A computing power optimization method and system for cloud computing data center transmission
By optimizing the network topology structure and virtual machine migration algorithm of cloud computing data centers, using software-defined network technology and convolutional neural network model, the problem of high energy consumption in data centers is solved, and resource utilization and energy-saving effects are improved.
Patent Information
- Application Number
- CN202411237726.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-05
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2044-09-05
AI Technical Summary
The high energy consumption problems of cloud computing data centers lead to waste of electricity and environmental impacts, and it is difficult for the existing technology to effectively manage the efficient utilization of computing resources and storage resources.
By building a mathematical model for transmission in cloud computing data centers, optimizing network topology and link aggregation, using software-defined network technology to separate control planes and data planes, and building a predictive convolutional neural network model of migration strategy, optimized based on load data prediction results Virtual machine migration algorithm to achieve efficient utilization of resources.
It improves the utilization rate of data center resources, reduces the number of virtual machines used and the energy consumption of migration processes, reduces energy consumption and maintenance costs, and realizes energy saving in data centers.
Smart Images

Figure CN118764490B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of computing power resource configuration, and in particular to a computing power optimization method and system for cloud computing data center transmission. Background Art
[0002] Data centers are the core of cloud computing, managing tens of thousands of servers on the Internet. In a broad sense, data centers refer to complex facilities that integrate servers and communication equipment. They not only include computer systems and supporting equipment, but also data communication connections, environmental control equipment, monitoring equipment, and various safety devices. In a narrow sense, data centers are places where many companies on the Internet store equipment and data. It is another major invention in the IT industry, marking the organizational standardization of information technology applications.
[0003] In recent years, with the development of data centers, high energy consumption in data centers has gradually become a prominent problem. Especially with the advent of the cloud computing era, data centers are no longer just a simple place for unified hosting and maintenance of servers. They have evolved into a collection of high-performance computers that integrate large-scale data computing and storage. More computing resources and storage resources are concentrated in the cloud, which brings greater challenges to the efficient management of energy consumption. The high energy consumption problem of data centers not only causes waste of electricity and unstable system operation, but also has adverse effects on the environment. Therefore, it is urgent to develop new technologies to solve the high energy consumption problem of data centers, including considering the design of hardware, software, and algorithms, especially considering energy consumption as an important parameter in the design and management of data centers, rather than just considering the performance parameters of data centers. Summary of the invention
[0004] In order to solve the above technical problems, a method and system for optimizing computing power for cloud computing data center transmission are provided. This technical solution solves the problems raised in the above background technology.
[0005] In order to achieve the above purpose, the technical solution adopted by the present invention is:
[0006] A computing power optimization method for cloud computing data center transmission, comprising:
[0007] Construct a mathematical model of cloud computing data center transmission;
[0008] Optimize the transmission mathematical model of cloud computing data center from the perspectives of network bandwidth optimization and virtualization resource scheduling;
[0009] By introducing the leaf-spine topology and adaptively extended network structure, the topology of cloud computing data center network is optimized;
[0010] Through link aggregation and load balancing technology, at least one physical link is merged into a logical link, and the data transmission load is evenly distributed;
[0011] Use software-defined networking technology to separate the network control plane from the data plane and automatically manage the network architecture of cloud computing data centers;
[0012] Construct a predictive convolutional neural network model for the migration strategy and obtain load data prediction results;
[0013] Based on the load data prediction results, the minimum number selection algorithm is used to optimize the virtual machine migration algorithm;
[0014] Output the optimized cloud computing data center transmission mathematical model.
[0015] Preferably, the optimization of the topology of the cloud computing data center network by introducing a leaf-spine topology and an adaptively extended network structure specifically includes:
[0016] Establishing a leaf layer switch group, wherein the leaf layer switch group is used to process access of network data;
[0017] Establishing a spine layer switch group, wherein the spine layer switch group is used to process data transmission between leaf layer switch groups;
[0018] Establish a link between each switch in the spine switch group and the leaf switch group;
[0019] Divide the data center network into at least one module, each module including at least one leaf layer switch and the same number of spine layer switches;
[0020] Determine in real time whether the data transmission rate of the data center is greater than the preset threshold. If so, no output is made. If not, a new module is established, which includes at least one leaf layer and the same number of spine layer switches.
