Server hardware resource intelligent optimization method and system based on adaptive neural network

Abstract

The invention relates to the technical field of server hardware resource optimization, in particular to a server hardware resource intelligent optimization method and system based on an adaptive neural network, and the method comprises the steps: obtaining hardware resource information and system performance indexes of a server; constructing topological resource mapping based on the hardware resource information and the system performance indexes; executing a group theory resource scheduling algorithm according to the topological resource mapping; performing matrix resource state prediction based on a result of the group theory resource scheduling algorithm; executing number theory resource allocation optimization according to a result of matrix resource state prediction; updating the weight of the adaptive neural network based on a number theory resource allocation optimization result; according to the method, an optimized server hardware resource configuration scheme is output, and by introducing topological resource mapping, the method can comprehensively and accurately represent a complex relationship among multi-dimensional resources, so that a solid foundation is provided for subsequent optimization decision making; the application of the group theory resource scheduling algorithm enables the resource scheduling operation to be more flexible and systematized.

Description

Technical Field

[0001] The present invention relates to the technical field of server hardware resource optimization, and more specifically, to a server hardware resource intelligent optimization method and system based on an adaptive neural network. Background Art

[0002] With the rapid development of cloud computing and big data technology, the scale and complexity of data centers are increasing, and the efficient management and optimization of server hardware resources has become an increasingly severe challenge. Traditional server resource management methods mainly rely on static configuration and simple threshold trigger mechanisms, which are difficult to cope with the complex and changing load conditions of modern data centers.

[0003] In recent years, machine learning technology has been widely used in the field of server resource optimization. Some researchers have proposed resource allocation methods based on prediction models, which predict future resource requirements by analyzing historical data. However, these methods often only consider a single or a few resource dimensions, making it difficult to fully grasp the complex relationship between server resources. Other researchers have tried to use deep learning models to optimize resource allocation, but these models usually require a large amount of training data and computing resources, and face challenges in real-time and interpretability in practical applications.

[0004] In addition, most existing resource optimization methods adopt isolated optimization strategies and lack systematic consideration of the mutual impact between resources. For example, simply optimizing CPU utilization may lead to bottlenecks in memory or network bandwidth. At the same time, these methods usually find it difficult to balance multiple goals such as resource utilization, energy efficiency, and service quality, and often lose sight of one while focusing on another in practical applications.

[0005] Another common problem is that existing methods are difficult to adapt to rapidly changing load environments. The workload of data centers may fluctuate dramatically due to factors such as time, user behavior, and special events. Static optimization strategies are difficult to adjust in time to adapt to these changes. Although some adaptive methods have been proposed, they are often slow to respond and have difficulty making fast and accurate decisions in complex and changing environments.

[0006] In the face of these challenges, there is an urgent need for an intelligent server resource optimization method that can comprehensively consider multi-dimensional resources, balance multiple optimization goals, and adapt to dynamic environments. In response to this demand, the present invention proposes an intelligent optimization method and system for server hardware resources based on an adaptive neural network. Summary of the invention

[0007] The present invention proposes a server hardware resource intelligent optimization method and system based on an adaptive neural network, aiming to solve the following main problems existing in the prior art: incomplete resource representation, single optimization target, insufficient adaptability, opaque decision-making process, etc.

[0008] The present invention provides a server hardware resource intelligent optimization method based on an adaptive neural network, comprising:

[0009] The acquisition steps include:

[0010] Obtain server hardware resource information and system performance indicators;

[0011] Processing steps include:

[0012] Based on the hardware resource information and system performance indicators, construct a topology resource mapping;

[0013] Executing a group theory resource scheduling algorithm according to the topology resource mapping;

[0014] Based on the result of the group theory resource scheduling algorithm, matrix resource status prediction is performed;

[0015] According to the result of the matrix resource status prediction, performing number theory resource allocation optimization;

[0016] Based on the result of the number theory resource allocation optimization, updating the weights of the adaptive neural network;

[0017] Output steps include:

[0018] Output the optimized server hardware resource configuration plan.

[0019] Preferably, the obtaining of the hardware resource information and system performance indicators of the server specifically includes:

[0020] Get CPU usage, memory usage, disk I / O rate and network traffic;

[0021] Get application response time, system throughput, and server energy consumption.

[0022] Preferably, the constructing of topology resource mapping specifically includes:

[0023] Mapping the hardware resource information to a high-dimensional topological space;

[0024] Use spherical harmonics as topological basis functions;

[0025] Compute topological mapping functions.

