Calculation power analysis method and system for shared server and storage medium

Abstract

The invention discloses a computing power analysis method and system for a shared server and a storage medium. The computing power analysis method comprises the steps that S1, a shared server running state data set is obtained through a monitoring module; s2, preprocessing the operation state data set of the shared server; s3, the dynamic analysis model evaluates the current state and the change trend of the multi-dimensional calculation power index at the same time; s4, applying the dynamic analysis model of the computing power of the shared server to the real-time operation state data set of the shared server to generate an evaluation result of the computing power state; s5, optimizing server resource scheduling through a crown porcupine optimization algorithm; s6, applying the computing power optimization strategy of the shared server to a shared server environment; and S7, generating a prediction result by using the historical operation state data and the calculation power state evaluation result. According to the method, the trend prediction model is constructed through the crown porcupine optimization algorithm, so that the change direction and strength of the future computing power demand can be sensed in advance, and a prospective reference is provided for resource scheduling.

Description

Technical Field

[0001] The present invention relates to the technical field of shared servers, and in particular to a computing power analysis method, system and storage medium for shared servers. Background Art

[0002] With the rapid development of cloud computing and virtualization technology, shared servers, as an important part of the resource pool, play an important role in the cloud computing environment. The computing resources of shared servers mainly include CPU, GPU, memory, storage device I / O and network bandwidth. Resources meet the computing needs of multiple users and multiple tasks through dynamic scheduling. With the diversification of user needs and the increase in task complexity, the management and allocation of computing resources of shared servers face unprecedented challenges, and the problems of resource waste and low utilization are gradually emerging.

[0003] At present, most shared server resource management systems rely on traditional static analysis and scheduling methods. Traditional methods usually manage computing power based on predefined fixed strategies or simple resource allocation models. In this mode, the system periodically collects the usage status of server resources and allocates resources according to the set rules. However, traditional technologies have obvious defects:

[0004] Traditional methods often find it difficult to dynamically adapt to changes in resource demand. In scenarios with multiple tasks running concurrently and frequent load fluctuations, some resource dimensions are prone to overload while other resources are idle. This unbalanced resource utilization pattern leads to a waste of overall computing power and reduces the service capabilities of shared servers.

[0005] Existing computing power analysis methods are mostly based on offline statistics or static models and cannot respond to the complex operating environment of shared servers in real time. When the task load changes dynamically, traditional methods cannot quickly capture the changing trend of computing power demand, and scheduling decisions are delayed, which further aggravates the irrationality of resource allocation.

[0006] Traditional resource management methods usually rely on simple threshold judgments or fixed policy rules, and are unable to achieve intelligent global optimization among multi-dimensional computing power indicators. Faced with complex resource constraints and multi-objective requirements, existing technologies are unable to effectively improve the overall computing power utilization and operating efficiency of shared servers.

[0007] In summary, the existing technologies have significant deficiencies in resource utilization, real-time performance, optimization capabilities and trend prediction, and are unable to meet the needs of modern shared server environments for efficient computing power management. An innovative computing power analysis and optimization method is urgently needed to solve the existing defects. Summary of the invention

[0008] One object of the present invention is to propose a computing power analysis method, system and storage medium for a shared server. The present invention constructs a trend prediction model through the crown porcupine optimization algorithm, which can perceive the changing direction and intensity of future computing power demand in advance and provide a forward-looking reference for resource scheduling.

[0009] A computing power analysis method for a shared server according to an embodiment of the present invention includes the following steps:

[0010] S1. Obtain the shared server operation status data set through the monitoring module;

[0011] S2. Preprocessing the shared server operation status data set;

[0012] S3. Based on the preprocessed shared server operation status dataset, a shared server computing power dynamic analysis model is constructed using the dynamic modeling capability of the crown porcupine optimization algorithm. The dynamic analysis model simultaneously evaluates the current status and change trend of multi-dimensional computing power indicators;

[0013] S4. Apply the shared server computing power dynamic analysis model to the real-time shared server operation status data set to generate an evaluation result of the computing power status;

[0014] S5. According to the evaluation results of the computing power status of the shared server, the server resource scheduling is optimized by the crown porcupine optimization algorithm to generate a shared server computing power optimization strategy;

[0015] S6. Apply the shared server computing power optimization strategy to the shared server environment, dynamically adjust the operating parameters of the computing resources according to the optimization strategy, and collect the adjusted shared server operating status data in real time;

[0016] S7. Utilize the historical operation status data and computing power status evaluation results, combined with the crown porcupine optimization algorithm, to predict the future computing power demand change trend of the shared server and generate the prediction results.

[0017] Optionally, the S1 specifically includes:

[0018] S11. Using the real-time monitoring module deployed in the shared server, continuously collect and record the CPU operation status data of the server to generate the CPU utilization data set D CPU ,The CPU utilization data set records the CPU occupancy ratio of the shared ,server at each time point with the time point as the index, reflecting the ,work intensity and resource consumption of the CPU when executing ,tasks;

[0019] S12. Continuously sample the GPU operating status through the monitoring module to form a GPU occupancy rate data set D GPU,The GPU occupancy data set is indexed by time point and covers the dynamic usage of GPU resources during the ,server operation;

[0020] S13. Obtain memory usage status through the monitoring module and generate a memory usage rate data set D MEM ,The memory usage dataset records the memory allocation and usage ratio of the ,shared server at each time point, reflecting the immediate ,status and potential bottlenecks of resource allocation;

[0021] S14. Continuously monitor the read and write rates of the storage device, collect the read and write rate data of the storage device at different time points through the monitoring module, and construct a storage device read and write rate data set D including read performance and write performance IO ,The storage device read and write rate data set is in time order, reflecting the I / O performance of the storage device under high-load tasks;

[0022] S15. Collect the utilization of network resources through the monitoring module to generate a network bandwidth utilization data set D NET ,The network bandwidth utilization dataset uses time points as indexes, and records the ,network bandwidth occupancy ratio of shared servers at different time points, ,reflecting the changes in the demand for network resources and the ,real-time allocation of tasks. ,The fluctuation of network bandwidth utilization directly reveals the ,performance bottleneck of the server when transmitting large amounts of data;

[0023] S16 integrates the CPU utilization data set, GPU occupancy data set, memory usage data set, storage device read and write rate data set, and network bandwidth utilization data set of S11-S15 to form a complete shared server operation status data set:

[0024] D RUN ={D CPU ,D GPU ,D MEM ,D IO ,D NET}.