[0021] Preferably, the step of merging at least one physical link into one logical link and evenly distributing the data transmission load by link aggregation and load balancing technology specifically includes:
[0022] Subtract the highest frequency and the lowest frequency of each physical link that can pass through the channel to obtain the channel bandwidth;
[0023] Obtain channel signal average power and channel noise average power;
[0024] Use Shannon's formula to calculate the capacity of each channel;
[0025] Determine whether the channel capacity is greater than the amount of data transmitted by the client. If so, no output is made. If not, the current physical link is merged with at least one semi-idle physical link into a logical link through link aggregation technology.
[0026] Get the reserved channel capacity value;
[0027] Determine whether the channel capacity is greater than the sum of the client transmission data volume and the reserved channel capacity value. If so, output that the current physical link is a semi-idle physical link. If not, no output is made.
[0028] The Shannon formula is: ,
[0029] In the formula, is the channel capacity, is the channel bandwidth, is the average power of the channel signal, is the average channel noise power.
[0030] Preferably, the use of software defined network technology to separate the network control plane from the data plane and automatically manage the cloud computing data center network architecture specifically includes:
[0031] Deploy software-defined network technology controllers in cloud computing data centers and be responsible for the functional implementation of the control plane;
[0032] Upgrade traditional network equipment to software-defined network technology-compatible network equipment and be responsible for the function implementation of the data plane;
[0033] Use the OpenFlow network protocol to define the interface and information format between the controller and network devices, and implement data communication between the network control plane and the data plane;
[0034] The network policies formulated by the software-defined network technology controller are dynamically adjusted according to customer business needs, and the network policies include router policies, security policies and migration policies.
[0035] Preferably, the prediction convolutional neural network model of the migration strategy is constructed to obtain the load data prediction result, which specifically includes:
[0036] Collect historical load data, including CPU usage, memory usage, and network traffic;
[0037] The historical load data is normalized using a normalization formula to obtain normalized data;
[0038] Get the convolution kernel matrix;
[0039] Set filter window width;
[0040] Convolution of normalized data using filter function;
[0041] Select the maximum value among the convolution processing results of each window to replace the data at the original window position of the normalized data;
[0042] Summarize the replaced data and output the prediction results of the prediction convolutional neural network model of the migration strategy;
[0043] The normalization formula is: ,
[0044] In the formula, is the i-th normalized data, is the i-th historical load data, , They are the maximum and minimum values of historical load data respectively;
[0045] The filtering function is: ,
[0046] In the formula, is the convolution result of the i-th window, is the hyperbolic tangent function, is the convolution kernel matrix, is the matrix composed of all data from the i-th data to the i+h-1-th data, h is the filter window width, is the smoothing factor;
[0047] The hyperbolic tangent function is: ,
[0048] In the formula, Enter data for the hyperbolic tangent function, Output data for the hyperbolic tangent function.
[0049] Preferably, the optimizing the virtual machine migration algorithm by using the minimum number selection algorithm based on the load data prediction result specifically includes:
[0050] S101: Setting upper and lower limits of host resource utilization;
[0051] S102: Calculating resource utilization of each host based on load data prediction results;
[0052] S103: Determine whether the resource utilization of each host is less than the lower limit resource utilization of the host. If so, put all virtual machines in the host into the migration virtual machine set S1. If not, do not output.
[0053] S104: Determine whether the resource utilization of each host is greater than the upper limit resource utilization of the host. If so, put all virtual machines in the host into the set of virtual machines to be migrated S2. If not, do not output.
