Computing power resource partitioning method and device based on hypergraph clustering, equipment and medium

Abstract

The invention provides a computing power resource partitioning method and device based on hypergraph clustering, equipment and a medium. The method comprises the following steps: constructing a computing power resource hypergraph in a computing power network resource side scene; determining an equipment degree matrix and a hyperedge degree matrix of the computing power resource hypergraph, and converting the computing power resource hypergraph into an equipment Laplacian matrix required by hypergraph clustering by using the equipment degree matrix and the hyperedge degree matrix; carrying out eigendecomposition on the equipment Laplacian matrix to obtain an equipment eigenvector; and in combination with the graph cutting target function, performing hypergraph clustering processing on the computing power resource hypergraph based on the feature vector of the device Laplacian matrix to obtain a target resource partitioning result corresponding to the computing power resource hypergraph. According to the method, the problems of low resource utilization rate, high scheduling complexity, poor resource dynamics, lack of effective partitioning strategies and the like existing in the existing computing power resource partitioning can be obviously improved.

Description

Technical Field

[0001] The present invention relates to the technical field of computing power resource partitioning, and in particular, to a computing power resource partitioning method, device, equipment and medium based on hypergraph clustering. Background Art

[0002] With the rapid development of emerging technologies such as big data, artificial intelligence, cloud computing, and the Internet of Things, computing power resources (computing capabilities) have become the key infrastructure of modern information society. Current computing tasks are gradually migrating from traditional local servers to cloud and distributed architectures, and technologies such as cloud computing and edge computing are constantly popularized, driving the further expansion of computing power resources. Computing power resources not only include the server clusters in data centers, but also cover distributed computing units such as edge nodes and intelligent terminals.

[0003] To meet different computing requirements, the types and scales of computing power resources are becoming increasingly rich. Cloud computing platforms provide on-demand computing power resources through virtualization technology, while edge computing reduces latency and improves real-time processing capabilities by processing data locally. These technical architectures help to greatly improve computing efficiency and flexibility, and at the same time achieve on-demand allocation and dynamic expansion of resources.

[0004] Although the infrastructure of computing power resources has made great progress in terms of scale, flexibility, and diversity, there are still the following deficiencies: (1) Low resource utilization: In traditional computing power resource management, due to uneven task scheduling and unbalanced computing loads, some nodes are often overloaded while other nodes are idle. This unreasonable resource allocation not only causes waste of resources, but also limits the efficiency of the overall system.

[0005] (2) High scheduling complexity: As the distributed characteristics of computing power resources become more and more obvious, it becomes more and more complex to effectively schedule and allocate various types of computing power resources. Especially when the resource pool contains heterogeneous computing units (such as CPUs, GPUs, FPGAs, etc.), the scheduling algorithm needs to consider the type of computing unit, the requirements of computing tasks, and real-time constraints at the same time.

[0006] (3) Poor resource dynamics: The demand for computing power resources is highly dynamic and often fluctuates with changes in business needs. Traditional static resource allocation strategies are difficult to adapt to this dynamic change, unable to quickly respond when the computing power demand surges, and also difficult to release redundant resources in a timely manner when the demand decreases, resulting in resource waste or performance bottlenecks.

[0007] (4) Lack of effective partitioning strategy: In scenarios where multiple tenants or multiple services share the same computing power resources, how to effectively partition and isolate resources to ensure that different users or tasks can use computing power resources fairly and reasonably is still an urgent problem to be solved. Current partitioning methods often rely on coarse-grained strategies and cannot fully utilize the fine-grained characteristics of resources, resulting in insufficient accuracy in scheduling and management. Summary of the Invention

[0008] In view of this, the purpose of the present invention is to provide a computing power resource partitioning method, device, equipment and medium based on hypergraph clustering, which can significantly improve problems existing in the existing computing power resource partitioning, such as low resource utilization rate, high scheduling complexity, poor resource dynamics, and lack of effective partitioning strategies.

[0009] In the first aspect, the present invention provides a computing power resource partitioning method based on hypergraph clustering, including: Construct a computing power resource hypergraph in the scenario of the computing power network resource side, where the computing power resource hypergraph includes a computing power resource set and the hyperedge relationship between multiple server devices included therein; Determine the device degree matrix and hyperedge degree matrix of the computing power resource hypergraph, and use the device degree matrix and hyperedge degree matrix to convert the computing power resource hypergraph into a device Laplacian matrix required for hypergraph clustering; Perform eigen-decomposition on the device Laplacian matrix to obtain device eigenvectors, and the row vectors of the device eigenvectors are the eigenvectors of the device Laplacian matrix; Combined with the cut graph objective function, based on the eigenvectors of the device Laplacian matrix, perform hypergraph clustering processing on the computing power resource hypergraph to obtain the target resource partitioning result corresponding to the computing power resource hypergraph, and each computing power resource partition in the target resource partitioning result includes multiple server devices.

[0010] In an implementation manner, determining the device degree matrix of the computing power resource hypergraph includes: Perform the following operations on any two server devices in the computing power resource hypergraph: If there is no hyperedge relationship between the two server devices, determine that the device weight between the two server devices is 0; If there is a hyperedge relationship between the two server devices, determine the device weight between the two server devices according to the attribute information of the two server devices; According to the device weights between the server devices, determine the device degree matrix corresponding to the computing power resource hypergraph, and the degree corresponding to each server device in the device degree matrix is: the sum value of all device weights with hyperedge relationships with the server device.

[0011] In an implementation manner, determining the hyperedge degree matrix of the computing power resource hypergraph includes: Perform the following operations on any hyperedge relationship in the computing power resource hypergraph: Based on the communication rate between any two server devices connected in the hyperedge relationship and the number of all server devices connected by the hyperedge relationship, determine the hyperedge weight corresponding to the hyperedge relationship. According to the hyperedge weight corresponding to each hyperedge relationship, determine the hyperedge degree matrix corresponding to the computing power resource hypergraph, where the hyperedge degree matrix includes the degree corresponding to each hyperedge relationship.

