A data backup method based on cloud computing

Abstract

The present invention discloses a data backup method based on cloud computing, which relates to the field of data backup technology. The method includes: obtaining multiple data backup task queues of multiple cloud computing processors; traversing multiple data backup task queues to identify data backup features and obtain multiple data backup feature set sequences; obtaining an execution log set; identifying cloud computing processor loads and obtaining multiple cloud computing processor load tolerance intervals; identifying cloud computing processors to be deployed; mapping and matching cloud computing processors; and using multiple cloud computing processors to perform data backup according to multiple updated data backup task queues. The present invention solves the technical problems of the prior art in that the data backup task features are not accurately grasped, the cloud computing processor load assessment is inaccurate, and the task and processor resource allocation is unreasonable, and achieves the technical effect of effectively improving the data backup efficiency, resource utilization and stability of the cloud computing system.

Description

Technical Field

[0001] The present invention relates to the technical field of data backup, and in particular to a data backup method based on cloud computing. Background Art

[0002] With the rapid development of information technology, the amount of data has shown explosive growth, and the importance of data backup in enterprises and various organizations has become increasingly prominent. The rise of cloud computing technology has provided powerful computing and storage capabilities for data backup, but it has also brought a series of challenges. In traditional data backup methods, there is often a lack of refined management of data backup tasks. The task feature identification is not accurate enough, and it is impossible to deeply understand the key information such as the data size, data format, and priority of each task, resulting in a lack of targeted task scheduling. The load management of cloud computing processors is also insufficient. It is difficult to accurately evaluate the working status of the processor under different task loads, and its load tolerance range cannot be dynamically determined, which easily causes some processors to be overloaded while other processor resources are idle. This unbalanced load distribution not only reduces resource utilization, but may also cause task execution delays and affect the timeliness of data backup.

[0003] The existing technology has technical problems such as inaccurate grasp of data backup task characteristics, inaccurate cloud computing processor load assessment, and unreasonable allocation of tasks and processor resources. Summary of the invention

[0004] The present application provides a data backup method based on cloud computing, which is used to solve the technical problems in the prior art of inaccurate grasp of data backup task characteristics, inaccurate cloud computing processor load assessment, and unreasonable allocation of tasks and processor resources.

[0005] In view of the above problems, the present application provides a data backup method based on cloud computing, the method comprising:

[0006] Acquire multiple data backup task queues of multiple cloud computing processors; traverse the multiple data backup task queues to perform data backup feature identification to obtain multiple data backup feature set sequences; acquire multiple historical cloud computing processor execution log sets of the multiple cloud computing processors in a preset historical window; perform cloud computing processor load identification on the multiple historical cloud computing processor execution log sets to obtain multiple cloud computing processor load tolerance intervals; identify cloud computing processors to be deployed according to the multiple data backup feature set sequences and the multiple cloud computing processor load tolerance intervals to obtain multiple high-load cloud computing processors and multiple low-load cloud computing processors, wherein each high-load cloud computing processor includes a subsequence of a data backup feature set to be called out, and each low-load cloud computing processor includes a redundant load tolerance interval; map and match the multiple high-load cloud computing processors and the multiple low-load cloud computing processors according to the multiple subsequences of the data backup feature set to be called out and the multiple redundant load tolerance intervals to obtain multiple deployment mapping groups; map and adjust the multiple subsequences of the data backup feature set to be called out based on the multiple deployment mapping groups, and update the multiple data backup task queues according to the adjustment results to obtain multiple updated data backup task queues; and use the multiple cloud computing processors to perform data backup according to the multiple updated data backup task queues.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0008] Acquire multiple data backup task queues of multiple cloud computing processors; traverse the multiple data backup task queues to identify data backup features and obtain multiple data backup feature set sequences; acquire multiple historical cloud computing processor execution log sets of the multiple cloud computing processors in a preset historical window; identify the cloud computing processor loads of the multiple historical cloud computing processor execution log sets and obtain multiple cloud computing processor load tolerance intervals; identify the cloud computing processors to be deployed and obtain multiple high-load cloud computing processors and multiple low-load cloud computing processors; map and match the multiple high-load cloud computing processors and the multiple low-load cloud computing processors to obtain multiple deployment mapping groups; map and adjust the multiple data backup feature set subsequences to be called out based on the multiple deployment mapping groups, and update the multiple data backup task queues according to the adjustment results to obtain multiple updated data backup task queues; use the multiple cloud computing processors to perform data backup according to the multiple updated data backup task queues. The technical effect of effectively improving the data backup efficiency, resource utilization and stability of the cloud computing system is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0010] Figure 1 A schematic diagram of a data backup method based on cloud computing provided in an embodiment of the present application;

[0011] Figure 2 A schematic diagram of a flow chart of obtaining load tolerance intervals of multiple cloud computing processors in a cloud computing-based data backup method provided in an embodiment of the present application. DETAILED DESCRIPTION

[0012] The present application provides a data backup method based on cloud computing, which is used to solve the technical problems in the prior art of inaccurate grasp of data backup task characteristics, inaccurate cloud computing processor load assessment, and unreasonable allocation of tasks and processor resources.

[0013] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0014] Examples, such as Figure 1 As shown, the present application provides a data backup method based on cloud computing, the method comprising:

[0015] Step S100: Acquire multiple data backup task queues of multiple cloud computing processors.