[0054] S105: Calculate the resource utilization rate of the host that exceeds the upper threshold and the resource utilization rate of the virtual machine in S2;
[0055] S106: Subtract the resource utilization rate of the virtual machine in S2 from the resource utilization rate that exceeds the upper threshold to obtain a matching parameter;
[0056] S107: Determine whether the matching parameters of each virtual machine are non-negative. If so, output that the resource utilization of the host can be lower than the upper threshold after migrating the virtual machine, and put the matching parameters of the virtual machine into the matching parameter set S3. If not, do not output;
[0057] S108: Determine whether the matching parameters of all virtual machines in S2 are less than zero. If so, put the virtual machine with the largest resource utilization in S2 into set S1 and return to step S106. If not, do not output.
[0058] S109: Sort the matching parameters in set S3, put the virtual machine with the smallest parameter value into the migration virtual machine set S1, and the algorithm ends.
[0059] Furthermore, a computing power optimization system for cloud computing data center transmission is proposed, which is used to implement the computing power optimization method for cloud computing data center transmission as described above, including:
[0060] A transmission model building module, wherein the transmission model building module is used to build a transmission mathematical model of a cloud computing data center;
[0061] A transmission model optimization module, which is used to optimize the transmission mathematical model of the cloud computing data center from the two perspectives of network bandwidth optimization and virtualization resource scheduling;
[0062] A network bandwidth optimization module, which is used to optimize the topology of the cloud computing data center network by introducing a leaf-spine topology and an adaptively extended network structure, merge at least one physical link into a logical link through link aggregation and load balancing technology, and evenly distribute the data transmission load, separate the network control plane from the data plane using software-defined network technology, and automatically manage the cloud computing data center network architecture;
[0063] A virtualization resource scheduling module, wherein the virtualization resource scheduling module is used to construct a predictive convolutional neural network model of a migration strategy, obtain a load data prediction result, and optimize a virtual machine migration algorithm using a minimum number selection algorithm based on the load data prediction result;
[0064] The optimization model output module is used to output the optimized cloud computing data center transmission mathematical model.
[0065] Optionally, the network bandwidth optimization module specifically includes:
[0066] A structure optimization unit, wherein the structure optimization unit is used to optimize the topology structure of the cloud computing data center network by introducing a leaf-spine topology structure and an adaptively extended network structure;
[0067] A link aggregation unit, the link aggregation unit is used to merge at least one physical link into a logical link and evenly distribute the data transmission load through link aggregation and load balancing technology;
[0068] A plane separation unit is used to separate the network control plane from the data plane by using software defined network technology, so as to automatically manage the network architecture of the cloud computing data center.
[0069] Optionally, the virtualization resource scheduling module specifically includes:
[0070] A data prediction unit, wherein the data prediction unit is used to construct a prediction convolutional neural network model of the migration strategy to obtain a load data prediction result;
[0071] A virtual machine migration unit is used to optimize the virtual machine migration algorithm based on the load data prediction result by using a minimum number selection algorithm.
[0072] Compared with the prior art, the present invention has the following beneficial effects:
[0073] Based on the energy-saving architecture of the existing data center, improvements are made to improve the resource utilization of the data center. When providing external services, the virtual machine migration process is optimized to reduce the number of virtual machines in use, improve the resource utilization of virtual machines, reduce the energy consumption of the migration process, and achieve energy saving in the data center. By setting up a network bandwidth optimization module and a virtualization resource scheduling module, energy consumption is reduced, and the maintenance cost of the data center is also reduced invisibly, thereby improving resource utilization. BRIEF DESCRIPTION OF THE DRAWINGS
[0074] Figure 1 This is a flow chart of a computing power optimization method for cloud computing data center transmission according to the present invention;
[0075] Figure 2 A flow chart of a method for optimizing a topology structure of a cloud computing data center network according to the present invention;
[0076] Figure 3 It is a flow chart of the link aggregation and load balancing technical method of the present invention;
[0077] Figure 4 The flowchart of the software defined network technology method of the present invention is as follows;
[0078] Figure 5 A flow chart of a predictive convolutional neural network model method for constructing a migration strategy of the present invention;
[0079] Figure 6 This is a flow chart of the minimum number selection algorithm method of the present invention. DETAILED DESCRIPTION
[0080] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art may think of other obvious variations.
