[0188] Using the same case data, respectively adopt the traditional scheduling method (static resource allocation method) and the method based on the present invention for resource scheduling, and compare their performances in evaluation indicators. Design a variety of test scenarios, such as different task load intensities (light load, medium load, heavy load), different degrees of demand uncertainty (low, medium, high), different network topologies (simple topology, complex topology), etc., to test the performance of this technical route in different scenarios.

[0189] Table 1 Comparison Table of Evaluation Indicators

[0190] Evaluation index Traditional scheduling method Method of the present invention Improvement rate Computing resource utilization rate 72％ 85％ ↑13％ Network resource utilization rate 65％ 80％ ↑15％ Task completion rate 88％ 96％ ↑8％ Average task completion time (seconds) 320 260 ↓18.75％ System cost (seconds) 1200 980 ↓18.33％ System resilience 0.85 0.92 ↑7％

[0191] Table 2 Performance Test Results in Different Scenarios

[0192]

[0193]

[0194] As can be seen from Table 1 and Table 2, the method of the present invention significantly improves the resource utilization efficiency, especially more prominent under high load conditions; the optimized scheduling algorithm can complete tasks more efficiently, and the task completion rate is generally higher than that of traditional methods; the system cost of adopting the method of the present invention is significantly reduced, effectively saving computing and network resources; under high demand fluctuation conditions, the method of the present invention maintains high system stability and exhibits excellent robustness.

[0195] The above has described in detail one embodiment of the present invention, but the content described is only the preferred embodiment of the present invention and cannot be considered as limiting the scope of implementation of the present invention. All equivalent changes and improvements made according to the scope of the application of the present invention should still fall within the scope covered by the claims of the present invention.

[0196] It should be understood that various forms of processes shown above can be used, reordering, adding or deleting steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in the present invention can be achieved, and no limitations are imposed herein.

[0197] The above specific implementation manners do not constitute a limitation on the protection scope of the present invention. Those skilled in the art should understand that various modifications, combinations, sub - combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-layer scheduling optimization method for computing-network convergence based on dynamic requirements, characterized in that It includes the following steps: S1: Collect relevant data on computing power and network resources and perform preprocessing. The relevant data includes network resource dynamic demand data, task demand data, computing power resource data, and network resource data; S2: According to the preprocessed relevant data, construct a network resource model for converting the discrete computing power and network resource allocation problem into a continuous optimization problem; S3: Apply the constructed network resource model to the actual computing-network fusion system, combine multi-level dynamic scheduling strategies for computing power and network resources, and solve the multi-level scheduling optimization problem of computing-network fusion through an optimized genetic algorithm, accurately perceive and analyze task demands and real-time load status, and achieve dynamic scheduling and optimized allocation of computing power and network resources.

2. The multi-layer scheduling optimization method for computing-network convergence based on dynamic requirements according to claim 1, wherein Among them, Step S2 further includes: Conduct a sensitivity analysis of interference characteristics and evaluate their impact on resilience indicators; By changing interference characteristic parameters, observe the change trend of resilience indicators to determine the sensitivity of the computing-network fusion system to different interference characteristics; According to the sensitivity, provide guidance for the design and adjustment of multi-level scheduling strategies.

3. The multi-layer scheduling optimization method for computing-network convergence based on dynamic requirements according to claim 1 or 2, characterized in that Among them, The construction of the network resource model includes: Determine the modeling scope, including determining the scope of network resources, computing power resources, and task demands; Define decision variables, which include the allocation amount of computing power resources, the allocation amount of network bandwidth, and the computing nodes to which tasks are assigned; Determine the constraint conditions, which include the capacity limit of computing power devices, the upper limit of network bandwidth, and the task execution time limit; Construct an objective function with the goal of minimizing resource costs while considering the degree of task demand satisfaction and system load balancing.

4. The multi-layer scheduling optimization method for computing-network convergence based on dynamic requirements according to claim 3, wherein, The design of the multi-level dynamic scheduling strategy for computing power and network resources includes: According to the type of computing power resources, network topology structure, as well as the scale and characteristics of tasks, divide into multiple scheduling levels and formulate scheduling strategies for each level; Among them, the multiple scheduling levels include the global resource scheduling layer, the regional resource scheduling layer, and the local resource scheduling layer; the global resource scheduling layer is responsible for overall resource coordination and strategic planning; the regional resource scheduling layer focuses on resource optimization and task allocation within the region; the local resource scheduling layer performs precise resource scheduling for specific business scenarios.