[0016] Specifically, through the deep integration interface with the cloud computing platform, a communication connection with multiple cloud computing processors is established. This connection ensures that the data backup task related information in the task management module responsible for each cloud computing processor can be obtained. For each cloud computing processor, its task queue management system stores a large number of data backup tasks waiting to be executed. These tasks come from a wide range of sources, from different enterprise users, different business systems or different applications. Send a special data acquisition request instruction to each cloud computing processor. These instructions follow the standard communication protocol of the cloud computing platform to ensure that the intention of obtaining the task queue is accurately conveyed. After receiving the request, the cloud computing processor will perform permission verification and data encapsulation processing on its own task queue. Permission verification ensures that the requester has the legal authority to obtain the task queue information and prevents unauthorized access. Data encapsulation is to organize the task information in the task queue according to a predetermined data format, including key information such as the unique identifier of the task, the identifier of the user or system to which the task belongs, the creation time of the task, the priority of the task, and the data storage path involved in the task. After the above processing, the cloud computing processor will return the packaged multiple data backup task queue information to the system one by one through a secure data transmission channel. When receiving these data backup task queues, it will perform integrity checks and data verification to ensure that the received information is accurate, complete and has not been tampered with. Once all the task queues of the cloud computing processors are successfully acquired and passed the verification, these task queues will be integrated into the local cache of the system as the basic data for subsequent data backup feature identification and task scheduling adjustment, laying a solid foundation for the smooth implementation of the entire data backup process and ensuring comprehensive control and effective management of diversified data backup tasks in the cloud computing environment.

[0017] Step S200: traverse the multiple data backup task queues to perform data backup feature identification to obtain multiple data backup feature set sequences.

[0018] Specifically, a deep traversal operation is performed on the obtained multiple data backup task queues. First, starting from the first data backup task queue, each data backup task in the queue is carefully analyzed one by one. For each task, a specially constructed data backup feature recognition program is started, which uses a series of complex algorithms and models to mine the key features in the task. It will parse the task description information and identify the data types involved in the task, such as structured data (database records, etc.), unstructured data (documents, images, audio, etc.), and semi-structured data (XML, JSON files, etc.), because different types of data have different backup methods and resource requirements. At the same time, the program will extract the priority tag of the task. This information is crucial for determining the execution order of the tasks in the future. High-priority tasks need to be allocated more resources to ensure that the backup is completed quickly, while low-priority tasks can be processed when resources are relatively abundant. In addition, the data volume of the task is also obtained, which directly affects the time and storage space required for backup. The data storage location and storage format of the task are also key identification objects. They determine the way data is read and transmitted. For example, data stored on the local disk and data stored in the cloud distributed storage system have different access speeds and transmission protocols. The program will also identify the backup frequency requirements of the task, such as daily backup, weekly backup or monthly backup, so as to arrange the backup plan reasonably. After completing the identification of all features of a task, these features are combined into a data backup feature set, which comprehensively describes the characteristics of this task. With the completion of the traversal of all tasks in the first task queue, a data backup feature set sequence is obtained, which reflects the feature distribution of all tasks in the task queue. Then, the other data backup task queues are traversed in the same way, and the above data backup feature identification process is repeated to generate a corresponding data backup feature set sequence for each task queue. Finally, multiple data backup feature set sequences are obtained, which constitute a comprehensive description of the data backup task characteristics in the entire cloud computing environment, providing a detailed and accurate basis for the subsequent accurate task scheduling and resource allocation according to task characteristics, ensuring that the data backup process can be carried out efficiently and orderly to meet the diverse needs of different tasks.

[0019] Step S300: Acquire a plurality of historical cloud computing processor execution log sets of the plurality of cloud computing processors in a preset historical window.

[0020] Specifically, the time range of the preset historical window is determined. This time range is set according to the requirements of performance analysis and task scheduling optimization of cloud computing processors, such as the past week, month or a specific time period. Then, for each cloud computing processor, a special log acquisition request instruction is sent to its log management system, and the instruction clearly specifies the historical time window range to be obtained. After receiving the request, the cloud computing processor will start the log retrieval program, which locates and filters the execution log records that meet the time window requirements in the local storage or distributed storage system of the processor. These execution log records contain rich information, such as the timestamps when the processor processes various tasks, and records the specific time when the task starts and completes processing. These timestamps can be used to calculate the task processing time and then analyze the processing efficiency of the processor. The log also records in detail the resource usage of the processor during the execution of the task, including CPU usage, memory usage, disk I / O operation frequency, and network bandwidth usage. These resource usage data can intuitively reflect the resource consumption mode of the processor under different task loads, and provide key data for evaluating the load capacity of the processor. At the same time, the execution log will also record the result status of the task execution, such as whether the task was successfully completed, whether an error occurred, and the type and cause of the error, which helps to analyze the stability and reliability of task execution. After each cloud computing processor completes the retrieval and screening of its own historical execution logs, it will package the execution log collection that meets the preset historical window requirements to ensure the integrity and accuracy of the data. Then, these log collections are returned to the system through a secure and reliable data transmission channel. When receiving these historical cloud computing processor execution log collections from multiple cloud computing processors, strict data verification and integrity checks will be performed to ensure that each log record is accurate and not missing. Once the log collections of all processors are successfully obtained and verified, these log collections are integrated and stored by the system as an important data basis for subsequent cloud computing processor load identification and task scheduling optimization, providing strong support for a comprehensive understanding of the processor's past performance and optimizing future task allocation.

[0021] Step S400: performing cloud computing processor load identification on the plurality of historical cloud computing processor execution log sets to obtain a plurality of cloud computing processor load tolerance intervals.