[0025] Optionally, S3 specifically includes the following steps:

[0026] S31. Using the crown porcupine optimization algorithm on the preprocessed shared server operation status dataset D NORMALIZED The multi-layer structure combines the five computing power dimensions of CPU, GPU, memory, storage I / O and network bandwidth to form an initial multi-layer nested model. The multi-layer nested structure of the initial multi-layer nested model simultaneously captures the local characteristics and global synergy of each computing power dimension:

[0027]

[0028] K means that the shared server computing power is divided into K layers, and each layer model With the crown porcupine optimization algorithm as the core, Θ (k) Represents the initial parameter set of the model at this layer, including the population size, initial position, and iteration strategy of the computing power dimension;

[0029] S32. Based on the initial multi-layer nested model, define the multi-objective dynamic optimization objective function for CPU, GPU, memory, storage I / O and network bandwidth in each layer:

[0030] F(x,t)=∑ d∈{CPU,GPU,MEM,IO,NET} w d (t)·f d (x,t);

[0031] Among them, x represents the computing power allocation status of the current crested porcupine individual, t is the discrete time series index, and w d (t) is the time variable weight for the computing power dimension d, which is dynamically adjusted as the shared server operation status and task requirements change. d (x, t) is a comprehensive description of the resource utilization balance, load balance, and bottleneck situation in the computing power dimension d, which is used to measure the degree of achievement of the optimization goal of this dimension at time t;

[0032] S33. In the multi-layer nested model, the interaction behaviors between crested porcupines were dynamically constrained to simulate the real decision-making process of crested porcupines in resource competition and cooperation. The constraint mechanism includes:

[0033]

[0034] Among them, p i ,p j They represent the positions of two crested porcupine individuals in the multidimensional computing power index space, sim(p i ,p j ) represents the similarity between the two in terms of computing power configuration, γ sim is the collaboration triggering threshold, C(p i ,p j ) represents the impact of collaborative operations on individual positions, λ c is the collaboration intensity, u bottleneck (p i ) indicates that in individual p i The bottleneck utilization rate in the current state, u th is the bottleneck threshold, λ b is the obstacle avoidance strength, represents the local gradient in the computing power dimension d, w priority (t) is the time-variable weight of the task priority, δ p Indicates the priority trigger threshold, λp is the priority intensity, f priority (p i ,t) is the gradient function of the resource demand of the high priority task;

[0035] The dynamic constraint mechanism handles three scenarios in multi-dimensional computing power allocation: resource collaboration, bottleneck avoidance, and priority allocation;

[0036] S34. For the crowned porcupine individuals in each layer of the multi-layer nested model, the individuals are combined and iterated to cooperate, avoid obstacles or allocate priority according to the dynamic constraint output results. The multi-layer feedback mechanism transmits messages between dynamic analysis models according to the iteration completion and convergence of each layer, so that the optimization process of the five computing power dimensions is coordinated with each other and gradually approaches the overall optimum:

[0037]

[0038] in, represents the computing power configuration of crested porcupine individual i in the k-th layer model at the t-th generation iteration, α (k) and β (k) are the adjustment parameters of the k-th layer model in the local gradient and global guidance direction, D is the total number of multi-objective dimensions, Ω(P (k) ) represents the same layer crested porcupine individual group P (k) Global information aggregation operation;

[0039] S35. After the multi-objective dynamic optimization convergence conditions are met, the shared server computing power dynamic analysis model is output, and the dynamic analysis model and the corresponding crested porcupine individual iteration trajectory are stored in a distributed storage medium:

[0040]

[0041] in, Indicates that the crowned porcupine individuals in the k-th layer model converge at the final generation T end The optimal computing power configuration when n k is the number of crested porcupine individuals in this layer, Λ constraints is the set of dynamic constraints formed during the optimization process, F final Represents the final set of multi-objective function values.

[0042] Optionally, the S4 specifically includes the following steps:

[0043] S41. Collect the shared server operation status data set D in real time through the monitoring module REALTIME Dynamic analysis model M of shared server computing power CPO-dynamic Matching, calculate the mapping position Φ(x,t) of the real-time running status in the model;

[0044] S42. Dynamic analysis model of computing power based on matching M CPO-dynamic Evaluate the utilization of the real-time shared server operation status data and calculate the utilization of each computing power dimension:

[0045]

[0046] Among them, U d (t) represents the utilization rate of computing power dimension d at time t, R d (t) is the real-time computing power requirement, C d is the total computing capacity of this dimension;

[0047] S43. Based on the real-time computing power utilization U(t) and computing power dynamic analysis model M CPO-dynamic Analyze the load distribution of each computing power dimension of the current shared server, define the load distribution function and generate the shared server load heat map based on the load distribution function:

[0048]

[0049] Among them, L d (t) represents the overall load distribution of computing power dimension d, n d is the total number of tasks distributed on this dimension, is the computing power requirement of the i-th task at time t;

[0050] S44. Detect the bottleneck of computing resources by combining real-time data and load distribution:

[0051]

[0052] Among them, B d (t) indicates whether there is a bottleneck in the computing power dimension d at time t, η d is the bottleneck threshold;

[0053] S45. Comprehensive current computing power utilization set U(t) and load distribution function L d (t) and bottleneck function B(t) to generate a complete computing power status evaluation result:

[0054] E(t)={U(t),L d (t), B(t)}.

