Resource Dynamic Scheduling System and Method for Off-site Disaster Recovery Cloud Systems

Abstract

This invention relates to the field of off-site disaster recovery technology, and specifically discloses a resource dynamic scheduling system and method for off-site disaster recovery cloud systems. The system includes: a historical data parsing module, used to generate a multi-dimensional static state matrix and a multi-dimensional dynamic action matrix of the system to be recovered based on historical traceability data of multiple system state indicators; a fluctuation risk determination module, used to predict the multi-dimensional state fluctuation risk of the system to be recovered within a future scheduling cycle based on the multi-dimensional static state matrix and the multi-dimensional dynamic action matrix; and a backup path adjustment module, used to obtain the resource dynamic scheduling result of the off-site disaster recovery cloud system based on the multi-dimensional state fluctuation risk of the system to be recovered within a future scheduling cycle and the real-time load status of all off-site disaster recovery backup nodes at all levels in the off-site disaster recovery cloud system; thereby achieving reasonable resource allocation during off-site disaster recovery backup and improving resource utilization efficiency.

Description

Technical Field

[0001] This invention relates to the field of off-site disaster recovery technology, and in particular to a dynamic resource scheduling system and method for off-site disaster recovery cloud systems. Background Technology

[0002] In today's highly digitalized era, the importance of data to enterprises and organizations is self-evident. Enterprises' various business operations heavily rely on data storage, processing, and application; data loss or system failures can lead to severe economic losses and business interruptions. Off-site disaster recovery cloud systems have emerged to address this need. By establishing disaster recovery backup centers in different geographical locations, they ensure that business operations continue uninterrupted and data is not lost when the primary system encounters catastrophic events (such as natural disasters, human error, or cyberattacks). Dynamic resource scheduling in off-site disaster recovery cloud systems is crucial. Reasonable dynamic resource scheduling can flexibly allocate resources of off-site disaster recovery backup nodes according to changes in the status of the system to be recovered, improving the efficiency and reliability of the disaster recovery system. This not only reduces downtime caused by system failures and ensures business continuity but also optimizes resource utilization and reduces operating costs. With the continuous development of cloud computing technology and the increasing demands of enterprises for data security and business continuity, dynamic resource scheduling systems and methods for off-site disaster recovery cloud systems have broad application prospects and are expected to become a key technology for ensuring enterprise data security and stable business operation.

[0003] However, existing resource scheduling methods for off-site disaster recovery cloud systems have many shortcomings. In analyzing the status of disaster recovery systems, it is difficult to comprehensively and systematically grasp the historical status and behavioral patterns of the systems to be recovered. Due to the lack of in-depth analysis of historical data, it is impossible to accurately predict the fluctuations in the status risks of the systems to be recovered during future scheduling cycles. This leads to insufficient assessment of potential risks, inability to conduct reasonable dynamic resource scheduling, resulting in unreasonable resource allocation, failure to fully leverage the advantages of off-site disaster recovery cloud systems, potential resource waste, or ineffective protection of critical business operations, and an inability to meet enterprises' needs for efficient and reliable operation of disaster recovery systems.

[0004] Therefore, this invention proposes a dynamic resource scheduling system and method for off-site disaster recovery cloud systems. Summary of the Invention

[0005] This invention provides a dynamic resource scheduling system and method for off-site disaster recovery cloud systems. The system includes: analyzing historical data of multiple system status indicators of the disaster recovery system to generate a multi-dimensional static state matrix and a multi-dimensional dynamic action matrix. This comprehensively and systematically structures the historical data, providing a solid data foundation for subsequent in-depth analysis and providing a clear and quantitative presentation of the system's past states and actions. Based on these two matrices, the system predicts multi-dimensional state fluctuation risks within future scheduling cycles, enabling early detection of potential system instability factors. This allows the operations and maintenance team to plan countermeasures in advance, avoiding serious impacts of potential risks on the disaster recovery system and ensuring its stability and reliability. Combining the predicted fluctuation risks with the real-time load status of off-site disaster recovery backup nodes at various levels within the off-site disaster recovery cloud system, the system derives dynamic resource scheduling results, achieving scientific and real-time resource scheduling. It considers both potential future system risks and the actual load of current backup nodes, thereby optimizing backup paths, rationally allocating resources, improving resource utilization efficiency, and ensuring that the off-site disaster recovery cloud system can operate efficiently and stably under various complex situations, effectively guaranteeing data security and timely disaster recovery.

[0006] This invention provides a dynamic resource scheduling system for off-site disaster recovery cloud systems, comprising:

[0007] The historical data parsing module is used to generate a multi-dimensional static state matrix and a multi-dimensional dynamic action matrix of the system to be recovered from disaster based on historical data of multiple system status indicators of the system to be recovered from disaster.

[0008] The volatility risk determination module is used to predict the multidimensional state volatility risk of the system to be recovered from in future scheduling cycles based on the multidimensional static state matrix and multidimensional dynamic action matrix of the system to be recovered from.

[0009] The backup path adjustment module is used to obtain the dynamic resource scheduling results of the disaster recovery cloud system based on the multi-dimensional state fluctuation risk of the disaster recovery system in the future scheduling cycle and the real-time load status of all disaster recovery backup nodes at all levels in the disaster recovery cloud system.

[0010] Optionally, the historical data parsing module includes:

[0011] The historical data acquisition submodule is used to acquire historical data of resource performance indicators, system status indicators, disaster recovery action indicators, and environmental configuration indicators of the system to be disaster recovered, as historical data of multiple system status indicators of the system to be disaster recovered.

[0012] The feature dimension contribution rate determination submodule is used to analyze the historical trace data of each static status indicator in the historical trace data of multiple system status indicators of the disaster recovery system to determine the contribution of each feature dimension of each static status indicator.

[0013] The static feature representation module is used to generate a multi-dimensional static state matrix of the disaster recovery system based on the feature value sequence of the feature dimension with the maximum contribution of all static state indicators in historical traceability data.

[0014] The dynamic feature representation module is used to generate a multi-dimensional dynamic action matrix of the system to be recovered from disasters based on the historical traceability data of all kinds of dynamic status index values ​​of multiple system status indicators of the system to be recovered from disasters.

[0015] Optionally, the feature dimension contribution rate determination submodule includes:

[0016] The feature standardization unit is used to generate a feature dimension standard matrix for each static state indicator based on the historical traceability data of each static state indicator in the historical traceability data of multiple system state indicators of the disaster recovery system.

[0017] The feature decomposition processing unit is used to determine the principal component feature sequence and the corresponding principal component loading matrix of the covariance matrix of the feature dimension standard matrix of each static state index.

[0018] The eigenvalue contribution analysis unit is used to calculate the principal component eigenvalue contribution rate vector of the covariance matrix of the characteristic dimension standard matrix of the corresponding static state index based on the principal component eigenvalue sequence.

[0019] The load vector determination unit is used to determine the load vector of each feature dimension of the corresponding static state index based on the principal component load matrix of the feature dimension standard matrix of each static state index.

[0020] The contribution rate determination unit is used to take the sum of the products of the squares of all values ​​in the load vector of each feature dimension of each static state index and the values ​​at the same position in the corresponding principal component eigenvalue contribution rate vector as the contribution rate of each feature dimension of each static state index.

[0021] Optionally, the feature normalization unit includes:

[0022] The feature dimension value determination subunit is used to determine all feature dimension values ​​of each static state indicator at each moment within the historical tracing period in the historical tracing data of each static state indicator in the historical tracing data of multiple system state indicators of the disaster recovery system.

