CN117880291A - Multi-cloud resource load balancing method, device, equipment and medium based on GPT technology - Google Patents
Multi-cloud resource load balancing method, device, equipment and medium based on GPT technology Download PDFInfo
- Publication number
- CN117880291A CN117880291A CN202311697248.3A CN202311697248A CN117880291A CN 117880291 A CN117880291 A CN 117880291A CN 202311697248 A CN202311697248 A CN 202311697248A CN 117880291 A CN117880291 A CN 117880291A
- Authority
- CN
- China
- Prior art keywords
- cloud
- load
- data
- resource
- load balancing
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 57
- 238000005516 engineering process Methods 0.000 title claims abstract description 43
- 238000012544 monitoring process Methods 0.000 claims abstract description 76
- 238000013468 resource allocation Methods 0.000 claims abstract description 64
- 238000004422 calculation algorithm Methods 0.000 claims description 34
- 238000012549 training Methods 0.000 claims description 24
- 238000000605 extraction Methods 0.000 claims description 16
- 238000005457 optimization Methods 0.000 claims description 16
- 238000007781 pre-processing Methods 0.000 claims description 16
- 238000004590 computer program Methods 0.000 claims description 15
- 238000003860 storage Methods 0.000 claims description 15
- 238000010606 normalization Methods 0.000 claims description 13
- 238000004364 calculation method Methods 0.000 claims description 11
- 238000004140 cleaning Methods 0.000 claims description 11
- 238000013480 data collection Methods 0.000 claims description 9
- 238000001514 detection method Methods 0.000 claims description 4
- 230000000694 effects Effects 0.000 abstract description 7
- 238000012545 processing Methods 0.000 description 11
- 238000010586 diagram Methods 0.000 description 7
- 230000008569 process Effects 0.000 description 7
- 230000008859 change Effects 0.000 description 6
- 230000004044 response Effects 0.000 description 6
- 238000004458 analytical method Methods 0.000 description 3
- 238000009826 distribution Methods 0.000 description 3
- 238000005206 flow analysis Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 238000007726 management method Methods 0.000 description 3
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 238000002790 cross-validation Methods 0.000 description 2
- 230000003247 decreasing effect Effects 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000005856 abnormality Effects 0.000 description 1
- 238000003491 array Methods 0.000 description 1
- CLOMYZFHNHFSIQ-UHFFFAOYSA-N clonixin Chemical compound CC1=C(Cl)C=CC=C1NC1=NC=CC=C1C(O)=O CLOMYZFHNHFSIQ-UHFFFAOYSA-N 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L67/00—Network arrangements or protocols for supporting network services or applications
- H04L67/01—Protocols
- H04L67/10—Protocols in which an application is distributed across nodes in the network
- H04L67/1001—Protocols in which an application is distributed across nodes in the network for accessing one among a plurality of replicated servers
- H04L67/1004—Server selection for load balancing
- H04L67/1008—Server selection for load balancing based on parameters of servers, e.g. available memory or workload
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/14—Network analysis or design
- H04L41/145—Network analysis or design involving simulating, designing, planning or modelling of a network
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/14—Network analysis or design
- H04L41/147—Network analysis or design for predicting network behaviour
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Computer Hardware Design (AREA)
- General Engineering & Computer Science (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
本申请提供一种基于GPT技术的多云资源负载均衡方法、装置、设备及介质，该方法通过负载平衡监测模型根据云服务器群的当前负载数据对云服务器群中各云资源的负载运行数据进行预测，从而根据预测负载数据，计算各云资源的资源分配权重，以根据各云资源对应的资源分配权重，调整各云资源的资源份额，以调整各云资源之间的负载平衡。通过对多个云资源的负载进行实时监测和预测，并根据负载情况进行资源分配和调度，以实现更好的负载均衡和资源利用率，提供更加高效的负载均衡效果，提高混合云负载平衡模型的负载均衡可靠性。
The present application provides a multi-cloud resource load balancing method, device, equipment and medium based on GPT technology. The method predicts the load operation data of each cloud resource in the cloud server group according to the current load data of the cloud server group through a load balancing monitoring model, thereby calculating the resource allocation weight of each cloud resource according to the predicted load data, and adjusting the resource share of each cloud resource according to the resource allocation weight corresponding to each cloud resource, so as to adjust the load balance between each cloud resource. By real-time monitoring and prediction of the load of multiple cloud resources, and resource allocation and scheduling according to the load situation, better load balancing and resource utilization can be achieved, providing a more efficient load balancing effect, and improving the load balancing reliability of the hybrid cloud load balancing model.
Description
Technical Field
The application relates to the technical field of hybrid cloud load balancing, in particular to a method, a device, equipment and a medium for balancing multi-cloud resource load based on a GPT technology.
Background
Cloud computing is an important technology which is rapidly developed in recent years, and currently comprises four deployment modes of public cloud, private cloud, community cloud and hybrid cloud mixed by the three modes. Public cloud computing resources are abundant and low in investment, but security is risky. The security of private cloud data and the availability of the system are controlled by the private cloud data, so that the security is greatly improved, but the construction requires a large amount of investment. The community cloud may serve different units with the same demand. The hybrid cloud consists of two or more different types of cloud, and the smooth processing of inter-cloud data and programs is realized by combining the advantages of cloud computing of different types of cloud promotion. Considering that some users wish to have rich resources of public cloud and do not wish to deploy some private information on public cloud, the advantages of hybrid cloud are reflected. In the hybrid cloud scenario, as the number of network communications and information processing increases, the amount of tasks increases, and therefore, improving the efficiency of load balancing becomes a key issue of concern.