[0055] Optionally, the S5 specifically includes the following steps:

[0056] S51. According to the computing power status evaluation result E(t) combined with the crown porcupine optimization algorithm, define the resource scheduling optimization objective function:

[0057] F opt (x) = ∑ d∈{CPU,GPU,MEM,IO,NET} wd (t)·[α d ·U d (t)+β d ·L d (t)+γ d ·(1-B d (t))];

[0058] Among them, x represents the current allocation status of server resources, w d (t) is the dynamic weight of computing power dimension d, α d ,β d ,γ d They are the adjustment coefficients for real-time utilization, load distribution, and smoothing bottlenecks, respectively;

[0059] S52. Optimizing the objective function F based on resource scheduling opt (x) Construct the initial population P of the crown porcupine optimization algorithm. Each individual p in the initial population P i Represents a resource scheduling scheme whose initial position is generated by random distribution in the computing power allocation space that satisfies the constraints:

[0060]

[0061] in, Indicates the resource allocation ratio of computing power dimension d in individual i;

[0062] S53. Optimize the objective function F according to resource scheduling opt (x) Using the rules of the crested porcupine optimization algorithm, iteratively update the position of each individual:

[0063]

[0064] in, and represents the resource allocation status of individual i in the tth and t+1th generations, respectively, and x best is the individual's best historical resource allocation state, x g is the global optimal resource allocation state of the population, To optimize the gradient of the objective function and guide the individual moving direction, γ 1 ,γ 2 ,γ 3 are adjustment coefficients, which control the weights of history guidance, global guidance, and gradient optimization respectively;

[0065] S54. According to the optimized resource scheduling plan Dynamically generate task allocation strategies for shared servers:

[0066] Task reallocation: Combining real-time utilization U d(t) and the dynamic weights w of each dimension in the optimization objective function d (t), prioritize resources to tasks on high-weight dimensions;

[0067] Virtual machine configuration adjustment: dynamically adjust the virtual machine's computing power quota based on the optimization results Requirements:

[0068]

[0069] Load balancing strategy update: According to the load distribution L in the optimization objective function d (t) and bottleneck detection result B d (t), dynamically adjust task allocation to low-load nodes to optimize overall load balancing;

[0070] S55. Store the optimized resource scheduling scheme and the corresponding optimization objective function value as a shared server computing power optimization strategy set S opt .

[0071] Optionally, the S7 specifically includes the following steps:

[0072] S71. Extract the historical operation status data and computing power status evaluation results of the shared server, and construct a historical data set containing multi-dimensional computing power indicators. The historical data set includes the real-time utilization, load distribution and bottleneck status of each computing power dimension at different time points;

[0073] S72. By analyzing the historical data set and the current evaluation results, set the computing power demand change trend analysis target, which combines the utilization rate change speed, load distribution stability and bottleneck state persistence of each computing power dimension to measure the direction and intensity of the future computing power demand change trend;

[0074] S73. Under the guidance of the computing power demand change trend analysis goal, a population of the crested porcupine optimization algorithm is constructed, each population individual represents a possible future computing power demand distribution state, and the initial population is randomly generated according to the statistical law of historical data and current computing power distribution;

[0075] S74. Utilize the iterative mechanism of the crown porcupine optimization algorithm to optimize the distribution of computing power requirements of individuals in the population based on historical data and real-time evaluation results, so that it gradually approaches the global optimal trend. Through the collaboration, obstacle avoidance and priority guidance mechanisms among individuals, dynamically simulate the growth or decline trend of future requirements of different computing power dimensions.

[0076] S75. Generate the prediction results of the future computing power demand change trend based on the optimized population. The prediction results include the utilization change trend of each computing power dimension in the future time period, the load distribution possibility and the potential risk of bottleneck state.

[0077] A computing power analysis system for a shared server, used to implement computing power analysis for a shared server, the system comprising:

[0078] The monitoring module is used to collect the shared server operation status data in real time. The monitoring module interacts directly with the server hardware resources and obtains the computing power status in real time through the data interface;

[0079] A data processing module is used to pre-process the operating status data collected by the monitoring module;

[0080] Dynamic analysis modeling module, used to build a dynamic analysis model of shared server computing power based on the crown porcupine optimization algorithm;

[0081] The computing power status assessment module is used to apply the dynamic analysis model to the real-time operation status data to generate the computing power status assessment results;

[0082] Resource scheduling optimization module, which is used to generate resource scheduling optimization strategies for shared servers based on computing power status evaluation results and combined with the crown porcupine optimization algorithm;

[0083] The computing power demand prediction module is used to predict the future computing power demand change trend of the shared server based on the historical operation status data and the current computing power status evaluation results, combined with the crown porcupine optimization algorithm;

[0084] Storage module, used to store data and model results generated by each module;

[0085] The control module is used to coordinate the operation of the above modules, receive real-time data input from the monitoring module, and transmit the data to the data processing module and the dynamic analysis modeling module.

[0086] A computer-readable storage medium stores a computer program, and the computer program is executed by a processor to perform the steps of the computing power analysis method for a shared server.

[0087] The beneficial effects of the present invention are:

[0088] (1) The present invention adopts an improved crowned porcupine optimization algorithm to construct a shared server computing power dynamic analysis model, which can capture the coordinated change relationship of different computing power dimensions in real time, and accurately evaluate the computing power status through dynamic modeling. By simulating the multi-dimensional behavioral characteristics of the crowned porcupine, the model is dynamically updated while optimizing resource scheduling, which significantly improves the real-time and accuracy of computing power analysis.

[0089] (2) Based on the computing power status evaluation results and combined with the multi-objective optimization function, the present invention dynamically generates task allocation, virtual machine configuration adjustment and load balancing strategies through the crown porcupine optimization algorithm. Compared with the traditional resource scheduling method based on fixed rules, the multi-objective optimization mechanism can realize the intelligent allocation and balance of resources under the condition of complex changes in task load, avoiding resource conflicts and idle problems caused by scheduling delays.

[0090] (3) The present invention combines historical computing power status data and current computing power evaluation results, and constructs a trend prediction model through the crown porcupine optimization algorithm. It can perceive the direction and intensity of changes in future computing power demand in advance, and provide a forward-looking reference for resource scheduling. By optimizing the population iteration method to simulate the trend of changes in computing power demand, the accuracy and robustness of the prediction are significantly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0091] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0092] Figure 1 A flowchart of a computing power analysis method, system and storage medium for a shared server proposed by the present invention. DETAILED DESCRIPTION

[0093] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0094] refer to Figure 1 , a computing power analysis method for a shared server, comprising the following steps:

[0095] S1. Obtain the shared server operation status data set through the monitoring module;

[0096] S2. Preprocessing the shared server operation status data set;

[0097] S3. Based on the preprocessed shared server operation status dataset, a shared server computing power dynamic analysis model is constructed using the dynamic modeling capability of the crown porcupine optimization algorithm. The dynamic analysis model simultaneously evaluates the current status and change trend of multi-dimensional computing power indicators;

[0098] S4. Apply the shared server computing power dynamic analysis model to the real-time shared server operation status data set to generate an evaluation result of the computing power status;

[0099] S5. According to the evaluation results of the computing power status of the shared server, the server resource scheduling is optimized by the crown porcupine optimization algorithm to generate a shared server computing power optimization strategy;

[0100] S6. Apply the shared server computing power optimization strategy to the shared server environment, dynamically adjust the operating parameters of the computing resources according to the optimization strategy, and collect the adjusted shared server operating status data in real time;

[0101] S7. Utilize the historical operation status data and computing power status evaluation results, combined with the crown porcupine optimization algorithm, to predict the future computing power demand change trend of the shared server and generate the prediction results.