[0023] The feature dimension value standardization subunit is used to standardize the same feature dimension values ​​among all feature dimension values ​​at all times during the historical tracing period for each static state indicator, so as to obtain the standard values ​​of all feature dimensions for each static state indicator at all times during the historical tracing period.

[0024] The feature dimension standard value matrix sub-unit is used to generate the feature dimension standard matrix of each static state indicator based on all feature dimension standard values ​​at all times within the historical tracing period.

[0025] Optionally, the eigenvalue decomposition unit includes:

[0026] The eigenvalue decomposition subunit is used to perform eigenvalue decomposition on the covariance matrix of the standard matrix of the feature dimensions of each static state index to obtain all eigenvalues ​​and eigenvectors.

[0027] The eigenvalue filtering subunit is used to sort all the eigenvalues ​​of the covariance matrix of the feature dimension standard matrix of each static state index from largest to smallest, obtain the eigenvalue sequence of the covariance matrix of the feature dimension standard matrix of the corresponding static state index, and take the sequence of the first preset eigenvalues ​​in the eigenvalue sequence as the principal component eigenvalue sequence.

[0028] The loading matrix generation sub-unit is used to generate the principal component loading matrix based on the eigenvectors corresponding to all eigenvalues ​​in the principal component feature sequence, thereby generating the feature dimension standard matrix of the corresponding static state index.

[0029] Optionally, the volatility risk determination module includes:

[0030] The impact factor determination submodule is used to mine the impact factors between each static state indicator and each dynamic state indicator based on the multidimensional static state matrix and multidimensional dynamic action matrix of the disaster recovery system, and to build a multidimensional impact factor matrix based on the impact factors between all static state indicators and all dynamic state indicators.

[0031] The influence path marking submodule is used to mark the extreme unidirectional influence path of each static state indicator and each dynamic state indicator in the multidimensional influence factor matrix.

[0032] The Basic Volatility Factor Determination Submodule is used to analyze the historical data of each static state indicator and each dynamic state indicator to determine the basic volatility factor of each static state indicator in the future scheduling cycle.

[0033] The comprehensive volatility factor determination submodule is used to analyze the comprehensive volatility factor of each static state indicator and each dynamic state indicator in the future scheduling period based on the basic volatility factor and the corresponding extreme unidirectional influence path of each static state indicator in the future scheduling period.

[0034] The multidimensional state fluctuation risk determination submodule is used to determine the multidimensional state fluctuation risk of the disaster recovery system in the future scheduling period based on the comprehensive fluctuation factor of all static state indicators and all dynamic state indicators in the future scheduling period.

[0035] Optionally, the process by which the backup path adjustment module obtains the real-time load status of all off-site disaster recovery backup nodes at all levels in the off-site disaster recovery cloud system also includes:

[0036] The real-time load entropy value of all off-site disaster recovery backup nodes at all levels in the off-site disaster recovery cloud system is obtained by normalizing the real-time load value of all off-site disaster recovery backup nodes at all levels in the off-site disaster recovery cloud system, and is used as the real-time load status of the corresponding off-site multi-level disaster recovery backup nodes.

[0037] Optionally, the backup path adjustment module includes:

[0038] The transmission delay and exposure risk determination submodule is used to determine the transmission delay of each disaster recovery data backup task of the system to be disaster recovery in each off-site disaster recovery backup node, and to determine the exposure risk value of each disaster recovery data backup task of the system to be disaster recovery in each off-site disaster recovery backup node.

[0039] The weight allocation determination submodule is used to determine the weight allocation values ​​of transmission delay, risk exposure value, and node load at each level based on the multi-dimensional state fluctuation risk of the disaster recovery system in the future scheduling cycle.

[0040] The backup path cost determination submodule is used to calculate the backup path cost between each disaster recovery data backup task and each off-site disaster recovery backup node based on the transmission delay, exposure risk value, node load of each disaster recovery data backup task of the system to be disaster recovery in each off-site disaster recovery backup node, as well as the weight allocation value of transmission delay, risk exposure value, and node load under the corresponding off-site disaster recovery backup node's level.

[0041] The resource dynamic scheduling submodule is used to obtain the resource dynamic scheduling result of the off-site disaster recovery cloud system based on the backup path cost between each disaster recovery data backup task and each off-site disaster recovery backup node.

[0042] Optionally, the resource dynamic scheduling submodule includes:

[0043] The optimal backup path determination unit is used to determine the optimal backup path between all disaster recovery data backup tasks and all off-site disaster recovery backup nodes based on the backup path cost between each disaster recovery data backup task and each off-site disaster recovery backup node.

[0044] The resource dynamic scheduling unit is used to dynamically schedule and deploy resources for all disaster recovery data backup tasks within the future scheduling cycle based on the optimal backup path between all disaster recovery data backup tasks and all off-site disaster recovery backup nodes, and obtain the resource dynamic scheduling results of the off-site disaster recovery cloud system.

[0045] This invention provides a method for dynamic resource scheduling in off-site disaster recovery cloud systems, comprising:

[0046] Based on historical data of multiple system status indicators of the disaster recovery system, a multi-dimensional static state matrix and a multi-dimensional dynamic action matrix of the disaster recovery system are generated.

[0047] Based on the multidimensional static state matrix and multidimensional dynamic action matrix of the disaster recovery system, the multidimensional state fluctuation risk of the disaster recovery system in future scheduling cycles is predicted.

[0048] Based on the multidimensional state fluctuation risk of the disaster recovery system in the future scheduling cycle and the real-time load status of all off-site disaster recovery backup nodes at all levels in the off-site disaster recovery cloud system, the dynamic scheduling results of the off-site disaster recovery cloud system are obtained.

[0049] The beneficial effects of this invention compared to existing technologies are as follows: By analyzing historical data of multiple system status indicators of the disaster recovery system, a multi-dimensional static state matrix and a multi-dimensional dynamic action matrix are generated. This comprehensively and systematically structures the historical data, providing a solid data foundation for subsequent in-depth analysis and providing a clear and quantitative presentation of the system's past states and actions. Furthermore, based on these two matrices, the multi-dimensional state fluctuation risks within future scheduling cycles can be predicted, allowing for early detection of potential system instability factors. This enables the operations and maintenance team to plan countermeasures in advance, avoiding serious impacts of potential risks on the disaster recovery system and ensuring its stability and reliability. Combining the predicted fluctuation risks with the real-time load status of each level of off-site disaster recovery backup nodes in the off-site disaster recovery cloud system, dynamic resource scheduling results are obtained, achieving scientific and real-time resource scheduling. This approach considers both potential future system risks and the actual load of current backup nodes, thereby optimizing backup paths, rationally allocating resources, improving resource utilization efficiency, and ensuring that the off-site disaster recovery cloud system can operate efficiently and stably in various complex situations, effectively guaranteeing data security and timely disaster recovery.

[0050] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in this application.

[0051] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0052] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0053] Figure 1 This is a schematic diagram of a resource dynamic scheduling system for a disaster recovery cloud system in an embodiment of the present invention;

[0054] Figure 2 This is a schematic diagram of the historical data parsing module in an embodiment of the present invention. Detailed Implementation

[0055] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0056] This invention provides an implementation method for a dynamic resource scheduling system for off-site disaster recovery cloud systems, including:

[0057] The historical data parsing module is used to generate a multi-dimensional static state matrix and a multi-dimensional dynamic action matrix of the system to be recovered from disaster based on historical data of multiple system status indicators of the system to be recovered from disaster.