[0026] Preferably, the execution of the group theory resource scheduling algorithm specifically includes:

[0027] Define resource scheduling groups;

[0028] Based on the resource scheduling group, transforming the topology mapping;

[0029] Generate an optimized topology map.

[0030] Preferably, the matrix resource status prediction specifically includes:

[0031] Construct resource status matrix;

[0032] Define the state transfer function based on the Toeplitz matrix;

[0033] Compute the predicted resource status matrix.

[0034] Preferably, the performing number theory resource allocation optimization specifically includes:

[0035] Optimize resource allocation using Riemann zeta function;

[0036] Calculate the optimal resource allocation plan.

[0037] Preferably, the updating of the weights of the adaptive neural network specifically includes:

[0038] Define a loss function based on von Neumann entropy;

[0039] Calculate the gradient of the neural network weights;

[0040] Update the neural network weights.

[0041] As a preference, it also includes:

[0042] Based on the optimized server hardware resource configuration scheme, the hardware resource allocation of the server is dynamically adjusted.

[0043] As a preference, it also includes:

[0044] The acquisition step, the processing step and the output step are periodically repeated to achieve continuous optimization.

[0045] The server hardware resource intelligent optimization system based on adaptive neural network for executing the method comprises:

[0046] An acquisition module is used to obtain the hardware resource information and system performance indicators of the server;

[0047] Processing module, including:

[0048] A topology mapping submodule, used to construct a topology resource mapping based on the hardware resource information and system performance indicators;

[0049] A group theory scheduling submodule, used for executing a group theory resource scheduling algorithm according to the topology resource mapping;

[0050] A matrix prediction submodule, used for predicting the matrix resource status based on the result of the group theory resource scheduling algorithm;

[0051] A number theory optimization submodule, used for performing number theory resource allocation optimization according to the result of the matrix resource state prediction;

[0052] A neural network updating submodule, used for updating the weights of the adaptive neural network based on the result of the number theory resource allocation optimization;

[0053] The output module is used to output the optimized server hardware resource configuration plan.

[0054] The beneficial effects of the present invention are mainly reflected in the following aspects:

[0055] First, by introducing topological resource mapping, this method can fully and accurately represent the complex relationship between multi-dimensional resources, providing a solid foundation for subsequent optimization decisions. The application of group theory resource scheduling algorithm makes resource scheduling operations more flexible and systematic, and can better cope with complex scheduling scenarios.

[0056] Secondly, the combination of matrix resource state prediction and number theory resource allocation optimization not only improves the accuracy of resource state prediction, but also realizes multi-objective collaborative optimization. This method can improve resource utilization while taking into account energy efficiency and service quality, achieving a balanced improvement of multiple key indicators.

[0057] More importantly, the adaptive neural network weight update mechanism of the present invention gives the system the ability to continuously learn and self-optimize. This enables the method to quickly adapt to load changes and maintain efficient resource allocation in a dynamic environment. At the same time, by introducing the quantum concept of loss function, the method considers deeper system state information during the optimization process, improving the accuracy and robustness of decision-making.

[0058] In addition, the method of the present invention achieves the superposition and complementarity of technical effects through the organic combination of various innovative steps. For example, the comprehensive resource representation provided by the topological mapping provides a richer information basis for group theory scheduling, and the continuous optimization of the adaptive neural network can continuously improve the quality of the topological mapping. This cyclic optimization mechanism enables the system performance to be continuously improved over time.

[0059] In summary, the server hardware resource intelligent optimization method and system based on adaptive neural network provided by the present invention not only solves many problems existing in the prior art, but also achieves significant improvements in resource utilization, energy efficiency, service quality, etc. This innovative solution provides a new technical path for the intelligent management of data centers, which has important theoretical value and broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1The figure is a flow chart of the method of the present invention.

[0061] Figure 2 This is a flow chart of obtaining hardware resource information and system performance indicators of the present invention.

[0062] Figure 3 A flow chart is constructed for the topological resource mapping of the present invention.

[0063] Figure 4 This is a flow chart of the group theory resource scheduling algorithm of the present invention.

[0064] Figure 5 This is a flow chart of matrix resource status prediction of the present invention.

[0065] Figure 6 This is a flowchart of number theory resource allocation optimization of the present invention.

[0066] Figure 7 This is a flow chart of updating the weights of the adaptive neural network of the present invention.