[0081] Reference Figure 1 As shown, a computing power optimization method for cloud computing data center transmission includes:
[0082] Construct a mathematical model of cloud computing data center transmission;
[0083] Optimize the transmission mathematical model of cloud computing data center from the perspectives of network bandwidth optimization and virtualization resource scheduling;
[0084] By introducing the leaf-spine topology and adaptively extended network structure, the topology of cloud computing data center network is optimized;
[0085] Through link aggregation and load balancing technology, at least one physical link is merged into a logical link, and the data transmission load is evenly distributed;
[0086] Use software-defined networking technology to separate the network control plane from the data plane and automatically manage the network architecture of cloud computing data centers;
[0087] Construct a predictive convolutional neural network model for the migration strategy and obtain load data prediction results;
[0088] Based on the load data prediction results, the minimum number selection algorithm is used to optimize the virtual machine migration algorithm;
[0089] Output the optimized cloud computing data center transmission mathematical model.
[0090] Reference Figure 2 As shown in the figure, by introducing the leaf-spine topology and the adaptively extended network structure, the topology of the cloud computing data center network is optimized, including:
[0091] Establishing a leaf layer switch group, wherein the leaf layer switch group is used to process access of network data;
[0092] Establishing a spine layer switch group, wherein the spine layer switch group is used to process data transmission between leaf layer switch groups;
[0093] Establish a link between each switch in the spine switch group and the leaf switch group;
[0094] Divide the data center network into at least one module, each module including at least one leaf layer switch and the same number of spine layer switches;
[0095] Determine in real time whether the data transmission rate of the data center is greater than the preset threshold. If so, no output is made. If not, a new module is established, which includes at least one leaf layer and the same number of spine layer switches.
[0096] The leaf-spine topology is a data center network architecture consisting of an upper-layer spine switch and a lower-layer leaf switch. The leaf-spine topology adds a horizontal network structure parallel to the backbone vertical network structure, and adds a corresponding switching network to this horizontal structure, which is similar to the traditional three-layer network architecture. This can also separate broadcast domains. The leaf-spine topology has the advantages of high reliability, high performance and scalability, and is favored by many large data centers and cloud computing networks.
[0097] Reference Figure 3 As shown, at least one physical link is merged into one logical link through link aggregation and load balancing technology, and the data transmission load is evenly distributed, specifically including:
[0098] Subtract the highest frequency and the lowest frequency of each physical link that can pass through the channel to obtain the channel bandwidth;
[0099] Obtain channel signal average power and channel noise average power;
[0100] Use Shannon's formula to calculate the capacity of each channel;
[0101] Determine whether the channel capacity is greater than the amount of data transmitted by the client. If so, no output is made. If not, the current physical link is merged with at least one semi-idle physical link into a logical link through link aggregation technology.
[0102] Get the reserved channel capacity value;
[0103] Determine whether the channel capacity is greater than the sum of the client transmission data volume and the reserved channel capacity value. If so, output that the current physical link is a semi-idle physical link. If not, no output is made.
[0104] The Shannon formula is: ,
[0105] In the formula, is the channel capacity, is the channel bandwidth, is the average power of the channel signal, is the average channel noise power.
[0106] Link aggregation is to aggregate the links between two devices and use them as a logical link. Link aggregation can increase the bandwidth of the link. In theory, through link aggregation, the bandwidth of an aggregated port can be made the sum of the bandwidths of all member ports. Link aggregation can improve the reliability of the network. If a port configured with link aggregation fails, the traffic of the member port will be switched to the member link, ensuring the reliability of network transmission. Link aggregation can also achieve traffic load balancing, evenly distributing the traffic to all member links, so that each member link can minimize the risk of traffic blocking the link.
[0107] Reference Figure 4 As shown in the figure, the network control plane is separated from the data plane by using software-defined networking technology, and the network architecture of cloud computing data center is automatically managed, including:
[0108] Deploy software-defined network technology controllers in cloud computing data centers and be responsible for the functional implementation of the control plane;
[0109] Upgrade traditional network equipment to software-defined network technology-compatible network equipment and be responsible for the function implementation of the data plane;
[0110] Use the OpenFlow network protocol to define the interface and information format between the controller and network devices, and implement data communication between the network control plane and the data plane;
[0111] The network policies formulated by the software-defined network technology controller are dynamically adjusted according to customer business needs, and the network policies include router policies, security policies and migration policies.