[0012] In one implementation, converting the computing power resource hypergraph into a device Laplacian matrix required for hypergraph spectral clustering by using the device degree matrix and the hyperedge degree matrix includes: Perform an inverse square root operation on the device degree matrix and an inverse operation on the hyperedge degree matrix. Construct a hyperedge weight matrix according to the hyperedge weight corresponding to each hyperedge relationship in the computing power resource hypergraph. Based on the device degree matrix after the inverse square root operation, the hyperedge degree matrix after the inverse operation, the hyperedge weight matrix, and the adjacency matrix, convert the computing power resource hypergraph into a device Laplacian matrix required for hypergraph spectral clustering.

[0013] In one implementation, performing eigenvalue decomposition on the device Laplacian matrix to obtain device eigenvectors includes: Determine the hyperedge Laplacian matrix. Based on the device Laplacian matrix and the hyperedge Laplacian matrix, perform singular value decomposition to determine the device eigenvectors.

[0014] In one implementation, combining the cut graph objective function and based on the eigenvectors of the device Laplacian matrix, performing hypergraph spectral clustering processing on the computing power resource hypergraph to obtain the target resource partition result corresponding to the computing power resource hypergraph includes: Perform hypergraph spectral clustering processing on the computing power resource hypergraph to obtain the current resource partition result. Combining the cut graph objective function and based on the eigenvectors of the device Laplacian matrix, determine the function value corresponding to the current resource partition result. With the goal of maximizing the function value, continue to perform hypergraph spectral clustering processing on the computing power resource hypergraph based on the eigenvectors of the device Laplacian matrix to obtain the target resource partition result corresponding to the computing power resource hypergraph.

[0015] In one implementation, combining the cut graph objective function and based on the eigenvectors of the device Laplacian matrix, determining the function value corresponding to the current resource partition result includes: For any computing power resource partition in the current resource partition result, determine the indicator vector corresponding to the computing power resource partition according to whether each server device included in the computing power resource hypergraph belongs to the computing power resource partition. Construct an indication vector matrix based on the indication vectors corresponding to each computing power resource partition; Normalize the indication vector matrix, and determine the function value corresponding to the current resource partition result according to the normalized indication vector and the eigenvector of the device Laplacian matrix.

[0016] In a second aspect, the present invention also provides a computing power resource partitioning device based on hypergraph spectral clustering, including: A hypergraph construction module, configured to construct a computing power resource hypergraph in the scenario of the computing power network resource side, where the computing power resource hypergraph includes a computing power resource set and the hyperedge relationship between multiple server devices included therein; A matrix conversion module, configured to determine the device degree matrix and the hyperedge degree matrix of the computing power resource hypergraph, and convert the computing power resource hypergraph into a device Laplacian matrix required for hypergraph spectral clustering by using the device degree matrix and the hyperedge degree matrix; An eigen - decomposition module, configured to perform eigen - decomposition on the device Laplacian matrix to obtain device eigenvectors, and the row vectors of the device eigenvectors are the eigenvectors of the device Laplacian matrix; A hypergraph spectral clustering module, configured to combine the cut graph objective function, and perform hypergraph spectral clustering processing on the computing power resource hypergraph based on the eigenvectors of the device Laplacian matrix to obtain a target resource partition result corresponding to the computing power resource hypergraph, and each computing power resource partition in the target resource partition result includes multiple server devices.

[0017] In a third aspect, the present invention also provides an electronic device, including a processor and a memory, where the memory stores computer - executable instructions that can be executed by the processor, and the processor executes the computer - executable instructions to implement the method according to any one of the first aspect.

[0018] In a fourth aspect, the present invention also provides a computer - readable storage medium, where the computer - readable storage medium stores computer - executable instructions, and when the computer - executable instructions are called and executed by a processor, the computer - executable instructions cause the processor to implement the method according to any one of the first aspect.

[0019] A method, device, equipment and medium for partitioning computing power resources based on hypergraph spectral clustering provided by the present invention first constructs a hypergraph of computing power resources in the scenario of the computing power network resource side. The hypergraph of computing power resources includes a set of computing power resources and the hyperedge relationships between multiple server devices included therein. Then, the device degree matrix and the hyperedge degree matrix of the hypergraph of computing power resources are determined, and the hypergraph of computing power resources is converted into a device Laplacian matrix required for hypergraph spectral clustering by using the device degree matrix and the hyperedge degree matrix. Next, the device Laplacian matrix is subjected to eigenvalue decomposition to obtain device eigenvectors, and the row vectors of the device eigenvectors are the eigenvectors of the device Laplacian matrix. Finally, in combination with the cut graph objective function, based on the eigenvectors of the device Laplacian matrix, hypergraph spectral clustering processing is performed on the hypergraph of computing power resources to obtain a target resource partitioning result corresponding to the hypergraph of computing power resources. Each computing power resource partition in the target resource partitioning result includes multiple server devices. The above method uses a hypergraph to describe the complex relationships between server devices in the scenario of the computing power network resource side, converts it into a device Laplacian matrix required for hypergraph spectral clustering by using the device degree matrix and the hyperedge degree matrix, performs eigenvalue decomposition on it to reduce the complexity of clustering, and on this basis, performs hypergraph spectral clustering processing in combination with the cut graph objective function to obtain the target resource partitioning result. The present invention can significantly improve the problems existing in the existing computing power resource partitioning, such as low resource utilization rate, high scheduling complexity, poor resource dynamics, and lack of effective partitioning strategies.

[0020] Other features and advantages of the present invention will be described in the following specification, and, in part, will be obvious from the specification, or will be understood by implementing the present invention. The objectives and other advantages of the present invention are achieved and obtained by the structures specifically pointed out in the specification, the claims, and the drawings.