[0058] The volatility risk determination module is used to predict the multidimensional state volatility risk of the system to be recovered from in future scheduling cycles based on the multidimensional static state matrix and multidimensional dynamic action matrix of the system to be recovered from.

[0059] The backup path adjustment module is used to obtain the dynamic resource scheduling results of the disaster recovery cloud system based on the multi-dimensional state fluctuation risk of the disaster recovery system in the future scheduling cycle and the real-time load status of all disaster recovery backup nodes at all levels in the disaster recovery cloud system.

[0060] In this embodiment, the disaster recovery system refers to a system that needs to be protected against disasters in a different location and may face the risk of data loss or system failure due to disaster events. It includes various types of business data and operating programs, such as the core business system of a bank, which stores key data such as customer account information and transaction records and runs related business logic programs.

[0061] In this embodiment, historical traceability data is a record of various system status indicators of the system to be disaster recovery over a period of time, including resource performance, system status, disaster recovery actions, and environment configuration. It can reflect the past operating status and behavior patterns of the system, such as the CPU utilization rate of the server, the number of online users, the disaster recovery backup operation time, and the network bandwidth configuration of the bank's core business system in the past.

[0062] In this embodiment, the multidimensional static state matrix of the disaster recovery system is generated based on the feature value sequence of the feature dimension of the maximum contribution of all static state indicators in the historical traceability data of multiple system state indicators of the disaster recovery system. The multidimensional dynamic action matrix is ​​generated based on all kinds of dynamic state indicator values. They comprehensively and systematically structure the historical data to provide a basis for in-depth analysis. For example, in an e-commerce disaster recovery system, the multidimensional static state matrix reflects the static configuration of different servers such as storage capacity and computing power, while the multidimensional dynamic action matrix includes dynamic behavior indicator values ​​such as changes in the number of orders processed and the frequency of user login and logout.

[0063] In this embodiment, the future scheduling cycle is a pre-set future time period during which resource scheduling planning needs to be carried out for the disaster recovery system. For example, the next week is set as the future scheduling cycle. The e-commerce disaster recovery system may face higher business pressure due to promotional activities, so resource scheduling needs to be planned in advance.

[0064] In this embodiment, the multi-dimensional state fluctuation risk of the disaster recovery system in the future scheduling cycle is derived from the analysis of the multi-dimensional static state matrix and the multi-dimensional dynamic action matrix, which comprehensively considers the influencing factors, influence paths, and basic fluctuation factors of various indicators. The system may face state instability risks in multiple dimensions in the future scheduling cycle. For example, the e-commerce disaster recovery system may experience risks in performance, data security, and business continuity in the future scheduling cycle, such as longer server response time, data transmission loss, and inaccessibility for users in some regions.

[0065] In this embodiment, all off-site disaster recovery backup nodes at all levels of the off-site disaster recovery cloud system refer to the nodes included in each level of the multi-level architecture of the off-site disaster recovery cloud system, which are used to receive and store backup data of the system to be recovered from disaster and provide support when the main system fails. Each node is distributed in different geographical locations and has different resource configurations and load capacities. For example, the off-site disaster recovery cloud system is divided into three levels: regional, provincial, and municipal. The three regional nodes have high storage capacity and powerful computing capabilities, the ten provincial nodes focus on fast data transmission, and the several municipal nodes are responsible for the initial processing and caching of local data.

[0066] In this embodiment, the dynamic resource scheduling result of the off-site disaster recovery cloud system is the final resource allocation scheme derived by comprehensively considering the multi-dimensional state fluctuation risk of the system to be disaster recovered during the future scheduling cycle and the real-time load status of all off-site disaster recovery backup nodes at all levels of the off-site disaster recovery cloud system. It determines the allocation node and resource allocation method for each disaster recovery data backup task, thereby improving resource utilization efficiency. For example, by calculating the backup path cost, the optimal backup path is determined, and the critical business data backup tasks are allocated to nodes with low regional load and low transmission latency. The resource allocation of other nodes is also adjusted. This set of resource allocation decisions is the dynamic resource scheduling result.

[0067] In another possible implementation, to comprehensively and accurately construct a multidimensional matrix from historical data of different types of indicators, and to provide a detailed and reliable data foundation for subsequent analysis, a historical data parsing module is proposed, including:

[0068] The historical data acquisition submodule is used to acquire historical data of resource performance indicators, system status indicators, disaster recovery action indicators, and environmental configuration indicators of the system to be disaster recovered, as historical data of multiple system status indicators of the system to be disaster recovered.

[0069] The feature dimension contribution rate determination submodule is used to analyze the historical trace data of each static status indicator in the historical trace data of multiple system status indicators of the disaster recovery system to determine the contribution of each feature dimension of each static status indicator.

[0070] The static feature representation module is used to generate a multi-dimensional static state matrix of the disaster recovery system based on the feature value sequence of the feature dimension with the maximum contribution of all static state indicators in historical traceability data.

[0071] The dynamic feature representation module is used to generate a multi-dimensional dynamic action matrix of the system to be recovered from disasters based on the historical traceability data of all kinds of dynamic status index values ​​of multiple system status indicators of the system to be recovered from disasters.

[0072] In this embodiment, the historical data of the resource performance indicators, system status indicators, disaster recovery action indicators, and environment configuration indicators of the disaster recovery system refers to the recorded data of the disaster recovery system over a period of time regarding resource performance (such as server CPU utilization, memory usage, etc.), system status (such as number of online users, system response time, etc.), disaster recovery actions (such as backup frequency, recovery time, etc.), and environment configuration (such as network bandwidth, storage device specifications, etc.). For example, the average server CPU utilization, daily peak online users, weekly backup frequency, and network bandwidth adjustment records of a company's disaster recovery system over the past month.

[0073] In this embodiment, static status indicators are indicators that reflect the relative stability of the system among multiple system status indicators of the disaster recovery system. They do not change rapidly over time, such as the hardware configuration parameters of the server and the database architecture design, which are relatively fixed over a period of time.

[0074] In this embodiment, generating a multidimensional static state matrix of the disaster recovery system based on the feature value sequence of the feature dimension with the greatest contribution of all static state indicators in historical traceability data refers to analyzing each static state indicator in historical traceability data, determining the feature dimension with the greatest impact on the system state of each indicator, extracting the feature values ​​of these feature dimensions and arranging them in a certain order to form a sequence, and then constructing a multidimensional matrix to present the static characteristics of the system from multiple dimensions. For example, after analyzing the static indicators of multiple servers, a multidimensional static state matrix is ​​constructed using the feature values ​​of the feature dimension with the greatest contribution of server computing power, storage capacity, etc.

[0075] In this embodiment, all kinds of dynamic status index values ​​refer to the specific values ​​of various indicators that reflect the dynamic changes of the disaster recovery system during its operation. These index values ​​will change with time or business operations, such as the specific values ​​of indicators such as the number of orders generated and the number of user logins and logouts in the transaction system as business operations change in real time.

[0076] In this embodiment, generating a multidimensional dynamic action matrix for the disaster recovery system based on all dynamic status indicator values ​​from the historical traceability data of multiple system status indicators of the disaster recovery system refers to organizing and structuring various dynamic status indicator values ​​in the historical traceability data of the disaster recovery system to form a matrix to display the dynamic behavior pattern of the system. For example, constructing a multidimensional dynamic action matrix from the dynamic indicator values ​​such as the number of product views, the number of times items are added to the shopping cart, and the number of successful payments for each sub-period of an e-commerce system within a certain period (each row of data contains the dynamic indicator values ​​such as the number of product views, the number of times items are added to the shopping cart, and the number of successful payments for each sub-period) for analyzing the dynamic changes in the system's business.