[0022] Specifically, a comprehensive and in-depth analysis is carried out on the obtained multiple historical cloud computing processor execution log sets. For each log set, a preliminary screening is carried out with the preset processing efficiency range as the index, and the log records with processing efficiency in different ranges are classified and sorted to form multiple historical screening cloud computing processor execution log sets. This step helps to focus on the operating status of the processor at different efficiency levels, laying the foundation for the subsequent accurate identification of load conditions. Then, with the load as the key index, multiple historical screening load sets are accurately extracted from the historical screening cloud computing processor execution log set. These sets contain detailed load data of the processor during the execution of various tasks, such as changes in key indicators such as CPU load value, memory load, and disk I / O load. Subsequently, each historical screening load set is traversed for intensive search. First, all load data in the set are processed by average, and multiple historical screening load means are obtained. These means can reflect the average status of the load level represented by the set. Taking these historical screening load means as the starting center point, a specially designed center update iteration model is called to iteratively update the starting center point. The model uses complex algorithm logic to continuously adjust the center point position according to the distribution of load data in the set until the preset number of times is met, thereby obtaining multiple target center points that better represent the load concentration trend. The historical screening loads corresponding to these target center points are determined as multiple central historical screening loads, which are key reference points in the load distribution. Finally, combined with the pre-set load tolerance radius, multiple cloud computing processor load tolerance intervals are constructed with the central historical screening load as the center. These intervals clearly define the reasonable range of load that each cloud computing processor can withstand under normal operating conditions, providing a scientific and quantitative basis for judging whether the processor is in a high-load or low-load state. At the same time, it also provides an important reference standard for the subsequent identification of cloud computing processors to be deployed and task scheduling adjustments, ensuring the reasonable allocation and efficient use of cloud computing resources.

[0023] Step S500: Identify the cloud computing processors to be deployed based on the multiple data backup feature set sequences and the multiple cloud computing processor load tolerance intervals to obtain multiple high-load cloud computing processors and multiple low-load cloud computing processors, wherein each high-load cloud computing processor includes a subsequence of the data backup feature set to be deployed, and each low-load cloud computing processor includes a redundant load tolerance interval.

[0024] Specifically, a comprehensive comparison and analysis is performed on the multiple data backup feature set sequences obtained previously and the load tolerance intervals of multiple cloud computing processors. For each feature set of a data backup task, key elements such as data size, data type, and backup priority are carefully considered. For example, for tasks with large data volume and high backup priority, the required computing resources and storage resources are relatively large. Then, by comparing the load tolerance interval of the cloud computing processor, it is determined whether the current load state of each processor can meet the needs of the task. If a processor's resource occupancy is close to or exceeds the upper limit of the load tolerance interval when executing an existing task, and assigning a new task to it may cause overall performance degradation or task execution delay, then the processor is identified as a high-load cloud computing processor. For these high-load cloud computing processors, the data backup features in their pending task queues are further analyzed, and the task features that can be transferred to other processors are extracted to form a subsequence of the data backup feature set to be transferred out. These subsequences contain task features such as relatively low priority, small data volume, or low real-time requirements, so that resources can be allocated more reasonably in the subsequent deployment process.

[0025] On the contrary, if the current load of the processor is far lower than the upper limit of its load tolerance interval, and there are enough redundant resources to process more tasks, that is, its load is in the lower part of the load tolerance interval, then the processor is identified as a low-load cloud computing processor. For low-load cloud computing processors, calculate its remaining load capacity relative to the upper limit of the load tolerance interval, that is, the redundant load tolerance interval. This interval indicates the range of capabilities of the processor to undertake additional tasks without affecting performance, providing a quantitative basis for the subsequent acceptance of tasks transferred from high-load processors. Through such a detailed analysis and identification process, it is possible to accurately distinguish between multiple high-load cloud computing processors and multiple low-load cloud computing processors, providing a clear target object for subsequent mapping matching and task scheduling adjustment, ensuring that cloud computing resources can be optimally configured between processors in different load states, improving the overall data backup efficiency, and ensuring the smooth execution of data backup tasks.

[0026] Step S600: mapping and matching the multiple high-load cloud computing processors and the multiple low-load cloud computing processors according to multiple to-be-called data backup feature set subsequences and multiple redundant load tolerance intervals to obtain multiple deployment mapping groups.

[0027] Specifically, a greedy algorithm is used to perform mapping and matching operations. First, for each high-load cloud computing processor, the subsequence of its data backup feature set to be called out is analyzed to extract key features, such as the amount of data, priority, and required computing resources of the task, and the corresponding weights are assigned to form a task feature vector. At the same time, the redundant load tolerance interval of the low-load cloud computing processor is quantitatively evaluated to determine its ability indicators to undertake tasks, such as the remaining CPU computing power, memory space, and network bandwidth, and to construct a load capacity vector. Then, starting from the first high-load cloud computing processor, all low-load cloud computing processors are traversed to calculate the matching degree between the task feature vector of the high-load processor and the load capacity vector of each low-load processor. The matching degree calculation comprehensively considers the difference between the task requirements and the processor capabilities, for example, by calculating the weighted distance between the two vectors (the smaller the distance, the higher the matching degree), and selects the low-load cloud computing processor with the highest matching degree as the preliminary matching object. Then, check whether the load of the low-load cloud computing processor after undertaking this task is still within a reasonable redundant load tolerance interval. If satisfied, the pair of high-load and low-load cloud computing processors are combined into a deployment mapping group, and the remaining load capacity indicator of the low-load processor is updated. If not satisfied, the low-load processor is skipped, and the next low-load processor with the second highest matching degree and meeting the load requirements is searched until a suitable match is found or all low-load processors are traversed. Repeat the above process, match each high-load cloud computing processor in turn, and continuously build a deployment mapping group. Throughout the process, the current local optimal matching selection is always taken as the principle, and the task deployment of all high-load processors is gradually completed. Finally, multiple deployment mapping groups are obtained, and the reasonable transfer of high-load processor tasks to low-load processors is realized, the utilization efficiency of the overall cloud computing resources is optimized, and the execution effect of data backup tasks is improved.

[0028] Step S700: mapping and adjusting the plurality of to-be-called data backup feature set subsequences based on the plurality of deployment mapping groups, and updating the plurality of data backup task queues according to the adjustment results to obtain a plurality of updated data backup task queues.