[0021] To make the above objectives, features, and advantages of the present invention more obvious and understandable, the following specific preferred embodiments are given, and in conjunction with the accompanying drawings, the following detailed description is provided. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for use in the description of the specific embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0023] Figure 1 It is a schematic flowchart of a method for partitioning computing power resources based on hypergraph spectral clustering provided by an embodiment of the present invention; Figure 2 It is a schematic overall flowchart of a method for partitioning computing power resources based on hypergraph spectral clustering provided by an embodiment of the present invention; Figure 3 Schematic structural diagram of a computing power resource partitioning device based on hypergraph clustering provided by an embodiment of the present invention; Figure 4 Schematic structural diagram of an electronic device provided by an embodiment of the present invention. Detailed implementation manners

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0025] Currently, the prior art has the following deficiencies: (1) Low resource utilization rate: Due to uneven task scheduling and unbalanced computing loads, some nodes are overloaded, while other nodes are idle, resulting in serious resource waste and limited overall system efficiency. (2) High scheduling complexity: With the increase in distributed and heterogeneous resources (such as CPUs, GPUs, FPGAs, etc.), scheduling algorithms need to consider multiple factors simultaneously, and the matching of computing tasks and resource types becomes increasingly complex, increasing the management difficulty. (3) Insufficient dynamic response: Traditional static resource allocation strategies cannot effectively handle the dynamic changes in computing power resource requirements, and resource scheduling cannot be quickly adjusted, resulting in performance bottlenecks during peak hours and resource waste during off-peak hours. (4) Insufficient resource partitioning strategy: The prior art lacks effective fine-grained partitioning methods, and resource partitioning and isolation in a multi-tenant environment are difficult to achieve precisely and fairly, affecting the overall efficiency of the system and the rationality of resource allocation.

[0026] Based on this, the embodiments of the present invention provide a computing power resource partitioning method, device, equipment, and medium based on hypergraph clustering, which can significantly improve problems such as low resource utilization rate, high scheduling complexity, poor resource dynamics, and lack of effective partitioning strategies existing in the existing computing power resource partitioning.

[0027] Among them, a hypergraph is a data structure that extends the edges of a traditional graph to hyperedges. In a hypergraph, a hyperedge can connect multiple vertices, thus being able to better describe multivariate relationships. Compared with the traditional graph structure, a hypergraph can represent complex relationships more comprehensively, especially showing strong modeling capabilities in multi-dimensional data analysis and processing scenarios.

[0028] In the partitioning and management of computing power resources, the hypergraph clustering algorithm has the following significant advantages: (1) Ability to adapt to high-dimensional complex relationships: During the allocation process of computing power resources, multiple resource units (such as computing nodes, storage units, network bandwidth, etc.) are often interrelated in a complex manner. Hypergraph clustering can well express and process such complex multivariate relationships. Hyperedges can not only connect multiple nodes but also describe the multi-dimensional interaction information between nodes, thus fully considering the cooperation of each resource unit during resource partitioning. (2) More refined resource partitioning: The hypergraph clustering algorithm can perform partitioning based on the correlation between resource units, generating a more detailed and accurate partitioning result than traditional graph clustering algorithms. In computing power resource scheduling, this refined partitioning helps improve resource utilization and reduce computing bottlenecks and resource waste caused by improper resource partitioning. (3) Flexibility of dynamic partitioning: The hypergraph structure can flexibly adapt to the dynamic changes in resource requirements. When the demand for certain computing power resources surges, the hypergraph clustering algorithm can quickly re-partition the resources and ensure that the partitioning of computing power resources can respond to changing needs at any time by dynamically adjusting the hyperedge structure. (4) Support for heterogeneous resource management: Hypergraph clustering can not only partition and manage homogeneous resources (such as CPU clusters) but also effectively partition heterogeneous resources (such as CPUs, GPUs, FPGAs, etc.). Through the hypergraph structure, the joint management and scheduling of heterogeneous resources can be achieved to ensure the coordinated operation of various resources. (5) Reduction of computational complexity: Traditional computing power resource scheduling algorithms often rely on classical clustering algorithms in graph theory, and these algorithms have a high computational complexity. Especially when dealing with large-scale distributed computing power resources, the amount of computation is too large to be scheduled in real time. Hypergraph clustering can significantly reduce the computational complexity while maintaining an efficient clustering effect by utilizing the multi-node correlation of hyperedges, improving the scalability and practicality of the algorithm.

[0029] For the convenience of understanding this embodiment, first, a method for partitioning computing power resources based on hypergraph spectral clustering disclosed in the embodiments of the present invention will be introduced in detail. Refer to Figure 1 the flowchart of a method for partitioning computing power resources based on hypergraph spectral clustering shown in Step S102: Construct a computing power resource hypergraph in the scenario of the computing power network resource side.

[0030] Among them, the computing power resource hypergraph includes the hyperedge relationship between the computing power resource set and multiple server devices included therein. The computing power resource set includes multiple server devices and their corresponding attribute information. The attribute information may include server hardware information, network interface information, server performance metrics, etc. The server hardware information includes the type of CPU (Central Processing Unit), the number of cores, frequency, memory capacity, storage type (such as HDD mechanical hard disk or SSD solid state drive) and its capacity, etc. The network interface information includes the number and bandwidth of network interfaces (such as Gigabit Ethernet, 10GbE, etc.). The server performance metrics include the computing power evaluated through benchmark tests (expressed in floating-point operations per second or instructions per second), the I / O performance of storage devices (such as read and write speeds and latency), and the energy efficiency ratio (such as computing power per watt), etc.

[0031] In one example, a preliminary clustering process can be performed on the computing power resource set, and the server devices in the same clustering cluster are connected by hyperedges. In another example, the engineer can also determine the hyperedge relationship between each server device according to his own experience to construct the computing power resource hypergraph.

[0032] Step S104, determine the device degree matrix and hyperedge degree matrix of the computing power resource hypergraph, and use the device degree matrix and hyperedge degree matrix to convert the computing power resource hypergraph into the device Laplacian matrix required for hypergraph spectral clustering.

[0033] Among them, the device degree matrix is also the matrix of the degree of each server device in the computing power resource hypergraph, and the hyperedge degree matrix is also the matrix of the degree of each hyperedge relationship in the computing power resource hypergraph. In one example, determine the device weights between any two server devices in the computing power resource hypergraph to construct the device degree matrix; determine the hyperedge matrix of any hyperedge relationship in the computing power resource hypergraph to construct the hyperedge degree matrix and hyperedge weight matrix; determine the device Laplacian matrix based on the device degree matrix, hyperedge matrix, hyperedge weight matrix and adjacency matrix.

[0034] Step S106, perform eigen-decomposition on the device Laplacian matrix to obtain device eigenvectors.