The current hybrid cloud load balancing method is divided into DNS load balancing, hardware-based load balancing, software-based load balancing and Yun Yuansheng load balancing. The load balancing based on the software is more flexible, the proportion of public cloud to private cloud can be dynamically adjusted, the expansion is easy, higher load capacity can be realized, and custom configuration and optimization can be carried out according to actual demands. In practical application, the load balancing based on software needs to be selected and optimized according to specific service requirements and technical conditions, meanwhile, factors of safety, performance, reliability and the like need to be considered, the advantages and disadvantages of various schemes are comprehensively evaluated, and the most suitable scheme is selected to realize the hybrid cloud load balancing. How to monitor cloud resources in real time, reasonably allocate and schedule resources of public cloud and private cloud according to load conditions, and efficiently evaluate and select the most suitable scheme to perform load balancing is an important problem for optimizing system resources and user services.
Common hybrid cloud load balancing algorithms include polling algorithms, least connection algorithms, and the like. Different algorithms exhibit different advantages in terms of load condition, performance requirements and expandability, and the same problems exist to different extents. The scholars propose to combine various task scheduling algorithms to achieve better effect and find the optimal path.
The existing load balancing method is difficult to process multi-source and fast-changing hybrid cloud loads when facing the hybrid cloud load balancing scene, an optimal hybrid cloud resource allocation scheme is found, and with the expansion of the hybrid cloud scene, the efficiency and the expandability of the two schemes are limited, so that challenges brought by load balancing in the hybrid cloud environment are difficult to meet. How to apply the deep learning model to adaptively process complex and changeable load modes in the mixed cloud environment and achieve a better load balancing effect is a problem to be solved.
Therefore, how to improve the load balancing reliability of the hybrid cloud load balancing model is a technical problem to be solved.
Disclosure of Invention
The application provides a method, a device, equipment and a storage medium for balancing multi-cloud resource load based on GPT technology, aiming at improving the load balancing reliability of a hybrid cloud load balancing model.
In a first aspect, the present application provides a method for balancing multi-cloud resource load based on GPT technology, where the method includes:
based on a pre-trained load balance monitoring model and current load data of a cloud server group, predicting load operation data of each cloud resource in the cloud server group to obtain predicted load data corresponding to each cloud resource;
calculating resource allocation weights of the cloud resources based on the predicted load data;
and carrying out resource allocation on the cloud resources based on the resource allocation weights corresponding to the cloud resources so as to realize load balance among the cloud resources.
In a second aspect, the present application further provides a device for balancing a multi-cloud resource load based on a GPT technology, where the device for balancing a multi-cloud resource load based on a GPT technology includes:
the load data prediction module is used for predicting load operation data of each cloud resource in the cloud server group based on a pre-trained load balance monitoring model and current load data of the cloud server group to obtain predicted load data corresponding to each cloud resource;
the weight calculation module is used for calculating the resource allocation weight of each cloud resource based on the predicted load data;
And the resource allocation module is used for allocating resources to the cloud resources based on the resource allocation weights corresponding to the cloud resources so as to realize load balance among the cloud resources.
In a third aspect, the present application further provides a computer device, where the computer device includes a processor, a memory, and a computer program stored on the memory and executable by the processor, where the computer program, when executed by the processor, implements the steps of a GPT technology-based multi-cloud resource load balancing method as described above.
In a fourth aspect, the present application further provides a computer readable storage medium, where a computer program is stored on the computer readable storage medium, where the computer program, when executed by a processor, implements the steps of the multi-cloud resource load balancing method based on GPT technology as described above.
The application provides a method, a device, equipment and a storage medium for balancing load of cloud resources based on a GPT technology, wherein the method comprises the steps of predicting load operation data of cloud resources in a cloud server group based on a pre-trained load balance monitoring model and current load data of the cloud server group to obtain predicted load data corresponding to the cloud resources; calculating resource allocation weights of the cloud resources based on the predicted load data; and carrying out resource allocation on the cloud resources based on the resource allocation weights corresponding to the cloud resources so as to realize load balance among the cloud resources. By means of the method, load operation data of cloud resources in the cloud server group are predicted according to current load data of the cloud server group through the load balance monitoring model, so that resource allocation weights of the cloud resources are calculated according to the predicted load data, resource share of the cloud resources is adjusted according to the resource allocation weights corresponding to the cloud resources, and load balance among the cloud resources is adjusted. The load of the cloud resources is monitored and predicted in real time, and the resources are distributed and scheduled according to the load condition, so that better load balancing and resource utilization rate are realized, a more efficient load balancing effect is provided, and the load balancing reliability of the hybrid cloud load balancing model is improved.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are some embodiments of the present application, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
Fig. 1 is a schematic structural diagram of an embodiment of a GPT-based hybrid cloud load balancing system provided in the present application;
FIG. 2 is a schematic flow chart of a resource monitoring module and model adjustment provided in the present application;
fig. 3 is a schematic flow chart of a second embodiment of a multi-cloud resource load balancing method based on the GPT technology provided in the present application;
fig. 4 is a schematic flow chart of a second embodiment of a multi-cloud resource load balancing method based on the GPT technology provided in the present application;
fig. 5 is a schematic flow chart of a third embodiment of a multi-cloud resource load balancing method based on the GPT technology provided in the present application;
FIG. 6 is a schematic flow chart of hybrid cloud load data collection, preprocessing and model training provided by the present application;
fig. 7 is a schematic structural diagram of a first embodiment of a GPT technology-based multi-cloud resource load balancing apparatus provided in the present application;
Fig. 8 is a schematic block diagram of a computer device according to an embodiment of the present application.