[0102] In this implementation, S1 specifically includes:

[0103] S11. Using the real-time monitoring module deployed in the shared server, continuously collect and record the CPU operation status data of the server to generate the CPU utilization data set D CPU ,The CPU utilization data set records the CPU occupancy ratio of the shared ,server at each time point with the time point as the index, reflecting the ,work intensity and resource consumption of the CPU when executing ,tasks;

[0104] S12. Continuously sample the GPU operating status through the monitoring module to form a GPU occupancy rate data set D GPU ,The GPU occupancy data set is indexed by time point and covers the dynamic usage of GPU resources during the ,server operation;

[0105] S13. Obtain memory usage status through the monitoring module and generate a memory usage rate data set D MEM ,The memory usage dataset records the memory allocation and usage ratio of the ,shared server at each time point, reflecting the immediate ,status and potential bottlenecks of resource allocation;

[0106] S14. Continuously monitor the read and write rates of the storage device, collect the read and write rate data of the storage device at different time points through the monitoring module, and construct a storage device read and write rate data set D including read performance and write performance IO ,The storage device read and write rate data set is in time order, reflecting the I / O performance of the storage device under high-load tasks;

[0107] S15. Collect the utilization of network resources through the monitoring module to generate a network bandwidth utilization data set D NET ,The network bandwidth utilization dataset uses time points as indexes, and records the ,network bandwidth occupancy ratio of shared servers at different time points, ,reflecting the changes in the demand for network resources and the ,real-time allocation of tasks. ,The fluctuation of network bandwidth utilization directly reveals the ,performance bottleneck of the server when transmitting large amounts of data;

[0108] S16 integrates the CPU utilization data set, GPU occupancy data set, memory usage data set, storage device read and write rate data set, and network bandwidth utilization data set of S11-S15 to form a complete shared server operation status data set:

[0109] D RUN ={D CPU ,D GPU ,D MEM ,D IO ,D NET}.

[0110] In this implementation, S3 specifically includes the following steps:

[0111] S31. Using the crown porcupine optimization algorithm on the preprocessed shared server operation status dataset D NORMALIZED The multi-layer structure combines the five computing power dimensions of CPU, GPU, memory, storage I / O and network bandwidth to form an initial multi-layer nested model. The multi-layer nested structure of the initial multi-layer nested model simultaneously captures the local characteristics and global synergy of each computing power dimension:

[0112]

[0113] K means that the shared server computing power is divided into K layers, and each layer model With the crown porcupine optimization algorithm as the core, Θ (k) Represents the initial parameter set of the model at this layer, including the population size, initial position, and iteration strategy of the computing power dimension;

[0114] S32. Based on the initial multi-layer nested model, define the multi-objective dynamic optimization objective function for CPU, GPU, memory, storage I / O and network bandwidth in each layer:

[0115] F(x,t)=∑ d∈{CPU,GPU,MEM,IO,NET} w d (t)·f d (x,t);

[0116] Among them, x represents the computing power allocation status of the current crested porcupine individual, t is the discrete time series index, and w d (t) is the time variable weight for the computing power dimension d, which is dynamically adjusted as the shared server operation status and task requirements change. d (x, t) is a comprehensive description of the resource utilization balance, load balance, and bottleneck situation in the computing power dimension d, which is used to measure the degree of achievement of the optimization goal of this dimension at time t;

[0117] S33. In the multi-layer nested model, the interaction behaviors between crested porcupines were dynamically constrained to simulate the real decision-making process of crested porcupines in resource competition and cooperation. The constraint mechanism includes:

[0118]

[0119] Among them, p i ,p j They represent the positions of two crested porcupine individuals in the multidimensional computing power index space, sim(p i ,p j ) represents the similarity between the two in terms of computing power configuration, γ sim is the collaboration triggering threshold, C(p i ,p j ) represents the impact of collaborative operations on individual positions, λ c is the collaboration intensity, u bottleneck (p i ) indicates that in individual p i The bottleneck utilization rate in the current state, u th is the bottleneck threshold, λ b is the obstacle avoidance strength, represents the local gradient in the computing power dimension d, w priority (t) is the time-variable weight of the task priority, δ p Indicates the threshold for priority triggering, λ p is the priority intensity, f priority (p i ,t) is the gradient function of the resource demand of the high priority task;

[0120] The dynamic constraint mechanism handles three scenarios in multi-dimensional computing power allocation: resource collaboration, bottleneck avoidance, and priority allocation;

[0121] S34. For the crowned porcupine individuals in each layer of the multi-layer nested model, the individuals are combined and iterated to cooperate, avoid obstacles or allocate priority according to the dynamic constraint output results. The multi-layer feedback mechanism transmits messages between dynamic analysis models according to the iteration completion and convergence of each layer, so that the optimization process of the five computing power dimensions is coordinated with each other and gradually approaches the overall optimum:

[0122]

[0123] in, represents the computing power configuration of crested porcupine individual i in the k-th layer model at the t-th generation iteration, α (k) and β (k) are the adjustment parameters of the k-th layer model in the local gradient and global guidance direction, D is the total number of multi-objective dimensions, Ω(P (k) ) represents the same layer crested porcupine individual group P (k)Global information aggregation operation;

[0124] S35. After the multi-objective dynamic optimization convergence conditions are met, the shared server computing power dynamic analysis model is output, and the dynamic analysis model and the corresponding crested porcupine individual iteration trajectory are stored in a distributed storage medium:

[0125]

[0126] in, Indicates that the crowned porcupine individuals in the k-th layer model converge at the final generation T end The optimal computing power configuration when n k is the number of crested porcupine individuals in this layer, Λ constraints is the set of dynamic constraints formed during the optimization process, F final Represents the final set of multi-objective function values.