[0077] In another possible implementation, in order to accurately determine the contribution of each feature dimension of each static state index to the system state and provide a quantitative basis for the generation of the multidimensional static state matrix, a feature dimension contribution rate determination submodule is proposed, including:

[0078] The feature standardization unit is used to generate a feature dimension standard matrix for each static state indicator based on the historical traceability data of each static state indicator in the historical traceability data of multiple system state indicators of the disaster recovery system.

[0079] The feature decomposition processing unit is used to determine the principal component feature sequence and the corresponding principal component loading matrix of the covariance matrix of the feature dimension standard matrix of each static state index.

[0080] The eigenvalue contribution analysis unit is used to calculate the principal component eigenvalue contribution rate vector of the covariance matrix of the characteristic dimension standard matrix of the corresponding static state index based on the principal component eigenvalue sequence.

[0081] The load vector determination unit is used to determine the load vector of each feature dimension of the corresponding static state index based on the principal component load matrix of the feature dimension standard matrix of each static state index.

[0082] The contribution rate determination unit is used to take the sum of the products of the squares of all values ​​in the load vector of each feature dimension of each static state index and the values ​​at the same position in the corresponding principal component eigenvalue contribution rate vector as the contribution rate of each feature dimension of each static state index.

[0083] In this embodiment, the principal component eigenvalue contribution rate vector, which is calculated based on the principal component eigenvalue sequence to obtain the covariance matrix of the characteristic dimension standard matrix of the corresponding static state index, refers to a vector obtained by performing specific calculations on the principal component eigenvalue sequence. Each value in this vector represents the proportion of the corresponding principal component eigenvalue in the sum of all eigenvalues ​​of the covariance matrix, thereby reflecting the degree of contribution of each principal component to the overall variance. For example, by dividing each principal component eigenvalue by the sum of all principal component eigenvalues, the principal component eigenvalue contribution rate vector is obtained, which is used to evaluate the relative importance of each principal component to the covariance matrix of the characteristic dimension standard matrix of the static state index.

[0084] In this embodiment, the load vector for each feature dimension of the corresponding static state index is determined based on the principal component load matrix of the feature dimension standard matrix of each static state index. This means that the vector related to each feature dimension is extracted according to the principal component load matrix. This vector reflects the strength of the linear relationship between the corresponding feature dimension and each principal component. For example, the vectors corresponding to different feature dimensions are extracted from the principal component load matrix column by column as the load vector of each feature dimension, which shows the degree of correlation between the feature dimension and the principal component.

[0085] In this embodiment, the sum of the products of the squares of all values ​​in the load vector of each feature dimension of each static state index and the values ​​at the same position in the corresponding principal component eigenvalue contribution rate vector is used as the contribution of each feature dimension of each static state index. This is a calculation method to quantify the contribution of each feature dimension to the static state index. That is, the load vector and the principal component eigenvalue contribution rate vector are combined through specific operations, and the relationship between the feature dimension and the principal component and the importance of the principal component are comprehensively considered to obtain the contribution of each feature dimension. For example, after squaring the elements of the load vector of a certain feature dimension, multiplying them by the corresponding elements of the principal component eigenvalue contribution rate vector, and then summing them, the contribution of that feature dimension to the static state index is obtained.

[0086] In another possible implementation, in order to transform the original feature dimension values ​​of static state indicators into a unified standard form to facilitate subsequent feature decomposition and analysis, a feature standardization unit is proposed, including:

[0087] The feature dimension value determination subunit is used to determine all (feature dimension) feature dimension values ​​of each static state indicator at each moment within the historical tracing period in the historical tracing data of each static state indicator in the historical tracing data of multiple system state indicators of the disaster recovery system.

[0088] The feature dimension value standardization subunit is used to standardize the same feature dimension values ​​among all feature dimension values ​​at all times during the historical tracing period for each static state indicator, so as to obtain the standard values ​​of all feature dimensions for each static state indicator at all times during the historical tracing period.

[0089] The feature dimension standard value matrix sub-unit is used to generate the feature dimension standard matrix of each static state indicator based on all feature dimension standard values ​​at all times within the historical tracing period.

[0090] In this embodiment, the same feature dimension values ​​of each static state indicator at all times within the historical tracing period are standardized to obtain the standard values ​​of all feature dimensions of each static state indicator at all times within the historical tracing period. This means that the values ​​of the same feature dimension at different times under the same static state indicator are processed according to a specific standardization method (such as Z-score standardization) so that the mean of all feature dimension values ​​of each feature dimension is 0 and the standard deviation is 1, so that these values ​​have a unified dimension and comparable scale, eliminating the influence of different feature dimension values ​​due to different dimensions and orders of magnitude. For example, the server CPU utilization rate (a feature dimension value) recorded at different times is processed through a standardization formula to obtain the standard value of the feature dimension under the same standard, which is convenient for subsequent analysis and comparison.

[0091] In this embodiment, generating a feature dimension standard matrix for each static state indicator based on the standard values ​​of all feature dimensions at all times within the historical tracing period means arranging and combining all the standardized feature dimension standard values ​​according to certain rules (such as using time as rows and feature dimensions as columns) to form a matrix structure. Such a matrix can clearly and intuitively present the standardized data of each static state indicator at different times and under different feature dimensions. For example, the standard values ​​of various feature dimensions such as server CPU utilization and memory usage at different times can be arranged into a matrix in chronological order as the feature dimension standard matrix of the static state indicator (each row of data represents the standard value of various feature dimensions such as server CPU utilization and memory usage at each time), providing a foundation for further analysis of the relationship between feature dimensions and data characteristics.

[0092] In another possible implementation, in order to extract key principal component features and loading matrices from the standard matrix covariance matrix of static state indices, and to prepare for the analysis of feature contributions, a feature decomposition unit is proposed, including:

[0093] The eigenvalue decomposition subunit is used to perform eigenvalue decomposition on the covariance matrix of the standard matrix of the feature dimensions of each static state index to obtain all eigenvalues ​​and eigenvectors.

[0094] The eigenvalue filtering subunit is used to sort all the eigenvalues ​​of the covariance matrix of the feature dimension standard matrix of each static state index from largest to smallest, obtain the eigenvalue sequence of the covariance matrix of the feature dimension standard matrix of the corresponding static state index, and take the sequence of the first preset eigenvalues ​​in the eigenvalue sequence as the principal component eigenvalue sequence.

[0095] The loading matrix generation sub-unit is used to generate the principal component loading matrix based on the eigenvectors corresponding to all eigenvalues ​​in the principal component feature sequence, thereby generating the feature dimension standard matrix of the corresponding static state index.

[0096] In this embodiment, "preset number" refers to a specific number that is set in advance. When performing eigenvalue processing on the covariance matrix of the feature dimension standard matrix of each static state indicator, the first "preset number" eigenvalues ​​are selected to form the principal component feature sequence. The setting of this number depends on the needs of data feature extraction and the purpose of subsequent analysis. For example, when analyzing static state indicators related to server performance, based on experience or previous experiments, the first 3 eigenvalues ​​are selected in advance to construct the principal component feature sequence in order to retain key information for subsequent analysis.