[0029] Specifically, operations are carried out according to the mapping relationships determined by multiple deployment mapping groups. For each high-load cloud computing processor in each deployment mapping group, the tasks in the subsequence of the data backup feature set to be called out are accurately located according to the mapping rules. Then, these tasks are separated from the task queue of the original high-load processor and prepared to be migrated to the corresponding low-load cloud computing processor. During the migration process, the relevant attributes and execution status information of the task are fully recorded and packaged to ensure the data integrity and traceability of the task during the transfer process. For each task to be called out, its task identifier, data storage location pointer, current execution progress (if any) and task dependency and other key information are recorded in detail and transferred together. Then, these tasks are inserted into the task queue of the target low-load cloud computing processor in sequence according to the predetermined rules and order. When inserting, according to the task queue scheduling strategy of the low-load processor, such as priority sorting, first-come-first-served, etc., the newly migrated tasks are reasonably placed in the corresponding position in the queue. If the target low-load processor adopts a priority scheduling strategy, it will be inserted into the appropriate priority queue level according to the priority of the task to ensure that high-priority tasks can be processed first. With the completion of task migration and insertion operations in all deployment mapping groups, the original multiple data backup task queues are fully updated according to the new distribution of tasks. For each cloud computing processor, the task composition and order in its task queue have changed, and the updated task queue is more in line with the load balancing requirements of the cloud computing processor. The task queue data structure of each processor is rebuilt, and the new task list, task status information and related scheduling parameters are integrated and stored, and finally multiple updated data backup task queues are obtained. These updated task queues lay a solid foundation for the subsequent use of cloud computing processors for efficient and stable data backup operations, effectively improving the performance and resource utilization of the entire data backup system.

[0030] Step S800: Utilize the multiple cloud computing processors to perform data backup according to the multiple update data backup task queues.

[0031] Specifically, the data backup operation is started based on multiple updated data backup task queues, and each cloud computing processor sequentially obtains tasks from its corresponding updated task queue. The processor first parses the detailed information of the task, including the storage location of the data, the backup target location, the data format, and the specific requirements of the backup. According to the nature and requirements of the task, the processor allocates corresponding computing resources, such as the number of CPU cores, memory space, and network bandwidth. For tasks with large data volumes, more CPU resources are allocated for data compression and transmission processing, while ensuring sufficient memory to cache data to increase the speed of data reading and writing. During the backup process, the processor performs operations according to the predetermined data backup strategy, such as full backup or incremental backup. If it is a full backup, the processor will read the source data completely and transmit it to the backup target location; if it is an incremental backup, only the part of the data that has changed since the last backup is transmitted, effectively reducing the data transmission volume and backup time. During the data transmission process, the data transmission speed is monitored in real time to ensure that it is stable within a reasonable range. If the transmission speed is found to be too slow, the processor will automatically adjust the transmission parameters, such as increasing the number of network transmission threads or optimizing the data compression algorithm, to improve the transmission efficiency. At the same time, it pays close attention to storage utilization. When the space utilization of the backup target storage location approaches the preset threshold, it will issue a warning in time and take corresponding measures according to the preset storage management strategy, such as automatically cleaning up expired backup data or migrating part of the backup data to other storage devices. During the entire backup process, the execution process and status of each task are recorded in detail through log records, including task start time, end time, data transmission volume, whether it is successfully completed, and other information. At the same time, the dashboard presents key indicators such as backup progress, the load of each processor, and the use of storage resources to the administrator in the form of intuitive charts and data displays. Based on this information, the overall status of data backup work can be understood in real time, and possible problems can be discovered and handled in a timely manner to ensure the efficiency, stability and reliability of data backup work and effectively protect the security and integrity of data.

[0032] In a possible implementation, step S200 further includes:

[0033] Step S210: construct a data backup task feature identifier, and use the data backup task feature identifier to identify the multiple data backup task queues to obtain the multiple data backup feature set sequences.

[0034] Specifically, the data backup task feature identifier is constructed by using a deep learning framework and selecting a suitable neural network structure as the basis, namely, a convolutional neural network (CNN). According to the characteristics of the data backup task, the input layer, hidden layer, and output layer structure of the network are designed. The input layer receives the task description information in the data backup task queue. After preprocessing, this information is converted into a format suitable for neural network processing, such as word vector representation or numerical encoding of the text description of the task. Then, multiple convolution kernels and pooling layers are configured in the hidden layer. The convolution kernel is used to extract local features in the task description, such as keyword features related to the task type, numerical pattern features of the task priority, etc. The pooling layer performs dimensionality reduction processing on the extracted features, reducing the amount of data while retaining key information. By stacking multiple hidden layers, the network's understanding and abstraction ability of task features are continuously deepened. Then, a node structure corresponding to the expected data backup feature set sequence is designed in the output layer to output the feature classification results or feature vector representation of the task. After the construction is completed, the recognizer is trained using a large amount of sample data. The sample data includes various data backup tasks with known features. By continuously adjusting the network parameters, the recognizer can accurately identify key features from the tasks in the task queue, such as the data volume range of the task, the periodic law of task execution, and the data format category involved in the task. After training to convergence, the trained data backup task feature recognizer is applied to multiple actual data backup task queues. The recognizer analyzes each task in the queue one by one, extracts valuable features from various attributes and description information of the task, and combines these features into multiple data backup feature set sequences according to the predetermined feature classification or encoding method. These sequences accurately reflect the feature distribution of tasks in each task queue, providing an important basis for subsequent reasonable task scheduling and resource allocation based on task characteristics, thereby improving the efficiency and accuracy of the data backup process.

[0035] In a possible implementation, step S210 further includes:

[0036] Step S211: Acquire multiple sample data backup tasks and multiple sample data backup features as a sample training data set.