[0127] In this implementation, S4 specifically includes the following steps:

[0128] S41. Collect the shared server operation status data set D in real time through the monitoring module REALTIME Dynamic analysis model M of shared server computing power CPO-dynamic Matching, calculate the mapping position Φ(x,t) of the real-time running status in the model;

[0129] S42. Dynamic analysis model of computing power based on matching M CPO-dynamic Evaluate the utilization of the real-time shared server operation status data and calculate the utilization of each computing power dimension:

[0130]

[0131] Among them, U d (t) represents the utilization rate of computing power dimension d at time t, R d (t) is the real-time computing power requirement, C d is the total computing capacity of this dimension;

[0132] S43. Based on the real-time computing power utilization U(t) and computing power dynamic analysis model M CPO-dynamic Analyze the load distribution of each computing power dimension of the current shared server, define the load distribution function and generate the shared server load heat map based on the load distribution function:

[0133]

[0134] Among them, L d (t) represents the overall load distribution of computing power dimension d, n d is the total number of tasks distributed on this dimension, is the computing power requirement of the i-th task at time t;

[0135] S44. Detect the bottleneck of computing resources by combining real-time data and load distribution:

[0136]

[0137] Among them, B d (t) indicates whether there is a bottleneck in the computing power dimension d at time t, η d is the bottleneck threshold;

[0138] S45. Comprehensive current computing power utilization set U(t) and load distribution function L d (t) and bottleneck function B(t) to generate a complete computing power status evaluation result:

[0139] E(t)={U(t),L d (t), B(t)}.

[0140] In this implementation, S5 specifically includes the following steps:

[0141] S51. According to the computing power status evaluation result E(t) combined with the crown porcupine optimization algorithm, define the resource scheduling optimization objective function:

[0142] F opt (x) = ∑ d∈{CPU,GPU,MEM,IO,NET} w d (t)·[α d ·U d (t)+β d ·L d (t)+γ d ·(1-B d (t))];

[0143] Among them, x represents the current allocation status of server resources, w d (t) is the dynamic weight of computing power dimension d, α d ,β d ,γ d They are the adjustment coefficients for real-time utilization, load distribution, and smoothing bottlenecks, respectively;

[0144] S52. Optimizing the objective function F based on resource scheduling opt (x) Construct the initial population P of the crown porcupine optimization algorithm. Each individual p in the initial population P i Represents a resource scheduling scheme whose initial position is generated by random distribution in the computing power allocation space that satisfies the constraints:

[0145]

[0146] in, Indicates the resource allocation ratio of computing power dimension d in individual i;

[0147] S53. Optimize the objective function F according to resource scheduling opt (x) Using the rules of the crested porcupine optimization algorithm, iteratively update the position of each individual:

[0148]

[0149] in, and represents the resource allocation status of individual i in the tth and t+1th generations, respectively, and x best is the individual's best historical resource allocation state, x g is the global optimal resource allocation state of the population, To optimize the gradient of the objective function and guide the individual moving direction, γ 1 ,γ 2 ,γ 3 are adjustment coefficients, which control the weights of history guidance, global guidance, and gradient optimization respectively;

[0150] S54. According to the optimized resource scheduling plan Dynamically generate task allocation strategies for shared servers:

[0151] Task reallocation: Combining real-time utilization U d (t) and the dynamic weights w of each dimension in the optimization objective function d (t), prioritize resources to tasks on high-weight dimensions;

[0152] Virtual machine configuration adjustment: dynamically adjust the virtual machine's computing power quota based on the optimization results Requirements:

[0153]

[0154] Load balancing strategy update: According to the load distribution L in the optimization objective function d (t) and bottleneck detection result B d (t), dynamically adjust task allocation to low-load nodes to optimize overall load balancing;

[0155] S55. Store the optimized resource scheduling scheme and the corresponding optimization objective function value as a shared server computing power optimization strategy set S opt .

[0156] In this implementation, S7 specifically includes the following steps:

[0157] S71. Extract the historical operation status data and computing power status evaluation results of the shared server, and construct a historical data set containing multi-dimensional computing power indicators. The historical data set includes the real-time utilization, load distribution and bottleneck status of each computing power dimension at different time points;

[0158] S72. By analyzing the historical data set and the current evaluation results, set the computing power demand change trend analysis target, which combines the utilization rate change speed, load distribution stability and bottleneck state persistence of each computing power dimension to measure the direction and intensity of the future computing power demand change trend;

[0159] S73. Under the guidance of the computing power demand change trend analysis goal, a population of the crested porcupine optimization algorithm is constructed, each population individual represents a possible future computing power demand distribution state, and the initial population is randomly generated according to the statistical law of historical data and current computing power distribution;

[0160] S74. Utilize the iterative mechanism of the crown porcupine optimization algorithm to optimize the distribution of computing power requirements of individuals in the population based on historical data and real-time evaluation results, so that it gradually approaches the global optimal trend. Through the collaboration, obstacle avoidance and priority guidance mechanisms among individuals, dynamically simulate the growth or decline trend of future requirements of different computing power dimensions.

[0161] S75. Generate the prediction results of the future computing power demand change trend based on the optimized population. The prediction results include the utilization change trend of each computing power dimension in the future time period, the load distribution possibility and the potential risk of bottleneck state.

[0162] A computing power analysis system for a shared server, used to implement computing power analysis for a shared server, the system comprising:

[0163] The monitoring module is used to collect the shared server operation status data in real time. The monitoring module interacts directly with the server hardware resources and obtains the computing power status in real time through the data interface;

[0164] A data processing module is used to pre-process the operating status data collected by the monitoring module;

[0165] Dynamic analysis modeling module, used to build a dynamic analysis model of shared server computing power based on the crown porcupine optimization algorithm;

[0166] The computing power status assessment module is used to apply the dynamic analysis model to the real-time operation status data to generate the computing power status assessment results;

[0167] Resource scheduling optimization module, which is used to generate resource scheduling optimization strategies for shared servers based on computing power status evaluation results and combined with the crown porcupine optimization algorithm;

[0168] The computing power demand prediction module is used to predict the future computing power demand change trend of the shared server based on the historical operation status data and the current computing power status evaluation results, combined with the crown porcupine optimization algorithm;

[0169] Storage module, used to store data and model results generated by each module;

[0170] The control module is used to coordinate the operation of the above modules, receive real-time data input from the monitoring module, and transmit the data to the data processing module and the dynamic analysis modeling module.