[0097] In this embodiment, the principal component loading matrix, which generates the standard matrix of the feature dimensions of the corresponding static state index based on the eigenvectors corresponding to all eigenvalues ​​in the principal component feature sequence, refers to combining the eigenvectors corresponding to each eigenvalue in the principal component feature sequence in a certain way to form the principal component loading matrix. This matrix reflects the linear relationship between the principal components and the original feature dimensions, and helps to analyze the degree and direction of influence of each principal component on the original feature dimensions. For example, when analyzing the static state index of network bandwidth configuration, the eigenvectors corresponding to the eigenvalues ​​in the principal component feature sequence are arranged to generate the principal component loading matrix. Through this matrix, it can be seen how different principal components act on the various feature dimensions related to network bandwidth configuration.

[0098] In another possible implementation, in order to accurately predict the multidimensional state fluctuation risk of the disaster recovery system during future scheduling cycles by analyzing the interaction and fluctuation of static and dynamic indicators of the system, a fluctuation risk determination module is proposed, including:

[0099] The impact factor determination submodule is used to mine the impact factors between each static state indicator and each dynamic state indicator based on the multidimensional static state matrix and multidimensional dynamic action matrix of the disaster recovery system, and to build a multidimensional impact factor matrix based on the impact factors between all static state indicators and all dynamic state indicators.

[0100] The influence path marking submodule is used to mark the extreme unidirectional influence path of each static state indicator and each dynamic state indicator in the multidimensional influence factor matrix.

[0101] The Basic Volatility Factor Determination Submodule is used to analyze the historical data of each static state indicator and each dynamic state indicator to determine the basic volatility factor of each static state indicator in the future scheduling cycle.

[0102] The comprehensive volatility factor determination submodule is used to analyze the comprehensive volatility factor of each static state indicator and each dynamic state indicator in the future scheduling period based on the basic volatility factor and the corresponding extreme unidirectional influence path of each static state indicator in the future scheduling period.

[0103] The multidimensional state fluctuation risk determination submodule is used to determine the multidimensional state fluctuation risk of the disaster recovery system in the future scheduling period based on the comprehensive fluctuation factor of all static state indicators and all dynamic state indicators in the future scheduling period.

[0104] In this embodiment, by analyzing the system's stable state characteristics represented by the multidimensional static state matrix and the system's dynamic behavior information presented by the multidimensional dynamic action matrix, the Pearson correlation coefficient can be used to find the quantitative relationship between static state indicators (such as relatively fixed characteristic indicators like server configuration) and dynamic state indicators (such as indicators that change over time like business operation frequency). For example, the analysis reveals that for every certain increase in server computing power (static indicator), the business processing speed (dynamic indicator) will increase by a specific percentage. This value is the influencing factor between them.

[0105] Assume we have a static status metric "Server Core Count S1" and a dynamic status metric "Processing Speed ​​D1". We collected daily data on server core count and processing speed over the past week (7 samples in total). The server core counts were [4, 4, 6, 6, 8, 8, 8], and the processing speeds were [100, 110, 150, 160, 200, 210, 220]. Therefore, the impact factor is approximately 0.986.

[0106] Following the rule of using static state indicators as rows and dynamic state indicators as columns, all calculated impact factors are filled into a matrix to form a multidimensional impact factor matrix.

[0107] In this embodiment, marking the limiting unidirectional influence path for each static and dynamic state indicator in the multidimensional influence factor matrix means identifying and marking the path of maximum influence transmission from each static state indicator to each dynamic state indicator, or conversely, without considering reverse influences, within the multidimensional influence factor matrix. This helps to clarify the strongest influence link between static and dynamic indicators in a specific direction. For example, identifying and marking the path with the greatest influence from the static indicator of "server computing power" to the dynamic indicator of "business processing speed" in the multidimensional influence factor matrix can assist in the analysis of key influence links.

[0108] In the multidimensional influence factor matrix, for each static state index Si to each dynamic state index Dj, the path is searched by starting from Si and moving along the direction with the largest matrix element value until Dj is reached, recording the path taken. This can be implemented using depth-first search (DFS) or breadth-first search (BFS) algorithms. Assuming the matrix is ​​M (m×n), starting from M[i][1], each time the current position is selected to move right or down, choosing the direction with the larger value, until M[i][n is reached.

[0109] Suppose we have a simple 2×2 multidimensional influence factor matrix [0.5 0.6 0.8 0.7]. Starting from the first row and first column (corresponding to the influence factor from static index S1 to dynamic index D1), we compare 0.5 and 0.6 and choose 0.6. Then we compare 0.6 and 0.7 and choose 0.7. The ultimate unidirectional influence path from S1 to D2 is the path through (1,1), (2,1), and (2,2).

[0110] In this embodiment, the basic volatility factor for each static and dynamic state indicator in the future scheduling cycle is analyzed based on historical data. This means predicting the degree of natural fluctuation of these indicators in the future scheduling cycle by studying their past changes and based on information such as trends and fluctuation amplitudes in historical data. Time series analysis methods can be used, such as calculating the standard deviation of historical data, and using the standard deviation as an estimate of the basic volatility factor. For example, by analyzing historical data of server CPU utilization (static indicator) and business transaction frequency (dynamic indicator), the possible fluctuation range of CPU utilization in the future scheduling cycle is quantified as the basic volatility factor, and the possible fluctuation degree of business transaction frequency is also quantified as the corresponding basic volatility factor.

[0111] In this embodiment, the comprehensive fluctuation factor of each static and dynamic state indicator in the future scheduling period is analyzed based on the basic fluctuation factor and the corresponding extreme one-way influence path. This means combining the basic fluctuation factor (the natural fluctuation degree of the indicator itself) and the extreme one-way influence path (the maximum transmission link of influence between indicators), comprehensively considering the interaction between static and dynamic indicators and their own fluctuations, to obtain a quantitative factor that can more comprehensively reflect the fluctuation of the indicator in the future scheduling period. For example, for server memory usage (a static indicator), considering its own basic fluctuation factor and the impact of changes in business data volume (a dynamic indicator) through the extreme one-way influence path, the comprehensive fluctuation factor of memory usage is comprehensively obtained, which more accurately reflects its future fluctuation. For example, the sum of the basic fluctuation factor of the starting indicator of the extreme one-way influence path of each static or dynamic state indicator and the cumulative value of all influence factors on the corresponding extreme one-way influence path, and the basic fluctuation factor of the static or dynamic state indicator, is taken as the comprehensive fluctuation factor of the corresponding static or dynamic state indicator in the future scheduling period.

[0112] In this embodiment, the multidimensional state fluctuation risk of the system to be recovered from disaster is determined based on the comprehensive fluctuation factors of all static and dynamic state indicators within the future scheduling period. This involves integrating and analyzing the comprehensive fluctuation factors of all static and dynamic state indicators to assess the degree of state instability risk that the system may face within the future scheduling period from multiple dimensions (such as performance and business dimensions). For example, by summarizing and analyzing the comprehensive fluctuation factors of server-related static indicators and business operation-related dynamic indicators, the fluctuation risk of the system to be recovered from disaster is determined in multiple dimensions such as performance and business continuity, providing a basis for taking countermeasures.

[0113] The pre-defined calculation method for state fluctuation risk in multi-dimensional state fluctuation risk (e.g., system load) involves weighted summation of multiple or all of the various static and dynamic state indicators to obtain the state fluctuation risk for that dimension. All state fluctuation risks are then aggregated to obtain the multi-dimensional state fluctuation risk. Different dimensions (e.g., system load, data security, business continuity) have their own specific pre-defined calculation methods. These methods are based on a deep understanding of the business scenarios, technical architecture, and potential risk factors involved in that dimension. For example, in the system load dimension, the focus might be on static state indicators such as server CPU utilization and memory usage, as well as dynamic state indicators such as task processing rate and resource request frequency. Multiple or all static and dynamic state indicators related to that dimension are clearly defined. For instance, in the system load dimension, static state indicators might include server hardware configuration parameters (e.g., number of CPU cores, memory capacity), while dynamic state indicators might include real-time CPU utilization and memory utilization.