The realization, functional characteristics and advantages of the present application will be further described with reference to the embodiments, referring to the attached drawings.
Detailed Description
The following description of the embodiments of the present application will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are some, but not all, of the embodiments of the present application. All other embodiments, which can be made by one of ordinary skill in the art without undue burden from the present disclosure, are within the scope of the present disclosure.
The flow diagrams depicted in the figures are merely illustrative and not necessarily all of the elements and operations/steps are included or performed in the order described. For example, some operations/steps may be further divided, combined, or partially combined, so that the order of actual execution may be changed according to actual situations.
Some embodiments of the present application are described in detail below with reference to the accompanying drawings. The following embodiments and features of the embodiments may be combined with each other without conflict.
As shown in fig. 1, fig. 1 is a GPT-based hybrid cloud load balancing system provided in an embodiment of the present application, where the system includes a GPT-based flow control module, a management module, and a resource monitoring module.
The flow control module based on the GPT is used in the scenes such as network flow control, load balancing and the like.
The management module comprises modules such as a capacity reduction module, a capacity expansion module and the like. And receiving the demand condition of load balancing on resources, and distributing cloud resources according to the resource distribution instruction sent by the flow control module.
An embodiment of the resource monitoring module is shown in fig. 2, where the resource monitoring module is configured to monitor and manage a load condition of each cloud resource.
In one embodiment, the GPT-based flow control module includes three sub-modules: the system comprises a flow analysis and prediction module, an equalization configuration module and a data collection module.
In the mixed cloud service scene, the data collection module is used for collecting network traffic data and server state data, and the traffic analysis prediction module can analyze and model historical traffic data to predict network traffic conditions in a future period of time.
According to the result of the flow analysis and prediction module, the balanced configuration module dynamically adjusts the configuration of the load equalizer, so that the network flow can be more balanced and shared to different servers, and the performance and reliability of the network are improved.
The data collection module also records the data fed back by the resource monitoring module and is used for training and optimizing the flow analysis and prediction module and decision and control of the balanced configuration module. Meanwhile, the flow control module based on the GPT can also detect and report network faults and safety events, so that a system administrator can be helped to find and solve the problems in time, and the network safety and stable operation of enterprises are guaranteed.
Referring to fig. 3, fig. 3 is a flowchart of a first embodiment of a method for balancing multi-cloud resource load based on GPT technology provided in the present application.
As shown in fig. 3, the method for balancing the cloud resources load based on the GPT technology includes steps S101 to S103.
S101, predicting load operation data of each cloud resource in a cloud server group based on a pre-trained load balance monitoring model and current load data of the cloud server group, and obtaining predicted load data corresponding to each cloud resource;
in an embodiment, a pre-trained GPT model is used to predict future load conditions according to current load conditions such as CPU utilization, memory utilization, disk utilization, etc., to determine which cloud service providers or private cloud resources should be allocated more load.
In an embodiment, resource allocation is performed on different cloud service provider resources or private cloud resources according to the result of load prediction. If it is predicted that the load of a certain cloud resource will increase significantly, more resources should be allocated to that cloud resource. If it is predicted that the load of a certain cloud resource will drop significantly, the number of resources allocated to the cloud resource should be reduced appropriately, so as to achieve load balancing between different cloud resources.
S102, calculating the resource allocation weight of each cloud resource based on the predicted load data;
further, determining a weighting coefficient of each index data corresponding to each cloud resource in each cloud server group based on the current service scene type; and carrying out weighted calculation on each normalized index data corresponding to each cloud resource based on the weighting coefficient of each index data corresponding to each cloud resource to obtain the resource allocation weight corresponding to each cloud resource.
In one embodiment, after model training is completed, the cloud server farm is load balanced using the model. And setting the weight value of the server as a basis of load balancing, and distributing the traffic to the corresponding server. While using the original load balancing method (e.g., a polling algorithm or a weighted polling algorithm) as a comparison to facilitate evaluation of the model's effectiveness. The practicality and performance of the model are evaluated by comparing the load balancing effect, response time, error rate and other indicators of the two methods.
For example, assume that a server farm with 4 cloud servers is load balanced, a GPT model and an original polling algorithm are used for load balancing, and indexes such as response time and error rate of each server are recorded. The weight value for each server is calculated using the following formula:
W(i)＝W1*n1(i)+W2*n2(i)+W3*n3(i) (1)
wherein w1, w2 and w3 represent weight parameters of the CPU utilization, the memory utilization and the disk utilization, and n1 (i), n2 (i) and n3 (i) represent normalized values of the CPU utilization, the memory utilization and the disk utilization of the i-th server, respectively.
And S103, carrying out resource allocation on the cloud resources based on the resource allocation weights corresponding to the cloud resources so as to realize load balance among the cloud resources.
In an embodiment, the traffic may be distributed to the corresponding servers according to the weight values to achieve load balancing.
The embodiment provides a multi-cloud resource load balancing method based on a GPT technology, which predicts load operation data of cloud resources in a cloud server group according to current load data of the cloud server group through a load balancing monitoring model, so that resource allocation weights of the cloud resources are calculated according to the predicted load data, and resource share of the cloud resources is adjusted according to the resource allocation weights corresponding to the cloud resources to adjust load balancing among the cloud resources. The load of the cloud resources is monitored and predicted in real time, and the resources are distributed and scheduled according to the load condition, so that better load balancing and resource utilization rate are realized, a more efficient load balancing effect is provided, and the load balancing reliability of the hybrid cloud load balancing model is improved.