[0037] Step S212: using the sample training data set to perform supervised training on a framework built based on a convolutional neural network until the training reaches convergence, thereby obtaining the trained data backup task feature identifier.

[0038] Specifically, we extensively collect and organize multiple sample data backup tasks and the corresponding sample data backup features to construct a sample training data set. The sample data backup tasks come from various sources, covering different business fields, different data types, and actual backup scenarios of different scales. For example, we obtain backup samples of structured data from the regular database backup tasks of an enterprise, whose characteristics include large data volume, high requirements for data consistency, and fixed backup cycles; we obtain backup samples of unstructured data from the enterprise office document backup scenarios, such as diverse document types, relatively small data volume but large number of files, and backup frequency related to employee office habits. At the same time, we also collect backup task samples in some special scenarios, such as the backup of financial transaction data with extremely high real-time requirements, which are characterized by large dynamic changes in data volume, the need to respond quickly to backup needs, and extremely strict requirements for data integrity and accuracy. For each sample data backup task, multiple sample data backup features are accurately extracted through expert knowledge, historical data analysis, and actual business needs research, such as the data size, data format, backup frequency, priority, data storage location, data security level and other key attributes of the task. These tasks and features are combined into sample training data pairs to form a rich, diverse and representative sample training data set, providing sufficient and high-quality data foundation for subsequent convolutional neural network training.

[0039] The framework based on the convolutional neural network is supervised and trained using the constructed sample training data set. First, the relevant information of the sample data backup task (such as task description, task parameters, etc.) is preprocessed and converted into a format suitable for the convolutional neural network input, such as converting the text-based task description into a word vector sequence or a numerical feature matrix. These preprocessed task data are input into the input layer of the convolutional neural network. The convolution layer in the network slides on the data through the convolution kernel to automatically extract local features in the task data, such as specific data format patterns, numerical feature patterns of task priorities, etc. The pooling layer downsamples these local features, reduces the data dimension while retaining key feature information, and enhances the generalization ability of the network. Multiple fully connected layers in the hidden layer further integrate and abstract features and learn complex relationships between features. In the output layer, according to the predefined sample data backup feature category or value range, the corresponding output node structure is designed, such as predicting the feature category to which the task belongs through a multi-classification output node, or predicting the specific feature value of the task (such as the specific numerical range of the data volume) through a regression output node. During the training process, the known sample data backup features corresponding to the sample data backup task are used as supervisory signals to calculate the loss function value between the network output and the real features. The back propagation algorithm is used to propagate the error from the output layer to each layer of the network, and the weights and bias parameters in the network are continuously adjusted to gradually reduce the loss function value. The training is continued iteratively until the performance indicators of the network (such as accuracy, loss value, etc.) reach a stable state, that is, the training converges. At this time, a trained data backup task feature identifier is obtained, which can accurately identify the key data backup feature set sequence from the new data backup task, providing strong support for subsequent task scheduling and resource allocation.

[0040] In one possible implementation, Figure 2 As shown, step S400 also includes:

[0041] Step S410: Filter the plurality of historical cloud computing processor execution log sets by using a preset processing efficiency interval as an index to obtain a plurality of historical filtered cloud computing processor execution log sets.

[0042] Step S420: Taking the load as an index, extracting the plurality of historically filtered cloud computing processor execution log sets to obtain a plurality of historically filtered load sets.

[0043] Step S430: traverse the multiple historical screening load sets to perform intensive search to determine the load tolerance intervals of the multiple cloud computing processors.

[0044] Specifically, firstly, based on the preset processing efficiency interval as the key index, the multiple historical cloud computing processor execution log sets obtained are carefully screened. This preset processing efficiency interval is determined based on the long-term observation and analysis of the performance of the cloud computing processor, for example, it is set to the number of tasks processed per second within a certain range or the average time consumed for task processing within a specific time interval. For each historical cloud computing processor execution log set, the task processing efficiency data corresponding to each log record is checked one by one, and the log records with processing efficiency within the preset interval are screened out to form a new subset. The same screening operation is performed on all historical cloud computing processor execution log sets, and finally multiple historical filtered cloud computing processor execution log sets are obtained. These filtered log sets focus on the operation of the processor within a specific processing efficiency range, providing a more targeted data basis for subsequent more accurate analysis of the load situation, excluding data that may interfere with extreme cases such as excessively high or low processing efficiency, and helping to more accurately grasp the load characteristics of the processor under normal working conditions.

[0045] Based on the obtained multiple historically filtered cloud computing processor execution log sets, in-depth extraction operations are performed with load as the core index. Go deep into each historically filtered cloud computing processor execution log set, carefully search and extract various data records related to the load, such as key load indicators such as CPU load rate, memory usage, disk I / O load, and network bandwidth usage. These load data can fully reflect the resource consumption status of the processor in the process of executing tasks, and are the key basis for determining its load tolerance range. By extracting the load data of all historically filtered cloud computing processor execution log sets, multiple historically filtered load sets are obtained. Each set contains a load data sequence of a cloud computing processor at different times within a specific processing efficiency range. These sequences record in detail the dynamic changes of the processor load, providing rich data materials for subsequent intensive searches and load tolerance range determination, so that the analysis of the processor load capacity can be based on actual operation data and is more in line with actual work scenarios.