[0067] Figure 8 It is a logic block diagram of the system of the present invention. DETAILED DESCRIPTION

[0068] Please refer to Figure 1-8 The present invention provides a server hardware resource intelligent optimization method based on an adaptive neural network and a system thereof. The method realizes efficient optimization of server hardware resources.

[0069] First, the method of the present invention includes an acquisition step, a processing step and an output step. In the acquisition step, the system acquires the hardware resource information and system performance indicators of the server. This information is the basis for the subsequent optimization process.

[0070] Specifically, in one embodiment of the present invention, the hardware resource information obtained includes CPU usage, memory occupancy, disk I / O rate and network traffic. These parameters fully reflect the hardware resource usage of the server. Preferably, the sampling frequency of CPU usage can be set to once per second to capture changes in CPU load in a timely manner. Memory occupancy can be sampled every 5 seconds because memory usage usually changes relatively slowly. Disk I / O rate and network traffic can be sampled every 10 seconds to balance data accuracy and system overhead.

[0071] At the same time, this method also obtains system performance indicators, including application response time, system throughput, and server energy consumption. These indicators are crucial for evaluating the overall performance of the system. For example, a threshold value, such as 200 milliseconds, can be set for application response time. If it exceeds this value, it is considered that the system performance needs to be optimized. System throughput can be measured by the number of requests processed per second, and server energy consumption can be obtained through real-time power consumption monitoring.

[0072] In the processing step, the method of the present invention first constructs a topological resource mapping based on the acquired hardware resource information and system performance indicators. This step is one of the core innovations of the present invention. The topological resource mapping uses a high-dimensional topological space to represent complex resource relationships. This representation method can better capture the interdependence and influence between resources.

[0073] In server hardware resource optimization, it is crucial to understand and analyze the relationships between various hardware resources (such as CPU, memory, disk I / O, network traffic, etc.). Traditional linear models are difficult to capture these complex relationships, while topology provides a tool to describe these complex relationships.

[0074] The process of building a topology resource map can be described by the following mathematical formula:

[0075] ,

[0076] in, is the topological mapping function, For hardware resource space, is the target topological space, For specific resource allocation, is the weight of the ith basis function, is the ith topological basis function.

[0077] In this scenario, suppose we need to analyze the relationship between CPU usage and memory usage. By using spherical harmonics as basis functions, we can map these variables into a high-dimensional space. For example, if Represents the combination of CPU usage and memory usage at a specific point in time. The nonlinear relationship between these two variables can be captured. This mapping helps identify potential bottlenecks or redundancies, thereby guiding the reallocation of resources.

[0078] Suppose on a web server, it is found that there is a certain periodic correlation between CPU usage and memory usage. By applying the above formula, we can better understand this correlation and adjust the load balancing strategy based on the analysis results to improve the overall performance.

[0079] The present invention preferably uses spherical harmonics as topological basis functions, which are defined as follows:

[0080] ,

[0081] in, is the associated Legendre polynomial, and is an integer parameter that satisfies . is a spherical harmonic function, is the polar angle, is the azimuth, is a complex exponential function (where is an imaginary unit).

[0082] The benefit of using spherical harmonics is that they form an orthogonal basis on the sphere and can efficiently represent any function on the sphere, which is very useful for describing the multidimensional characteristics of server resources.

[0083] Through this topological mapping, the hardware resource status of the server can be mapped into a high-dimensional space, thereby better capturing the complex relationship between resources. For example, the correlation between CPU usage and memory occupancy can be better represented in this high-dimensional space.

[0084] Resource scheduling is one of the key issues in server management. Group theory provides a new perspective to understand and optimize the resource scheduling process.

[0085] Next, the method executes the group-theoretic resource scheduling algorithm according to the constructed topological resource mapping. The introduction of the group-theoretic resource scheduling algorithm is another innovation of the present invention. It regards the resource scheduling operation as an element on the group, so that resource scheduling can be performed more flexibly and systematically.

[0086] The group theory resource scheduling algorithm can be expressed by the following mathematical formula:

[0087] ,

[0088] in, For the resource scheduling group, For specific scheduling operations, Indicates group action. is the group action map, Mapping of topological resources.

[0089] In practical applications, if you need to balance CPU and memory usage among multiple server nodes, you can consider this problem as a rotation operation in two-dimensional space. For example, when the CPU usage on a certain node reaches a peak, you can calculate the appropriate rotation matrix To reallocate tasks to other nodes to avoid overload.

[0090] Consider a distributed database system where data is distributed across multiple nodes. By applying a group-theoretic resource scheduling algorithm, query tasks can be dynamically redistributed based on the current load of each node, ensuring that the workload of all nodes is as even as possible, improving the system's response speed and stability.