[0112] Software-defined networking technology is a network management method that supports dynamically programmable network configuration, improves network performance and management efficiency, and enables network services to provide flexible customization capabilities like cloud computing. Software-defined networking technology is a new network architecture that decouples network control logic from underlying forwarding hardware. The control plane is responsible for intelligent decision-making and policy implementation of network traffic, while the data plane focuses on forwarding data packets according to the instructions of the control plane. This separation design makes network management more flexible and efficient.
[0113] Reference Figure 5 As shown in the figure, a prediction convolutional neural network model of the migration strategy is constructed, and the load data prediction results obtained include:
[0114] Collect historical load data, including CPU usage, memory usage, and network traffic;
[0115] The historical load data is normalized using a normalization formula to obtain normalized data;
[0116] Get the convolution kernel matrix;
[0117] Set filter window width;
[0118] Convolution of normalized data using filter function;
[0119] Select the maximum value among the convolution processing results of each window to replace the data at the original window position of the normalized data;
[0120] Summarize the replaced data and output the prediction results of the prediction convolutional neural network model of the migration strategy;
[0121] The normalization formula is: ,
[0122] In the formula, is the i-th normalized data, is the i-th historical load data, , They are the maximum and minimum values of historical load data respectively;
[0123] The filtering function is: ,
[0124] In the formula, is the convolution result of the i-th window, is the hyperbolic tangent function, is the convolution kernel matrix, is the matrix composed of all data from the i-th data to the i+h-1-th data, h is the filter window width, is the smoothing factor;
[0125] The hyperbolic tangent function is: ,
[0126] In the formula, Enter data for the hyperbolic tangent function, Output data for the hyperbolic tangent function.
[0127] Since convolutional neural networks have powerful spatial feature learning capabilities, they are often used in related fields such as computer vision. Existing convolutional neural networks usually include input layer, convolution layer, pooling layer, fully connected layer and output layer. The unique training method of sparse connection and parameter sharing is the biggest advantage of the network. Among them, the design of convolution layer and pooling layer is extremely critical. Usually, the underlying structure learns the local features of the data, while the deep structure of the network can learn the global features.
[0128] Reference Figure 6 As shown, based on the load data prediction results, the minimum number selection algorithm is used to optimize the virtual machine migration algorithm, specifically including:
[0129] S101: Setting upper and lower limits of host resource utilization;
[0130] S102: Calculating resource utilization of each host based on load data prediction results;
[0131] S103: Determine whether the resource utilization of each host is less than the lower limit resource utilization of the host. If so, put all virtual machines in the host into the migration virtual machine set S1. If not, do not output.
[0132] S104: Determine whether the resource utilization of each host is greater than the upper limit resource utilization of the host. If so, put all virtual machines in the host into the set of virtual machines to be migrated S2. If not, do not output.
[0133] S105: Calculate the resource utilization rate of the host that exceeds the upper threshold and the resource utilization rate of the virtual machine in S2;
[0134] S106: Subtract the resource utilization rate of the virtual machine in S2 from the resource utilization rate that exceeds the upper threshold to obtain a matching parameter;
[0135] S107: Determine whether the matching parameters of each virtual machine are non-negative. If so, output that the resource utilization of the host can be lower than the upper threshold after migrating the virtual machine, and put the matching parameters of the virtual machine into the matching parameter set S3. If not, do not output;
[0136] S108: Determine whether the matching parameters of all virtual machines in S2 are less than zero. If so, put the virtual machine with the largest resource utilization in S2 into set S1 and return to step S106. If not, do not output.
[0137] S109: Sort the matching parameters in set S3, put the virtual machine with the smallest parameter value into the migration virtual machine set S1, and the algorithm ends.