5. The multi-layer scheduling optimization method for computing-network convergence based on dynamic requirements according to claim 4, wherein, Among them, The design of the optimized genetic algorithm includes: S321: Encode chromosomes using integer encoding. The chromosome contains gene sequences of task allocation, resource allocation amount, and execution time. Each gene represents a relevant attribute of a task or resource, and genetic operations and constraint processing are performed through integer encoding; S322: Consider the network resource dynamic time window limit and satisfaction requirements when initializing the population; among them, the population represents a set of resource allocation schemes; S323: Sort chromosomes according to the gene excellence value, select chromosomes with higher gene excellence to enter the next generation population, and at the same time retain the elite individuals in the previous generation population to ensure that excellent genes are not lost and accelerate the convergence of the algorithm; among them, the gene excellence value represents the excellence degree of the resource allocation scheme, and the elite individual represents the chromosome with the highest gene excellence; S324: Simulate the recombination process of resource allocation schemes and exchange partial paths of chromosomes using a specific crossover method; S325: Perform mutation operations to prevent the algorithm from falling into local optimality, including: randomly select a parental chromosome, randomly select two gene positions on the chromosome, exchange the gene values at these two positions to form a mutated offspring chromosome; S326: Solve the network fusion multi-layer scheduling optimization problem, including: taking the objective function of the constructed network resource model as the objective function of the algorithm, incorporating the constraint conditions of the constructed network resource model into the feasibility judgment of the algorithm, and solving the network fusion multi-layer scheduling optimization problem through steps S321 to S325. By continuously iteratively optimizing the chromosome population, gradually approaching the optimal resource allocation scheme; generate a new population in each iteration, calculate the gene excellence value of each chromosome in the new population, and retain the chromosomes with higher gene excellence values to enter the next generation until the termination condition of gene excellence value convergence is met.

6. The multi-layer scheduling optimization method for computing-network convergence based on dynamic requirements according to claim 5, wherein Among them, The S322: When initializing the population, consider the network resource dynamic time window limit and satisfaction requirement, including: Generate an initial feasible chromosome according to the task demand time window and resource availability, ensuring that the resource allocation scheme represented by each chromosome is feasible under time and resource constraints; Remove the chromosomes that do not meet the requirements through a screening factor, and retain the high-quality chromosomes that meet the time window and resource limitations to form an initial population.

7. The multi-layer scheduling optimization method for computing-network convergence based on dynamic requirements according to claim 5, wherein, Among them, The step S323: Sort the chromosomes according to the gene excellence value, select the chromosomes with higher gene excellence to enter the next generation population, and at the same time retain the elite individuals in the previous generation population to ensure that excellent genes are not lost and accelerate the convergence of the algorithm, including: Calculate the gene excellence value of each chromosome through a gene excellence function, and the gene excellence function comprehensively considers resource utilization rate, task completion time, and cost; Select the resource allocation scheme with better gene excellence value to enter the next round of screening, and retain the elite individuals.

8. The method for optimizing multi-layer scheduling of computing and network convergence based on dynamic requirements according to claim 7, wherein, Among them, The gene excellence function expression is: Value(x) = α × ResourceUtilization(x) + β × TaskCompletionRate(x) - γ × Cost(x) Among them, x is the chromosome, that is, it represents a resource allocation scheme; α, β, γ are weight coefficients used to adjust the importance of each factor in gene excellence evaluation; ResourceUtilization(x) is the resource utilization rate; TaskCompletionRate(x) is the task completion rate, CompletedTasks(x) is the number of tasks completed on time when using the resource allocation scheme x, and TotalTasks is the total number of tasks; Cost(x) is the cost of the resource allocation scheme x, and its calculation method is to calculate the sum of resource costs and network resource costs.

9. The multi-layer scheduling optimization method for computing-network convergence based on dynamic requirements according to claim 7, wherein The selecting the resource allocation scheme with better gene excellence value to enter the next round of screening and retaining the elite individuals includes: After calculating the gene excellence value of each chromosome, sort the chromosomes from largest to smallest according to the gene excellence value; Calculate the probability of each chromosome being selected: where G is the population size, representing the sum of the excellent gene values of the entire population; among them, Value(x g ) represents the excellent gene value of the g-th chromosome x g . Randomly generate a number r between 0 and 1. If then select chromosome x g to enter the next generation population while retaining elite individuals. Among them, represents the sum of the selection probabilities from the first chromosome to the g-th chromosome in the population, which is used to determine in which interval the random number r falls; represents the sum of the selection probabilities from the first chromosome to the (g - 1)-th chromosome in the population, and is used to determine whether the currently considered g-th chromosome should be selected.

10. The method for optimizing multi-layer scheduling of computing-network convergence based on dynamic requirements according to claim 1, wherein The algorithm performance evaluation indicators include resilience indicators, computing power resource utilization rate, task completion rate, average task completion time, and system cost; The resilience indicators include robustness, redundancy, flexibility, and rapidity, which are used to measure the resilience of the system under interference; among them, Robustness, which is used to measure the ability of the system to resist interference, is calculated as the ratio of the decline in system performance during interference to the performance without interference; Redundancy, which is used to reflect the substitutability of system elements, is measured by calculating the proportion of substitutable elements in the system or the availability of backup resources; Flexibility, which is used to evaluate the system's ability to diagnose and solve problems, is quantified based on the adjustment of the system's resource allocation strategy and the re-planning of the task execution path when facing interference; Rapidity, which is used to reflect the timeliness of the system's recovery function, measures the time required for the system to recover from interference to the normal service level.

Images