[0091] In practical applications, the special orthogonal group SO(n) can be selected as the resource scheduling group, which is defined as:

[0092] ,

[0093] here, represents a set of real matrices of order n, is the identity matrix, represents the determinant of matrix A, is a set of real matrices of order n.

[0094] The advantage of using the SO(n) group is that it can represent rotations in n-dimensional space, which is very similar to the operations used to schedule resources. For example, when you need to balance the load between CPU and memory, you can think of it as a rotation operation in two-dimensional space.

[0095] By executing the group theory resource scheduling algorithm, an optimized topology mapping can be obtained This optimized mapping reflects the new state after resource scheduling and provides a basis for subsequent optimization steps.

[0096] In summary, the method of the present invention realizes efficient representation and flexible scheduling of server hardware resources by innovatively introducing the concepts of topology and group theory. Compared with traditional resource optimization methods, this method can better capture the complex relationship between resources, thereby achieving more accurate resource optimization. In practical applications, this method can significantly improve the resource utilization of the server, reduce resource waste, and improve the overall performance of the system. After executing the group theory resource scheduling algorithm, the method of the present invention further performs matrix resource state prediction based on its results. This step makes full use of the advantages of matrix theory and can more accurately predict future resource states.

[0097] In order to achieve more accurate resource prediction and scheduling, it is very important to use historical data for status prediction. Specifically, matrix resource status prediction can be expressed by the following mathematical formula:

[0098] ,

[0099] in, is the resource status matrix at time t, is the state transfer function, This is the optimized topology mapping.

[0100] In this method, by constructing the Toeplitz matrix , past data points can be effectively used to predict future resource requirements. For example, when predicting CPU usage in the next 5 minutes, considering the data from the previous 10 time points can help capture potential trends and patterns.

[0101] Suppose you want to predict the disk I / O rate of a server in the future. By collecting I / O rate data every 10 seconds over the past period of time, constructing the corresponding Toeplitz matrix, and using the above formula to make predictions. This allows you to take measures in advance, such as increasing the cache size or adjusting the read and write strategy, to cope with the upcoming peak load.

[0102] In a preferred embodiment of the present invention, is defined as a special function based on the Toeplitz matrix:

[0103] ,

[0104] here, Is The generated Toeplitz matrix. The choice of the Toeplitz matrix is ​​based on its special structural characteristics, which can effectively capture the time correlation of resource status. For example, when predicting CPU usage, the usage at the current moment is usually closely related to the usage at the previous moments, and this relationship can be well represented by the Toeplitz matrix.

[0105] Preferably, the method of the present invention can select to use the data of the most recent 10 time points when constructing the Toeplitz matrix. The selection of this value is based on experience, which can capture sufficient historical information without introducing too much computational overhead.

[0106] Resource allocation must not only consider current needs but also foresee possible future changes. Number theory provides a unique perspective for optimization.

[0107] Next, the method performs number-theoretic resource allocation optimization based on the results of the matrix resource state prediction. This step innovatively introduces concepts from number theory, especially the Riemann zeta function, to optimize resource allocation. Number-theoretic resource allocation optimization can be expressed by the following mathematical formula:

[0108]

[0109] in, For optimized resource allocation, is the Riemann zeta function, is a complex parameter, For the matrix The vector of diagonal elements of . Allocation vector for resources to be optimized, belonging to integer vector space , Represents the L2 norm (Euclidean norm) of a vector.

[0110] Assume that there is a data center with three server nodes, and the number of CPU cores, memory capacity, and disk / O bandwidth need to be allocated. The current forecast results show that the future resource requirements are: Number of cores: 5; Memory capacity: 16GB; Disk I / O bandwidth: 100MB / s;

[0111] These requirements can be represented as vectors .

[0112] Now, we need to find an integer vector , so that As close as possible to [5,16,100]. Assume that ,but At this point, we need to find an integer vector ,For example , so that In this way, the resource allocation plan can be dynamically adjusted based on the current resource forecast and historical data to ensure the efficient operation of the system.

[0113] Choose the right parameters when resources are limited The Riemann zeta function can be made to exhibit different characteristics, thereby simulating different resource allocation strategies. For example, when When it is close to 1, the function value increases rapidly, indicating that resources are very tight; when resources are sufficient, a larger value, so that the function value approaches 1.