[0171] A computer-readable storage medium stores a computer program, and the computer program is executed by a processor to perform steps of a computing power analysis method for a shared server.

[0172] Embodiment 1:

[0173] In the embodiment, a cloud service provider operates a group of large shared server clusters to support the online business of hundreds of enterprise users. The enterprise users are involved in e-commerce, financial computing and real-time data analysis scenarios, and the demand for computing resources fluctuates greatly. During the specific time periods of the Double Eleven promotion and financial settlement peak, the resource load of the server cluster rises sharply, resulting in serious problems of unbalanced computing power allocation and task congestion.

[0174] In this embodiment, the shared server cluster consists of 500 servers, each of which has the following resources: 32-core CPU, 4 GPU cards, 128GB memory, 1TB storage device and 10Gbps network bandwidth. The method of the present invention is used to perform computing power analysis and optimization management on this group of shared servers:

[0175] The monitoring module collects the server's operating status data in real time to form a data set. During the peak period of the Double Eleven promotion, data is collected once a second, including CPU utilization, GPU occupancy, memory usage, storage device read and write rates, and network bandwidth occupancy. The collected data is cleaned and normalized by the data processing module to remove noise and outliers to ensure data accuracy.

[0176] Then, a dynamic analysis model of shared server computing power is constructed based on the real-time collected data. The dynamic analysis modeling module uses the crown porcupine optimization algorithm to map the multi-dimensional computing power data into a computing power distribution model. During the model construction process, the dynamic relationship between different computing power dimensions is captured by simulating the resource competition and collaborative behavior of the server, and resource utilization imbalance and potential bottlenecks are identified. In the embodiment, at a certain moment, the average GPU utilization in the cluster is monitored to be 92%, while the CPU and memory utilization are only 55% and 40% respectively, which indicates that the GPU resources are close to the bottleneck and other resources are idle.

[0177] Next, the computing power status assessment module applies the dynamic analysis model to real-time data to generate computing power status assessment results. The assessment results show that there is a large room for optimization in the allocation of CPU and memory during the peak promotion period, while the load distribution of GPU and network bandwidth is unbalanced. The GPU utilization rate of some nodes reaches 95%, while that of other nodes is only 50%. The bottleneck of network bandwidth causes some high-traffic tasks to fail to be completed in time.

[0178] The resource scheduling optimization module generates an optimization strategy based on the computing power status evaluation results. Specifically, through the dynamic iteration of the crown porcupine optimization algorithm, tasks are reallocated to low-load nodes to adjust the virtual machine configuration, and bandwidth is prioritized in high-load tasks. In the embodiment, high-computing intensive tasks originally concentrated on certain nodes are dispersed to more nodes, and bandwidth priority weights are set for certain GPU tasks to reduce resource competition.

[0179] Finally, combining historical data and current evaluation results, the computing power demand prediction module predicts the future trend of computing power demand changes. In the embodiment, due to the gradual decrease in user shopping behavior in the next two hours, it is predicted that the CPU and GPU utilization rates will decrease by 10% and 15% respectively, and the storage I / O demand will increase by 20%. Based on these prediction results, the system adjusts the resource allocation strategy in advance to smoothly respond to task changes.

[0180] In order to verify the effect of the method of the present invention, a comparative test was conducted with the traditional static resource management method. The test data was derived from the operating data of the above scenario, and the statistical cycle was 24 hours. The following Table 1 is a detailed comparison result:

[0181] Table 1 Comparison of test results of the present invention and traditional static resource management method

[0182]

[0183]

[0184] With the support of the method of the present invention, the utilization rates of CPU, GPU and memory increased by 33.1%, 54.0% and 17.3% respectively, and the average task completion rate increased from 86.2% to 97.4%. In particular, during the peak promotion period, the response time of the traditional method was as high as 678ms, while the method of the present invention reduced it to 324ms after optimization, significantly improving the user experience. In addition, the 7 downtime nodes caused by overload in the traditional method were completely eliminated, further ensuring the stability of the system.

[0185] The method of the present invention improves the accuracy of computing power demand prediction, effectively schedules resources in advance, and reduces performance bottlenecks caused by a surge in tasks. Overall, the efficiency and intelligence of the method of the present invention greatly surpass traditional methods, and significantly improves the computing power management efficiency and operational stability of shared servers.

[0186] The present invention adopts an improved crowned porcupine optimization algorithm to construct a shared server computing power dynamic analysis model, which can capture the coordinated change relationship of different computing power dimensions in real time, and accurately evaluate the computing power status through dynamic modeling. By simulating the multi-dimensional behavioral characteristics of the crowned porcupine, the model is dynamically updated while optimizing resource scheduling, which significantly improves the real-time and accuracy of computing power analysis.

[0187] Based on the computing power status evaluation results and combined with the multi-objective optimization function, the present invention dynamically generates task allocation, virtual machine configuration adjustment and load balancing strategies through the crown porcupine optimization algorithm. Compared with the traditional resource scheduling method based on fixed rules, the multi-objective optimization mechanism can realize the intelligent allocation and balance of resources under the condition of complex changes in task load, avoiding resource conflicts and idle problems caused by scheduling delays.

[0188] The present invention combines historical computing power status data and current computing power assessment results, and constructs a trend prediction model through the crown porcupine optimization algorithm. It can perceive the changing direction and intensity of future computing power demand in advance, and provide a forward-looking reference for resource scheduling. By optimizing the population iteration method to simulate the changing trend of computing power demand, the accuracy and robustness of the prediction are significantly improved.