[0114] Then, according to a preset calculation method, a corresponding weight is determined for each indicator. The weight is usually determined based on the relative importance of the indicator to the risk of state fluctuations in that dimension. This may be determined through expert experience, historical data analysis, machine learning algorithms, etc. For example, analysis of historical data reveals that changes in CPU utilization have a significant impact on system load fluctuations, so it is given a high weight, such as 0.6; while memory capacity, although a static indicator, also has a certain impact on system load, so it is given a weight of 0.2.

[0115] The default calculation method determines how to weight these indicators based on their correlation with system load fluctuations.

[0116] In another possible implementation, in order to characterize the real-time load status of off-site disaster recovery backup nodes in a more comprehensive and quantitative way and provide more accurate information for dynamic resource scheduling, a process is proposed in which the backup path adjustment module obtains the real-time load status of all off-site disaster recovery backup nodes at all levels in the off-site disaster recovery cloud system, which also includes:

[0117] The real-time load entropy value of all off-site disaster recovery backup nodes at all levels in the off-site disaster recovery cloud system is obtained by normalizing the real-time load value of all off-site disaster recovery backup nodes at all levels in the off-site disaster recovery cloud system, and is used as the real-time load status of the corresponding off-site multi-level disaster recovery backup nodes.

[0118] In this embodiment, the real-time load entropy value of all off-site disaster recovery backup nodes at all levels of the off-site disaster recovery cloud system is obtained by normalizing the real-time load values ​​of all off-site disaster recovery backup nodes at all levels of the off-site disaster recovery cloud system. This means that the real-time load value of each off-site disaster recovery backup node at each level of the off-site disaster recovery cloud system is first converted into a normalized value through a specific normalization method (such as mapping the load value to the 0-1 range) to eliminate the impact of differences in the dimensions or orders of magnitude of the load values ​​of different nodes. Then, using these normalized values, the real-time load entropy value of each node is calculated according to the principle of entropy calculation (entropy is used to measure the uncertainty or disorder of a system). For example, the negative value of the product of the normalized value of all off-site disaster recovery backup nodes and the base-2 logarithm of the corresponding normalized value is taken as the real-time load entropy value of the corresponding off-site multi-level disaster recovery backup node.

[0119] In another possible implementation, in order to comprehensively consider multiple factors to determine the cost of the backup path, and thus scientifically and rationally obtain the dynamic resource scheduling results of the off-site disaster recovery cloud system, a backup path adjustment module is proposed, including:

[0120] The transmission delay and exposure risk determination submodule is used to determine the transmission delay of each disaster recovery data backup task of the system to be disaster recovery in each off-site disaster recovery backup node, and to determine the exposure risk value of each disaster recovery data backup task of the system to be disaster recovery in each off-site disaster recovery backup node.

[0121] The weight allocation determination submodule is used to determine the weight allocation values ​​of transmission delay, risk exposure value, and node load at each level based on the multi-dimensional state fluctuation risk of the disaster recovery system in the future scheduling cycle.

[0122] The backup path cost determination submodule is used to calculate the backup path cost between each disaster recovery data backup task and each off-site disaster recovery backup node based on the transmission delay, exposure risk value, node load of each disaster recovery data backup task of the system to be disaster recovery in each off-site disaster recovery backup node, as well as the weight allocation value of transmission delay, risk exposure value, and node load under the corresponding off-site disaster recovery backup node's level.

[0123] The resource dynamic scheduling submodule is used to obtain the resource dynamic scheduling result of the off-site disaster recovery cloud system based on the backup path cost between each disaster recovery data backup task and each off-site disaster recovery backup node.

[0124] In this embodiment, the transmission delay of each disaster recovery data backup task of the disaster recovery system in each off-site disaster recovery backup node is determined. This means that for each disaster recovery data backup task of the disaster recovery system to be backed up to each backup node of the off-site disaster recovery cloud system, the time required to transmit the data from the disaster recovery system to the corresponding off-site disaster recovery backup node is calculated by analyzing factors such as network conditions, data volume, and node processing capabilities. For example, considering that the network bandwidth of a certain backup node is limited and the amount of data to be backed up is large, it is estimated that it may take 10 minutes to transmit a certain disaster recovery data backup task to that node. This 10 minutes is the transmission delay of the task at this node.

[0125] Transmission delay = Data volume (in bytes) ÷ (Network bandwidth (in bytes / second) + Average data processing rate of nodes (in bytes / second). Here, we consider the combined impact of network bandwidth and node processing capacity on data transmission. The larger the data volume, the lower the network bandwidth and node processing rate, and the higher the transmission delay.

[0126] In this embodiment, determining the exposure risk value of each disaster recovery data backup task in each off-site disaster recovery backup node of the disaster recovery system refers to assessing the degree of risk of data leakage, loss, or damage that each disaster recovery data backup task may face during transmission and storage at a specific off-site disaster recovery backup node, and quantifying it into a numerical value. For example, if the network security of some backup nodes is poor, the possibility of data leakage due to network attacks during transmission or storage is relatively high. After assessment, the exposure risk value of the disaster recovery data backup task at this node is set to 0.8 (assuming the risk value ranges from 0 to 1, with higher values ​​indicating greater risk).

[0127] Exposure risk values ​​can be determined using methods such as the Analytic Hierarchy Process (AHP). First, a hierarchical model is established, with risk factors such as data leakage, loss, and corruption as the criterion layer, and different risk scenarios or threats as the solution layer. Then, a judgment matrix is ​​constructed using methods such as expert scoring to calculate the weight of each factor, and finally, the exposure risk value is derived by combining these factors. For example, regarding data leakage risk, experts, based on experience and relevant data, consider the weight of network attack as w1=0.6, storage device failure as w2=0.3, and human error as w3=0.1. Next, the risk level under different risk scenarios is scored (0-1 points). Assuming the network attack risk score is r1=0.8, the storage device failure risk score is r2=0.4, and the human error risk score is r3=0.3, then the exposure risk value Rij=w1r1+w2r2+w3r3=0.6×0.8+0.3×0.4+0.1×0.3=0.48+0.12+0.03=0.63 (assuming the risk value range is 0-1, the higher the value, the greater the risk).

[0128] In this embodiment, the weighted values ​​of transmission delay, risk exposure value, and node load at each level are determined based on the multi-dimensional state fluctuation risk of the system to be recovered during future scheduling cycles. This means that, based on the potential state fluctuation risks that the system to be recovered may face in the future, the impact of these three factors on the disaster recovery effect is weighed, and a corresponding weight is assigned to each factor at different levels of the off-site disaster recovery cloud system. For example, if the data security fluctuation risk is large in the future scheduling cycle, the weight of the risk exposure value may be relatively increased; if the system performance fluctuation risk is prominent, the weight of transmission delay may be increased. For example, at the regional level, based on the multi-dimensional state fluctuation risk assessment, the weight of transmission delay is determined to be 0.4, the weight of risk exposure value is 0.4, and the weight of node load is 0.2.