[0138] When there is a host in the data center that needs to migrate virtual machines, the virtual machines in the host need to be migrated. However, there are many virtual machines in the host. Migrating any virtual machine in the host will have an impact on the data center. If the virtual machine with the highest resource utilization in the host is selected, although the resource rate of the host is reduced, the migrated virtual machine is very large, which increases the amount of information during migration and correspondingly increases energy consumption. If the virtual machine with the lowest resource utilization in the host is selected, it is possible that after migrating a virtual machine, the utilization of the host cannot be reduced to the threshold range. In this case, another virtual machine needs to be migrated, which increases the migration energy consumption and fails to achieve the purpose of reducing energy consumption. After using the above algorithm for the hosts in the data center, it is guaranteed that the number of migrated virtual machines is minimized, reducing the migration overhead and energy consumption of virtual machines, and can also ensure that the source host runs in a state closest to the resource utilization threshold, thereby reducing energy consumption.
[0139] Further, based on the same inventive concept as the above-mentioned computing power optimization method for cloud computing data center transmission, the present solution also proposes a computing power optimization system for cloud computing data center transmission, including:
[0140] A transmission model building module, wherein the transmission model building module is used to build a transmission mathematical model of a cloud computing data center;
[0141] A transmission model optimization module, which is used to optimize the transmission mathematical model of the cloud computing data center from the two perspectives of network bandwidth optimization and virtualization resource scheduling;
[0142] A network bandwidth optimization module, which is used to optimize the topology of the cloud computing data center network by introducing a leaf-spine topology and an adaptively extended network structure, merge at least one physical link into a logical link through link aggregation and load balancing technology, and evenly distribute the data transmission load, separate the network control plane from the data plane using software-defined network technology, and automatically manage the cloud computing data center network architecture;
[0143] A virtualization resource scheduling module, wherein the virtualization resource scheduling module is used to construct a predictive convolutional neural network model of a migration strategy, obtain a load data prediction result, and optimize a virtual machine migration algorithm using a minimum number selection algorithm based on the load data prediction result;
[0144] The optimization model output module is used to output the optimized cloud computing data center transmission mathematical model.
[0145] The network bandwidth optimization module specifically includes:
[0146] A structure optimization unit, wherein the structure optimization unit is used to optimize the topology structure of the cloud computing data center network by introducing a leaf-spine topology structure and an adaptively extended network structure;
[0147] A link aggregation unit, the link aggregation unit is used to merge at least one physical link into a logical link and evenly distribute the data transmission load through link aggregation and load balancing technology;
[0148] A plane separation unit is used to separate the network control plane from the data plane by using software defined network technology, so as to automatically manage the network architecture of the cloud computing data center.
[0149] The virtualization resource scheduling module specifically includes:
[0150] A data prediction unit, wherein the data prediction unit is used to construct a prediction convolutional neural network model of the migration strategy to obtain a load data prediction result;
[0151] A virtual machine migration unit is used to optimize the virtual machine migration algorithm based on the load data prediction result by using a minimum number selection algorithm.
[0152] Furthermore, the present solution also proposes a computer-readable storage medium on which a computer-readable program is stored. When the computer-readable program is called, the above-mentioned computing power optimization method for cloud computing data center transmission is executed.
[0153] It is understandable that the storage medium may be a magnetic medium, such as a floppy disk, a hard disk, or a magnetic tape; an optical medium, such as a DVD; or a semiconductor medium, such as a solid state drive (SSD).
[0154] In summary, the advantages of the present invention are: based on the energy-saving architecture of the existing data center, improvements are made in improving the resource utilization of the data center. When providing external services, the virtual machine migration process is optimized to reduce the number of virtual machines in use, improve the resource utilization of the virtual machines, reduce the energy consumption of the migration process, and achieve energy saving in the data center. By setting a network bandwidth optimization module and a virtualization resource scheduling module, energy consumption is reduced, and the maintenance cost of the data center is also reduced invisibly, thereby improving resource utilization.
[0155] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions only describe the principles of the present invention. The present invention may be subject to various changes and improvements without departing from the spirit and scope of the present invention. These changes and improvements fall within the scope of the present invention. The scope of protection claimed by the present invention is defined by the attached claims and their equivalents.