[0114] In a cloud computing environment, the number of CPU cores and memory capacity allocated to each virtual machine are dynamically adjusted according to the number and type of current user requests. By adjusting the parameters of the Riemann zeta function , can flexibly respond to changes in resource demand in different time periods, maximizing resource utilization while ensuring service quality.

[0115] The effectiveness of all the above algorithms depends on high-quality data input. The data acquisition step includes real-time monitoring of various server indicators (such as CPU usage, memory occupancy, etc.) and regular sampling of these data. These data are not only used for current resource optimization decisions, but also provide a basis for future predictions. Through sophisticated data processing and analysis, we can ensure more accurate resource scheduling, reduce unnecessary resource waste, and improve overall system efficiency.

[0116] The Riemann zeta function is defined as follows:

[0117] ,

[0118] is the Riemann zeta function, is a complex parameter of the form ,in and They are The real and imaginary parts, Re plural The real part of , is a positive integer used for each term in the summation, For the specific expression of each term, we can write or .

[0119] The method of the present invention chooses to use the Riemann zeta function because of its unique mathematical properties. In particular, when When taking different values, exhibit different characteristics, which can be used to simulate different resource allocation strategies. When it approaches 1, the function value increases rapidly, which can be used to indicate resource shortage; When it is large, the function value approaches 1, which can indicate sufficient resources.

[0120] Preferably, in one embodiment of the present invention, The value range of can be set between 1.1 and 2. This range is selected based on a large number of experimental results and can achieve good optimization results in most cases.

[0121] After completing the number theory resource allocation optimization, the method of the present invention updates the weights of the adaptive neural network based on the optimization results. This step is the core of the entire method, which ensures that the optimization process can continuously adapt to the changing environment. The adaptive neural network weight update can be expressed by the following mathematical formula:

[0122] ,

[0123] in, is the neural network weight at time t, is the learning rate, is the loss function, is the gradient of the loss function with respect to the weight, is the loss function.

[0124] Assume that a data center needs to allocate resources such as CPU cores, memory capacity, and disk I / O bandwidth. Current forecast results show that future resource requirements are: CPU cores: 5; memory capacity: 16GB; disk I / O bandwidth: 100MB / s;

[0125] These requirements can be represented as vectors .

[0126] Assume that the weight matrix of the current neural network is for:

[0127]

[0128] Need to be based on the loss function To update these weights. Assume that the current loss function gradient for:

[0129]

[0130] The learning rate is , then the new weight matrix The calculation is as follows:

[0131]

[0132] Substituting the values:

[0133]

[0134] Calculation results:

[0135]

[0136] In this process, As the optimized resource allocation vector, it represents the best allocation of each resource, and the quantity guides the adjustment direction of the weight. By continuously updating the weight, the neural network can eventually predict and optimize resource allocation more accurately.

[0137] An innovation of the present invention is that it introduces the von Neumann entropy in quantum mechanics to define the loss function:

[0138] ,

[0139] here, is the density matrix, is the Hamiltonian, is the inverse temperature parameter, is the regularization parameter. is the von Neumann entropy, which is used to quantify the uncertainty of the system, For the matrix and The Frobenius norm square between is used to measure the difference between two matrices. The Hamiltonian The exponential form of is used to describe the energy distribution of the system.

[0140] Adaptive neural networks (ANNs) are able to continuously adjust their weights based on feedback in a dynamic environment to optimize resource allocation and improve system performance. This ability makes ANNs very suitable for intelligent optimization of server hardware resources.

[0141] By using the von Neumann entropy in quantum mechanics as part of the loss function, the complexity and uncertainty of the system can be better captured. This helps to achieve more accurate optimization in the resource scheduling process. For example, in a highly loaded data center, the status of resources such as CPU and memory are monitored in real time, and the weights of the neural network are adjusted according to these statuses to achieve optimal resource allocation.

[0142] Suppose a data center needs to dynamically adjust the resource allocation of each virtual machine based on the current workload. By applying the above formula, the weights of the neural network can be continuously updated according to the actual resource usage, so as to more accurately predict future resource needs and make corresponding adjustments. For example, if the CPU usage of a virtual machine suddenly increases, the system can increase the CPU allocation of the virtual machine by adjusting the weight to prevent overload.

[0143] Preferably, in the method of the present invention, the learning rate It can be set to 0.01, which usually strikes a good balance between convergence speed and stability. Can be set to 1, which provides sufficient quantum effects in most cases. Regularization parameter It can be set to 0.001 to prevent overfitting.