[0046] The data in each historically filtered load set is preprocessed. The load data is normalized so that its value range is between [0, 1], which helps to improve the efficiency and accuracy of neural network training. Then, an autoencoder neural network architecture is constructed. The autoencoder consists of two parts: an encoder and a decoder. The encoder part compresses the input load data sequence into a low-dimensional representation, and the decoder part attempts to reconstruct the original load data sequence from this low-dimensional representation. The load data in multiple historically filtered load sets are used as input to train the autoencoder neural network. During the training process, the error between the load data reconstructed by the decoder and the original input data is minimized by adjusting the weights and biases of the network. The mean square error (MSE) is used as the loss function, and the network parameters are continuously optimized through the back propagation algorithm. During the training process, the autoencoder learns the key features and intrinsic patterns in the load data and maps the load data to a low-dimensional feature space, which can effectively represent the main information of the load data. After the training is completed, the encoder part is used to convert the load data in each historically filtered load set into a low-dimensional feature representation. Then, cluster analysis is performed on these low-dimensional features, and the K-means clustering algorithm is used to cluster similar load features into one category. For each cluster, its center value and radius are calculated (which can be determined based on the distance distribution from the data points in the cluster to the center). These center values ​​and radius can be used to define the load tolerance range of the cloud computing processor. The center value represents the typical characteristics of this type of load, and the radius represents the fluctuation range of the load under normal circumstances. According to actual needs and the performance characteristics of cloud computing processors, the obtained load tolerance range is appropriately adjusted and optimized. For example, the interval radius can be appropriately expanded or reduced according to the stability requirements of the system; or the processor load tolerance range corresponding to different types of tasks can be set differently according to the priority of the business. In this way, based on the autoencoder neural network and clustering analysis, the load tolerance range of the cloud computing processor can be determined more accurately, providing a more scientific and reasonable basis for subsequent task scheduling and resource allocation, and improving the overall data backup efficiency and stability of the cloud computing system. At the same time, this method can automatically adapt to the complex change pattern of load data, reduce manual intervention, and improve the degree of intelligence.

[0047] In a possible implementation, step S430 further includes:

[0048] Step S431: performing mean processing on the multiple historical screening load sets respectively to obtain multiple historical screening load mean values;

[0049] Step S432: using the averages of the multiple historical screening loads as multiple starting center points, calling the center update iteration model to iteratively update the multiple starting center points until a preset number of times is met, obtaining multiple target center points, and using the historical screening loads corresponding to the multiple target center points as multiple center historical screening loads;

[0050] Step S433: Obtain a preset load tolerance radius, and construct the load tolerance intervals of the plurality of cloud computing processors in combination with the historical screening loads of the plurality of centers.

[0051] Specifically, the mean processing operation is carried out for multiple historical filtered load sets. For each historical filtered load set, the set contains a series of load data records of the cloud computing processor within a specific processing efficiency range, such as key indicators such as CPU load rate, memory usage, disk I / O load at different times. The corresponding indicator values ​​of all load data in the set are summed and then divided by the number of data points to obtain the mean of each historical filtered load set. For example, for a historical filtered load set that contains a CPU load rate recorded every 10 seconds in the past hour, the 36 CPU load rate data are added and divided by 36 to obtain the CPU load mean of the set. The same mean calculation operation is performed on various load indicators (such as memory usage, disk I / O load, etc.) in each historical filtered load set, and finally multiple historical filtered load means are obtained. These means can represent the average load level of each cloud computing processor within the corresponding processing efficiency range to a certain extent, provide a basic reference point for the subsequent determination of the load tolerance range, help to initially grasp the typical load status of the processor, eliminate the interference of some local fluctuations on the overall load evaluation, and make the subsequent analysis more focused on the long-term stable load characteristics of the processor.

[0052] The average values ​​of multiple historical filtered loads are used as multiple starting center points, which are the initial reference positions for subsequent iterative updates. Then, the center update iteration model is called to start the iteration process. In each iteration, the model calculates the distance metric between each historical filtered load data point and the current starting center point. The calculation method of the distance metric depends on the algorithm used, such as Euclidean distance, Manhattan distance, etc., to measure the degree of deviation of each load data point from the center point. According to the calculated distance metric, each historical filtered load data point is assigned to the category or cluster to which the nearest starting center point belongs. After the data point is assigned, the center point is recalculated for each category or cluster. The calculation method is to add the corresponding load values ​​of all historical filtered load data points in the cluster, and then divide it by the number of data points in the cluster to obtain the new center point coordinates. These new center points are the updated starting center points. Repeat the above distance calculation, data point assignment and center point update process to continuously adjust the position of the starting center point. This iterative process continues until the preset number of times is met. When the preset number of times is reached, the iteration stops, and the multiple center points obtained at this time are the target center points. After multiple iterations of optimization, these target center points can better reflect the distribution characteristics and centralization trends of historical screening load data. Finally, the historical screening loads corresponding to these target center points are determined as multiple central historical screening loads. These central historical screening loads become the key basis for the subsequent construction of cloud computing processor load tolerance intervals. They represent representative load states in the entire historical screening load data set, which is of great significance for accurately defining the load range that the processor can withstand during normal operation, and provides important data support for ensuring the stable and efficient operation of cloud computing systems under data backup tasks and other workloads.

[0053] Obtain the preset load tolerance radius, which is a key parameter pre-set based on an in-depth understanding of the performance of cloud computing processors, past operating experience, and comprehensive consideration of the characteristics of data backup tasks. It aims to determine the load fluctuation range that the processor can withstand around the central historical filtering load, so as to ensure that the system can tolerate the load changes caused by normal business fluctuations to a certain extent, while maintaining a stable operating state. Then, with each central historical filtering load as the core, the load tolerance interval is constructed in combination with the obtained preset load tolerance radius. For each central historical filtering load, the range of the preset load tolerance radius is expanded in the upper and lower directions. For example, if a key indicator (such as CPU load rate) of a central historical filtering load is a specific value, with this value as the center, the value of the preset load tolerance radius is increased upward as the upper limit of the interval, and the value of the preset load tolerance radius is subtracted downward as the lower limit of the interval, so as to determine the load tolerance interval range corresponding to the indicator. The same operation is performed on various key indicators (such as memory usage, disk I / O load, etc.) of each central historical filtering load, and finally multiple cloud computing processor load tolerance intervals are constructed. These intervals together constitute a complete system that describes the load range that cloud computing processors can tolerate under normal operation, providing clear quantitative standards for subsequent judgment of whether the processor is in a reasonable load state. When the processor is performing data backup tasks or other tasks, its actual load fluctuates within these intervals, which can be regarded as a normal operating state; if it exceeds these intervals, it indicates that the processor may be facing overload or insufficient resource utilization. At this time, it is necessary to make corresponding task scheduling adjustments or resource optimization measures according to the system-set strategy to ensure the overall stability and efficiency of the cloud computing system, and ensure that data backup tasks can be completed smoothly and reliably. At the same time, it also provides a scientific and reasonable basis for resource management and task allocation in the cloud computing environment.