[0189] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A computing power analysis method for a shared server, characterized in that: The steps include: S1. Obtain the shared server operation status data set through the monitoring module; S2. Preprocessing the shared server operation status data set; S3. Based on the preprocessed shared server operation status dataset, a shared server computing power dynamic analysis model is constructed using the dynamic modeling capability of the crown porcupine optimization algorithm. The dynamic analysis model simultaneously evaluates the current status and change trend of multi-dimensional computing power indicators; S4. Apply the shared server computing power dynamic analysis model to the real-time shared server operation status data set to generate an evaluation result of the computing power status; S5. According to the evaluation results of the computing power status of the shared server, the server resource scheduling is optimized by the crown porcupine optimization algorithm to generate a shared server computing power optimization strategy; S6. Apply the shared server computing power optimization strategy to the shared server environment, dynamically adjust the operating parameters of the computing resources according to the optimization strategy, and collect the adjusted shared server operating status data in real time; S7. Utilize the historical operation status data and computing power status evaluation results, combined with the crown porcupine optimization algorithm, to predict the future computing power demand change trend of the shared server and generate the prediction results.

2. A computing power analysis method for a shared server according to claim 1, characterized in that: The S1 specifically includes: S11. Using the real-time monitoring module deployed in the shared server, continuously collect and record the CPU operation status data of the server to generate the CPU utilization data set D CPU ,The CPU utilization data set records the CPU occupancy ratio of the shared ,server at each time point with the time point as the index, reflecting the ,work intensity and resource consumption of the CPU when executing ,tasks; S12. Continuously sample the GPU operating status through the monitoring module to form a GPU occupancy rate data set D GPU ,The GPU occupancy data set is indexed by time point and covers the dynamic usage of GPU resources during the ,server operation; S13. Obtain memory usage status through the monitoring module and generate a memory usage rate data set D MEM ,The memory usage dataset records the memory allocation and usage ratio of the ,shared server at each time point, reflecting the immediate ,status and potential bottlenecks of resource allocation; S14. Continuously monitor the read and write rates of the storage device, collect the read and write rate data of the storage device at different time points through the monitoring module, and construct a storage device read and write rate data set D including read performance and write performance IO ,The storage device read and write rate data set is in time order, reflecting the I / O performance of the storage device under high-load tasks; S15. Collect the utilization of network resources through the monitoring module to generate a network bandwidth utilization data set D NET ,The network bandwidth utilization dataset uses time points as indexes, and records the ,network bandwidth occupancy ratio of shared servers at different time points, ,reflecting the changes in the demand for network resources and the ,real-time allocation of tasks. ,The fluctuation of network bandwidth utilization directly reveals the ,performance bottleneck of the server when transmitting large amounts of data; S16 integrates the CPU utilization data set, GPU occupancy data set, memory usage data set, storage device read and write rate data set, and network bandwidth utilization data set of S11-S15 to form a complete shared server operation status data set: D RUN ={D CPU ,D GPU ,D MEM ,D IO ,D NET }。 3. A computing power analysis method for a shared server according to claim 1, characterized in that: The S3 specifically includes the following steps: S31. Using the crown porcupine optimization algorithm on the preprocessed shared server operation status dataset D NORMALIZED The multi-layer structure combines the five computing power dimensions of CPU, GPU, memory, storage I / O and network bandwidth to form an initial multi-layer nested model. The multi-layer nested structure of the initial multi-layer nested model simultaneously captures the local characteristics and global synergy of each computing power dimension: K means that the shared server computing power is divided into K layers, and each layer model With the crown porcupine optimization algorithm as the core, Θ (k) Represents the initial parameter set of the model at this layer, including the population size, initial position, and iteration strategy of the computing power dimension; S32. Based on the initial multi-layer nested model, define the multi-objective dynamic optimization objective function for CPU, GPU, memory, storage I / O and network bandwidth in each layer: F(x,t)=∑ d∈{CPU,GPU,MEM,IO,NET} w d (t)·f d (x,t); Among them, x represents the computing power allocation status of the current crested porcupine individual, t is the discrete time series index, and w d (t) is the time variable weight for the computing power dimension d, which is dynamically adjusted as the shared server operation status and task requirements change. d (x, t) is a comprehensive description of the resource utilization balance, load balance, and bottleneck situation in the computing power dimension d, which is used to measure the degree of achievement of the optimization goal of this dimension at time t; S33. In the multi-layer nested model, the interaction behaviors between crested porcupines were dynamically constrained to simulate the real decision-making process of crested porcupines in resource competition and cooperation. The constraint mechanism includes: (1) Collaboration trigger: (2) Obstacle avoidance trigger: When (3) Priority trigger: When Among them, p i ,p j They represent the positions of two crested porcupine individuals in the multidimensional computing power index space, sim(p i ,p j ) represents the similarity between the two in terms of computing power configuration, γ sim is the collaboration triggering threshold, C(p i ,p j ) represents the impact of collaborative operations on individual positions, λ c is the collaboration intensity, u bottleneck (p i ) indicates that in individual p i The bottleneck utilization rate in the current state, u th is the bottleneck threshold, λ b is the obstacle avoidance strength, represents the local gradient in the computing power dimension d, w priority (t) is the time-variable weight of the task priority, δ p Indicates the priority trigger threshold, λ p is the priority intensity, f priority (p i ,t) is the gradient function of the resource demand of the high priority task; The dynamic constraint mechanism handles three scenarios in multi-dimensional computing power allocation: resource collaboration, bottleneck avoidance, and priority allocation; S34. For the crowned porcupine individuals in each layer of the multi-layer nested model, the individuals are combined and iterated to cooperate, avoid obstacles or allocate priority according to the dynamic constraint output results. The multi-layer feedback mechanism transmits messages between dynamic analysis models according to the iteration completion and convergence of each layer, so that the optimization process of the five computing power dimensions is coordinated with each other and gradually approaches the overall optimum: in, represents the computing power configuration of crested porcupine individual i in the k-th layer model at the t-th generation iteration, α (k) and β (k) are the adjustment parameters of the k-th layer model in the local gradient and global guidance direction, D is the total number of multi-objective dimensions, Ω(P (k) ) represents the same layer crested porcupine individual group P (k) Global information aggregation operation; S35. After the multi-objective dynamic optimization convergence conditions are met, the shared server computing power dynamic analysis model is output, and the dynamic analysis model and the corresponding crested porcupine individual iteration trajectory are stored in a distributed storage medium: in, Indicates that the crowned porcupine individuals in the k-th layer model converge at the final generation T end The optimal computing power configuration when n k is the number of crested porcupine individuals in this layer, Λ constraints is the set of dynamic constraints formed during the optimization process, F final Represents the final set of multi-objective function values.