[0144] It is worth noting that the method of the present invention also includes dynamically adjusting the server hardware resource allocation based on the optimized server hardware resource configuration scheme. This step ensures that the optimization results can be actually applied to the server system. For example, if the optimization results show that a certain application requires more CPU resources, the system will increase the CPU time slice allocated to the application accordingly.

[0145] In addition, the method of the present invention further comprises periodically repeating the acquisition step, the processing step and the output step to achieve continuous optimization. Preferably, this period can be set to 5 minutes. The selection of this time interval is based on the consideration of the speed of server load change, which can respond to load changes in a timely manner without introducing additional system overhead due to overly frequent optimization.

[0146] In general, the method of the present invention realizes the intelligent optimization of server hardware resources by innovatively combining the concepts of topology, group theory, matrix theory, number theory and quantum mechanics. This method can not only accurately capture the complex relationship between resources, but also adapt to dynamically changing environments, thereby achieving more efficient and accurate resource allocation. In practical applications, this method can significantly improve the resource utilization of the server, reduce resource waste, improve the overall performance of the system, and provide strong technical support for the efficient operation of the data center. The present invention also provides a server hardware resource intelligent optimization system based on an adaptive neural network. The system is designed to implement the various steps of the aforementioned method, thereby achieving efficient optimization of server hardware resources.

[0147] Specifically, the system includes an acquisition module 1, a processing module 2 and an output module 3. These modules work together to complete the intelligent optimization process of server hardware resources.

[0148] The acquisition module 1 is responsible for acquiring the hardware resource information and system performance indicators of the server. In a preferred embodiment of the present invention, the acquisition module 1 may include multiple sensor submodules, such as a CPU usage sensor submodule 11, a memory occupancy sensor submodule 12, a disk I / O rate sensor submodule 13, and a network traffic sensor submodule 14. These sensor submodules are responsible for collecting corresponding hardware resource information. At the same time, the acquisition module 1 may also include a performance indicator acquisition submodule 15 for collecting system performance indicators such as application response time, system throughput, and server energy consumption.

[0149] The processing module 2 is the core of the system, which includes multiple submodules, each of which corresponds to a key step in the method. Specifically, the processing module 2 includes a topological mapping submodule 21, a group theory scheduling submodule 22, a matrix prediction submodule 23, a number theory optimization submodule 24 and a neural network update submodule 25.

[0150] The topology mapping submodule 21 is used to construct a topology resource mapping based on the hardware resource information and system performance indicators provided by the acquisition module 1. In one embodiment of the present invention, the topology mapping submodule 21 can use a high-performance GPU for parallel computing to accelerate the construction process of the topology mapping. Preferably, an NVIDIA Tesla V100 GPU can be used, which has 5120 CUDA cores and can efficiently process large-scale matrix operations.

[0151] The group theory scheduling submodule 22 is responsible for executing the group theory resource scheduling algorithm according to the topology resource mapping. In a preferred embodiment of the present invention, the group theory scheduling submodule 22 can adopt a distributed computing architecture to utilize the computing power of multiple servers to process group theory operations in parallel. For example, the Apache Spark distributed computing framework can be used, which can effectively process large-scale data sets and complex mathematical operations.

[0152] The matrix prediction submodule 23 performs matrix resource status prediction based on the result of the group theory resource scheduling algorithm. In one embodiment of the present invention, the matrix prediction submodule 23 can use a dedicated matrix operation accelerator, such as Google's Tensor Processing Unit (TPU). TPU is specially optimized for large-scale matrix operations and can significantly improve the speed and efficiency of matrix prediction.

[0153] The number theory optimization submodule 24 performs number theory resource allocation optimization according to the result of matrix resource state prediction. In a preferred embodiment of the present invention, the number theory optimization submodule 24 can use a quantum annealing algorithm to solve the optimization problem. For example, a D-Wave quantum annealing machine can be used, which can find an approximate optimal solution to a complex optimization problem in a relatively short time.

[0154] The neural network update submodule 25 is responsible for updating the weights of the adaptive neural network based on the results of the number theory resource allocation optimization. In one embodiment of the present invention, the neural network update submodule 25 can use FPGA (Field-Programmable Gate Array) to accelerate the training and update process of the neural network. For example, IntelStratix 10 FPGA can be used, which can provide up to 10 TFLOPS of floating-point computing performance, which is very suitable for the rapid update of the neural network.