[0054] In a possible implementation, step S433 further includes:

[0055] Step S4331: Obtain a central update iteration model, wherein the central update iteration model includes a central iteration formula, wherein the central iteration formula is:

[0056]

[0057] Among them, l(x) is the update center point, N(x) is the multiple historical filtering loads in each historical filtering load set, x i For each historical filter load set, the i-th historical filter load, x is the mean of the historical filter load, Q(x i -x) is a weight kernel function built based on the Gaussian function.

[0058] Specifically, in this formula, l(x) represents the updated center point. It is calculated by weighted summing of the data points in the historical filtered load set. N(x) represents multiple historical filtered loads in each historical filtered load set. These load data points are the basic data source for the center point update calculation. i For each historical filter load set, the i-th historical filter load is taken, that is, each load data point in the set is traversed to participate in the calculation. x is the mean of the historical filter load, which is used as the initial center point reference position. And Q(x i -x) is a weight kernel function built based on the Gaussian function. The role of this weight kernel function is crucial. It assigns weights according to the distance between each historical filtered load data point and the current center point (that is, the historical filtered load mean). Data points that are closer will be assigned larger weights, which means that when calculating the updated center point, these data points have a greater impact on the position of the new center point; while data points that are farther away are assigned smaller weights, and their impact on the position of the new center point is relatively small. Through this weighting method, the updated center point can be more inclined to be close to areas with denser data distribution, thereby more accurately reflecting the concentration trend of the load data. During the iteration process, this formula is used to calculate a new center point each time, and then the new center point is used as the starting point for the next iteration. This process is repeated until the preset number of times is met. Through multiple iterations, the center point gradually converges to a position that can better represent the overall distribution characteristics of the historical screening load data. The historical screening load corresponding to the target center point is the central historical screening load. These central historical screening loads play a key reference role in constructing the tolerance range of cloud computing processor loads, and provide a scientific basis for accurately defining the processor load range, which helps to achieve efficient and reasonable scheduling of data backup tasks in cloud computing systems and optimization of overall performance.

[0059] In a possible implementation, step S800 further includes:

[0060] Step S810: construct a preset monitoring window.

[0061] Step S820: monitoring the processing efficiency of multiple cloud computing processors in the preset monitoring window, and obtaining a backup warning instruction according to the monitoring result.

[0062] Specifically, a preset monitoring window is constructed. This preset monitoring window is a specific time range setting, and its purpose is to continuously observe and analyze the operation status of the cloud computing processor within a period of time. During the construction process, it is necessary to clarify the start time and end time of the window, set it to start from the current moment, and continuously monitor the next hour, day or week, etc. The specific duration is determined based on the system's timeliness requirements for data backup tasks, the performance fluctuation characteristics of the cloud computing processor, and past experience. At the same time, determine the monitoring time interval, that is, how often the relevant data of the processor is collected, such as every 10 seconds, 1 minute or 5 minutes. Multiple time nodes are divided in this specific time window, and data sampling will be performed on the cloud computing processor at each node. In addition, it is necessary to set the accuracy requirements of data collection to ensure that the acquired data can accurately reflect the actual status of the processor, such as the statistics of the number of processing tasks are accurate to an integer number of tasks, and the measurement of resource utilization is accurate to several decimal places. Through the setting of these parameters, the construction of the preset monitoring window is completed, laying a basic framework for the subsequent processing efficiency monitoring work, and being able to obtain the operation data of the cloud computing processor with a certain frequency and accuracy within a clear time range, so as to fully and accurately grasp its performance dynamics.

[0063] The processing efficiency of multiple cloud computing processors is monitored using the built preset monitoring window. At each time node of the monitoring window, a series of key performance indicator data are obtained by communicating with the management interface of the cloud computing processor to evaluate its processing efficiency. These indicators include but are not limited to the number of data backup tasks completed per unit time, the average processing time of each task, the CPU utilization rate of the processor, the memory utilization rate, and the disk I / O operation rate. For example, the number of data backup tasks completed by the cloud computing processor in a monitoring time interval is calculated and compared with the data of the previous time interval. If the number of completed tasks is found to be significantly reduced, it may indicate a decrease in processing efficiency. At the same time, the average processing time of each task is analyzed. If the time continues to increase, it also indicates that the processor may face a performance bottleneck. The obtained performance indicator data is analyzed and processed in real time. The current data is compared with the historical data to observe its changing trend. If it is found that some indicators exceed the preset normal range, such as the CPU utilization rate exceeds 90% for a long time, the average task processing time exceeds a certain proportion of the normal level, etc., a backup warning instruction will be generated according to the preset rules. These warning instructions contain detailed abnormal information, such as the abnormal cloud computing processor identification, the abnormal performance indicator name and its specific value, the time when the abnormality occurred, etc. Warning instructions can be issued in various forms so that the abnormal situation of the cloud computing processor can be known in time and corresponding measures can be taken, such as adjusting the task allocation strategy, increasing processor resources or troubleshooting system failures, etc., to ensure that data backup tasks can be executed efficiently and stably in the cloud computing environment and to ensure the security and integrity of data.