[0129] Let the multidimensional state fluctuation risk vector of the system to be disaster recovery in the future scheduling cycle be R=(R1,R2,⋯,Rk), representing the fluctuation risk in different dimensions, such as R1 representing data security dimension risk, R2 representing system performance dimension risk, etc. The impact of transmission delay, risk exposure value, and node load on the disaster recovery effect can be determined by establishing a linear regression model. Let the weight of transmission delay be wT, the weight of risk exposure value be wR, the weight of node load be wL, and the disaster recovery effect index be E, then the model can be established as E=wT×T+wR×R+wL×L+ϵ (ϵ is the error term). Through historical data or simulated data, wT, wR, and wL can be solved using methods such as least squares, and wT+wR+wL=1 can be satisfied. Assuming that at the regional level, after analyzing a large amount of historical disaster recovery data and risk assessment data, it is found that when the data security dimension fluctuation risk R1 is high, the risk exposure value has a greater impact on the disaster recovery effect; when the system performance dimension fluctuation risk R2 is high, the transmission delay has a greater impact on the disaster recovery effect. Linear regression analysis yielded the following weights: transmission delay weight wT = 0.4, risk exposure weight wR = 0.4, and node load weight wL = 0.2. Therefore, at the regional level, the weights for transmission delay, risk exposure, and node load are determined to be 0.4, 0.4, and 0.2, respectively.

[0130] In this embodiment, based on the transmission delay, exposure risk value, node load, and weight allocation values ​​of the transmission delay, risk exposure value, and node load of each disaster recovery data backup task in the disaster recovery system at each off-site disaster recovery backup node, the backup path cost between each disaster recovery data backup task and each off-site disaster recovery backup node is calculated. This means multiplying the transmission delay, exposure risk value, and node load of each disaster recovery data backup task at the corresponding node by its corresponding weight value at the node's level, and then adding these three products together. The resulting value is the backup path cost between the disaster recovery data backup task and the off-site disaster recovery backup node. For example, if a disaster recovery data backup task has a transmission latency of 15 minutes (assuming it has been quantified) at a certain off-site disaster recovery backup node, an exposure risk value of 0.6, and a node load quantification value of 0.7, and the weight allocation values ​​of transmission latency, risk exposure value, and node load at the node's level are 0.3, 0.5, and 0.2 respectively, then the backup path cost = 15 × 0.3 + 0.6 × 0.5 + 0.7 × 0.2 = 4.5 + 0.3 + 0.14 = 4.94. This 4.94 represents the comprehensive cost of selecting this node for this disaster recovery data backup task, which is used for subsequent dynamic resource scheduling decisions.

[0131] In another possible implementation, in order to determine the optimal backup path based on the backup path cost and complete the dynamic scheduling and deployment of resources, thereby achieving reasonable allocation of resources in the off-site disaster recovery cloud system, a resource dynamic scheduling submodule is proposed, including:

[0132] The optimal backup path determination unit is used to determine the optimal backup path between all disaster recovery data backup tasks and all off-site disaster recovery backup nodes based on the backup path cost between each disaster recovery data backup task and each off-site disaster recovery backup node.

[0133] The resource dynamic scheduling unit is used to dynamically schedule and deploy resources for all disaster recovery data backup tasks within the future scheduling cycle based on the optimal backup path between all disaster recovery data backup tasks and all off-site disaster recovery backup nodes, and obtain the resource dynamic scheduling results of the off-site disaster recovery cloud system.

[0134] In this embodiment, determining the optimal backup path between all disaster recovery data backup tasks and all off-site disaster recovery backup nodes based on the backup path cost between each disaster recovery data backup task and each off-site disaster recovery backup node means comparing the backup path costs of all off-site disaster recovery backup nodes corresponding to each disaster recovery data backup task, and selecting the off-site disaster recovery backup node with the lowest backup path cost for each disaster recovery data backup task. The correspondence between these selected nodes and tasks constitutes the optimal backup path between all disaster recovery data backup tasks and all off-site disaster recovery backup nodes. For example, if the backup path costs of disaster recovery data backup task A at nodes 1, 2, and 3 are 3, 2, and 4 respectively, then the optimal backup path for task A corresponds to node 2. By performing this comparison and selection for all disaster recovery data backup tasks, the overall optimal backup path can be determined.

[0135] In this embodiment, based on the optimal backup path between all disaster recovery data backup tasks and all off-site disaster recovery backup nodes, resources are dynamically scheduled and deployed for all disaster recovery data backup tasks within the future scheduling period to obtain the resource dynamic scheduling result of the off-site disaster recovery cloud system. This means that, according to the previously determined optimal backup path, all disaster recovery data backup tasks within the future scheduling period are allocated to the corresponding off-site disaster recovery backup nodes according to their respective optimal backup paths. At the same time, based on factors such as node load and task priority, various resources such as network resources and storage resources are rationally allocated to complete the dynamic scheduling and deployment of resources, ultimately forming the resource dynamic scheduling result of the off-site disaster recovery cloud system, achieving reasonable allocation and efficient utilization of resources. For example, according to the optimal backup path, disaster recovery data backup tasks with large data volumes are allocated to nodes with strong storage and processing capabilities, and sufficient network bandwidth resources are allocated to them to ensure that the backup tasks are completed efficiently. These specific arrangements of resource allocation and task deployment constitute the resource dynamic scheduling result.

[0136] This invention also provides an implementation method for a dynamic resource scheduling method for off-site disaster recovery cloud systems, including:

[0137] Based on historical data of multiple system status indicators of the disaster recovery system, a multi-dimensional static state matrix and a multi-dimensional dynamic action matrix of the disaster recovery system are generated.

[0138] Based on the multidimensional static state matrix and multidimensional dynamic action matrix of the disaster recovery system, the multidimensional state fluctuation risk of the disaster recovery system in future scheduling cycles is predicted.

[0139] Based on the multidimensional state fluctuation risk of the disaster recovery system in the future scheduling cycle and the real-time load status of all off-site disaster recovery backup nodes at all levels in the off-site disaster recovery cloud system, the dynamic scheduling results of the off-site disaster recovery cloud system are obtained.

[0140] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of this invention and its equivalents, this invention also intends to include these modifications and variations.