Claims
1. A computing power optimization method for cloud computing data center transmission, characterized in that: include: Construct a mathematical model of cloud computing data center transmission; Optimize the transmission mathematical model of cloud computing data center from the perspectives of network bandwidth optimization and virtualization resource scheduling; By introducing the leaf-spine topology and adaptively extended network structure, the topology of cloud computing data center network is optimized; Through link aggregation and load balancing technology, at least one physical link is merged into a logical link, and the data transmission load is evenly distributed; Use software-defined networking technology to separate the network control plane from the data plane and automatically manage the network architecture of cloud computing data centers; Construct a predictive convolutional neural network model for the migration strategy and obtain load data prediction results; Based on the load data prediction results, the minimum number selection algorithm is used to optimize the virtual machine migration algorithm; Output optimized cloud computing data center transmission mathematical model; The optimization of the virtual machine migration algorithm by using the minimum number selection algorithm based on the load data prediction result specifically includes: S101: Setting upper and lower limits of host resource utilization; S102: Calculating resource utilization of each host based on load data prediction results; S103: Determine whether the resource utilization of each host is less than the lower limit resource utilization of the host. If so, put all virtual machines in the host into the migration virtual machine set S1. If not, do not output. S104: Determine whether the resource utilization of each host is greater than the upper limit resource utilization of the host. If so, put all virtual machines in the host into the set of virtual machines to be migrated S2. If not, do not output. S105: Calculate the resource utilization rate of the host that exceeds the upper threshold and the resource utilization rate of the virtual machine in S2; S106: Subtract the resource utilization rate of the virtual machine in S2 from the resource utilization rate that exceeds the upper threshold to obtain a matching parameter; S107: Determine whether the matching parameters of each virtual machine are non-negative. If so, output that the resource utilization of the host can be lower than the upper threshold after migrating the virtual machine, and put the matching parameters of the virtual machine into the matching parameter set S3. If not, do not output; S108: Determine whether the matching parameters of all virtual machines in S2 are less than zero. If so, put the virtual machine with the largest resource utilization in S2 into set S1 and return to step S106. If not, do not output. S109: Sort the matching parameters in set S3, put the virtual machine with the smallest parameter value into the migration virtual machine set S1, and the algorithm ends.
2. The computing power optimization method for cloud computing data center transmission according to claim 1 is characterized in that: The optimization of the topology structure of the cloud computing data center network by introducing the leaf-spine topology structure and the adaptively extended network structure specifically includes: Establishing a leaf layer switch group, wherein the leaf layer switch group is used to process access of network data; Establishing a spine layer switch group, wherein the spine layer switch group is used to process data transmission between leaf layer switch groups; Establish a link between each switch in the spine switch group and the leaf switch group; Divide the data center network into at least one module, each module including at least one leaf layer switch and the same number of spine layer switches; Determine in real time whether the data transmission rate of the data center is greater than the preset threshold. If so, no output is made. If not, a new module is established, which includes at least one leaf layer and the same number of spine layer switches.
3. The computing power optimization method for cloud computing data center transmission according to claim 2 is characterized in that: The method of combining at least one physical link into one logical link and evenly distributing the data transmission load by link aggregation and load balancing technology specifically includes: Subtract the highest frequency and the lowest frequency of each physical link that can pass through the channel to obtain the channel bandwidth; Obtain channel signal average power and channel noise average power; Use Shannon's formula to calculate the capacity of each channel; Determine whether the channel capacity is greater than the amount of data transmitted by the client. If so, no output is made. If not, the current physical link is combined with at least one semi-idle physical link into a logical link through link aggregation technology. Get the reserved channel capacity value; Determine whether the channel capacity is greater than the sum of the client transmission data volume and the reserved channel capacity value. If so, output that the current physical link is a semi-idle physical link. If not, no output is made. The Shannon formula is: , In the formula, is the channel capacity, is the channel bandwidth, is the average power of the channel signal, is the average channel noise power.
4. The computing power optimization method for cloud computing data center transmission according to claim 3 is characterized in that: The use of software-defined network technology to separate the network control plane from the data plane and automatically manage the cloud computing data center network architecture specifically includes: Deploy software-defined network technology controllers in cloud computing data centers and be responsible for the functional implementation of the control plane; Upgrade traditional network equipment to software-defined network technology-compatible network equipment and be responsible for the function implementation of the data plane; Use the OpenFlow network protocol to define the interface and information format between the controller and network devices, and implement data communication between the network control plane and the data plane; The network policies formulated by the software-defined network technology controller are dynamically adjusted according to customer business needs, and the network policies include router policies, security policies and migration policies.
5. The computing power optimization method for cloud computing data center transmission according to claim 4 is characterized in that: The prediction convolutional neural network model of the migration strategy is constructed to obtain the load data prediction result, which specifically includes: Collect historical load data, including CPU usage, memory usage, and network traffic; The historical load data is normalized using a normalization formula to obtain normalized data; Get the convolution kernel matrix; Set filter window width; Convolve the normalized data using a filter function; Select the maximum value among the convolution processing results of each window to replace the data at the original window position of the normalized data; Summarize the replaced data and output the prediction results of the prediction convolutional neural network model of the migration strategy; The normalization formula is: , In the formula, is the i-th normalized data, is the i-th historical load data, , They are the maximum and minimum values of historical load data respectively; The filtering function is: , In the formula, is the convolution result of the i-th window, is the hyperbolic tangent function, is the convolution kernel matrix, is the matrix composed of all data from the i-th data to the i+h-1-th data, h is the filter window width, is the smoothing factor; The hyperbolic tangent function is: , In the formula, Enter data for the hyperbolic tangent function, Output data for the hyperbolic tangent function.
6. A computing power optimization system for cloud computing data center transmission, used to implement a computing power optimization method for cloud computing data center transmission as described in any one of claims 1 to 5, characterized in that: include: A transmission model building module, wherein the transmission model building module is used to build a transmission mathematical model of a cloud computing data center; A transmission model optimization module, which is used to optimize the transmission mathematical model of the cloud computing data center from the two perspectives of network bandwidth optimization and virtualization resource scheduling; A network bandwidth optimization module, which is used to optimize the topology of the cloud computing data center network by introducing a leaf-spine topology and an adaptively extended network structure, merge at least one physical link into a logical link through link aggregation and load balancing technology, and evenly distribute the data transmission load, separate the network control plane from the data plane using software-defined network technology, and automatically manage the cloud computing data center network architecture; A virtualization resource scheduling module, wherein the virtualization resource scheduling module is used to construct a predictive convolutional neural network model of a migration strategy, obtain a load data prediction result, and optimize a virtual machine migration algorithm using a minimum number selection algorithm based on the load data prediction result; The optimization model output module is used to output the optimized cloud computing data center transmission mathematical model.
7. A computing power optimization system for cloud computing data center transmission according to claim 6, characterized in that: The network bandwidth optimization module specifically includes: A structure optimization unit, wherein the structure optimization unit is used to optimize the topology structure of the cloud computing data center network by introducing a leaf-spine topology structure and an adaptively extended network structure; A link aggregation unit, the link aggregation unit is used to merge at least one physical link into a logical link and evenly distribute the data transmission load through link aggregation and load balancing technology; A plane separation unit is used to separate the network control plane from the data plane by using software defined network technology, so as to automatically manage the network architecture of the cloud computing data center.
8. The computing power optimization system for cloud computing data center transmission according to claim 7, characterized in that: The virtualization resource scheduling module specifically includes: A data prediction unit, wherein the data prediction unit is used to construct a prediction convolutional neural network model of the migration strategy to obtain a load data prediction result; A virtual machine migration unit is used to optimize the virtual machine migration algorithm based on the load data prediction result by using a minimum number selection algorithm.
Citation Information
Patent Citations
Scheduling method and system for optimizing cloud data center
CN112269632A
Cloud computing data center energy consumption optimization method and system
CN117113235A
Energy-efficient traffic scheduling algorithm for SDN data center network based on multi-layer virtual topology
US20210111994A1