[0064] In a possible implementation, step S600 further includes:

[0065] Step S610: Each deployment mapping group includes at least one high-load cloud computing processor and one low-load cloud computing processor.

[0066] Specifically, in the resource scheduling process of the data backup task, the construction of the deployment mapping group follows specific rules, where each deployment mapping group contains at least one high-load cloud computing processor and one low-load cloud computing processor. This requirement is based on the consideration of the overall performance optimization and load balancing of the cloud computing system. During the operation of the high-load cloud computing processor, its resource occupancy rate is high, and its ability to process tasks gradually approaches saturation. For example, when processing large-scale data backup tasks, the high-load processor may face long-term high CPU occupancy, tight memory resources, and frequent disk I / O, which leads to a decrease in task processing speed and may even affect the completion of the task on time. The low-load cloud computing processor has relatively abundant resources, and there is a certain margin in its CPU usage, memory free space, and disk I / O bandwidth. By combining at least one high-load processor and one low-load processor into a deployment mapping group, effective complementarity and reallocation of resources can be achieved. In the deployment mapping group, some tasks in the subsequence of the data backup feature set to be called out on the high-load processor can be reasonably migrated to the redundant load tolerance interval of the low-load processor for processing. This combination helps improve the resource utilization of the entire cloud computing system and avoids the situation where some processors are overloaded while other processor resources are idle. At the same time, it also helps to shorten the overall completion time of data backup tasks, improve response speed and reliability, and ensure that the cloud computing system can run stably and efficiently in complex data backup scenarios to meet the growing data backup needs.

[0067] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0068] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

[0069] This specification and the drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A data backup method based on cloud computing, characterized in that: The method comprises: Acquire multiple data backup task queues of multiple cloud computing processors; Traversing the multiple data backup task queues to perform data backup feature identification, and obtaining multiple data backup feature set sequences; Acquire a plurality of historical cloud computing processor execution log sets of the plurality of cloud computing processors in a preset historical window; Performing cloud computing processor load identification on the plurality of historical cloud computing processor execution log sets to obtain a plurality of cloud computing processor load tolerance intervals; Identify the cloud computing processors to be deployed according to the multiple data backup feature set sequences and the multiple cloud computing processor load tolerance intervals to obtain multiple high-load cloud computing processors and multiple low-load cloud computing processors, wherein each high-load cloud computing processor includes a subsequence of the data backup feature set to be deployed, and each low-load cloud computing processor includes a redundant load tolerance interval; According to multiple to-be-called data backup feature set subsequences and multiple redundant load tolerance intervals, mapping and matching the multiple high-load cloud computing processors and the multiple low-load cloud computing processors are performed to obtain multiple deployment mapping groups; Based on the multiple deployment mapping groups, mapping adjustment is performed on the multiple subsequences of the data backup feature sets to be called out, and the multiple data backup task queues are updated according to the adjustment results to obtain multiple updated data backup task queues; The plurality of cloud computing processors are utilized to perform data backup according to the plurality of update data backup task queues.

2. The data backup method based on cloud computing according to claim 1, characterized in that: A data backup task feature identifier is constructed, and the multiple data backup task queues are identified using the data backup task feature identifier to obtain the multiple data backup feature set sequences.

3. The data backup method based on cloud computing according to claim 2, characterized in that: include: Acquire multiple sample data backup tasks and multiple sample data backup features as a sample training data set; The framework constructed based on the convolutional neural network is supervised and trained using the sample training data set until the training converges, thereby obtaining the trained data backup task feature identifier.

4. The data backup method based on cloud computing according to claim 1, characterized in that: Performing cloud computing processor load identification on the plurality of historical cloud computing processor execution log sets to obtain a plurality of cloud computing processor load tolerance intervals includes: Performing log screening on the plurality of historical cloud computing processor execution log sets using a preset processing efficiency interval as an index to obtain a plurality of historical screened cloud computing processor execution log sets; Taking the load as an index, extracting the plurality of historically screened cloud computing processor execution log sets to obtain a plurality of historically screened load sets; The plurality of historically filtered load sets are traversed to perform intensive search to determine the plurality of cloud computing processor load tolerance intervals.

5. The data backup method based on cloud computing according to claim 4, characterized in that: Traversing the multiple historical screening load sets to perform intensive search to determine the load tolerance intervals of the multiple cloud computing processors includes: Performing mean processing on the multiple historical screening load sets respectively to obtain multiple historical screening load mean values; Taking the averages of the multiple historical screening loads as multiple starting center points, calling the center update iteration model to iteratively update the multiple starting center points until a preset number of times is met, obtaining multiple target center points, and taking the historical screening loads corresponding to the multiple target center points as multiple center historical screening loads; A preset load tolerance radius is obtained, and the load tolerance intervals of the plurality of cloud computing processors are constructed in combination with the historical screening loads of the plurality of centers.

6. The data backup method based on cloud computing according to claim 5, characterized in that: include: A center update iteration model is obtained, wherein the center update iteration model includes a center iteration formula, wherein the center iteration formula is: Among them, l(x) is the update center point, N(x) is the multiple historical filtering loads in each historical filtering load set, x i For each historical filter load set, the i-th historical filter load, x is the mean of the historical filter load, Q(x i -x) is a weight kernel function built based on the Gaussian function.

7. The data backup method based on cloud computing according to claim 1, characterized in that: include: Construct preset monitoring windows; The processing efficiency of multiple cloud computing processors is monitored in the preset monitoring window, and a backup warning instruction is obtained according to the monitoring results.

8. The data backup method based on cloud computing according to claim 1, characterized in that: Each provisioning mapping group includes at least one high-load cloud computing processor and one low-load cloud computing processor.

Images