4. A computing power analysis method for a shared server according to claim 1, characterized in that: The S4 specifically comprises the following steps: S41. Collect the shared server operation status data set D in real time through the monitoring module REALTIME Dynamic analysis model M of shared server computing power CPO-dynamic Matching, calculate the mapping position Φ(x,t) of the real-time running status in the model; S42. Dynamic analysis model of computing power based on matching M CPO-dynamic Evaluate the utilization of the real-time shared server operation status data and calculate the utilization of each computing power dimension: Among them, U d (t) represents the utilization rate of computing power dimension d at time t, R d (t) is the real-time computing power requirement, C d is the total computing capacity of this dimension; S43. Based on the real-time computing power utilization U(t) and computing power dynamic analysis model M CPO-dynamic Analyze the load distribution of each computing power dimension of the current shared server, define the load distribution function and generate the shared server load heat map based on the load distribution function: Among them, L d (t) represents the overall load distribution of computing power dimension d, n d is the total number of tasks distributed on this dimension, is the computing power requirement of the i-th task at time t; S44. Detect the bottleneck of computing resources by combining real-time data and load distribution: Among them, B d (t) indicates whether there is a bottleneck in the computing power dimension d at time t, η d is the bottleneck threshold; S45. Comprehensive current computing power utilization set U(t) and load distribution function L d (t) and bottleneck function B(t) to generate a complete computing power status evaluation result: E(t)={U(t),L d (t),B(t)}。 5. The computing power analysis method for a shared server according to claim 1, characterized in that: The S5 specifically includes the following steps: S51. According to the computing power status evaluation result E(t) combined with the crown porcupine optimization algorithm, define the resource scheduling optimization objective function: F opt (x)=∑ d∈{CPU,GPU,MEM,IO,NET} w d (t)·[a d ·U d (t)+β d ·L d (t)+γ d ·(1-B d (t))]; Among them, x represents the current allocation status of server resources, w d (t) is the dynamic weight of computing power dimension d, α d ,β d ,γ d They are the adjustment coefficients for real-time utilization, load distribution, and smoothing bottlenecks, respectively; S52. Optimizing the objective function F based on resource scheduling opt (x) Construct the initial population P of the crown porcupine optimization algorithm. Each individual p in the initial population P i Represents a resource scheduling scheme whose initial position is generated by random distribution in the computing power allocation space that satisfies the constraints: in, Indicates the resource allocation ratio of computing power dimension d in individual i; S53. Optimize the objective function F according to resource scheduling opt (x) Using the rules of the crested porcupine optimization algorithm, iteratively update the position of each individual: in, and represents the resource allocation status of individual i in the tth and t+1th generations, respectively, and x best is the individual's best historical resource allocation state, x g is the global optimal resource allocation state of the population, To optimize the gradient of the objective function and guide the individual movement direction, γ1, γ2, and γ3 are adjustment coefficients, which control the weights of historical guidance, global guidance, and gradient optimization respectively; S54. According to the optimized resource scheduling plan Dynamically generate task allocation strategies for shared servers: Task reallocation: Combining real-time utilization U d (t) and the dynamic weights w of each dimension in the optimization objective function d (t), prioritize resources to tasks on high-weight dimensions; Virtual machine configuration adjustment: dynamically adjust the virtual machine's computing power quota based on the optimization results Requirements: Load balancing strategy update: According to the load distribution L in the optimization objective function d (t) and bottleneck detection result B d (t), dynamically adjust task allocation to low-load nodes to optimize overall load balancing; S55. Store the optimized resource scheduling scheme and the corresponding optimization objective function value as a shared server computing power optimization strategy set S opt .

6. A computing power analysis method for a shared server according to claim 1, characterized in that: The S7 specifically comprises the following steps: S71. Extract the historical operation status data and computing power status evaluation results of the shared server, and construct a historical data set containing multi-dimensional computing power indicators. The historical data set includes the real-time utilization, load distribution and bottleneck status of each computing power dimension at different time points; S72. By analyzing the historical data set and the current evaluation results, set the computing power demand change trend analysis target, which combines the utilization rate change speed, load distribution stability and bottleneck state persistence of each computing power dimension to measure the direction and intensity of the future computing power demand change trend; S73. Under the guidance of the computing power demand change trend analysis goal, a population of the crested porcupine optimization algorithm is constructed, each population individual represents a possible future computing power demand distribution state, and the initial population is randomly generated according to the statistical law of historical data and current computing power distribution; S74. Utilize the iterative mechanism of the crown porcupine optimization algorithm to optimize the distribution of computing power requirements of individuals in the population based on historical data and real-time evaluation results, so that it gradually approaches the global optimal trend. Through the collaboration, obstacle avoidance and priority guidance mechanisms among individuals, dynamically simulate the growth or decline trend of future requirements of different computing power dimensions. S75. Generate the prediction results of the future computing power demand change trend based on the optimized population. The prediction results include the utilization change trend of each computing power dimension in the future time period, the load distribution possibility and the potential risk of bottleneck state.

7. A computing power analysis system for a shared server, used to implement the computing power analysis for a shared server as described in any one of claims 1 to 6, characterized in that: The system comprises: The monitoring module is used to collect the shared server operation status data in real time. The monitoring module interacts directly with the server hardware resources and obtains the computing power status in real time through the data interface; A data processing module is used to pre-process the operating status data collected by the monitoring module; Dynamic analysis modeling module, used to build a dynamic analysis model of shared server computing power based on the crown porcupine optimization algorithm; The computing power status assessment module is used to apply the dynamic analysis model to the real-time operation status data to generate the computing power status assessment results; Resource scheduling optimization module, which is used to generate resource scheduling optimization strategies for shared servers based on computing power status evaluation results and combined with the crown porcupine optimization algorithm; The computing power demand prediction module is used to predict the future computing power demand change trend of the shared server based on the historical operation status data and the current computing power status evaluation results, combined with the crown porcupine optimization algorithm; Storage module, used to store data and model results generated by each module; The control module is used to coordinate the operation of the above modules, receive real-time data input from the monitoring module, and transmit the data to the data processing module and the dynamic analysis modeling module.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the computing power analysis method for a shared server as described in any one of claims 1 to 6 are implemented.

Images