Referring to fig. 4, fig. 4 is a flow chart of a second embodiment of a method for balancing multi-cloud resource load based on GPT technology provided in the present application.
In this embodiment, based on the embodiment shown in fig. 3, after step S103, the method further includes:
s201, monitoring current load operation data corresponding to each monitoring index in each cloud resource based on preset monitoring indexes and data thresholds of the monitoring indexes;
in an embodiment, an initial monitoring index and a threshold are set, and the information of the cloud node resources in the monitoring system comprises information such as CPU utilization, memory utilization, disk utilization, network bandwidth, connection number and the like.
S202, when the current load operation data of the monitoring index exceeds the data threshold, adjusting the resource allocation weight of each cloud resource based on the prediction result of the load balance monitoring model and the load balance algorithm to obtain an adjusted resource allocation weight;
and S203, adjusting the distributed resources of the cloud resources based on the adjusted resource distribution weight so as to realize load balance among the cloud resources.
In an embodiment, after resource allocation, if an abnormality, i.e. too high or too low, is detected in the load of a certain cloud resource, the adjustment is performed according to the prediction result of the GPT model and the principle and parameters of the load balancing algorithm. If the GPT model predicts that the load of a cloud resource will increase beyond a threshold (where the threshold gets an empirical value during model training), then a corresponding scale of resources are added to the cloud resource in advance, including the number of instances deployed at the resource, expanding CPU/memory, adding additional cached nodes, etc. Otherwise, the reduction of the number of resources of the cloud resource is considered to fully utilize the idle resources. If the load is in a normal state, the performance and accuracy of the GPT model are continuously monitored.
In an embodiment, according to the load change condition after resource allocation, the load condition of each cloud resource is continuously monitored. If the load of a cloud resource is found to be too high or too low, the resource allocation scheme should be adjusted in time. The number of resources allocated to the cloud resources can be increased or reduced, the parameters of the load balancing algorithm can be replaced, the weight of the cloud resources can be increased or reduced, and the like.
It will be appreciated that monitoring load changes and timely adjustments are critical to achieving long-term stable load balancing. Meanwhile, the performance of the model needs to be monitored, if the prediction accuracy of the model is found to be reduced, new data should be timely used for model retraining, and the prediction capability of the model is improved. The automatic operation of the flow is realized, and a closed loop system is further formed, so that the dynamic balance of cloud resource loads is realized.
In an embodiment, according to the load change condition after the resource allocation, the load condition of public cloud and private cloud multi-resource is continuously monitored. The resource load monitoring formula is shown in the formula 2:
wherein L (t) represents timeIntegral load of t, L i (t) represents the load of a certain cloud resource at time t, L p (t) and L h (t) respectively representing the loads of public cloud and private cloud at time t, W i Weight representing a cloud, W p And W is h The weights of public clouds and private clouds are represented, respectively.
In one embodiment, when L (t) continues to be above a certain threshold for a period of time, then it is considered an abnormal allocation of resources.
In an embodiment, for a hybrid cloud resource anomaly, according to a predicted load level of a GPT model, increasing or decreasing the number of resources allocated to the cloud resource or replacing a load balancing algorithm parameter, a resource allocation adjustment formula is shown in formula 3:
ΔR i ＝α(L i (t)-L target ) (3)
wherein DeltaR i Representing resource allocation adjustment quantity, L, for cloud resource i i (t) represents the load of cloud resource i at time t, L target Representing the target load level, α is an adjustment factor.
In an embodiment, the method may also be improved by increasing or decreasing the weight of the cloud resource, where the weight adjustment formula is shown in formula 4:
W i_new ＝W i_old *(1+β*(L i (t)-L target )) (4)
wherein W is i_new Representing new weight of cloud resource, W i_old Represents the old weight of cloud resources, L target Is the target load level and β is the weight adjustment factor.
In an embodiment, if the load of a certain resource is higher, the weight of the resource is reduced, the use proportion of the resource is reduced, and the excessive use of the resource is avoided.
In an embodiment, cloud load information is monitored and collected in real time through a GPT model, complex load condition changes such as future trends of CPU utilization rate, memory utilization rate, disk utilization rate and the like can be predicted more accurately, resources are pre-allocated, and therefore a quicker and better load balancing effect is achieved.
In the embodiment, the actual load change condition is monitored in real time by training the special GPT resource monitoring model, and the resource allocation scheme is further dynamically adjusted according to the actual load change condition, so that flexible and dynamic management is provided for different hybrid cloud environments on the basis of ensuring the monitoring efficiency.
Referring to fig. 5, fig. 5 is a flow chart of a third embodiment of a method for balancing multi-cloud resource load based on GPT technology provided in the present application.
In this embodiment, based on the embodiment shown in fig. 3, before step S101, the method further includes:
s301, acquiring historical load data of the cloud server group;
in an embodiment, a Fluented data collector is used to collect load data of cloud service providers or private cloud resources, including indicators such as CPU utilization, memory utilization, disk utilization, and the like.
In one embodiment, for the case where load data cannot be provided, data is collected and processed through a custom script, and then a scripting language such as python, bash, etc. is used to communicate with the cloud resources through ssh. And the load data is stored in a MySQL database, so that subsequent processing and analysis are facilitated.