Claims

1. A resource dynamic scheduling system for a geographically separated disaster recovery cloud system, characterized in that, Comprise: The historical data analysis module is used for generating the multi-dimensional static state matrix and the multi-dimensional dynamic action matrix of the disaster recovery system based on the historical tracking data of the multiple system state indexes of the disaster recovery system to be prepared; The fluctuation risk determination module is used for predicting the multi-dimensional state fluctuation risk of the disaster recovery system to be prepared in the future scheduling period based on the multi-dimensional static state matrix and the multi-dimensional dynamic action matrix of the disaster recovery system to be prepared; The backup path adjustment module is used for obtaining the resource dynamic scheduling result of the off-site disaster recovery cloud system based on the multi-dimensional state fluctuation risk of the disaster recovery system to be prepared in the future scheduling period and the real-time load state of all off-site disaster recovery backup nodes under all levels in the off-site disaster recovery cloud system; The historical data analysis module comprises: The historical tracking data acquisition submodule is used for acquiring the historical tracking data of the resource performance indexes, the system state indexes, the disaster recovery action indexes and the environmental configuration indexes of the disaster recovery system to be prepared as the historical tracking data of the multiple system state indexes of the disaster recovery system to be prepared; The feature dimension contribution rate determination submodule is used for analyzing the contribution degree of each feature dimension of each static state index based on the historical tracking data of each static state index in the historical tracking data of the multiple system state indexes of the disaster recovery system to be prepared; The static feature representation module is used for generating the multi-dimensional static state matrix of the disaster recovery system based on the feature value sequence of the feature dimension with the maximum contribution degree of all static state indexes in the historical tracking data; The dynamic feature representation module is used for generating the multi-dimensional dynamic action matrix of the disaster recovery system based on all dynamic state index values in the historical tracking data of the multiple system state indexes of the disaster recovery system to be prepared; The fluctuation risk determination module comprises: The influence factor determination submodule is used for mining the influence factor between each static state index and each dynamic state index based on the multi-dimensional static state matrix and the multi-dimensional dynamic action matrix of the disaster recovery system to be prepared, and building a multi-dimensional influence factor matrix based on the influence factor between all static state indexes and all dynamic state indexes; The influence path marking submodule is used for marking the limit one-way influence path of each static state index and each dynamic state index in the multi-dimensional influence factor matrix; The basic fluctuation factor determination submodule is used for analyzing the basic fluctuation factor of each static state index and each dynamic state index in the future scheduling period based on the historical tracking data of each static state index and each dynamic state index; The comprehensive fluctuation factor determination submodule is used for analyzing the comprehensive fluctuation factor of each static state index and each dynamic state index in the future scheduling period based on the basic fluctuation factor of each static state index and each dynamic state index in the future scheduling period and the corresponding limit one-way influence path; The multi-dimensional state fluctuation risk determination submodule is used for determining the multi-dimensional state fluctuation risk of the disaster recovery system to be prepared in the future scheduling period based on the comprehensive fluctuation factor of all static state indexes and all dynamic state indexes in the future scheduling period; The backup path adjustment module comprises: The transmission delay and exposure risk determination submodule is configured to determine transmission delays of each disaster recovery data backup task of the to-be-disaster-recovered system in each offsite disaster recovery backup node and determine exposure risk values of each disaster recovery data backup task of the to-be-disaster-recovered system in each offsite disaster recovery backup node. The weight allocation determination submodule is configured to determine weight allocation values of the transmission delay, the risk exposure value, and the node load at each level based on the multi-dimensional state fluctuation risk of the to-be-disaster-recovered system in a future scheduling period. The backup path cost determination submodule is configured to calculate backup path costs between each disaster recovery data backup task and each offsite disaster recovery backup node based on the transmission delay, the exposure risk value, and the node load of the corresponding offsite disaster recovery backup node of each disaster recovery data backup task of the to-be-disaster-recovered system in each offsite disaster recovery backup node and the weight allocation values of the transmission delay, the risk exposure value, and the node load at the level to which the corresponding offsite disaster recovery backup node belongs. The resource dynamic scheduling submodule is configured to obtain a resource dynamic scheduling result of the offsite disaster recovery cloud system based on the backup path costs between each disaster recovery data backup task and each offsite disaster recovery backup node. 2.The resource dynamic scheduling system for geo-disaster tolerance cloud system according to claim 1, wherein, The feature dimension contribution rate determination submodule includes: The feature standardization unit is configured to generate a feature dimension standard matrix of each static state index based on historical tracking data of each static state index in historical tracking data of the multiple system state indexes of the to-be-disaster-recovered system. The feature decomposition processing unit is configured to determine a principal component feature sequence and a corresponding principal component loading matrix of a covariance matrix of the feature dimension standard matrix of each static state index. The eigenvalue contribution analysis unit is configured to calculate a principal component eigenvalue contribution rate vector of the covariance matrix of the feature dimension standard matrix of the corresponding static state index based on the principal component eigenvalue sequence. The loading vector determination unit is configured to determine a loading vector of each feature dimension of the corresponding static state index based on the principal component loading matrix of the feature dimension standard matrix of each static state index. The contribution rate determination unit is configured to take a sum value of products of squares of all values in the loading vector of each feature dimension of each static state index and values at the same positions in the corresponding principal component eigenvalue contribution rate vector as a contribution degree of each feature dimension of each static state index. 3.The resource dynamic scheduling system for geo-disaster tolerance cloud system according to claim 2, wherein, The feature standardization unit includes: The feature dimension value determination subunit is configured to determine all feature dimension values of each static state index at each time in a historical tracking period in historical tracking data of each static state index in historical tracking data of the multiple system state indexes of the to-be-disaster-recovered system. The feature dimension value standardization subunit is configured to perform standardization processing on the same feature dimension values in all feature dimension values of each static state index at all times in the historical tracking period to obtain all feature dimension standard values of each static state index at all times in the historical tracking period. The feature dimension standard value matrixing subunit is configured to generate a feature dimension standard matrix of each static state index based on all feature dimension standard values of each static state index at all times in the historical tracking period.

4. The resource dynamic scheduling system for geo-disaster tolerance cloud system according to claim 2, wherein, The feature decomposition unit includes: a characteristic decomposition subunit configured to perform characteristic decomposition on the covariance matrix of the characteristic dimension standard matrix of each static state index to obtain all characteristic values and characteristic vectors; an eigenvalue screening subunit configured to sort all characteristic values of the covariance matrix of the characteristic dimension standard matrix of each static state index from large to small to obtain a characteristic value sequence of the covariance matrix of the characteristic dimension standard matrix of the corresponding static state index, and take a sequence formed by the first preset characteristic values in the characteristic value sequence as a principal component characteristic sequence; a load matrix generation subunit configured to generate a principal component load matrix of the characteristic dimension standard matrix of the corresponding static state index based on the characteristic vectors corresponding to all characteristic values in the principal component characteristic sequence.

5. The resource dynamic scheduling system for geo-disaster tolerance cloud system according to claim 1, wherein, The process in which the backup path adjustment module obtains the real-time load states of all offsite disaster recovery backup nodes at all levels in the offsite disaster recovery cloud system further includes: The real-time load entropy values of all offsite disaster recovery backup nodes at all levels in the offsite disaster recovery cloud system are obtained based on the normalized values of the real-time load values of all offsite disaster recovery backup nodes at all levels in the offsite disaster recovery cloud system, as the real-time load states of the corresponding offsite multi-level disaster recovery backup nodes. 6.The resource dynamic scheduling system for geo-disaster tolerance cloud system according to claim 5, wherein, The resource dynamic scheduling sub-module includes: an optimal backup path determination unit configured to determine the optimal backup paths between all disaster data backup tasks and all offsite disaster recovery backup nodes based on the backup path costs between each disaster data backup task and each offsite disaster recovery backup node; a resource dynamic scheduling unit configured to perform resource dynamic scheduling and deployment on all disaster data backup tasks in a future scheduling period based on the optimal backup paths between all disaster data backup tasks and all offsite disaster recovery backup nodes, to obtain a resource dynamic scheduling result of the offsite disaster recovery cloud system.

7. A resource dynamic scheduling method for a geographically separated disaster recovery cloud system, characterized in that, The resource dynamic scheduling system for the offsite disaster recovery cloud system according to any one of claims 1 to 6 includes: generating a multi-dimensional static state matrix and a multi-dimensional dynamic action matrix of the system to be protected based on the historical tracking data of multiple system state indexes of the system to be protected; predicting the multi-dimensional state fluctuation risk of the system to be protected in a future scheduling period based on the multi-dimensional static state matrix and the multi-dimensional dynamic action matrix of the system to be protected; obtaining a resource dynamic scheduling result of the offsite disaster recovery cloud system based on the multi-dimensional state fluctuation risk of the system to be protected in the future scheduling period and the real-time load states of all offsite disaster recovery backup nodes at all levels in the offsite disaster recovery cloud system.

Images