[0155] The output module 3 is used to output the optimized server hardware resource configuration plan. In a preferred embodiment of the present invention, the output module 3 may include a visualization submodule 31 and an execution submodule 32. The visualization submodule 31 is responsible for intuitively displaying the optimization results to the system administrator in the form of charts or animations, while the execution submodule 32 is responsible for automatically applying the optimization plan to the server system.

[0156] It is worth noting that the system can also include a feedback module 4. The feedback module 4 is responsible for collecting the optimized system performance data and feeding the data back to the acquisition module 1 to form a closed-loop optimization system. This design enables the system to continuously learn and improve, thereby achieving continuous performance optimization.

[0157] Preferably, the system can adopt a modular design, and each module communicates through a standardized interface. This design makes the system have good scalability and maintainability. For example, if a more advanced optimization algorithm appears in the future, only the corresponding submodule needs to be replaced without changing the structure of the entire system.

[0158] In general, the server hardware resource intelligent optimization system based on adaptive neural network provided by the present invention realizes efficient and intelligent optimization of server hardware resources through the collaborative work of various functional modules. The system can not only accurately capture the complex relationship between resources, but also adapt to the dynamically changing environment, thereby achieving more efficient and accurate resource allocation. In practical applications, this system can significantly improve the resource utilization of data centers, reduce energy waste, and improve overall operational efficiency, providing strong technical support for the intelligent management of modern data centers.

[0159] In order to verify the superiority of the server hardware resource intelligent optimization method based on adaptive neural network and the system thereof, a set of tests was designed, including an embodiment and two comparative examples. These experiments were conducted in a large data center environment, involving 1,000 servers, and continuously observed for 30 days.

[0160] Embodiment 1 adopts the method of the present invention, including topological resource mapping, group theory resource scheduling, matrix resource state prediction, number theory resource allocation optimization and adaptive neural network weight update.

[0161] Comparative Example 1 adopts a traditional threshold-based resource allocation method, and triggers resource reallocation when resource utilization exceeds a preset threshold.

[0162] Comparative Example 2 adopts a simple machine learning method and uses a linear regression model to predict resource requirements and allocate them.

[0163] The following five metrics were selected to evaluate the performance of each method:

[0164] 1. Average resource utilization: Use system monitoring tools to collect data once a minute and take the average value for 30 days.

[0165] 2. Energy efficiency: Calculate the number of tasks processed per kilowatt-hour of electricity.

[0166] 3. Service quality (SLA achievement rate): Statistics show the proportion of requests that meet the service level agreement within 30 days.

[0167] 4. System response time: records the average response time of all requests.

[0168] 5. Optimization algorithm execution time: records the average time taken for each resource optimization decision.

[0169] The test results are shown in Table 1:

[0170] Table 1. Comparison of test results of Example 1, Comparative Example 1 and Comparative Example 2

[0171] index Example 1 Comparative Example 1 Comparative Example 2 Average resource utilization 78.5% 62.3% 70.1% Energy efficiency (number of tasks / kWh) 1250 950 1050 SLA Achievement Rate 99.7% 95.2% 97.8% System response time (ms) 85 150 120 Optimization algorithm execution time (s) 2.5 0.5 1.2

[0172] It can be seen from the test results in Table 1 that the method of the present invention (Example 1) is significantly superior to the traditional method (Comparative Example 1) and the simple machine learning method (Comparative Example 2) in most indicators.

[0173] In terms of resource utilization, the method of the present invention reaches 78.5%, which is 16.2 and 8.4 percentage points higher than Comparative Example 1 and Comparative Example 2, respectively. This shows that the method of the present invention can more effectively utilize server resources and reduce resource waste. High resource utilization means that the data center can use less hardware to process the same number of tasks, thereby reducing hardware costs and energy consumption.

[0174] The energy efficiency index shows that the method of the present invention can process 1250 tasks per kilowatt-hour, which is 31.6% and 19% higher than that of comparative examples 1 and 2, respectively. This result directly reflects the advantages of the method in energy conservation and emission reduction, which is conducive to reducing the operating costs and environmental impact of data centers.

[0175] In terms of service quality, the method of the present invention achieved an SLA achievement rate of 99.7%, which is much higher than the other two methods. This means that the method can provide users with more stable and reliable services, which helps to improve user satisfaction and corporate reputation.

[0176] System response time is an important indicator for measuring user experience. The method of the present invention controls the average response time to 85 milliseconds, which is 65 milliseconds and 35 milliseconds faster than Comparative Example 1 and Comparative Example 2, respectively. This significant performance improvement can directly improve user experience, which is particularly important for applications with high real-time requirements.