S302, based on a data preprocessing algorithm, performing data preprocessing on the historical load data to obtain a data feature vector;
In one embodiment, the data collection and preprocessing and model training embodiment is shown in FIG. 6.
Illustratively, assume that the model is used to load balance a set of cloud server clusters:
firstly, a Fluentd data collector is used for collecting load data of a cloud server group, wherein the load data comprises indexes such as CPU utilization rate, memory utilization rate, disk utilization rate and the like. And storing the data in a MySQL database, and cleaning the data by using an OpenRefine tool to ensure the accuracy and consistency of the data.
And secondly, carrying out minimum-maximum normalization processing on the data so as to facilitate subsequent feature extraction and model training. The HTTP traffic data provided by the load balancing dataset may also be directly utilized, three datasets, theLARDDataset, theCloudHarmonyDataset, theCERNETTrafficDataset being provided as references.
Finally, in the aspect of feature extraction, the CPU utilization rate, the memory utilization rate and the disk utilization rate are taken as features, and different weight parameters are set according to service requirements.
Further, based on a data cleaning algorithm, performing data cleaning on the historical load data to obtain at least one index data corresponding to each cloud resource in the cloud server; based on a data normalization algorithm, performing normalization processing on each index data corresponding to each cloud resource to obtain at least one piece of normalized index data corresponding to each cloud resource; and carrying out feature extraction on the normalized index data based on a feature extraction algorithm to obtain the data feature vector.
In one embodiment, the GPT model is first trained using historical data. In the training process, a cross entropy loss function and an Adam optimizer are used for model optimization, and parameters of the model are adjusted and optimized. And secondly, evaluating the model by using a cross verification method, calculating indexes such as accuracy, recall rate, precision, root mean square error and the like of the model, and adjusting and optimizing the model according to an evaluation result.
Preprocessing the collected data, wherein the data preprocessing mainly comprises the following three aspects: data cleaning, data normalization and feature extraction.
In an embodiment, data cleansing refers to performing preliminary data cleansing on inaccurate, incomplete and nonstandard data in collected load data in opendefined, and completing operations of data removal, null filling and outlier processing, so as to ensure data accuracy and consistency.
In one embodiment, data normalization refers to the performance and load metrics j of the data after cleaning, and calculates the minimum value min_j (as shown in equation 5) and the maximum value max_j (as shown in equation 6) for all servers i:
min j ＝min(x ij )(i＝1,2,...n)(5)
max j ＝max(x ij )(i＝1,2,...n)(6)
wherein x is ij Data representing the jth performance and load metrics of the ith server.
In one embodiment, the data of each server is subjected to a min-max normalization process, where the normalization formula is shown in formula 7:
x ij_norm ＝(x ij -min j )/(max j -min j )(7)
wherein x is ij_norm Normalized values representing the jth performance and load metrics of the ith server.
In an embodiment, the normalized value is calculated, and according to a certain weighting coefficient, the normalized value of each server i is weighted, so as to obtain a weight value of the server, where the weighted calculation formula is shown in formula 8:
wherein w is (1) 、w (2) .......w (n) Is the weighting coefficient of each performance and load index, x ij_norm Is the normalized value, w, of all performance and load metrics of the ith server i Is the weight value of the i-th server. And distributing the flow to the corresponding server according to the weight value of the server so as to facilitate subsequent data processing and analysis.
In one embodiment, feature extraction refers to the need to perform feature extraction on data to extract the most valuable information before the data is input into the GPT model. In load balancing, the CPU utilization, memory utilization, disk utilization, network bandwidth, connection number, response time, and error rate may be input as features into the model for training. And selecting specific indexes according to actual application scenes to perform feature extraction, so as to ensure that service requirements and performance requirements are met.
Wherein, the feature extraction formula is set as shown in the formula 9:
feature=α+β×memory utilization
+gamma disk utilization +delta network bandwidth
+ε number of connections +ζ response time +η error rate (9)
Wherein, alpha, beta, gamma, delta, epsilon, zeta and eta are weight parameters, and the range is between 0 and 1.
In an embodiment, the priority may be set in advance for different traffic scenarios. In a scenario where traffic demand is prominent, the weight occupied by the response time can be raised if the response time is critical to the traffic. If the error rate is more critical, the weight of eta is increased. In terms of performance, the weight of delta can be reduced under the condition of limited network bandwidth, the weight of gamma is reduced under the condition of limited disk, and the connection number weight is adjusted in the period of high and low peaks. Therefore, the feature vector is ensured to be more in line with the requirements of the current application scene, and the accuracy and the practicability of the GPT model are improved.
And S303, training a GPT model based on the data feature vector to obtain the load balance monitoring model.
Further, based on a preset data acquisition period, acquiring load operation data of the cloud server group in the preset data acquisition period; and retraining the load balance monitoring model based on the load operation data so as to realize model optimization of the load balance monitoring model.
In one embodiment, if the model is found to have degraded prediction accuracy, the model may be retrained using data collected during actual operation to improve its prediction ability.
The method is characterized in that a day is set as a time node, the newly acquired data is used as training data to update a load prediction model, and the performance of the model is evaluated by using methods such as cross validation and the like, and the model is optimized and adjusted to adapt to the change of the data and the change of business requirements.
Further, load prediction data of the load balance monitoring model on each cloud resource at a preset time and actual load data of each cloud resource at the preset time are obtained; calculating model prediction accuracy of the load balance detection model based on the load prediction data and the actual load data; when the model prediction precision is lower than a preset precision threshold, acquiring load operation data of the cloud server group in the preset data acquisition period; and retraining the load balance monitoring model based on the load operation data so as to realize model optimization of the load balance monitoring model.