[0035] Among them, the row vectors of the device eigenvectors are the eigenvectors of the device Laplacian matrix, and the row vectors of the node eigenvectors correspond to the low-dimensional embedding representation of a server device. In one example, first define a specific matrix according to the device degree matrix, hyperedge matrix, hyperedge weight matrix and adjacency matrix, define the hyperedge Laplacian matrix according to this matrix, and perform singular value decomposition on this matrix using the device Laplacian matrix and hyperedge Laplacian matrix to obtain the device eigenvectors.

[0036] Step S108: Based on the eigenvectors of the device Laplacian matrix and in combination with the graph-cut objective function, perform hypergraph spectral clustering on the computing power resource hypergraph to obtain the corresponding target resource partition result of the computing power resource hypergraph.

[0037] Among them, each computing power resource partition in the target resource partition result contains multiple server devices. In one example, with the minimization of the function value of the graph-cut objective function as the goal, hypergraph spectral clustering can be performed on the computing power resource hypergraph based on the eigenvectors of the device Laplacian matrix to obtain the final target resource partition result.

[0038] The computing power resource partitioning method based on hypergraph spectral clustering provided by the embodiments of the present invention uses a hypergraph to describe the complex relationships between server devices in the scenario of the computing power network resource side. By using the device degree matrix and the hyperedge degree matrix, it is converted into the device Laplacian matrix required for hypergraph spectral clustering, and its eigen-decomposition is performed to reduce the complexity of clustering. On this basis, hypergraph spectral clustering is performed in combination with the graph-cut objective function to obtain the target resource partition result. The present invention can significantly improve the problems existing in the existing computing power resource partitioning, such as low resource utilization rate, high scheduling complexity, poor resource dynamics, and lack of effective partitioning strategies.

[0039] For ease of understanding, the embodiments of the present invention provide a specific implementation manner of a computing power resource partitioning method based on hypergraph spectral clustering. Refer to Figure 2 the overall flowchart of a computing power resource partitioning method based on hypergraph spectral clustering shown in Step S202: Construct a computing power resource hypergraph in the scenario of the computing power network resource side.

[0040] In a computing power network, server devices have multiple key attributes, which together determine their performance, availability, and applicability. In terms of hardware, the CPU type, number of cores, frequency, and memory capacity of the server directly affect the computing performance. At the same time, the storage type (such as HDD or SSD) and its capacity are also crucial for the data processing speed. In addition, the number and bandwidth of network interfaces (such as Gigabit Ethernet, 10GbE, etc.) affect the data transmission rate and network throughput. Performance metrics include computing power evaluated through benchmark tests (expressed in floating-point operations per second or instructions per second), I / O performance of storage devices (such as read / write speed and latency), and energy efficiency ratio (such as computing power per watt).

[0041] A hypergraph is an extended concept in graph theory, aiming to represent complex relationships more flexibly. Different from the traditional graph structure that only contains edges and nodes, a hypergraph allows edges (called hyperedges) to connect two or more nodes, so it is suitable for representing multivariate relationships and multi-dimensional data.

[0042] In the scenario of the computing power network resource side, the embodiment of the present invention defines the computing power resource hypergraph as , where represents the set of computing power resources, represents the hyperedge relationship. Suppose there are n server devices in the set of computing power resources , the th server device's attribute information is expressed as . Among them, represents the storage capacity of the th server device, represents the GPU computing power of the th server device, represents whether the th server device is available, represents the number of GPUs of the th server device.

[0043] Before explaining steps S204 to S208, the embodiment of the present invention first explains hypergraph spectral clustering. Hypergraph spectral clustering is a clustering method that applies spectral clustering technology to hypergraphs. It combines the advantages of hypergraphs and spectral graph theory to solve the clustering problem of complex multi-dimensional data. Compared with traditional graph clustering, hypergraph spectral clustering can better capture the high-order relationships between multiple objects and is suitable for data sets that require analyzing multivariate relationships, such as social networks, recommendation systems, computing power resource scheduling, etc.

[0044] Similar to the Laplacian matrix of an ordinary graph, the Laplacian matrix of a hypergraph is the core tool of hypergraph spectral clustering. Since the hyperedges of a hypergraph can connect multiple nodes, constructing the Laplacian matrix of a hypergraph requires special processing. The common method is to transform the hypergraph into a "weighted bipartite graph" and then construct the corresponding Laplacian matrix based on this bipartite graph. On this basis, please continue to refer to the subsequent explanations of steps S204 to S208.

[0045] Step S204, determine the device degree matrix of the computing power resource hypergraph. Specifically, refer to the following steps 1.1 to 1.2: Step 1.1, perform the following operations for any two server devices in the computing power resource hypergraph: If there is no hyperedge relationship between the two server devices, determine the device weight between the two server devices to be 0; if there is a hyperedge relationship between the two server devices, determine the device weight between the two server devices according to the attribute information of the two server devices.

[0046] In the computing power resource hypergraph, for two server devices and that have no hyperedge relationship, = 0; For two server devices with a hyperedge relationship and , define as the device weight between the two server devices. Generally, the Gaussian kernel function RBF is used to construct the device weight : ; Among them, is the device weight, , are two server devices with a hyperedge relationship, is the standard deviation.

[0047] Step 1.2: Determine the device degree matrix corresponding to the computing power resource hypergraph according to the device weights between the server devices. The device degree matrix contains the degrees corresponding to each server device. The degree corresponding to a server device is the sum of the weights of all devices with a hyperedge relationship with the server device.

[0048] Specifically, for any server device in the computing power resource hypergraph, its degree is defined as the sum of the weights of all hyperedge relationships connected to it, that is: ; Using the definition of the degree of each server device, the embodiment of the present invention can obtain a degree device degree matrix , which is a diagonal matrix with values only on the main diagonal, corresponding to the degree of the i-th point in the i-th row, and is defined as follows: .

[0049] Step S206: Determine the hyperedge degree matrix of the computing power resource hypergraph. Specifically, refer to the following steps 2.1 to 2.2: Step 2.1: For any hyperedge relationship in the computing power resource hypergraph, perform the following operations: Based on the communication rate between any two server devices connected in the hyperedge relationship and the number of all server devices connected by the hyperedge relationship, determine the hyperedge weight corresponding to the hyperedge relationship.