[0177] It is worth noting that the method of the present invention is slightly inferior to the other two methods in terms of optimization algorithm execution time. This is because the method involves more complex calculation processes, including topological mapping, group theory scheduling, etc. However, considering the significant performance improvement brought by the method, this slightly increased calculation time is completely acceptable. Moreover, with the continuous improvement of hardware performance and further optimization of the algorithm, this gap is expected to be further narrowed.

[0178] In general, this set of experimental results strongly proves the superiority of the method of the present invention. It not only performs well in terms of resource utilization and energy efficiency, but also significantly improves the system response speed while ensuring high service quality. This comprehensive performance improvement fully reflects the innovative value of the present invention that integrates advanced concepts such as topology, group theory, matrix theory, number theory and quantum mechanics.

[0179] It is particularly worth mentioning that the method of the present invention performs well in dynamic and complex data center environments, thanks to its core mechanism of adaptive neural networks. Traditional methods often have difficulty coping with rapidly changing load conditions, while this method can better adapt to environmental changes and make smarter resource allocation decisions through continuous learning and optimization.

[0180] This set of experimental results not only verifies the theoretical innovation of this invention, but also proves its great potential in practical applications. With this method, data center operators can significantly improve resource utilization efficiency, reduce energy consumption, and provide users with better services. This is undoubtedly of great practical significance for the data center industry, which is currently facing challenges such as resource constraints, high energy consumption, and growing user demand.

[0181] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modification, replacement, and improvement made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A server hardware resource intelligent optimization method based on an adaptive neural network, characterized in that: include: The acquisition steps include: Obtain server hardware resource information and system performance indicators; Processing steps include: Based on the hardware resource information and system performance indicators, construct a topology resource mapping; Executing a group theory resource scheduling algorithm according to the topology resource mapping; Based on the result of the group theory resource scheduling algorithm, matrix resource status prediction is performed; According to the result of the matrix resource status prediction, performing number theory resource allocation optimization; Based on the result of the number theory resource allocation optimization, updating the weights of the adaptive neural network; Output steps include: Output the optimized server hardware resource configuration plan.

2. The method according to claim 1, characterized in that The obtaining of the server's hardware resource information and system performance indicators specifically includes: Get CPU usage, memory usage, disk I / O rate and network traffic; Get application response time, system throughput, and server energy consumption.

3. The method according to claim 1, characterized in that The construction of topology resource mapping specifically includes: Mapping the hardware resource information to a high-dimensional topological space; Use spherical harmonics as topological basis functions; Compute topological mapping functions.

4. The method according to claim 1, characterized in that: The execution group theory resource scheduling algorithm specifically includes: Define resource scheduling groups; Based on the resource scheduling group, transforming the topology mapping; Generate an optimized topology map.

5. The method according to claim 1, characterized in that The matrix resource status prediction specifically includes: Construct resource status matrix; Define the state transfer function based on the Toeplitz matrix; Compute the predicted resource status matrix.

6. The method according to claim 1, characterized in that The execution of number theory resource allocation optimization specifically includes: Optimize resource allocation using Riemann zeta function; Calculate the optimal resource allocation plan.

7. The method according to claim 1, characterized in that The updating of the weights of the adaptive neural network specifically includes: Define a loss function based on von Neumann entropy; Calculate the gradient of the neural network weights; Update the neural network weights.

8. The method according to claim 1, characterized in that Also includes: Based on the optimized server hardware resource configuration scheme, the hardware resource allocation of the server is dynamically adjusted.

9. The method according to claim 1, characterized in that: Also includes: The acquisition step, the processing step and the output step are periodically repeated to achieve continuous optimization.

10. A server hardware resource intelligent optimization system based on an adaptive neural network for executing the method according to any one of claims 1 to 9, characterized in that: include: The acquisition module is used to obtain the hardware resource information and system performance indicators of the server; Processing module, including: A topology mapping submodule, used to construct a topology resource mapping based on the hardware resource information and system performance indicators; A group theory scheduling submodule, used for executing a group theory resource scheduling algorithm according to the topology resource mapping; A matrix prediction submodule, used for predicting the matrix resource status based on the result of the group theory resource scheduling algorithm; A number theory optimization submodule, used for performing number theory resource allocation optimization according to the result of the matrix resource state prediction; A neural network update submodule, used for updating the weights of the adaptive neural network based on the result of the number theory resource allocation optimization; The output module is used to output the optimized server hardware resource configuration plan.

Images