In an embodiment, the prediction accuracy of the monitoring model can be selected to monitor the performance of the model, and when the prediction accuracy of the model is reduced, new data can be timely used for model retraining, so that the prediction capability of the model is improved. The model performance monitoring formula is shown as 10:
Wherein P is model (t) represents the prediction accuracy of the model at time t, D (t) represents the new data collected at time t, M GPT Representing the model, T represents the training algorithm.
In one embodiment, the training data may be trained using historical data or simulation data, and the model is parameter adjusted and optimized to improve the performance and generalization ability of the model. The loss function formula during training uses a cross entropy loss formula as shown in equation 11:
wherein y is true Is the true load value, y pred Is a model predictive value, and N is the number of training data.
In one embodiment, based on the loss function, an Adam optimizer is used to update parameters, thereby continuously reducing the loss value and optimizing the learning rate alpha of the process study The formula is shown in formula 12:
wherein alpha is initial For initial learning rate, decay rate For the attenuation rate, global step Representing the total training steps, decay steps In order to attenuate the frequency, the model update amplitude and the loss entropy dropping speed are affected, so that the aim of optimization is fulfilled.
In one embodiment, the model is evaluated for Accuracy (Accuracy), recall (Recall), precision (Precision), and root mean square error (root mean square error) using a cross-validation method, and the model is compared for correctness. And further adjusting parameters of the model, adjusting the structure and the parameters of the model, performing trial and error and optimization according to actual conditions, and monitoring and maintaining the model to ensure the stability and the reliability of the model.
In this embodiment, a proprietary GPT flow control model is trained, and load influencing factors and correlations under various hybrid cloud environments are learned and quantified, so that flow control can be more comprehensively performed according to load conditions under complex environments, and rapid changes of load conditions under hybrid multi-cloud environments are adapted. Compared with the fixed technical scheme, the method has stronger adaptability and expansibility. And processing and storing a large amount of complex load data through a real-time data collection module, uploading the acquired actual load data to update a training database by taking one day as a periodic node, and marking the special load resource balancing condition. When the GPT model is applied, new learning training is performed on special conditions, and then the load prediction model is updated, so that the accuracy and the robustness of the model and the processing capacity of the special load conditions are further improved.
Referring to fig. 7, fig. 7 is a schematic structural diagram of a first embodiment of a GPT technology-based cloudy resource load balancing device provided in the present application, where the GPT technology-based cloudy resource load balancing device is configured to execute the foregoing GPT technology-based cloudy resource load balancing method. The cloud resource load balancing device based on the GPT technology can be configured in a server.
As shown in fig. 7, the multi-cloud resource load balancing apparatus 400 based on the GPT technology includes: load data prediction module 401, weight calculation module 402, and resource allocation module 403.
The load data prediction module 401 is configured to predict load operation data of each cloud resource in the cloud server group based on a pre-trained load balance monitoring model and current load data of the cloud server group, so as to obtain predicted load data corresponding to each cloud resource;
a weight calculation module 402, configured to calculate a resource allocation weight of each cloud resource based on the predicted load data;
and the resource allocation module 403 is configured to allocate resources to the cloud resources based on the resource allocation weights corresponding to the cloud resources, so as to implement load balancing between the cloud resources.
In an embodiment, the multi-cloud resource load balancing apparatus 400 based on GPT technology further includes a resource adjustment module, including:
the current load data monitoring unit is used for monitoring current load operation data corresponding to each monitoring index in each cloud resource based on preset monitoring indexes and data thresholds of the monitoring indexes;
the adjusting resource allocation weight determining unit is used for adjusting the resource allocation weight of each cloud resource based on the prediction result of the load balance monitoring model and the load balance algorithm when the current load operation data of the monitoring index exceeds the data threshold value, so as to obtain an adjusting resource allocation weight;
And the resource adjusting unit is used for adjusting the distributed resources of the cloud resources based on the adjusted resource distribution weight so as to realize load balance among the cloud resources.
In an embodiment, the multi-cloud resource load balancing apparatus 400 based on the GPT technology further includes a model training module, including:
a historical load data acquisition unit, configured to acquire historical load data of the cloud server group;
the data preprocessing unit is used for preprocessing the data of the historical load data based on a data preprocessing algorithm to obtain a data feature vector;
and the model training unit is used for training the GPT model based on the data feature vector to obtain the load balance monitoring model.
In an embodiment, the multi-cloud resource load balancing apparatus 400 based on the GPT technology further includes a model optimization module, including:
the load operation data acquisition unit is used for acquiring load operation data of the cloud server group in a preset data acquisition period based on the preset data acquisition period;
and the model optimization first unit is used for retraining the load balance monitoring model based on the load operation data so as to realize model optimization on the load balance monitoring model.
In an embodiment, the model optimization module further comprises:
the load data acquisition unit is used for acquiring load prediction data of the load balance monitoring model on each cloud resource at a preset moment and actual load data of each cloud resource at the preset moment;
a model prediction accuracy calculation unit configured to calculate a model prediction accuracy of the load balance detection model based on the load prediction data and the actual load data;
the data period acquisition unit is used for acquiring load operation data of the cloud server group in the preset data acquisition period when the model prediction precision is lower than a preset precision threshold;
and the model optimization second unit is used for retraining the load balance monitoring model based on the load operation data so as to realize model optimization on the load balance monitoring model.