[0050] In practical applications, the server devices connected in the hyperedge relationship have a relatively close geographical distribution and similar attribute relationships, and it can be considered that the server devices in the same hyperedge relationship have similar communication conditions. Therefore, the hyperedge weight of the hyperedge relationship is expressed as the communication rate of the server devices connected by the hyperedge relationship, and the specific numerical setting is the average value of the communication rates between all server devices connected by the hyperedge relationship: ; Among them, Represents the server device and the server device The communication rate between them Represents the hyperedge relationship The total number of server devices connected That is, the server device and the server device Are different server devices

[0051] Step 2.2: Determine the hyperedge degree matrix corresponding to the computing power resource hypergraph according to the hyperedge weight corresponding to each hyperedge relationship. The hyperedge degree matrix contains the degree corresponding to each hyperedge relationship. Determine the hyperedge degree matrix For the specific process, please refer to the aforementioned step 1.2, and this embodiment of the present invention will not elaborate further

[0052] Step S208: Convert the computing power resource hypergraph into the device Laplacian matrix required for hypergraph spectral clustering by using the device degree matrix and the hyperedge degree matrix. Specifically, it includes the following steps 3.1 to step 3.3 Step 3.1: Perform an inverse square root operation on the device degree matrix and an inverse operation on the hyperedge degree matrix. Among them, the device degree matrix after the inverse square root operation is denoted as The hyperedge degree matrix after the inverse operation is denoted as

[0053] Step 3.2: Construct a hyperedge weight matrix according to the hyperedge weight corresponding to each hyperedge relationship in the computing power resource hypergraph. The hyperedge weight matrix is a diagonal matrix, and its diagonal elements are the hyperedge weights corresponding to each hyperedge relationship

[0054] Step 3.3: Based on the device degree matrix after the inverse square root operation, the hyperedge degree matrix after the inverse operation, the hyperedge weight matrix, and the adjacency matrix, convert the computing power resource hypergraph into the device Laplacian matrix required for hypergraph spectral clustering

[0055] The device Laplacian matrix of the computing power resource hypergraph captures the complex multi-dimensional relationships between server devices, thus providing a basis for clustering. Define the Laplacian matrix as

[0056] where D is the degree matrix of the computing power resource hypergraph represents the adjacency matrix. On this basis, this embodiment of the present invention provides the expression of the device Laplacian matrix as ; where is the device degree matrix is the hyperedge degree matrix​ is the adjacency matrix, and is the hyperedge weight matrix.

[0057] In practical applications, by calculating the eigenvalues and eigenvectors of the Laplacian matrix, the original high-dimensional data can be mapped to a low-dimensional space. The data points in this low-dimensional space retain the most important geometric and topological structures in the original data. Usually, only the eigenvectors corresponding to the first few smallest eigenvalues are selected for clustering.

[0058] Step S210, determine the hyperedge Laplacian matrix, and perform singular value decomposition based on the device Laplacian matrix and the hyperedge Laplacian matrix to determine the device eigenvectors.

[0059] Accelerate hypergraph clustering: Store the Laplacian matrix requires O( ) of storage space, and the cost of solving its corresponding eigenproblem is O( ), so for large graphs, it is computationally infeasible. For ease of description, a specific matrix is defined as follows: ; To distinguish it from the device Laplacian matrix , a new hyperedge Laplacian matrix is defined as: ; Perform singular value decomposition (SVD) on matrices and : ; ; The equation holds, where I represents the identity matrix. It is easy to infer that: ; ; where the diagonal elements of matrix are and 's corresponding eigenvalues. In other words, and share trailing eigenvalues. Given the hyperedge eigenvector , the corresponding device eigenvector can be calculated by the following formula: ; So far, a method for calculating the eigenvectors of a device has been provided based on the hypergraph Laplacian and SVD transformation, with relatively low time and space complexity. The time complexity of solving the eigenproblem of the device Laplacian matrix is O( ), and the saved space complexity is O( ), while the corresponding cost of the hyperedge Laplacian matrix is O( ), and the space complexity is O( ). Compared with directly solving the eigenproblem, the embodiments of the present invention propose to solve the eigenproblem. If m < n (the embodiments of the present invention will discuss the case where the number m of hyperedge relationships in the sampling part is relatively large), it will significantly reduce the computational cost from O( ) to O( ). Since the square root and inverse operations of the diagonal matrix T are simple and Z is sparse, the calculation is very efficient. Another advantage is that the conversion between and is accurate without any information loss.

[0060] Step S212: Combining the cut graph objective function, based on the eigenvectors of the device Laplacian matrix, perform hypergraph spectral clustering on the computing power resource hypergraph to obtain the target resource partitioning result corresponding to the computing power resource hypergraph.

[0061] Among them, the derivation process of the cut graph objective function is as follows: Using RatioCut to cut the computing power resource hypergraph, this cut graph method not only considers minimizing , but also simultaneously considers maximizing the number of server devices included in each computing power resource partition. The expression of the cut graph objective function is as follows: ; The embodiments of the present invention introduce the indicator vector For any vector , it is an n-dimensional vector (n is the number of samples). The embodiments of the present invention define as: ; That is, for any server device , if the server device does not belong to the computing power resource partition , then the indicator vector of the server device relative to the computing power resource partition is 0; if the server device belongs to the computing power resource partition , then the server device with respect to the computing power resource partition indicator vector is .

[0062] Then for there is: ; wherein, , represents the server device, represents the server device , the device weight between, represents the server device with respect to the computing power resource partition indicator vector, represents the server device with respect to the computing power resource partition indicator vector.

[0063] Based on this, the expression of the final cut graph objective function is as follows: ; wherein, is the trace of the indicator matrix . That is to say, the cut graph objective function of the embodiment of the present invention is actually to minimize . Note that , then the cut graph optimization objective of the embodiment of the present invention is: .

[0064] Based on the above cut graph objective function, the embodiment of the present invention provides a specific process of hypergraph spectral clustering, including the following steps 4.1 to step 4.3: Step 4.1, perform hypergraph spectral clustering processing on the computing power resource hypergraph to obtain the current resource partition result.

[0065] Step 4.2, in combination with the cut graph objective function, determine the function value corresponding to the current resource partition result based on the eigenvector of the device Laplacian matrix. Including the following (i) to (iii): (i) For any computing power resource partition in the current resource partition result, determine the indicator vector corresponding to the computing power resource partition according to whether each server device included in the computing power resource hypergraph belongs to the computing power resource partition. For the specific calculation formula, reference can be made to the foregoing , and the embodiments of the present invention will not elaborate herein.

[0066] (2) Construct an indicator vector matrix based on the indicator vectors corresponding to each computing power resource partition, and normalize the indicator vector matrix.

[0067] In practical applications, it is noted that each indicator vector in the indicator vector matrix is n-dimensional, and the value of each variable in the vector is 0 or , so there are kinds of values. If there are computing power resource partitions, there will be indicator vectors, and there are kinds of indicator vector matrices . Therefore, finding the indicator vector matrix that meets the above optimization objective is an NP-hard problem.

[0068] Pay attention to observing each optimization sub-objective in , where is an orthonormal basis, is a symmetric matrix. At this time, the maximum value of is the maximum eigenvalue of, and the minimum value is the minimum eigenvalue of. In spectral clustering, the objective of the embodiment of the present invention is to find the minimum eigenvalue and obtain the corresponding eigenvector. At this time, the corresponding bipartite cut graph effect is the best. That is to say, the embodiment of the present invention uses the idea of dimensionality reduction here to approximately solve this NP-hard problem.

[0069] By finding the k smallest eigenvalues of, the corresponding eigenvectors can be obtained. These eigenvectors form a -dimensional matrix, which is the indicator vector matrix . Generally, it is necessary to normalize the indicator vector matrix row by row, that is: .

[0070] (3) Determine the function value corresponding to the current resource partition result according to the normalized indicator vector and the eigenvector of the device Laplacian matrix. In specific implementation, substitute the normalized indicator vector and the eigenvector of the device Laplacian matrix (i.e., the device eigenvector ) into the aforementioned cut graph objective function, and the function value corresponding to the current resource partition result can be obtained.

[0071] Step 4.3, aiming at maximizing the function value, continue to perform hypergraph spectral clustering on the computing power resource hypergraph based on the eigenvectors of the device Laplacian matrix to obtain the target resource partition result corresponding to the computing power resource hypergraph.

[0072] In summary, the computing power resource partitioning method based on hypergraph spectral clustering provided by the embodiments of the present invention models the complex multi-dimensional relationship between computing power resources and tasks through a hypergraph structure, provides a comprehensive description of resource characteristics, and uses an advanced hypergraph clustering algorithm to achieve fine-grained partitioning of resources, significantly improving resource utilization. Therefore, the embodiments of the present invention at least have the following characteristics: 1. Improve resource utilization: Through the hypergraph spectral clustering algorithm, the embodiments of the present invention can accurately capture the complex multi-dimensional relationship between computing power resource units, realizing fine-grained resource partitioning and task scheduling. Compared with traditional methods, hypergraphs can reflect multi-dimensional associations through hyperedges (connecting multiple nodes), thereby optimizing resource allocation, reducing the problems of node overload and resource idleness, and significantly improving resource utilization.

[0073] 2. Simplify scheduling complexity: The hypergraph clustering algorithm simplifies the complex associations between multiple resource units into a hyperedge structure, making the scheduling algorithm more flexible and efficient when facing heterogeneous resources. It can automatically cluster and match according to the characteristics of resources and task requirements, thereby reducing complex calculations and resource matching operations and lowering the complexity of scheduling.

[0074] 3. Enhance dynamic response ability: The embodiments of the present invention provide a dynamic resource partitioning method based on a hypergraph structure, which can quickly respond to changes in computing power resource requirements. Through dynamic adjustment of the hypergraph, the embodiments of the present invention can re-partition resource clusters according to real-time changes in computing power requirements, ensuring that the system can quickly adjust in the face of fluctuating resource requirements, avoiding performance bottlenecks and resource waste.

[0075] 4. Fine-grained resource partitioning and isolation: The hypergraph clustering algorithm can perform fine-grained resource partitioning in a multi-tenant or multi-task environment. Through the clustering mechanism of the hypergraph, the relevance and synergy of different resource units are fully considered, realizing more accurate and fair resource allocation. Compared with existing coarse-grained partitioning strategies, the embodiments of the present invention are more flexible and adaptable, effectively improving the contradiction between resource isolation and sharing.

[0076] Based on the foregoing embodiments, the embodiments of the present invention provide a specific implementation manner of a computing power resource partitioning device based on hypergraph spectral clustering. Refer to Figure 3 the structural schematic diagram of a computing power resource partitioning device based on hypergraph spectral clustering shown in The hypergraph construction module 302 is used to construct a hypergraph of computing power resources in the scenario of the computing power network resource side. The hypergraph of computing power resources includes a set of computing power resources and the hyperedge relationships between multiple server devices included therein; The matrix conversion module 304 is used to determine the device degree matrix and the hyperedge degree matrix of the hypergraph of computing power resources, and convert the hypergraph of computing power resources into a device Laplacian matrix required for hypergraph spectral clustering by using the device degree matrix and the hyperedge degree matrix; The eigen - decomposition module 306 is used to perform eigen - decomposition on the device Laplacian matrix to obtain device eigen - vectors. The row vectors of the device eigen - vectors are the eigen - vectors of the device Laplacian matrix; The hypergraph spectral clustering module 308 is used to combine the graph - cut objective function, and based on the eigen - vectors of the device Laplacian matrix, perform hypergraph spectral clustering processing on the hypergraph of computing power resources to obtain the target resource partition result corresponding to the hypergraph of computing power resources. Each computing power resource partition in the target resource partition result contains multiple server devices.

[0077] The computing power resource partitioning device based on hypergraph spectral clustering provided by the embodiments of the present invention uses a hypergraph to describe the complex relationships between server devices in the scenario of the computing power network resource side, converts it into a device Laplacian matrix required for hypergraph spectral clustering by using the device degree matrix and the hyperedge degree matrix, performs eigen - decomposition on it to reduce the complexity of clustering, and on this basis, combines the graph - cut objective function to perform hypergraph spectral clustering processing to obtain the target resource partition result. The present invention can significantly improve the problems existing in the existing computing power resource partitioning, such as low resource utilization rate, high scheduling complexity, poor resource dynamics, and lack of effective partitioning strategies.

[0078] In one implementation manner, the matrix conversion module 304 is specifically used for: Perform the following operations on any two server devices in the hypergraph of computing power resources: If there is no hyperedge relationship between the two server devices, determine that the device weight between the two server devices is 0; If there is a hyperedge relationship between the two server devices, determine the device weight between the two server devices according to the attribute information of the two server devices; According to the device weights between server devices, determine the device degree matrix corresponding to the hypergraph of computing power resources. The device degree matrix contains the degree corresponding to each server device. The degree corresponding to a server device is: the sum value of all device weights of the devices having hyperedge relationships with the server device.

[0079] In one implementation manner, the matrix conversion module 304 is specifically used for: Perform the following operations on any one hyperedge relationship in the hypergraph of computing power resources: Based on the communication rate between any two server devices connected by the hyperedge relationship, and the number of all server devices connected by the hyperedge relationship, determine the hyperedge weight corresponding to the hyperedge relationship; Determine the hyperedge degree matrix corresponding to the computing power resource hypergraph according to the hyperedge weights corresponding to each hyperedge relationship, where the hyperedge degree matrix includes the degrees corresponding to each hyperedge relationship.

[0080] In one implementation, the matrix conversion module 304 is specifically configured to: Perform an inverse square root operation on the device degree matrix and an inverse operation on the hyperedge degree matrix; Construct a hyperedge weight matrix according to the hyperedge weights corresponding to each hyperedge relationship in the computing power resource hypergraph; Based on the device degree matrix after the inverse square root operation, the hyperedge degree matrix after the inverse operation, the hyperedge weight matrix, and the adjacency matrix, convert the computing power resource hypergraph into the device Laplacian matrix required for hypergraph spectral clustering.

[0081] In one implementation, the eigenvalue decomposition module 306 is specifically configured to: Determine the hyperedge Laplacian matrix; Perform singular value decomposition based on the device Laplacian matrix and the hyperedge Laplacian matrix to determine the device eigenvectors.

[0082] In one implementation, the hypergraph spectral clustering module 308 is specifically configured to: Perform hypergraph spectral clustering processing on the computing power resource hypergraph to obtain the current resource partitioning result; Combine the cut graph objective function and determine the function value corresponding to the current resource partitioning result based on the eigenvectors of the device Laplacian matrix; With the goal of maximizing the function value, continue to perform hypergraph spectral clustering processing on the computing power resource hypergraph based on the eigenvectors of the device Laplacian matrix to obtain the target resource partitioning result corresponding to the computing power resource hypergraph.

[0083] In one implementation, the hypergraph spectral clustering module 308 is specifically configured to: For any computing power resource partition in the current resource partitioning result, determine the indicator vector corresponding to the computing power resource partition according to whether each server device included in the computing power resource hypergraph belongs to the computing power resource partition; Construct an indicator vector matrix based on the indicator vectors corresponding to each computing power resource partition; Normalize the indicator vector matrix, and determine the function value corresponding to the current resource partitioning result according to the normalized indicator vectors and the eigenvectors of the device Laplacian matrix.

[0084] The device provided by the embodiments of the present invention has the same implementation principle and the same technical effects as those of the foregoing method embodiments. For the sake of brief description, for the parts not mentioned in the device embodiments, reference may be made to the corresponding contents in the foregoing method embodiments.

[0085] An embodiment of the present invention provides an electronic device. Specifically, the electronic device includes a processor and a storage device; a computer program is stored on the storage device, and when the computer program is run by the processor, it executes the method described in any one of the above-described embodiments.

[0086] Figure 4 FIG. 4 is a schematic structural diagram of an electronic device provided by an embodiment of the present invention. The electronic device 100 includes: a processor 40, a memory 41, a bus 42, and a communication interface 43. The processor 40, the communication interface 43, and the memory 41 are connected through the bus 42; the processor 40 is configured to execute an executable module stored in the memory 41, such as a computer program.

[0087] Among them, the memory 41 may include a high-speed random access memory (RAM, Random Access Memory), and may also include a non-volatile memory, such as at least one disk memory. Through at least one communication interface 43 (which can be wired or wireless), a communication connection between the system network element and at least one other network element can be realized, and the Internet, wide area network, local area network, metropolitan area network, etc. can be used.

[0088] The bus 42 may be an ISA bus, a PCI bus, an EISA bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. For the sake of representation, Figure 4 only a bidirectional arrow is used in FIG. 4, but it does not mean that there is only one bus or one type of bus.

[0089] Among them, the memory 41 is used to store a program. After receiving an execution instruction, the processor 40 executes the program. The method executed by the device defined by the flow process disclosed in any one of the foregoing embodiments of the present invention can be applied to the processor 40 or implemented by the processor 40.

[0090] The processor 40 may be an integrated circuit chip with the ability to process signals. In the implementation process, each step of the above method can be completed by the integrated logic circuit of the hardware in the processor 40 or the instructions in the form of software. The above-mentioned processor 40 may be a general-purpose processor, including a central processing unit (CPU for short), a network processor (NP for short), etc.; it may also be a digital signal processor (DSP for short), an application specific integrated circuit (ASIC for short), a field-programmable gate array (FPGA for short), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. It can implement or execute the various methods, steps and logic block diagrams disclosed in the embodiments of the present invention. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The steps of the method disclosed in combination with the embodiments of the present invention can be directly embodied as being executed and completed by a hardware decoding processor, or executed and completed by a combination of the hardware and software modules in the decoding processor. The software module may be located in a mature storage medium in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. This storage medium is located in the memory 41, and the processor 40 reads the information in the memory 41 and combines its hardware to complete the steps of the above method.

[0091] The computer program product of the readable storage medium provided by the embodiments of the present invention includes a computer-readable storage medium storing program code, and the instructions included in the program code can be used to execute the methods described in the foregoing method embodiments. For the specific implementation, reference can be made to the foregoing method embodiments and will not be elaborated here.

[0092] When the above-mentioned functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs.

[0093] Finally, it should be noted that the above-mentioned embodiments are only specific implementation manners of the present invention, used to illustrate the technical solutions of the present invention, rather than limiting them. The protection scope of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: any person skilled in the art within the technical scope disclosed by the present invention can still modify the technical solutions recorded in the foregoing embodiments, or can easily think of changes, or make equivalent replacements for some of the technical features; and these modifications, changes, or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A computing power resource partitioning method based on hypergraph clustering, characterized in that Including: Constructing a computing power resource hypergraph in the scenario of the computing power network resource side, where the computing power resource hypergraph includes a computing power resource set and the hyperedge relationship between multiple server devices included therein; Determining the device degree matrix and the hyperedge degree matrix of the computing power resource hypergraph, and converting the computing power resource hypergraph into a device Laplacian matrix required for hypergraph spectral clustering by using the device degree matrix and the hyperedge degree matrix; Performing eigenvalue decomposition on the device Laplacian matrix to obtain device eigenvectors, where the row vectors of the device eigenvectors are the eigenvectors of the device Laplacian matrix; Combined with the cut graph objective function, based on the eigenvectors of the device Laplacian matrix, performing hypergraph spectral clustering processing on the computing power resource hypergraph to obtain the target resource partition result corresponding to the computing power resource hypergraph, and each computing power resource partition in the target resource partition result includes multiple of the server devices.

2. The method for partitioning computing power resources based on hypergraph clustering according to claim 1, wherein Determining the device degree matrix of the computing power resource hypergraph includes: Performing the following operations on any two of the server devices in the computing power resource hypergraph: if there is no hyperedge relationship between the two server devices, determining that the device weight between the two server devices is 0; if there is a hyperedge relationship between the two server devices, determining the device weight between the two server devices according to the attribute information of the two server devices; Determining the device degree matrix corresponding to the computing power resource hypergraph according to the device weights between the server devices, where the device degree matrix includes the degree corresponding to each server device, and the degree corresponding to the server device is the sum value of all the device weights having a hyperedge relationship with the server device; 3. The method for partitioning computing power resources based on hypergraph clustering according to claim 1, wherein Determining the hyperedge degree matrix of the computing power resource hypergraph includes: Performing the following operations on any one hyperedge relationship in the computing power resource hypergraph: based on the communication rate between any two server devices connected by the hyperedge relationship and the number of all server devices connected by the hyperedge relationship, determining the hyperedge weight corresponding to the hyperedge relationship; Determining the hyperedge degree matrix corresponding to the computing power resource hypergraph according to the hyperedge weight corresponding to each hyperedge relationship, where the hyperedge degree matrix includes the degree corresponding to each hyperedge relationship.

4. The method for partitioning computing power resources based on hypergraph clustering according to claim 1, wherein Converting the computing power resource hypergraph into a device Laplacian matrix required for hypergraph spectral clustering by using the device degree matrix and the hyperedge degree matrix includes: Performing an inverse square root operation on the device degree matrix and an inverse operation on the hyperedge degree matrix; Constructing a hyperedge weight matrix according to the hyperedge weight corresponding to each hyperedge relationship in the computing power resource hypergraph; Based on the device degree matrix after the inverse square root operation, the hyperedge degree matrix after the inverse operation, the hyperedge weight matrix, and the adjacency matrix, converting the computing power resource hypergraph into a device Laplacian matrix required for hypergraph spectral clustering.

5. The method for partitioning computing power resources based on hypergraph clustering according to claim 1, wherein Performing eigenvalue decomposition on the device Laplacian matrix to obtain device eigenvectors includes: Determining the hyperedge Laplacian matrix; Performing singular value decomposition based on the device Laplacian matrix and the hyperedge Laplacian matrix to determine the device eigenvectors.

6. The method for partitioning computing power resources based on hypergraph clustering according to claim 1, wherein, Combined with the cut graph objective function, based on the eigenvectors of the device Laplacian matrix, perform hypergraph spectral clustering on the computing power resource hypergraph to obtain the target resource partitioning result corresponding to the computing power resource hypergraph, including: Perform hypergraph spectral clustering on the computing power resource hypergraph to obtain the current resource partitioning result; Combined with the cut graph objective function, based on the eigenvectors of the device Laplacian matrix, determine the function value corresponding to the current resource partitioning result; With the goal of maximizing the function value, continue to perform hypergraph spectral clustering on the computing power resource hypergraph based on the eigenvectors of the device Laplacian matrix to obtain the target resource partitioning result corresponding to the computing power resource hypergraph.

7. The method for partitioning computing power resources based on hypergraph clustering according to claim 6, wherein, Combined with the cut graph objective function, based on the eigenvectors of the device Laplacian matrix, determine the function value corresponding to the current resource partitioning result, including: For any computing power resource partition in the current resource partitioning result, determine the indication vector corresponding to the computing power resource partition according to whether each server device included in the computing power resource hypergraph belongs to the computing power resource partition; Construct an indication vector matrix based on the indication vectors corresponding to each computing power resource partition; Normalize the indication vector matrix, and determine the function value corresponding to the current resource partitioning result according to the normalized indication vector and the eigenvectors of the device Laplacian matrix.

8. An arithmetic power resource partitioning device based on hypergraph clustering, characterized in that, Including: A hypergraph construction module, configured to construct a computing power resource hypergraph in the scenario of the computing power network resource side, where the computing power resource hypergraph includes a computing power resource set and the hyperedge relationship between multiple server devices included therein; A matrix conversion module, configured to determine the device degree matrix and the hyperedge degree matrix of the computing power resource hypergraph, and use the device degree matrix and the hyperedge degree matrix to convert the computing power resource hypergraph into a device Laplacian matrix required for hypergraph spectral clustering; A feature decomposition module, configured to perform eigenvalue decomposition on the device Laplacian matrix to obtain device eigenvectors, and the row vectors of the device eigenvectors are the eigenvectors of the device Laplacian matrix; A hypergraph spectral clustering module, configured to combine the cut graph objective function, based on the eigenvectors of the device Laplacian matrix, perform hypergraph spectral clustering on the computing power resource hypergraph to obtain the target resource partitioning result corresponding to the computing power resource hypergraph, and each computing power resource partition in the target resource partitioning result includes multiple server devices.

9. An electronic device, characterized in that, Including a processor and a memory, where the memory stores computer-executable instructions that can be executed by the processor, and the processor executes the computer-executable instructions to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are called and executed by the processor, the computer-executable instructions cause the processor to implement the method according to any one of claims 1 to 7.

Images