In an embodiment, the data preprocessing unit includes:
the data cleaning unit is used for performing data cleaning on the historical load data based on a data cleaning algorithm to obtain at least one index data corresponding to each cloud resource in the cloud server;
the data normalization unit is used for carrying out normalization processing on the index data corresponding to each cloud resource based on a data normalization algorithm to obtain at least one piece of normalized index data corresponding to each cloud resource;
And the feature extraction unit is used for carrying out feature extraction on the normalized index data based on a feature extraction algorithm to obtain the data feature vector.
In one embodiment, the weight calculation module includes:
the weighting coefficient determining unit is used for determining the weighting coefficient of each index data corresponding to each cloud resource in each cloud server group based on the current service scene type;
the resource allocation weight obtaining unit is used for carrying out weighted calculation on the normalized index data corresponding to the cloud resources based on the weighting coefficients of the index data corresponding to the cloud resources to obtain the resource allocation weight corresponding to the cloud resources.
It should be noted that, for convenience and brevity of description, a person skilled in the art may clearly understand that, for the specific working process of the above-described apparatus and each module, reference may be made to the corresponding process in the foregoing embodiment of the multi-cloud resource load balancing method based on the GPT technology, which is not described herein again.
The apparatus provided by the above embodiments may be implemented in the form of a computer program which may be run on a computer device as shown in fig. 8.
Referring to fig. 8, fig. 8 is a schematic block diagram of a computer device according to an embodiment of the present application. The computer device may be a server.
With reference to FIG. 8, the computer device includes a processor, memory, and a network interface connected by a system bus, where the memory may include a non-volatile storage medium and an internal memory.
The non-volatile storage medium may store an operating system and a computer program. The computer program comprises program instructions which, when executed, cause the processor to perform any of a number of cloud resource load balancing methods based on GPT techniques.
The processor is used to provide computing and control capabilities to support the operation of the entire computer device.
The internal memory provides an environment for the execution of a computer program in a non-volatile storage medium, which when executed by a processor, causes the processor to perform any of a number of cloud resource load balancing methods based on GPT technology.
The network interface is used for network communication such as transmitting assigned tasks and the like. It will be appreciated by those skilled in the art that the structure shown in fig. 8 is merely a block diagram of some of the structures associated with the present application and is not limiting of the computer device to which the present application may be applied, and that a particular computer device may include more or fewer components than shown, or may combine certain components, or have a different arrangement of components.
It should be appreciated that the processor may be a central processing unit (CentralProcessingUnit, CPU), but may also be other general purpose processors, digital signal processors (DigitalSignalProcessor, DSP), application specific integrated circuits (ApplicationSpecificIntegratedCircuit, ASIC), field programmable gate arrays (Field-ProgrammableGateArray, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or the like. Wherein the general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
Wherein in one embodiment the processor is configured to run a computer program stored in the memory to implement the steps of:
based on a pre-trained load balance monitoring model and current load data of a cloud server group, predicting load operation data of each cloud resource in the cloud server group to obtain predicted load data corresponding to each cloud resource;
calculating resource allocation weights of the cloud resources based on the predicted load data;
and carrying out resource allocation on the cloud resources based on the resource allocation weights corresponding to the cloud resources so as to realize load balance among the cloud resources.
In an embodiment, after implementing the resource allocation weight corresponding to each cloud resource, the processor is further configured to implement:
monitoring current load operation data corresponding to each monitoring index in each cloud resource based on preset monitoring indexes and data thresholds of the monitoring indexes;
when the current load operation data of the monitoring index exceeds the data threshold, adjusting the resource allocation weight of each cloud resource based on the prediction result of the load balance monitoring model and the load balance algorithm to obtain an adjusted resource allocation weight;
and adjusting the allocation resources of the cloud resources based on the adjustment resource allocation weights so as to realize load balance among the cloud resources.
In an embodiment, before implementing the pre-training-based load balance monitoring model and current load data of a cloud server group, the processor predicts load operation data of each cloud resource in the cloud server group to obtain predicted load data corresponding to each cloud resource, the processor is further configured to implement:
Acquiring historical load data of the cloud server group;
based on a data preprocessing algorithm, carrying out data preprocessing on the historical load data to obtain a data feature vector;
and training the GPT model based on the data feature vector to obtain the load balance monitoring model.
In an embodiment, after implementing the resource allocation weight corresponding to each cloud resource, the processor is further configured to implement:
based on a preset data acquisition period, acquiring load operation data of the cloud server group in the preset data acquisition period;
and retraining the load balance monitoring model based on the load operation data so as to realize model optimization of the load balance monitoring model.
In an embodiment, after implementing the resource allocation weight corresponding to each cloud resource, the processor is further configured to implement:
load prediction data of the load balance monitoring model on each cloud resource at a preset moment and actual load data of each cloud resource at the preset moment are obtained;
Calculating model prediction accuracy of the load balance detection model based on the load prediction data and the actual load data;
when the model prediction precision is lower than a preset precision threshold, acquiring load operation data of the cloud server group in the preset data acquisition period;
and retraining the load balance monitoring model based on the load operation data so as to realize model optimization of the load balance monitoring model.
In an embodiment, when implementing the data preprocessing algorithm, the processor performs data preprocessing on the historical load data to obtain a data feature vector, the processor is configured to implement:
based on a data cleaning algorithm, carrying out data cleaning on the historical load data to obtain at least one index data corresponding to each cloud resource in the cloud server;
based on a data normalization algorithm, performing normalization processing on each index data corresponding to each cloud resource to obtain at least one piece of normalized index data corresponding to each cloud resource;
and carrying out feature extraction on the normalized index data based on a feature extraction algorithm to obtain the data feature vector.
In an embodiment, when implementing the calculating, based on the predicted load data, a resource allocation weight of each cloud resource, the processor is configured to implement:
determining a weighting coefficient of each index data corresponding to each cloud resource in each cloud server group based on the current service scene type;
and carrying out weighted calculation on each normalized index data corresponding to each cloud resource based on the weighting coefficient of each index data corresponding to each cloud resource to obtain the resource allocation weight corresponding to each cloud resource.
The embodiment of the application also provides a computer readable storage medium, wherein the computer readable storage medium stores a computer program, the computer program comprises program instructions, and the processor executes the program instructions to realize any cloud resource load balancing method based on the GPT technology.
The computer readable storage medium may be an internal storage unit of the computer device according to the foregoing embodiment, for example, a hard disk or a memory of the computer device. The computer readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk provided on the computer device, a smart memory card (SmartMediaCard, SMC), a secure digital (SecureDigital, SD) card, a flash memory card (FlashCard), etc.
While the invention has been described with reference to certain preferred embodiments, it will be understood by those skilled in the art that various changes and substitutions of equivalents may be made and equivalents will be apparent to those skilled in the art without departing from the scope of the invention. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.
Claims (10)
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311697248.3A CN117880291A (en) | 2023-12-12 | 2023-12-12 | Multi-cloud resource load balancing method, device, equipment and medium based on GPT technology |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311697248.3A CN117880291A (en) | 2023-12-12 | 2023-12-12 | Multi-cloud resource load balancing method, device, equipment and medium based on GPT technology |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117880291A true CN117880291A (en) | 2024-04-12 |
Family
ID=90590760
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311697248.3A Pending CN117880291A (en) | 2023-12-12 | 2023-12-12 | Multi-cloud resource load balancing method, device, equipment and medium based on GPT technology |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117880291A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118394592A (en) * | 2024-04-16 | 2024-07-26 | 广州视声智能股份有限公司 | A Paas platform based on cloud computing |
| CN118449875A (en) * | 2024-07-08 | 2024-08-06 | 江西省气象数据中心(江西省气象档案馆) | A self-adaptive cloud server monitoring data collection method, device and cloud server |
| CN119883663A (en) * | 2025-03-31 | 2025-04-25 | 湖南财政经济学院 | Dynamic resource allocation and load balancing method in cloud computing environment |
-
2023
- 2023-12-12 CN CN202311697248.3A patent/CN117880291A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118394592A (en) * | 2024-04-16 | 2024-07-26 | 广州视声智能股份有限公司 | A Paas platform based on cloud computing |
| CN118449875A (en) * | 2024-07-08 | 2024-08-06 | 江西省气象数据中心(江西省气象档案馆) | A self-adaptive cloud server monitoring data collection method, device and cloud server |
| CN119883663A (en) * | 2025-03-31 | 2025-04-25 | 湖南财政经济学院 | Dynamic resource allocation and load balancing method in cloud computing environment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN117997906B (en) | Node computing resource allocation method, network switching subsystem and intelligent computing platform | |
| CN117880291A (en) | Multi-cloud resource load balancing method, device, equipment and medium based on GPT technology | |
| CN112486687B (en) | Cloud platform workload prediction method based on multitask learning time sequence | |
| CN108733509A (en) | Method and system for data to be backed up and restored in group system | |
| CN112398700B (en) | Service degradation method and device, storage medium and computer equipment | |
| CN117972367B (en) | Data storage prediction method, data storage subsystem and intelligent computing platform | |
| US20230238801A1 (en) | Dynamic hosting capacity analysis framework for distribution system planning | |
| CN102413186B (en) | Resource scheduling method and device based on private cloud computing, and cloud management server | |
| CN114217974A (en) | Resource management method and system in cloud computing environment | |
| CN113902116A (en) | Deep learning model-oriented reasoning batch processing optimization method and system | |
| KR101630125B1 (en) | Method for resource provisioning in cloud computing resource management system | |
| Shariffdeen et al. | Adaptive workload prediction for proactive auto scaling in PaaS systems | |
| CN118193211A (en) | Distributed database cluster resource management method, apparatus, device, storage medium and program product | |
| WO2020206699A1 (en) | Predicting virtual machine allocation failures on server node clusters | |
| CN104932898B (en) | A kind of component selection method to be increased based on improvement multi-objective particle | |
| CN119294444A (en) | Transformer large model training method based on cloud-edge collaboration | |
| CN119383675B (en) | Dynamic bandwidth allocation method, system, electronic equipment and medium based on large model | |
| CN116827951A (en) | Availability prediction-based edge cloud cluster task scheduling method and device | |
| CN119645622A (en) | Dynamic load balancing system and method for multi-cluster intelligent expansion | |
| CN117354310A (en) | Node selection method, device, equipment and medium based on task collaboration | |
| CN118245205A (en) | Hybrid multi-cloud resource scheduling method based on GPT technology and genetic optimization | |
| CN117827617A (en) | Container cloud resource prediction method based on ARIMA-LSTM | |
| CN110971468A (en) | A Delayed Copy Incremental Container Checkpoint Processing Method Based on Dirty Page Prediction | |
| WO2023057955A1 (en) | Anomaly-aware cloud resource management system receiving external information, and including short- and long-term resource planning | |
| CN115562841A (en) | Cloud video service self-adaptive resource scheduling system and method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |